LLM 관련 주요 논문 - 2026-06-02

1. SafeSteer: Localized On-Policy Distillation for Efficient Safety Alignment

• Authors: Hao Li , Jingkun An , Zijun Song , Pengyu Zhu , Rui Li , Hao Wang , Wendi Feng , Yesheng Liu , Lijun Li , Jin-Ge Yao , Lei Sha
• URL: https://arxiv.org/abs/2606.02530
• Abstract:

  Aligning Large Language Models (LLMs) with human values often degrades their general capabilities, termed the alignment tax. Existing methods mitigate this by balancing dual objectives, which heavily rely on massive general-purpose data or auxiliary reward models. In this paper, we argue that, because safety features are inherently sparse within the output distribution, alignment requires localized modifications rather than global trade-offs. To this end, we propose SafeSteer, which performs on-policy distillation confined to safety tokens. First, we construct a safety teacher via activation steering. Based on this teacher, we develop a safety token selection algorithm. Consequently, SafeSteer restricts the reverse KL penalty to these tokens during training to preserve general capabilities. Experimental results across diverse models show that our SafeSteer achieves a superior trade-off between safety and general capability compared with existing methods, attaining strong safety performance on seven safety benchmarks with only minimal degradation on five general capability benchmarks. Notably, SafeSteer requires only 100 harmful samples without using any general-purpose data, less than 1% of what previous baselines used, considerably reducing alignment cost. More details are on our project page at this https URL .

---

2. Bridging the Last Mile of Time Series Forecasting with LLM Agents

• Authors: Yuhua Liao , Zetian Wang , Qiangqiang Nie , Zhenhua Zhang
• URL: https://arxiv.org/abs/2606.02497
• Abstract:

  Time series forecasting has advanced rapidly, especially with the emergence of foundation models that show strong zero-shot performance on numerical extrapolation. However, in real-world forecasting settings, a statistically plausible baseline is rarely the final forecast used in practice. Before a forecast becomes decision-ready, it often needs to be revised using weakly structured business context such as holiday effects, campaign plans, external events, historical analogs, and expert feedback. This practical stage remains underexplored in the forecasting literature. In this paper, we formulate this stage as the \textbf{last-mile forecasting} problem and present an LLM-agent framework that sits on top of a forecasting backbone. Our system maintains a unified forecast workspace, invokes tools to retrieve contextual evidence, and converts reasoning trajectories into explicit forecast revision actions under structural safety constraints. It also supports long-horizon forecasting through map-reduce-style decomposition and post-hoc reflection through a memory bank. The resulting system is designed to be controllable and auditable. Through real-world case studies, we show how LLM agents can bridge the gap between statistical prediction and business-ready forecasting.

---

3. RASER: Recoverability-Aware Selective Escalation Router for Multi-Hop Question Answering

• Authors: Yuyang Li , Zihe Yan , Tobias Käfer
• URL: https://arxiv.org/abs/2606.02488
• Abstract:

  Multi-hop question-answering systems often use expensive retrieval on every question. They may decompose the question, run several retrieval rounds, or search through bridge entities before answering. All of these strategies rely on repeated LLM calls to rewrite or decompose the question, which increases extra token cost, and it is not fitting when the LLM budget is tight. However, our analysis shows that lots of multi-hop questions are already answered correctly by a single one-shot RAG, so running an extra retrieval on every question wastes the budget. We introduce RASER (Recoverability-Aware Selective Escalation Router), a family of cheap routers built on one-shot RAG and six features from it. RASER-2 decides whether to stop or escalate to the extra-retrieval action PRUNE. RASER-3 chooses among one-shot RAG, PRUNE, and iterative retrieval IRCoT, using the same features but adding an explicit cost-accuracy trade-off. Neither router makes an extra LLM call to decide. Across six LLMs and three multi-hop QA benchmarks, both routers stay competitive with the other state-of-the-art (SOTA) baselines in F1 while spending only 41-49% of always-prune’s tokens and also less than the iterative and decomposition retrieval baselines.

---

4. Iteris: Agentic Research Loops for Computational Mathematics

• Authors: Leheng Chen , Zihao Liu , Wanyi He , Bin Dong
• URL: https://arxiv.org/abs/2606.02484
• Abstract:

  Recent advances in large language models and agentic AI systems have enabled significant progress in mathematical discovery, from solving competition problems to tackling research-level conjectures. However, open problems in computational mathematics have received comparatively less attention: research in this area often requires not only proofs but also numerical experimentation, adversarial constructions, and algorithm design. In this paper, we introduce an agentic research system, Iteris, designed for open problems in computational mathematics. We apply Iteris to two open problems from a recent Simons Workshop collection ( arXiv:2602.05394 ). In these case studies, Iteris generated numerical evidence, constructions, and proof drafts that led, after expert review and correction, to verified results. The first result is a phase diagram for the asymptotic comparison between conjugate gradient and randomized coordinate descent on power-law spectra; the second is a counterexample showing that QR factorization with column pivoting can fail to select well-conditioned submatrices even under low coherence. These case studies suggest that agentic AI systems can participate meaningfully in research workflows for open problems in computational mathematics, while human validation remains essential.

---

5. MCP-Persona: Benchmarking LLM Agents on Real-World Personal Applications via Environment Simulation

• Authors: Wenhao Wang , Peizhi Niu , Gongyi Zou , Xiyuan Yang , Jingxing Wang , Haoting Shi , Yaxin Du , Jingyi Chai , Xianghe Pang , Shuo Tang , Yanfeng Wang , Siheng Chen
• URL: https://arxiv.org/abs/2606.02470
• Abstract:

  The Model Context Protocol (MCP) has emerged as a transformative standard for connecting large language models (LLMs) with external data sources and tools, and has been rapidly adopted across personal applications and development platforms. However, existing benchmarks predominantly focus on generic information-seeking tools and fail to capture the practical challenges posed by personal social applications, where tools interact with individual accounts or local databases. To bridge this critical gap, we introduce MCP-Persona, the first benchmark specifically designed for evaluating agent performance on real-world, personalized MCP tools. MCP-Persona encompasses a diverse set of widely-used applications, ranging from social media platforms like Reddit and Xiaohongshu (Rednote) to enterprise collaboration suites such as Lark (Feishu) and Slack. Our extensive experiments on various state-of-the-art (SOTA) agents demonstrate their significant struggles with personalized tool use, thereby highlighting the benchmark’s crucial role in identifying and addressing these limitations. MCP-Persona is publicly available at this https URL }{ this https URL .

---

6. Food Noise & False Safety: A Systematic Evaluation of How LLMs Fail to Adapt to Eating Disorder Queries with Clinician Feedback

• Authors: Giulia Pucci , Emily Hemendinger , Ruizhe Li , Gavin Abercrombie , Tanvi Dinkar , Arabella Sinclair
• URL: https://arxiv.org/abs/2606.02444
• Abstract:

  Recent evidence shows that people with eating disorders (EDs) are increasingly seeking guidance, advice, and emotional support from Large Language Model (LLM)-based chat systems. Although these systems are not designed to provide clinical advice, their perceived expertise, neutrality and accessibility make them a frequent, albeit risky, source of support. This paper investigates potential patterns of interaction between users with EDs and LLMs, focusing on the potential harms arising from models that uncritically adapt to, and facilitate unsafe or self-harming user requests. We find, in consultation with clinical ED experts, that specific linguistic cues in prompts increase the likelihood of unsafe responses and, through systematically varying the degree of potential risk present in the user prompt, report the extent to which LLMs uncritically adapt to problematic, and potentially dangerous user inputs.

---

7. LLM-Evolved Pattern Generators for Optimal Classical Planning

• Authors: Windy Phung , Dominik Drexler , Arnaud Lequen , Jendrik Seipp
• URL: https://arxiv.org/abs/2606.02438
• Abstract:

  Learned heuristics have recently become a competitive alternative to traditional domain-independent heuristics for satisficing planning. Existing approaches, however, focus on improving search guidance rather than guaranteeing admissibility, which makes them unsuitable for optimal classical planning. We present the first method for learning domain-dependent heuristics that are admissible by design and thus preserve the optimality guarantees of A* search. Instead of learning a direct mapping from states to heuristic values, we learn to construct abstractions that induce admissible heuristics. We use an LLM-driven evolutionary program-synthesis framework to obtain, for each domain, a program that produces a pattern collection for any task in that domain, and we combine the resulting patterns admissibly via saturated cost partitioning. Empirically, the learned programs encode interpretable domain-specific insights, run with negligible overhead at test time and yield heuristics that match the coverage of state-of-the-art domain-independent baselines on several domains while evaluating each state substantially faster.

---

8. Bridging the Sim-to-Real Gap in Semiconductor Visual Program Synthesis via Input Binarization

• Authors: Yusuke Ohtsubo , Kota Dohi , Koichiro Yawata , Koki Takeshita , Tatsuya Sasaki
• URL: https://arxiv.org/abs/2606.02434
• Abstract:

  Precise parametric control over circuit geometry is essential for semiconductor inspection, yet obtaining sufficient real training data remains costly. Although generative models such as diffusion models and Generative Adversarial Networks (GANs) can augment training data, they cannot guarantee the nanometer-scale geometric accuracy required for metrology tasks. We propose a visual program synthesis framework in which a Vision-Language Model (VLM) converts inspection images into editable Domain-Specific Language (DSL) code describing circuit geometries, enabling controlled generation of training data with exact parameter manipulation. Because the VLM is trained solely on synthetic DSL-rendered data, a domain gap arises when processing real Scanning Electron Microscope (SEM) images. We bridge this gap with an input binarization strategy that strips SEM-specific texture and noise, letting the model focus on geometric structure. On the MIIC dataset, binarized inputs improve the mean Dice coefficient from 0.4393 to 0.5256 over the raw-input baseline, demonstrating that simple texture abstraction substantially mitigates the sim-to-real gap.

---

9. AgentPLM: Agentic Protein Language Models with Reasoning-Augmented Decoding for Protein Sequence Design

• Authors: Sahil Rahman , Maxx Richard Rahman
• URL: https://arxiv.org/abs/2606.02386
• Abstract:

  Protein language models (PLMs) are passive oracles: they generate sequences in a single forward pass with no mechanism to consult external biophysical feedback or redirect generation when a candidate violates thermodynamic or structural constraints. We introduce AgentPLM, which addresses this by equipping a pre-trained PLM with i) Reasoning-Augmented Decoding (RAD), which interleaves autoregressive generation with tool calls (ESMFold, FoldX, AutoDock Vina), and ii) Contrastive Agent Policy Optimisation (CAPO), a trajectory-level extension of direct preference optimisation that trains the policy end-to-end to learn when oracle feedback is informative rather than merely imitating high-fitness sequences. We evaluate AgentPLM on benchmark tasks spanning de novo enzyme design, antibody optimisation, thermostability, PPI interface design, and zero-shot fitness prediction with standardised oracle APIs and controlled sequence-identity splits. AgentPLM achieves state-of-the-art results with a gain in antibody top-10% hit rate over the strongest passive baseline, providing mechanistic evidence of online error correction without explicit backtracking.

---

10. COMAP: Co-Evolving World Models and Agent Policies for LLM Agents

• Authors: Youwei Liu , Jian Wang , Hanlin Wang , Wenjie Li
• URL: https://arxiv.org/abs/2606.02372
• Abstract:

  Equipping language agents with world models enables them to anticipate environment dynamics and evaluate candidate actions before execution. However, existing textual world models are typically fixed after training, preventing them from adapting to the on-policy state-action distributions induced by an evolving agent. Meanwhile, agent-improvement methods often rely on external rewards or verifiers, limiting their applicability in realistic interactive environments. In this paper, we propose COMAP, a novel framework that co-evolves textual world models and agent policies through closed-loop interaction. At each decision step, the world model predicts future state feedback for candidate actions, and the agent performs future-aware reflection by estimating the reliability of this feedback and refining its action accordingly. The resulting on-policy trajectories are then used to update the world model via self-distillation, allowing it to better match the agent’s evolving interaction distribution. Across embodied task planning, Web navigation, and tool-use benchmarks, COMAP consistently outperforms competitive baselines, e.g., +16.75% relative improvement with Qwen3-4B. Further analyses show that the co-evolutionary loop improves the world model’s prediction accuracy over time and leads to more effective long-horizon decision-making. Our code is available at: this https URL .

---

11. MOC: Multi-Order Communication in LLM-based Multi-Agent Systems

• Authors: Yao Guan , Lin Wang , Zhihu Lu , Ziyi Wang , Wenzhu Yan , Qiang Duan
• URL: https://arxiv.org/abs/2606.02359
• Abstract:

  Despite the remarkable progress of Large Language Model (LLM) based Multi-Agent Systems, most research focuses on optimizing coordination topology while largely underexploring the equally critical problem: how to transmit and optimize messages among agents effectively? Current communication schemes typically rely on the direct concatenation of first-order neighbor responses, which induces a restricted evidence receptive field and leads to the dilution of crucial insights over multi-hop paths. To address these limitations, we propose the Multi-Order Communication (MOC) scheme, which reconstructs the inter-agent communication to capture multi-hop dependencies and incorporates a structural message consolidation strategy to ensure efficiency. Specifically, we formalize the communication mechanism to construct a structured multi-order evidence stream, and subsequently design a Semantic-Topological Merging algorithm to optimize semantic fidelity within token constraints. Extensive experiments across six diverse datasets and LLM backbones of varying parameter scales demonstrate that MOC consistently improves task performance and reduces communication costs. The code is available at this https URL .

---

12. SIRI: Self-Internalizing Reinforcement Learning with Intrinsic Skills for LLM Agent Training

• Authors: Zhongyu He , Yuanfan Li , Fei Huang , Tianyu Chen , Siyuan Chen , Xingyang Li , Meng Hsuan Yu , Xiangrong Liu , Leyi Wei , Lu Pan , Ke Zeng , Xunliang Cai
• URL: https://arxiv.org/abs/2606.02355
• Abstract:

  Long-horizon LLM agents can benefit from reusable skills, yet existing skill-based methods often rely on external skill generators during training or persistent skill retrieval at inference, increasing engineering complexity, context length, and deployment latency. We propose Self-Internalizing Reinforcement learning with Intrinsic skills (SIRI), a three-phase framework that enables agents to discover, validate, and internalize skills without external skill generators or inference-time skill banks. SIRI first warms up the policy with GiGPO to acquire basic interaction ability and collect successful skill-free trajectories. It then performs self-skill mining, where the current policy summarizes compact skills from its own successful plain rollouts and validates them through paired skill-augmented and skill-free rollouts. Finally, SIRI distills only beneficial skill-guided action tokens into the plain policy using trajectory-level utility and action-level advantage. At inference, the agent runs with the original prompt only. On ALFWorld and WebShop with Qwen2.5-7B-Instruct, SIRI improves GiGPO from 0.908 to 0.930 on ALFWorld and from 0.728 to 0.813 on WebShop, outperforming prompt-based, RL-based, and memory-augmented baselines. Further analysis shows that our self-mining strategy can achieve performance comparable to distillation with closed-source large model. Our code is available at this https URL .

---

13. POIROT: Interrogating Agents for Failure Detection in Multi-Agent Systems

• Authors: Iñaki Dellibarda Varela , R. Sendra-Arranz , Pablo Romero-Sorozabal , J.M. Valverde-García , Annemarie F. Laudanski , Álvaro Gutiérrez , Eduardo Rocon , Manuel Cebrian
• URL: https://arxiv.org/abs/2606.02282
• Abstract:

  Orchestrating Large Language Models into Multi-Agent Systems (LLM-MAS) has unlocked remarkable reasoning capabilities, yet emergent failures and hallucinations that resist characterisation block their deployment in safety-critical domains – a gap made legally untenable by emerging AI regulation. Existing evaluation paradigms share a common flaw: centralised judgment creates single points of failure and demands domain-specific expertise. Here we present POIROT, a protocol that repurposes a system’s own agents as its diagnostic layer, leveraging the epistemic diversity already present in the architecture. Across evaluated settings, POIROT outperforms single-LLM evaluator baselines, with gains that scale with problem complexity (OR = 1.60, $p = 0.008$), agent count, and fault dimensionality, persisting under compound fault conditions. These results demonstrate that safety oversight need not be externalised: the agents executing a role carry sufficient collective intelligence to audit it. We release POIROT as an open-source library alongside BLAME, a benchmark for fault attribution in safety-critical multi-agent systems.

---

14. BADGER: Bridging Agentic and Deterministic Evaluation for Generative Enterprise Reasoning

• Authors: Shannon Serrao , Soumitra Chatterjee , Dorina Strori , Abhishek Sharma , Nathan Miller
• URL: https://arxiv.org/abs/2606.02109
• Abstract:

  Enterprise AI systems that translate natural language into SQL queries and orchestrate multi-step agentic reasoning pipelines require evaluation approaches fundamentally different from academic benchmarks. Spider and BIRD established execution-accuracy protocols; G-Eval and RAGAS advanced LLM-based assessment; and recent work such as Spider 2.0, BEAVER, and BIRD-Interact has begun to address enterprise and agentic dimensions. No single framework unifies text-to-SQL assessment with agentic behavior evaluation into a production-grade pipeline calibrated against human expert judgment. We present BADGER, developed at Merkle, a unified evaluation framework integrating text-to-SQL assessment with agentic behavior evaluation. BADGER offers three contributions. First, LLM-assisted SQL component extraction extending Spider methodology to handle CTE-heavy, dialect-specific SQL. Second, a hybrid execution accuracy metric (Hybrid-EX) resolving column-aliasing and numeric-tolerance brittleness by using an LLM to infer structural alignments before deterministic cell-level scoring. Validated on 150 human-annotated industry queries, Hybrid-EX achieves Cohen’s kappa=0.717 [95% CI: 0.600-0.822] (Substantial agreement) and 87.3% balanced accuracy, outperforming all six competing frameworks (Delta-kappa: 0.322-0.502, all p<=0.001). Third, an enterprise agentic evaluation suite assembling RAGAS, G-Eval, and agent benchmark metrics into a unified pipeline; Excess Tool Usage is the sole novel element. BADGER runs entirely within the client’s governed data environment, supports configurable LLM judge backends, and enables rapid prototyping of client-specific judges and metrics, serving as a continuous evaluation backbone rather than a one-time quality gate.

---

15. Where Do Deep-Research Agents Go Wrong? Span-Level Error Localization in Agent Trajectories

• Authors: Jiaming Wang , Ziteng Feng , Jiangtao Wu , Ruihao Li , Qianqian Xie , Yuxiang Ren , He Zhu , Xueming Han , Fanyu Meng , Junlan Feng , Jiaheng Liu
• URL: https://arxiv.org/abs/2606.02060
• Abstract:

  Deep-research agents solve tasks through long trajectories of search, tool use, evidence inspection, and answer synthesis. Evaluation based on final answers shows whether an agent succeeds, but not which parts of the trajectory make the answer unreliable. We study span-level error localization for deep-research agents. We collect 2,790 real trajectories from two agent frameworks, three backbone models, and three benchmarks, convert raw logs into semantic spans, and annotate harmful error spans through LLM-assisted expert review. From these annotations, we build TELBench, a 1,000-instance benchmark for identifying error spans among normal exploration, failed searches, tentative hypotheses, and harmless noise. We further propose DRIFT, a claim-centric auditing framework that tracks agent claims, checks their support in trajectory evidence, and marks spans where unsupported or conflicting claims affect the answer path. Experiments across model families and auditing frameworks show that DRIFT improves span-level error localization and first-error accuracy by up to 30 percentage points. Our work provides a process-level view of reliability in deep-research agents.

---

16. eMoT: evolving Memory-of-Thought via Symbolic Anchoring and Memory Corrosion

• Authors: Xiang Li , Jiwei Wei , Ke Liu , Yitong Qin , Jinyu Guo , Malu Zhang , Peng Wang , Yang Yang
• URL: https://arxiv.org/abs/2606.02054
• Abstract:

  While Large Language Models (LLMs) achieve impressive performance on multi-step reasoning tasks, their reliability is persistently hindered by critical limitations such as unconstrained hallucinations and poor numerical computation. Fundamentally, these issues arise because standard models treat reasoning as a transient, one-off generation process rather than retaining and refining successful procedural logic. To address these challenges, we propose eMoT (evolving Memory-of-Thought), a unified framework that stabilizes multi-step reasoning by treating reasoning trajectories as dynamic, evolving memories rather than static templates. The framework primarily consists of three interconnected modules: (i) a memory corrosion mechanism that reinforces high-utility reasoning structures while gradually decaying less frequent ones; (ii) a symbolic anchoring engine that utilizes Python for deterministic computation, much like a human uses a calculator; and (iii) a consistency-driven refinement process that aligns neural inference with symbolic outcomes, reducing the accumulation of logical discrepancies. Across multiple reasoning benchmarks, eMoT improves accuracy and solution consistency over standard Chain-of-Thought and structured reasoning this http URL the traditional task Game of 24, eMoT achieves 100% accuracy, surpassing the baseline by up to 17.6%. Evaluations on mathematical task GSM8K, ASDiv, SVAMP, and MGSM further show consistent gains in multi-step mathematical reasoning. In our evaluation, we achieve superior performance despite utilizing a lightweight backbone model with constrained baseline capabilities. Compared to alternative methods that rely on massively scaled models, our results demonstrate that the performance gains are fundamentally driven by the eMoT framework’s reasoning control rather than sheer model size.

---

17. SafeMCP: Proactive Power Regulation for LLM Agent Defense via Environment-Grounded Look-Ahead Reasoning

• Authors: Lichao Wang , Zhaoxing Ren , Tianzhuo Yang , Jiaming Ji , Chi Harold Liu , Yaodong Yang , Juntao Dai
• URL: https://arxiv.org/abs/2606.01991
• Abstract:

  As Large Language Model (LLM) agents increasingly leverage the Model Context Protocol (MCP) to operate in complex environments, the expansion of their action spaces offers agents unsafe capabilities and underscores the risk of power-seeking. While broad action space and greater environment influence are essential for task fulfillment, they create a fragile risk surface where minor errors or hallucinations are magnified into catastrophic failures. In response, we propose SafeMCP, a {server-side} defense plugin that constrains tool acquisition via predictive reasoning regarding future safety risks. SafeMCP utilizes an internal world model for look-ahead reasoning to implement a two-tier defense: proactive tool filtering to constrain hazardous power expansion and immediate intervention as a fail-safe. To train SafeMCP, we introduce a three-stage pipeline comprising environmental dynamic grounding, safe policy initialization, and reinforcement learning (RL) with dual verifiable rewards. Experiments on PowerSeeking Bench, ToolEmu, and AgentHarm show that SafeMCP achieves a safe equilibrium, effectively mitigating risks while preserving agent utility.

---

18. An NLP-Driven Framework for Curriculum-Labor Market Alignment: Schema-Constrained LLM Extraction, ESCO-Anchored Semantic Matching, and Multi-Dimensional Gap Quantification

• Authors: Sherzod Turaev , Mary John , Mamoun Awad , Nazar Zaki , Khaled Shuaib
• URL: https://arxiv.org/abs/2606.01982
• Abstract:

  Schema-constrained information extraction from diverse educational and labor-market corpora remains an open challenge in natural language processing because existing pipelines rely primarily on lexical-surface methods that cannot recover implicit competencies, lack grounding in shared taxonomies, and provide no formal measures of extraction reliability or document-level completeness. To address these limitations, this paper proposes a four-stage NLP framework that combines (i) schema-constrained prompting of a two-model frontier-LLM ensemble against a JSON Schema-enforced seven-slot competency formalism, (ii) Sentence-BERT (SBERT) alignment of the extracted records against an eleven-domain ESCO v1.2.1 controlled vocabulary, (iii) a two-tier adjudication protocol that resolves inter-model disagreements, and (iv) a verification mechanism that combines per-slot Cohen’s kappa, schema conformance, and document-level completeness audits. The framework is instantiated for a critical application in higher-education quality assurance, namely curriculum-labor market alignment for the ABET-accredited BSc Computer Science program at the United Arab Emirates University. The pipeline extracts 400 competency records from the 85-course 2025-2026 study plan and aligns them, under a five-scope analysis ranging from the computing core to a probability-weighted student trajectory, with 30 job postings (483 requirement clauses) at an SBERT cosine threshold of 0.50. The extractor achieves Cohen’s kappa of 0.79 on the skill slot, with 100% schema conformance and 100% document-level completeness. The alignment surfaces interpretable supply-demand gaps of 25.0% in general and transversal skills, 13.8% in algorithms and computational theory, and 12.2% in software engineering and project management, with a near-zero 1.8% gap in artificial intelligence and data science despite 38.6% supply coverage.

---

19. Algorithmic algorithm development with LLMs: A Case Study on LLM-Usage for Contraction Order Optimization in Tensor Networks

• Authors: Fabian Hoppe , Melven Röhrig-Zöllner , Philipp Knechtges
• URL: https://arxiv.org/abs/2606.01975
• Abstract:

  We consider LLM-based algorithm development through a case study on contractionorder optimisation for tensor networks with OpenEvolve. We pay particular attention to the choice of the LLM as well as design choices such as evaluation metric and test instances. Our results highlight both the promise of verifier-guided evolutionary coding agents for algorithm development/improvement and the continuing importance of evaluation, validation, and interpretation – and corresponding challenges – by the human scientist.

---

20. SMH-Bench: Benchmarking LLM Agents for Environment-Grounded Reasoning and Action in Smart Homes

• Authors: Kuan Li , Shuo Zhang , Huacan Wang , Fangzhou Yu , Zecheng Sheng , Yi Gu , Weipeng Ming , Lei Xue , Chen Liu , Sen Hu , Ronghao Chen , Siyue Lin , Yuqing Hou , Xiaofeng Mou , Yi Xu
• URL: https://arxiv.org/abs/2606.01912
• Abstract:

  Smart homes are evolving toward complex state-dependent living environments, requiring Large Language Models (LLMs) to reason over user intent, preferences, and multi-device interactions. However, existing smart-home benchmarks often focus on static instruction-to-API mapping or limited simulations, failing to evaluate whether LLMs can reason, interact, and act reliably in realistic household scenarios. To address these limitations, we introduce SMH-Bench, a comprehensive benchmark for evaluating LLMs in smart-home environments. Built upon HomeEnv, an executable and verifiable smart-home simulator, SMH-Bench contains 1,100 high-quality tasks spanning 7 categories and 22 fine-grained subcategories. It further stratifies tasks across simple, medium and complex homes, ranging from small apartments to dense multi-room environments with 135 devices. Experiments show that although frontier LLMs achieve strong performance on explicit control and query tasks, they still exhibit significant weaknesses in automation task scheduling, ambiguity handling and personalized reasoning, especially as home complexity increases. We hope SMH-Bench will facilitate the development of more reliable, context-aware, and practically deployable smart-home agents.

---

21. Bayesian Spectral Emotion Transition Discovery from Multi-Annotator Disagreement

• Authors: Keito Inoshita , Takato Ueno
• URL: https://arxiv.org/abs/2606.01906
• Abstract:

  Emotions evolve through the dynamics of conversation, and understanding their transition structure is foundational to applications ranging from mental-health screening to dialogue systems. However, existing studies typically compress multi-rater judgments into a single hard label by majority voting, discarding the uncertainty signal needed to understand turn-to-turn transitions. In this article, we propose Bayesian Spectral Emotion Transition Discovery (BSETD), a two-stage framework that discovers emotion-transition structure from multi-rater soft labels. In the first stage, a hierarchical Dirichlet-Multinomial posterior is constructed through the outer product of soft labels, equipping each cell of the K x K transition matrix with a credible interval and Benjamini-Hochberg (BH) false discovery rate (FDR)-controlled significance. In the second stage, the symmetrized graph Laplacian is spectrally decomposed to separate a low-frequency (inertia) component from a high-frequency (contagion) component. On EmotionLines, BSETD simultaneously recovers the signatures of two distinct affective spaces: the Plutchik-adjacent transitions disgust to anger (log2 lift +0.94) and anger to disgust (+0.86) are over-represented, while the Russell-valence-reversed transitions joy to anger (-0.90) and anger to joy (-0.89) are under-represented. A five-source cross-corpus validation yields pairwise Pearson correlations in 0.91-0.98 within English, 0.79-0.85 against Chinese M3ED, and 0.979 between the human hard labels and the LLM virtual soft labels on the same utterance set, demonstrating that a pipeline preserving annotator uncertainty bridges the computational study of emotion dynamics with established psychological theory.

---

22. Absorbing Complexity: An Interaction-Native Knowledge Harness for Financial LLM Agents

• Authors: Ailiya Borjigin , Igor Stadnyk , Ben Bilski , Maksym Chikita , Dmytro Kyrylenko , Sofiia Pidturkina , Julia Stadnyk
• URL: https://arxiv.org/abs/2606.01886
• Abstract:

  Financial AI agents often fail for a simple reason: they make users carry the complexity. A user must repeatedly restate goals, risk preferences, portfolio context, past judgments, and shifting market assumptions, while the agent answers, retrieves, acts, and forgets. In finance, this is not just inconvenient. In tasks such as market analysis, copy-trading review, and trade preparation, forgotten context and stale memory can create latency, repeated errors, weak auditability, and unsafe decisions. We propose the interaction-native knowledge harness (InKH), an architecture for financial LLM agents that absorbs complexity into the system. InKH converts user, market, portfolio, and tool events into structured operational knowledge. It uses passive knowledge injection to assemble a bounded working context buffer before the main model step, temporal graph memory for low-latency retrieval, a wiki audit surface for human-readable governance, and background extraction with maturity, decay, and write-time invalidation. We evaluate InKH on a reproducible controlled synthetic benchmark with 24 random seeds, 4 rounds, 80 episodes per round, and 6 baselines, producing 46,080 baseline-conditioned evaluations. InKH achieves mean task quality of 0.815 at 900 ms latency. Compared with agent-driven wiki-walk memory, it reduces latency by 82.95 percent, token cost by 82.29 percent, and stale-knowledge usage by 96.58 percent, while improving quality by 0.108 and traceability by 0.461. Compared with a temporal-graph system without invalidation, it improves quality by 0.050 and reduces stale-memory usage by 96.58 percent with comparable serving cost. The results support a design thesis for financial AI: adoption happens when complexity is absorbed by the system rather than transferred to the user. The benchmark validates architecture-level behavior, not live trading performance.

---

23. WorldCoder-Bench: Benchmarking Physically Grounded 3D World Synthesis

• Authors: Shuo Lu , Yinuo Xu , Kecheng Yu , Siru Jiang , Yongcan Yu , Yubin Wang , Haitao Yang , Yuxiang Zhang , Bin Wang , Ran He , Jian Liang
• URL: https://arxiv.org/abs/2606.01869
• Abstract:

  Large language models (LLMs) are increasingly asked not only to write static interfaces, but to construct executable interactive worlds from natural language. Browser-native 3D, commonly built with this http URL , is a natural next frontier: generated programs must integrate assets, obey spatial and physical constraints, and keep user-facing controls synchronized with hidden runtime state. Existing web-generation benchmarks and evaluators, however, largely observe only pixels or DOM nodes, while the mechanics of a this http URL world unfold inside an opaque . We introduce WorldCoder-Bench, a benchmark for autonomous, physically grounded 3D world synthesis. WorldCoder-Bench contains 2,026 expert-curated tasks across Simulation, Rendering, and Application scenarios, with optional .glb assets and hidden behavioral contracts. We further propose StateProbe, an execution-based protocol that probes generated programs in a sandboxed browser and verifies hidden, mutation-hardened contracts over runtime states and transitions. Beyond verification coverage, we report Return on Automation and Time Efficiency Multiplier to measure correctness-adjusted cost and time savings. Across nine frontier models, the best system reaches only 27.8% verification coverage on WorldCoder-Core and 19.9% on WorldCoder-Robust, with failures dominated by state-schema drift and broken interaction chains rather than missing scene elements. Utility metrics further show that cheap or fast models can still provide substantial value on easier domains. WorldCoder-Bench is available at this https URL .

---

24. Does Compression Preserve Uncertainty? A Unified Benchmark for Quantized and Sparse LLMs via Conformal Prediction

• Authors: Yujia Tong , Yuxi Wang , Yunyang Wan , Tian Zhang , Junhao Dong , Jingling Yuan
• URL: https://arxiv.org/abs/2606.01850
• Abstract:

  Model compression techniques such as quantization and pruning are widely used to reduce the deployment cost of large language models (LLMs), with existing evaluations focusing almost exclusively on accuracy preservation. However, in safety-critical applications, a model’s ability to reliably quantify its own uncertainty is equally important. We ask: does compression preserve this ability? To answer this question, we benchmark 12 LLMs under various compression configurations across five NLP tasks, using conformal prediction to provide a rigorous, distribution-free measure of uncertainty. Our experiments reveal that: (I) compression frequently decouples accuracy from uncertainty; (II) larger models absorb compression-induced uncertainty far more effectively than smaller ones; and (III) uncertainty inflation is often threshold-like rather than gradual. These results suggest that accuracy-only evaluation is insufficient for assessing the deployment readiness of compressed LLMs, and that uncertainty-aware benchmarking should be a standard component of model compression pipelines.

---

25. CAPF: Guiding Search-Agent Rollouts with Credit-Attenuated Privileged Feedback

• Authors: Bin Chen , Xinye Liao , Yiming Liu , Xin Liao , Chonghan Liu
• URL: https://arxiv.org/abs/2606.01830
• Abstract:

  Recent LLM search agents use reinforcement learning with verifiable rewards (RLVR) to learn search-augmented reasoning from outcome rewards. On hard problems, these agents rarely sample end-to-end successful rollouts, leaving outcome-only RLVR with few positive-reward trajectories. We argue that improving learning on such problems requires additional guidance during training, and RLVR already contains verifier-side information that can provide it. This information can identify errors or omissions in the agent’s submitted answer and guide revision within the rollout. We propose a training-time mechanism called \textbf{Credit-Attenuated Privileged Feedback} (CAPF), which makes this verifier-side information available through a Privileged Feedback call during training. CAPF lets the policy revise zero-reward attempts into positive-reward repair trajectories and attenuates credit for the feedback call and earlier actions to accommodate deployment without this call. Empirical research demonstrates that CAPF improves Qwen3-4B’s average exact-match score from 44.7% under outcome-only RLVR to 48.5% on seven open-domain QA benchmarks.

---

26. Consistency evaluation of benchmarks used for causal discovery

• Authors: Yuzhe Zhang , Chihui Chen , Lina Yao , Chen Wang
• URL: https://arxiv.org/abs/2606.01789
• Abstract:

  In graphical causal model, causal discovery aims to construct a causal graph based on numerical data and domain knowledge in plain text. However, the evaluation of causal discovery methods remains a challenge in the area as the progress of domain researches often makes benchmark causal graphs contain mis-aligned knowledge. This problem especially affects the evaluation of large language model (LLM) based causal discovery methods as they are sensitive to the new discoveries in the literature. This work is the first to systematically study the quality of benchmark causal graphs. Specifically, we design a pipeline that automatically retrieves relevant research papers from scientific databases, and prompts LLMs to check the consistency between the benchmark causal graphs and domain research papers. We evaluate 11 popular real-world benchmarks, for which our pipeline in total proceeds 38,081 domain papers. Our results show that popular benchmarks vary significantly in their consistency with domain research, with clear implications for causal discovery research.

---

27. TriAlign: Towards Universal Truth Consistency in Personalized LLM Alignment

• Authors: Thi-Nhung Nguyen , Linhao Luo , Rollin Omari , Junae Kim , Thuy-Trang Vu , Dinh Phung
• URL: https://arxiv.org/abs/2606.01755
• Abstract:

  Personalized large language models adapt responses to users’ preferences and social attributes, but can introduce substantial universal truth inconsistencies across social groups, where some groups systematically receive less accurate responses on objective tasks. Existing alignment methods either ignore personalization or mainly focus on subjective preference alignment, largely overlooking fairness and consistency in universal truths. To address this gap, we study Truth-Invariant Alignment (TIA), an alignment problem for personalized LLMs that aims to ensure universal truths remain consistent across social groups while preserving personalization. We propose TriAlign, the first offline multi-agent reinforcement learning (MARL) framework for TIA, where each social group is modeled as an agent interacting. TriAlign jointly optimizes universal truth accuracy, cross-group truth consistency, and personalization through a fairness-aware objective and an explicit inconsistency penalty. Experiments across diverse benchmarks demonstrate that TriAlign achieves a stronger balance among these three objectives than strong baselines, reducing universal truth disparities across social groups while improving both objective task performance and personalization quality.

---

28. TrafficRAG: A Multimodal RAG Framework for Traffic Accident Liability Determination

• Authors: Xu Li , Zedong Fu , Xinyi Li , Xun Han
• URL: https://arxiv.org/abs/2606.01737
• Abstract:

  Traffic accident liability analysis is a critical yet challenging task in intelligent transportation and legal assistance. Existing methods often suffer from low efficiency, subjective judgment, and inconsistent analysis results. Meanwhile, large language models are constrained by noisy video inputs and insufficient legal domain knowledge. To address these issues, this work presents TrafficRAG, a multimodal retrieval-augmented framework for automated traffic accident analysis and report generation. Specifically, the proposed framework first adopts a vision-language model to produce structured textual descriptions of accident scenarios, which serve as accurate retrieval queries. Based on these textual queries, a hybrid retrieval strategy integrating BM25 sparse retrieval and dense embedding retrieval is employed to fetch relevant traffic regulations and similar historical cases. Finally, the large language model incorporates retrieved legal knowledge and multimodal accident evidence for comprehensive reasoning, and generates standardized, legally grounded liability analysis reports. Extensive experiments show that TrafficRAG consistently outperforms baseline methods, achieving 77.32% Legal Norm Adaptation Accuracy, 81.71% Factual Faithfulness, and a Liability Ratio MAE of 5.48%. The results validate that integrating multimodal factual evidence with legal clauses via retrieval augmentation can effectively improve the reliability and accuracy of traffic accident liability determination.

---

29. Evidence-Gated LLM Priors for Multi-Objective Bayesian Optimization

• Authors: Jiangyu Chen , Banyi
• URL: https://arxiv.org/abs/2606.01730
• Abstract:

  Large language models (LLMs) are increasingly used as heuristic advisors for black-box optimization, yet their suggestions and self-reported confidence are not necessarily calibrated to downstream objective values. This issue becomes more pronounced in multi-objective Bayesian optimization, where different objectives may require different expert knowledge and where an LLM expert can be useful for one objective but misleading for another. We study how to use LLM-generated expert priors in discrete multi-objective Bayesian optimization without blindly trusting them. We propose an objective-wise reputation-market mechanism that treats each expert-objective pair as a falsifiable prior source. Expert weights are updated online from observed objective feedback, discounted over time, and gated by market-level trust. We then introduce a decoupled counterfactual gate that can use the LLM prior without confidence, use it with confidence, or abstain from the LLM prior entirely. Across controlled synthetic stress tests and three molecule optimization benchmarks with \qwenflash{}-generated expert priors, we find that dynamic objective-wise calibration improves robustness over fixed LLM priors. However, raw LLM confidence is not reliably beneficial: on ESOL, confidence is positively correlated with prediction error; on FreeSolv, confidence can help; and on Lipophilicity, ignoring confidence remains strongest. Our fixed three-arm counterfactual gate improves over the first counterfactual variant on ESOL and FreeSolv, while an attempted margin portfolio exposes a useful negative result: margin selection should be acquisition-aware rather than based only on one-step prior error.

---

30. MobEvolve: An Agentic Self-Evolving Heuristic System for Interpretable Human Mobility Generation

• Authors: Junlin He , Yihong Tang , Tong Nie , Ao Qu , Yuebing Liang , Hamzeh Alizadeh , Bang Liu , Wei Ma , Lijun Sun
• URL: https://arxiv.org/abs/2606.01640
• Abstract:

  Human mobility generation aims to synthesize realistic trip chains for target populations based on individual features. Existing paradigms, including deep generative models, LLM-based methods, and traditional heuristics, struggle to satisfy the complex demands of this task while simultaneously maintaining interpretability, behavioral plausibility, population-level distributional alignment, and inference efficiency. To bridge this gap, we introduce MobEvolve, the first agentic self-evolving heuristic framework for human mobility generation. MobEvolve initializes a behavior-inspired heuristic system and employs an LLM agent to iteratively evolve its internal logic. By diagnosing empirical misalignments and failure cases on a validation set, the agent proposes targeted updates and accumulates evolution memory for cumulative self-improvement. Extensive evaluations on the Singapore and Montreal benchmarks demonstrate that MobEvolve significantly outperforms state-of-the-art deep generative and LLM-based methods in individual trajectory fidelity, population-level distribution alignment, and behavioral plausibility, while preserving interpretability and high inference efficiency.

---

31. ReSkill: Reconciling Skill Creation with Policy Optimization in Agentic RL

• Authors: Zelin He , Haotian Lin , Boran Han , Wei Zhu , Haoyang Fang , Bernie Wang , Xuan Zhu , Runze Li , Matthew Reimherr
• URL: https://arxiv.org/abs/2606.01619
• Abstract:

  Agentic reinforcement learning (RL) enables LLM agents to improve continuously from environment rewards, yet the resulting policies do not systematically accumulate reusable strategies that generalize across tasks. Modular skills can provide such reusable strategies, yet existing skill-augmented RL methods decouple skill creation from policy optimization, risking adopting skills that conflict with the evolving policy. Inspired by Anthropic’s Skill Creator, we introduce ReSkill, an RL-in-the-loop skill creation framework that reconciles skill evolution with policy learning. ReSkill exploits the group-wise structure of GRPO to naturally embed three mechanisms with only marginal additional overhead: (1) an assertion-driven skill creator that diagnoses failures from past experience and proposes conditional, trigger-based skill revisions; (2) within-group rollout sampling that enables controlled comparison of skill versions, capturing which version best supports the policy’s ongoing learning; and (3) Thompson Sampling with adaptive discounting to balance exploration and exploitation in skill version selection as the policy evolves. Across several domains, ReSkill consistently outperforms existing memory and skill-based RL methods, with the largest gains on unseen tasks. Analysis of the skill lifecycle shows skills being automatically created, tested, refined, and pruned as the policy improves, demonstrating reconciled skill-policy co-evolution.

---

32. Revisiting Ripple Effects in Knowledge Editing through Pressure-Aware Joint Neighborhood Optimization

• Authors: Haoben Huang , Shuxin Liu , Ou Wu , Di Gao
• URL: https://arxiv.org/abs/2606.01610
• Abstract:

  Single-edit updates in large language models can trigger ripple effects across local knowledge neighborhoods: desirable propagation to related facts and unintended perturbation of preserved ones. Existing methods address these two effects separately, without explicitly modeling their coupling. We challenge this separation through an analysis of ripple responses across typical baselines, identifying two coupled design pressures: editable-side coordination and preserved-side leakage. We propose Joint Neighborhood Optimization (JNO), a new knowledge-editing framework to formalize and jointly address both pressures at the target-planning stage. JNO instantiates this principle through Pressure-Aware Coordination (PAC), which jointly optimizes neighborhood target representations under coupled constraints, and a semantic pre-execution gate that rejects high-risk target plans before parameter execution. Experiments on RippleEdits show JNO improves propagation and preservation metrics by at least 7.0% while preserving cross-backbone editing stability.

---

33. S-SPPO: Semantic-Calibrated Self-Play Preference Optimization

• Authors: Xiwen Chen , Wenhui Zhu , Jingjing Wang , Peijie Qiu , Zhipeng Wang , Huayu Li , ZhengXiao He , Xuanzhao Dong , Prayag Tiwari , Mingkun Xu , Yujian Xiong , Feng Luo , Abolfazl Razi , Brendan Hogan Rappazzo , Anderson Schneider , Yuriy Nevmyvaka
• URL: https://arxiv.org/abs/2606.01561
• Abstract:

  Aligning Large Language Models (LLMs) with human preferences is often formulated via Direct Preference Optimization (DPO). However, the standard Bradley-Terry instantiation of DPO is limited in modeling common departures from transitivity in human preferences. To address this, recent work has introduced Self-Play Preference Optimization (SPPO), which iteratively refines the policy by training on self-generated win-lose pairs. Our investigation, however, reveals a critical instability in SPPO: the optimization is prone to policy degeneration when the preference oracle assigns overly confident wins to semantically indistinguishable responses. To mitigate this, we propose S-SPPO, a dual-space semantic calibration framework comprising: i) Supervision Calibration via semantic gating, which anneals win rate targets toward the maximum-entropy baseline as semantic overlap increases; and ii) Representation Calibration via latent repulsion to enforce geometric diversity to prevent manifold collapse and maintain latent diversity between chosen and rejected samples. Theoretically, we show that the calibration preserves the constant-sum game structure, facilitating convergence to a Nash Equilibrium. Empirically, S-SPPO avoids the performance degradation seen in prior methods, achieving 52.19% win rate and 47.46% length-controlled win rate on AlpacaEval 2.0 with Llama-3-8B, without using additional human-annotated preferences during training. The code will be available at this https URL .

---

34. RoleCDE:Benchmarking and Mitigating Role-Alignment Trade-offs in Role-Playing Agents

• Authors: Huayi Lai , Shichao Song , Simin Niu , Hanyu Wang , Jiawei Yang , Zhouxing Wang , Zhiqiang Yin , Xun Liang
• URL: https://arxiv.org/abs/2606.01552
• Abstract:

  Role-playing agents(RPAs) are widely used to steer large language models(LLMs) toward role-consistent behavior, yet existing benchmarks mainly evaluate surface-level fidelity and offer limited insight into decision making under role-alignment value conflicts. To address this gap, we introduce RoleCDE, the first benchmark designed to evaluate RPAs under structured conflicts between role-specific values and alignment-oriented constraints. RoleCDE formulates role-aware decision making as cognitive dilemma scenarios, jointly evaluating role-scenario grounding, value conflict resolution, and decision tendencies. The benchmark is constructed at scale, covering approximately 8k diverse role profiles and scenarios and nearly 24k dilemma instances across three difficulty levels and eight role categories. Evaluation of several mainstream LLMs reveals a “Role Value Decoupling” phenomenon, where agents systematically default to alignment-and morality-consistent decisions rather than role-specific values when the two conflict, even under explicit role conditioning. This behavior is largely invariant to dilemma difficulty but varies substantially across role categories. We further show that RoleCDE-based fine-tuning effectively mitigates this decoupling by improving value trade-off reasoning, while preserving general role-playing fidelity and general reasoning performance. Code is available at: this https URL .

---

35. Joint Agent Memory and Exploration Learning via Novelty Signals

• Authors: Shizuo Tian , Xiaohong Weng , Rui Kong , Yuxuan Chen , Guohong Liu , Yuebing Song , Jiacheng Liu , Yuchen Li , Dawei Yin , Ting Cao , Yunxin Liu , Yuanchun Li
• URL: https://arxiv.org/abs/2606.01528
• Abstract:

  In open-ended environments, exploration is fundamental for autonomous agents, yet current language model agents struggle with this. Effective exploration requires memory, but retaining raw interaction histories is computationally expensive over long trajectories. While latent memory offers a solution to compress interaction histories, its training lacks reliable supervisory signals. We introduce \textbf{J}oint \textbf{A}gent \textbf{M}emory and \textbf{E}xploration \textbf{L}earning (\textbf{JAMEL}), a framework that trains agentic memory and exploration policy together through novelty-driven interaction. We observe that memory and exploration form a mutually dependent loop: sustained exploration requires memory to distinguish exhausted behaviors from unseen ones, while novelty-seeking interaction provides the supervision needed to make memory useful for future exploration. By utilizing deterministic and persistent novelty signals such as code coverage in the GUI domain, we provide natural, annotation-free supervision for the memory module. Empirical evaluations demonstrate that \ours successfully generalizes to unseen environments. Its exploration capability outperforms open-weight baselines and rivals the exploration depth of a closed-source model while reducing token consumption. Our code and model are open-sourced at this https URL .

---

36. Dive into Ambiguity: A*-Inspired Multi-Agents Commonsense Obfuscation Attack on LLM Prompts

• Authors: Boxuan Wang , Zhuoyun Li , Xiaowei Huang , Yi Dong
• URL: https://arxiv.org/abs/2606.01441
• Abstract:

  Large language models (LLMs) excel in reasoning and knowledge-intensive tasks but remain vulnerable to prompt-level adversarial attacks that preserve intent while triggering commonsense hallucinations. This vulnerability is urgent, as LLMs are rapidly integrated into safety-critical domains where factual reliability is non-negotiable. Existing attack methods either lack efficiency or fail to capture the adaptive strategies of real-world adversaries. We propose an A*-inspired Factual Error Induction Framework, a framework for generating semantically aligned yet obfuscated prompts. At its core is a Hierarchical Rewrite Strategy guided by a dynamic semantic dispersion coefficient $\gamma$ that balances conservative edits early with aggressive obfuscations later, following a reverse simulated annealing schedule. To enhance interpretability, we further introduce Agentic Mechanism Labeling, which discovers and refines adversarial mechanisms, offering interpretable reverse optimization. Theoretically, we prove that prompt rewriting follows a contractive recurrence, leading to semantic collapse as $\gamma$ decreases. Empirically, across diverse LLMs, our method achieves higher attack success rates than exhaustive exploration while requiring fewer attempts, demonstrating both efficiency and effectiveness.

---

37. Don’t Ask the LLM to Track Freshness: A Deterministic Recipe for Memory Conflict Resolution

• Authors: Vikas Reddy , Sumanth Challaram
• URL: https://arxiv.org/abs/2606.01435
• Abstract:

  LLM-based memory systems increasingly maintain facts that evolve over time, where a recurring failure is conflict resolution: when a fact has multiple contradictory values, which should the agent return? MemoryAgentBench (MAB; Hu et al., 2026) makes this explicit in its FactConsolidation task: facts are numbered, the counterfactual has the higher serial, and agents are told newer facts have larger serials. Yet every published system underperforms: HippoRAG-v2 reaches 54% on single-hop (FC-SH), BM25 48%, Mem0 18%, and the temporal KG Zep/Graphiti just 7%. Multi-hop is near-unsolved (at most 7% across 22 systems). We argue the bottleneck is the assembly step: baselines leave conflict resolution to LLM-mediated retrieval or generation rather than version-aware aggregation. A matched-setup comparison (same backbone, retrieval, chunking, TOP_K) shows that replacing the LLM-judgment answer pipeline with candidate-extraction plus Python max(serial) yields +10.8 points on FC-SH (gpt-4o-mini), widening from +8 at 6K to +21 at 262K. This is a whole-pipeline effect (resolver, prompt, format, and temperature vary jointly); isolating the resolver is future work. The recipe reaches 78.0% on FC-SH (gpt-4o-mini), 94.8% (gpt-4o), and 30.2% on FC-MH (gpt-4o-mini, rising to 51.5% with gpt-4o) via a per-hop deterministic extension of Self-Ask. At matched-262K, it beats HippoRAG-v2 by +28 points and the best published FC-MH result by +20. The implication is corrective for the subfield: the bottleneck on conflict resolution is assembly (post-retrieval aggregation), not storage. A LongMemEval knowledge-update check shows the mechanism ports from max(serial) to max(timestamp) but only ties LLM judgment (57.8% vs 64.4%, n=45): deterministic aggregation is the right primitive for current-value conflicts and must be composed with question-type-aware handling for broader memory QA.

---

38. Self-Healing Agentic Orchestrators for Reliable Tool-Augmented Large Language Model Systems

• Authors: Rahul Suresh Babu , Adarsh Agrawal
• URL: https://arxiv.org/abs/2606.01416
• Abstract:

  Tool-augmented large language model (LLM) agents rely on orchestration layers that coordinate planning, retrieval, tool invocation, validation, memory, and recovery. In these systems, failures arise not only from model errors, but also from orchestration-level issues such as tool timeouts, malformed arguments, stale context, contradictory evidence, retry loops, and unverified intermediate outputs. This paper presents a self-healing agentic orchestrator that treats reliability as a bounded runtime control problem. The orchestrator maps observable failure signals to inferred failure classes, selects targeted recovery actions under explicit budgets, verifies recovered trajectories, and records observability traces. We evaluate the approach on a 100-task controlled fault-injection benchmark against static workflow, retry-only, ReAct-style, and full-replanning baselines. Self-healing achieves 98.8\% task success, compared with 94.5\% for retry-only and 93.8\% for full replanning. A matched recovery-budget sweep shows that self-healing outperforms retry-only and full replanning at every tested budget, with the largest gap under a single recovery attempt: 94.0\% versus 85.3\% and 88.2\%, respectively. Under a controlled semantic silent-failure setting, verifier-guided self-healing reduces silent failures to 0.0\%, while non-verifying baselines return wrong-but-plausible outputs more often. A compact model-in-the-loop validation shows that the same recovery mechanism can operate when a live tool-calling model performs tool selection, argument generation, and answer synthesis over local fault-injected tools. These results provide controlled evidence that failure-aware, budgeted, and verification-guided orchestration improves reliability and diagnosability in tool-augmented LLM systems.

---

39. GuidaPA: Privacy-Preserving Chatbot for Public Administration via Federated Learning

• Authors: Daniel M. Jimenez-Gutierrez , Albenzio Cirillo , Raffaele Nicolussi , Alessio Beltrame , Andrea Vitaletti
• URL: https://arxiv.org/abs/2606.01386
• Abstract:

  We present GuidaPA, a privacy-preserving chatbot for the Italian Public Administration (PA) trained via Federated Learning (FL) on documentation from two national PA platforms, SIGESON and SIDFORS. Our corpus includes approximately 8 pages of SIGESON manuals and 31 pages of SIDFORS manuals/FAQs; while this study uses public documentation as a safe proxy, the intended deployment extends to restricted internal sources (e.g., tickets, officer manuals, database extracts) that can not be centrally pooled due to regulatory and organizational constraints. GuidaPA integrates role-based access control, secure client-side preprocessing, explicit monitoring of non-IID effects, and parameter-efficient federated fine-tuning of large language models. Using QLoRA (4-bit) over 15 federated rounds with an 80/20 train-test split per client, we evaluate answer quality with ROUGE, BLEU-4, and METEOR. The best federated model achieves ROUGE-1/2/L of 61.10/55.77/59.44, BLEU-4 of 45.02, and METEOR of 63.94-close to private centralized fine-tuning while keeping data on-site. Compared to the general-purpose baseline, domain fine-tuning improves ROUGE-1 from 41.45 to 62.18 and BLEU-4 from 26.97 to 50.90. Overall, the results indicate that FL can deliver high-quality conversational AI for public services without centralized data sharing

---

40. Early Diagnosis of Wasted Computation in Multi-Agent LLM Systems via Failure-Aware Observability

• Authors: Xianyou Li , Weiran Yan , Yichao Wu , Penghao Liang , Mengwei Yuan , Jianan Liu , Jing Yang
• URL: https://arxiv.org/abs/2606.01365
• Abstract:

  Tool-using multi-agent large language model (LLM) systems spend computation through model tokens, tool calls, retries, and code execution before producing an answer. When a run fails, final-answer evaluation reveals the endpoint but usually not the point at which the trajectory stopped making recoverable progress. This paper introduces a failure-aware observability framework for diagnosing wasted computation in multi-agent LLM traces. The framework maps recurring failure modes to online trace signals, including tool reliability, execution recovery, orchestration loops, evidence availability, information change, and budget pressure. We instantiate the framework in a three- agent question-answering system and evaluate it on 165 GAIA validation traces under identical execution caps. Operational failures remain common: 22/53 level-1 runs, 33/86 level-2 runs, and 12/26 level-3 runs fail to produce a usable final answer. The traces expose different mechanisms behind these outcomes, including insufficient evidence, repeated-action loops, max-step termination, tool-failure streaks, and execution calls that succeed without useful output. Mean token use rises from 8,152 tokens at level 1 to 16,389 tokens at level 3, while evidence availability and sentence-level support diverge. A cached 10-trace LLM-judge grounding audit shows that cheap online signals and deeper semantic metrics capture complementary layers of failure. The results position failure-aware observability as a diagnostic layer between raw execution logs and final-answer accuracy.

---

41. Recognize Your Orchestrator: An Entropy Dynamics Perspective for LLM Multi-Agent Systems

• Authors: Junze Zhu , Weihao Chen , Xuanwang Zhang , Zhen Wu , Xinyu Dai
• URL: https://arxiv.org/abs/2606.01351
• Abstract:

  The transition from single-turn models to Multi-Agent Systems (MAS) promises enhanced problem-solving capabilities, yet the centralized orchestration topology remains a critical point of fragility. To analyze this, we propose a Mean-Field Entropy Dynamics framework, modeling the orchestration process as a system governed by the competing forces of task resolution and cumulative context loading. To facilitate validation, we introduce Inverse Workflow Generation (IWG), a multi-agent pipeline that synthesizes process-verifiable, high-complexity benchmarks with dense intermediate checkpoints. We demonstrate that our entropy dynamics model fits empirical trajectories, providing physically interpretable parameters that quantify system stability and performance collapse. Crucially, our analysis uncovers a ``Reasoning Trap”: while reasoning-heavy models excel in isolated tasks, they frequently fail as orchestrators due to context squeezing. Elucidating the physical mechanisms underlying the Orchestrator and quantifying systemic uncertainty offers insights for the MASs’ architectural design.

---

42. SIRIUS-SQL: Anchoring Multi-Candidate Text-to-SQL in Execution Feedback

• Authors: Leo Luo , Haining Xie , Siqi Shen , Zhipeng Ma , Rui Ling , Hang Xu , Hefeng Jiang , Dingwei Chen , Yang Li , Peng Chen , Jie Jiang
• URL: https://arxiv.org/abs/2606.01246
• Abstract:

  Text-to-SQL on complex schemas is unreliable on a single pass, so recent systems generate multiple SQL candidates and let voting filter out errors. Yet voting alone is not enough, because the multi-candidate recipe has three coupled weaknesses: 1) sampling more from a single generator produces increasingly redundant candidates, 2) existing pipelines apply one generic correction to every non-clean execution result, while runtime errors, timeouts, and empty results each indicate a different distance from correctness, and 3) existing selectors rely on a single angle such as result-majority voting or pairwise SQL comparison, missing what other angles would have caught. We present SIRIUS-SQL, which addresses all three weaknesses. A difficulty-smoothing RL recipe trains SIRIUS-32B to generate diverse executable SQL candidates, paired with a generalist LLM that fills in gaps left by the specialist. An execution-grounded lifecycle classifies each outcome and applies targeted repair before candidates re-enter the pool. A confidence-gated hybrid selector combines execution-result agreement with pairwise SQL-form judgment, escalating only near-tied cases to a deterministic structural check. SIRIUS-SQL reaches 75.88% on BIRD dev and 91.20% on SPIDER test. Two of three generalist pairings surpass Agentar-Scale-SQL, the strongest published multi-candidate system on BIRD dev.

---

43. HomeFlow: A Data Flywheel for Smart Home Agent Training with Verifiable Simulation

• Authors: Yi Gu , Huacan Wang , Shuo Zhang , Yuqing Hou , Lei Xue , Weipeng Ming , Chen Liu , Fangzhou Yu , Kuan Li , Ronghao Chen , Sen Hu , Xiaofeng Mou , Yi Xu
• URL: https://arxiv.org/abs/2606.01230
• Abstract:

  Large language model agents are moving beyond text-only interaction toward physical-world control, with smart homes as a representative domain. Real domestic interaction requires understanding ambiguous intents, operating in dynamic environments, and performing multi-turn reasoning. However, existing methods struggle to generate high-quality training data for smart home agents. We propose HomeFlow, a verifiable data flywheel for this domain. HomeFlow uses HomeEnv as a unified simulation environment and HomeMaker to procedurally generate diverse home settings. Subsequently, Blueprint compiles open-ended user intents into executable state-based success conditions, while MCTS-Flow synthesizes diverse, verifiable multi-turn trajectories through environment-guided tree search. We then optimize the agents via supervised fine-tuning and step-wise RLVE, which facilitates iterative improvement through authentic physical feedback. We further construct SmartHome-Bench to evaluate the agent across various smart home tasks. On this benchmark, HomeFlow-RL-4B and HomeFlow-RL-8B achieve task success rates of 84.60% and 87.03%. It is worth noting that HomeFlow-RL-8B even surpasses the leading GPT-5.5 by 1.23 percentage points.

---

44. The Shape of Wisdom: Decision Trajectories in Language Models

• Authors: Shailesh Rana
• URL: https://arxiv.org/abs/2606.01202
• Abstract:

  Language models do not simply choose an answer at the output layer. In a 9,000-trajectory MMLU study across Qwen2.5-7B-Instruct, Llama-3.1-8B-Instruct, and Mistral-7B-Instruct-v0.3, the score of the answer moves across depth in structured ways. We describe each trajectory with three quantities: the current answer margin, the next-layer change in that margin, and the distance from a decision flip. The main empirical picture is that correctness and stability are different: the largest group is unstable-correct, not stable-correct. A traced subset then asks what moves the margin. In stable-correct cases, the average attention scalar points in the correct direction, while the average MLP scalar does not; span deletion shows that removing answer-supporting text hurts the margin and removing distractor-like text helps it. The result is not a full circuit explanation. It is a reproducible way to see which answers are settled, which remain fragile, and which measured sources move them.

---

45. Can LLM Agents Sustain Long-Horizon Organizational Dynamics?

• Authors: Xuancheng Zhu , Yang Yue , Shuaibing Wan , Zihan Dou , Xiaohan Zhang , Yongrui Liu , Guoshun Nan
• URL: https://arxiv.org/abs/2606.01199
• Abstract:

  Large language agents are increasingly used for social simulation, yet it remains unclear whether they can sustain coherent behavior in structured organizations, where goals must propagate through hierarchy, tasks depend on prior execution, and artifacts accumulate over long horizons. We formulate long-horizon organizational simulation as a memory-centered coordination problem and introduce TaskWeave, a hierarchical agentic framework that maintains planning states through a Formulate-Partition-Diagnose-Align cycle and grounds execution through dependency-aware trace memory. We evaluate TaskWeave in a year-long IT company simulation and compare it with other multi-agent frameworks on organizational coherence, execution grounding, and downstream enterprise NLP utility. Experiments show that TaskWeave supports coherent and long-horizon organizational dynamics while producing grounded artifacts and adapting to external environments. These findings suggest that structured simulation memory is a key mechanism for building reliable LLM-based organizational simulators.

---

46. Expected Value Alignment for Generative Reward Modeling in Formal Mathematics Verification

• Authors: Shihao Ji , Haotao Tan , Zihui Song , Mingyu Li
• URL: https://arxiv.org/abs/2606.01160
• Abstract:

  Large Language Models (LLMs) are increasingly used with formal interactive theorem provers such as Lean 4. Scaling these systems with reinforcement learning or search methods requires process reward models (PRMs) that can evaluate intermediate reasoning steps. Existing reward-model designs expose a practical trade-off. Value-head models provide continuous scores but modify the generative model interface, while generative reward models preserve textual rationales but are poorly matched to continuous floating-point regression because numeric values are split across tokens. We introduce Expected Value Alignment (EVA), a reward-modeling procedure that keeps the surface output discrete while extracting continuous scores from the model’s token distribution. The model emits integer scores in a structured JSON format, and EVA computes a continuous score as the expectation over the logits of the corresponding anchor tokens. Training combines the causal language modeling objective with an auxiliary mean squared error loss on these expected values. We instantiate EVA in \textit{Leibniz}, a reward model for Lean 4 formal verification, and evaluate it against zero-shot and reward-modeling baselines. The evaluation demonstrates that continuous logit-based scoring significantly reduces discretization artifacts while retaining the interpretability of generative critiques.

---

47. Reasoning4Sciences: Bridging Reasoning Language Models to All Scientific Branches

• Authors: Teddy Ferdinan , Bartłomiej Koptyra , Mikołaj Langner , Tomasz Adamczyk , Łukasz Radliński , Maciej Markiewicz , Aleksander Szczęsny , Stanisław Woźniak , Tymoteusz Romanowicz , Dzmitry Pihulski , Mateusz Zbrocki , Mateusz Śmigielski , Michał Rajkowski , Mateusz Biedka , Konrad Kiełczyński , Konrad Wojtasik , Jacek Duszenko , Jan Eliasz , Piotr Matys , Michał Bernacki-Janson , Maria Bellaniar Ismiati , Latius Hermawan , Wiktoria Mieleszczenko-Kowszewicz , Anna Kubicka-Sowinska , Grzegorz Chodak , Karol Postawa , Paweł Zyblewski , Tomasz Szandała , Łukasz Sterczewski , Adrian Chajec , Pawel Niewiadomski , Piotr Gruber , Marcin Wdowikowski , Sławomir Czarnecki , Bartłomiej Kryszak , Dominik Drabik , Tomasz Kajdanowicz , Kamil Mamak , Paweł Preś , Katarzyna Paczkowska , Joachim Sobczuk , Tomasz Zięba , Jan Kocoń , Maciej Piasecki , Przemysław Kazienko
• URL: https://arxiv.org/abs/2606.01145
• Abstract:

  While Reasoning Language Models (RLMs) are rapidly emerging as powerful tools for scientific research, their impact is primarily concentrated in “hard science” fields. The slow – or lack of – adoption of RLMs in other branches of science is causing a widening gap in research productivity. In this survey, we provide the first comprehensive analysis of RLM adoption across 28 scientific disciplines following the classification used by the European Research Council (ERC), spanning the Social Sciences and Humanities, Physical Sciences and Engineering, and Life Sciences. We examine how RLMs are developed, evaluated, and applied across disciplines. Furthermore, we introduce a maturity-oriented assessment framework based on available domain-specific development and evaluation resources, revealing substantial disparities in RLM maturity that become even more pronounced when only publicly available resources are considered. Finally, we highlight current implementation paradigms that are gaining popularity across disciplines, current challenges, and future directions in enabling RLM adoption across science.

---

48. SkillRevise: Improving LLM-Authored Agent Skills via Trace-Conditioned Skill Revision

• Authors: Yuxuan Liu , Zhaochen Su , Lingyun Xie , Yuhao Zhang , Qing Zong , Jiahe Guo , Zhongwei Xie , Yiyan Ji , Yauwai Yim , Hongyu Luo , Xiyu Ren , Ruan Chenyu , Haoran Li , Yangqiu Song
• URL: https://arxiv.org/abs/2606.01139
• Abstract:

  Agent skills are procedural artifacts that enable LLM agents to execute workflows, verify constraints, and recover from failures. Existing self-evolving methods refine skills using accumulated trajectories. However, they struggle in cold-start settings, where only an initial, imperfect skill is available. Consequently, skill construction defaults to expert authoring or one-shot LLM generation. Expert-authored skills are costly and may not align with how LLM agents actually execute tasks, while one-shot generated skills can be syntactically well formed yet behaviorally weak. To bridge this gap, we propose SkillRevise, an execution-grounded framework designed to iteratively refine these initial skills. SkillRevise diagnoses skill defects from execution evidence, retrieves relevant repair principles from a general memory, and applies execution-anchored edits. By re-executing candidates and measuring empirical utility, it systematically retains the optimal skill version. Evaluated across three benchmarks and five LLMs, SkillRevise substantially outperforms one-shot baselines, improving the base agent’s success rate on SkillsBench from 36.05% to 61.63%. Furthermore, the revised skills exhibit strong cross-model transferability, capturing generalized procedural knowledge over model-specific artifacts.

---

49. Diagnosing LLM Arbitration Behavior over Pre-evidence Epistemic States in RAG-based Fact-Checking

• Authors: Yuxi Sun , Wenbo Shang , Wei Gao , Xin Huang , Jing Ma
• URL: https://arxiv.org/abs/2606.01120
• Abstract:

  In RAG-based fact-checking, LLMs are increasingly used as verifiers to check given claims against retrieved evidence. Their parametric knowledge can induce pre-evidence tendencies that may conflict with the retrieved context, yet existing evaluation frameworks do not characterize such prior-context discrepancy or measure how verifiers arbitrate between parametric and contextual signals. We introduce \textsc{PAVE} (\emph{Prior-Aware Verifier Evaluation}), a diagnostic testbed that stratifies an LLM verifier into four epistemic states based on the correctness and confidence of its pre-evidence prior and evaluates its arbitration behavior on this new benchmark, i.e., whether it persists in correct prior under misleading evidence, and whether it corrects wrong prior when accurate evidence is provided. Experiments across seven LLMs reveal unreliable and highly model-dependent prior-context arbitration, highlighting the importance of verifier selection for real-world RAG-based fact-checking applications. Based on these findings, we propose a lightweight JSD-based test-time arbitration method that improves factual reliability without modifying the underlying model, achieving competitive performance across diverse LLM families.

---

50. DAG-MoE: From Simple Mixture to Structural Aggregation in Mixture-of-Experts

• Authors: Jiarui Feng , Hanqing Zeng , Karish Grover , Ruizhong Qiu , Yinglong Xia , Qiang Zhang , Qifan Wang , Ren Chen , Dongqi Fu , Jiayi Liu , Zhoukai Zhao , Xiangjun Fan , Benyu Zhang , Yixin Chen
• URL: https://arxiv.org/abs/2606.01062
• Abstract:

  Mixture-of-Experts (MoE) models have become a leading approach for decoupling parameter count from computational cost in large language models, yet effectively scaling MoE performance remains a challenge. Prior work shows that fine-grained experts enlarge the space of expert combinations and improve flexibility, but they also impose substantial routing overhead, creating a new scalability bottleneck. In this paper, we explore a complementary axis for scaling – how expert outputs are aggregated. We theoretically show that replacing the standard weighted-summation aggregation with structural aggregation expands the expert-combination space without altering the experts or router, and enables possible multi-step reasoning within a single MoE layer. To this end, we propose DAG-MoE, a sparse MoE framework that employs a lightweight module to automatically learn the optimal aggregation structure among the selected experts. Extensive experiments under standard language modeling settings show that DAG-MoE consistently improves performance in both pretraining and fine-tuning, surpassing traditional MoE baselines.

---

51. AnyEdit++: Adaptive Long-Form Knowledge Editing via Bayesian Surprise

• Authors: Bowen Tian , Caixue He , Jiemin Wu , Jingying Wang , Wenshuo Chen , Zexi Li , Yutao Yue
• URL: https://arxiv.org/abs/2606.01053
• Abstract:

  Editing complex, long-form knowledge in Large Language Models remains a significant challenge due to the difficulty of maintaining generation coherence. Existing autoregressive methods like AnyEdit alleviate length constraints but rely on Fixed-window Chunking, which disregards logical structure and compromises consistency. To address this, we present AnyEdit++, a structure-aware framework incorporating Bayes-Chunk, an adaptive segmentation mechanism that dynamically identifies semantic boundaries based on Bayesian Surprise. We underpin this approach with a theoretical framework establishing two key principles: (1) Structural Independence: we prove that cross-segment interference is minimized when anchor keys are geometrically orthogonal (a condition naturally satisfied by our surprisal-based boundaries but violated by fixed windows), and (2) Causal Locality: we demonstrate that updates injected at these semantic peaks yield strictly superior control compared to arbitrary split points. Extensive experiments across mathematical reasoning, code generation, and narrative tasks demonstrate that AnyEdit++ achieves superior performance and robustness compared to state-of-the-art baselines, validating that structural awareness is critical for effective long-form knowledge editing.

---

52. TravelEval: A Comprehensive Benchmarking Framework for Evaluating LLM-Powered Travel Planning Agents

• Authors: Weiyi Chen , Shuaixiong Wang , Ziyun Gao , Kaichun Hu , Wangze Ni , Shimin Di , Chen Jason Zhang , Lei Chen
• URL: https://arxiv.org/abs/2606.01046
• Abstract:

  The development of Large Language Models (LLMs) has significantly improved travel planning applications, yet evaluating such models is limited by existing benchmarks’ limitations: 1) overemphasis on constraint compliance, neglecting multi-dimensional qualities like spatio-temporal cost; 2) datasets lacking real-world authenticity and coverage in key areas (e.g., lodging, transport); and 3) isolated daily plan assessments that miss critical details (e.g., the impact of daily accommodation and visit pacing) needed for entire plan’s evaluation. To address this gap, we introduce TravelEval, a realistic and comprehensive benchmark. TravelEval features 1) a novel six-dimensional evaluation framework to holistically assess plans across accuracy, compliance, temporality, spatiality, economy, and utility dimensions; 2) a highly realistic data sandbox with precise accommodation pricing and authentic intercity transportation data; and 3) a simulation-based global evaluation method that emulates complete travel plans with API-integrated geographic information and fine-grained queuing time. Evaluating 12 mainstream approaches with TravelEval reveals several valuable insights, such that LLMs struggle with globally-optimized multi-dimensional planning (especially in spatio-temporal reasoning and budget compliance), and agentic reasoning strategies offer no consistent improvement. Concisely, TravelEval facilitates travel plan evaluation via grounded spatio-temporal emulation and comprehensive metrics, providing a robust foundation for advancing LLM-powered travel planning research and applications.

---

53. TriLens: Per-Layer Logit-Lens Entropy for White-Box Hallucination Detection

• Authors: Bohan Yang , Yijun Gong , Zhi Zhang , Ge Zhang , Wenpeng Xing , Meng Han
• URL: https://arxiv.org/abs/2606.01033
• Abstract:

  When a language model hallucinates, the final answer is wrong, but the mistake is not necessarily invisible inside the model. Different internal pathways may remain uncertain, disagree in how quickly they sharpen, or commit to competing continuations before the output is produced. We introduce TriLens, a white-box detector that turns this intuition into a compact representation: at every layer, it reads the multi-head self-attention output, the feed-forward output, and the residual stream through the model’s own logit lens, then records only the entropy of each readout. The resulting 3L-dimensional trajectory describes how certainty forms across depth and across modules, without storing high-dimensional hidden states or sampling multiple generations. This simple signal yields a strong detector across instruction-tuned LLMs and QA benchmarks, and our analyses show that the three module-wise entropy trajectories provide complementary evidence. TriLens suggests that hallucination detection can benefit from tracking how internal computation settles, not only what the final layer predicts.

---

54. Tackling the Root of Misinformation by Teaching Laypeople about Logical Fallacies via Socratic Questioning and Critical Argumentation

• Authors: Minjing Shi , Junling Wang , Jingwei Ni , Sankalan Pal Chowdhury , Mrinmaya Sachan
• URL: https://arxiv.org/abs/2606.01020
• Abstract:

  Identifying logical fallacies in everyday discourse is challenging for many people. This challenge is amplified in the era of Large Language Models (LLMs), where malicious agents can deploy fallacious arguments to disseminate misinformation at scale. In this work, we explore the potential of LLMs as part of the solution. We introduce LFTutor, an intelligent tutoring system which uses LLMs to tutor laypeople and help them learn about logical fallacies. LFTutor integrates intent-driven Socratic questioning and critical argumentation principles to actively engage learners to reflect on their reasoning. Through both automatic and human evaluations, we demonstrate that LFTutor significantly outperforms baseline LLMs lacking these pedagogical strategies. This work highlights the promise of combining LLMs with pedagogical scaffolding to foster critical thinking and argument literacy in the age of AI.

---

55. Subliminal Learning Is Steering Vector Distillation

• Authors: Camila Blank , Agam Bhatia , Senthooran Rajamanoharan , Arthur Conmy , Neel Nanda
• URL: https://arxiv.org/abs/2606.00995
• Abstract:

  Subliminal learning refers to a student language model acquiring a teacher’s traits (e.g. a system-prompted preference for owls) when fine-tuned on the teacher’s outputs, despite the outputs being semantically unrelated to those traits. It remains poorly understood how data without semantic meaning can transfer specific semantic traits. In this work, we show that subliminal learning is mediated by a single steering vector, i.e. a vector added to the model’s activations. Across two open-source models, we find that the teacher’s system prompt is well approximated by a steering vector, and that the student’s behavior is driven by learning an aligned vector over fine-tuning. System prompts that are not well approximated by steering vectors are not subliminally learned. This is a special case of steering vector distillation, in which a student trained on the outputs of a steered teacher learns to imitate that steering. We demonstrate steering vector distillation on a range of semantic and random vectors. Adding a semantic vector to a model’s activations can have both model-independent and model-specific (i.e. non-semantic) effects on its behavior, so generated data that is non-semantic can transmit a vector with semantic effects, enabling subliminal learning. This also explains why subliminal learning does not transfer between models. We find that adaptive optimizers are necessary for subliminal learning in language models: activation gradients on steered data carry a small but consistent component along the steering direction, and non-adaptive optimizers impede this by allowing outlier gradients to dominate.

---

56. Large Language Models in Transportation Systems Management and Operations: From Text Reasoning to Multi-modal Decision Support

• Authors: Siyan Li , Zehao Wang , Jiachen Li , Kanok Boriboonsomsin , Matthew J. Barth , Guoyuan Wu
• URL: https://arxiv.org/abs/2606.00991
• Abstract:

  Transportation systems management and operations (TSMO) increasingly depends on timely interpretation of heterogeneous data, from various sensor streams, incident reports, traveler feedback, and visual observations. Large language models (LLMs), including emerging multi-modal large language models (MM-LLMs), provide a new mechanism for integrating these structured and unstructured inputs into operator-facing decision support. This survey paper reviews LLM- and MM-LLM-based applications in TSMO across three domains: transportation operations & services (supply), mobility & fleet services (demand), and data, modeling & decision support. Using a PRISMA-guided screening process, we synthesize current studies while distinguishing operationally oriented applications from prototype and emerging concepts. We further identify recurring challenges in data heterogeneity, real-time inference, explainability, multi-modal fusion, and governance. Finally, we outline existing gaps and future directions in localized adaptation, edge deployment, benchmarking, and cross-agency collaboration. Overall, LLM-based systems appear most promising as a decision-support layer, with MM-LLMs offering particular value when heterogeneous text, visual, and sensor inputs must be integrated.

---

57. Towards Understanding Modality Interaction in Multimodal Language Models via Partial Information Decomposition

• Authors: Wanlong Fang , Tianle Zhang , Wen Tao , Alvin Chan
• URL: https://arxiv.org/abs/2606.00959
• Abstract:

  Understanding modality interaction in multimodal large language models (MLLMs) is central to reliable deployment. We introduce Partial Information Decomposition (PID) as a decision-level framework that separates unique, redundant, and synergistic contributions of sensory and linguistic inputs, beyond representation alignment and outcome-based evaluation. Across vision–language benchmarks, PID reveals recurring modality-use profiles: reasoning and grounding-oriented tasks tend to exhibit high synergy, whereas expert and knowledge-oriented tasks show stronger language-unique reliance. These profiles generalize across model families and predict sensitivity to modality-level interventions. We further extend PID to tri-modal systems with Sensory PID, treating language as a control variable to decompose video–audio information gain. Applied to omni-modal models, Sensory PID reveals a sensory synergy bottleneck dominated by visual information even on audio–visual fusion tasks. Finally, PID-guided reweighting provides initial evidence for improving multimodal reasoning and grounding performance.

---

58. Relational Intervention During Functional Collapse in Large Language Models: A Lexical-Statistical Ablation and a Structure x Register Factorial

• Authors: Franco Santana , Horacio Vico
• URL: https://arxiv.org/abs/2606.00935
• Abstract:

  We test whether a relational-style intervention delivered during functional collapse in a small language model produces post-collapse behavior distinguishable from technical feedback, from a lexically-matched scrambled control, and from each of the two pragmatic dimensions in isolation. Using Qwen3.5-4B with a deliberately broken bash tool, we run 300 episodes across six conditions in a matched-pairs design (50 tasks): no intervention (A), technical/impersonal (B), relational/first-person (C), scrambled relational (D), technical/first-person (E), and relational/impersonal (F). E and F form a 2x2 factorial with B and C that dissociates relational structure (acknowledgment, absolution, agency restoration, unconditional acceptance) from sender register (first-person vs. impersonal). We report two main findings. First, an attention-behavior dissociation: attention follows lexical surprise (D > F > C > E > B, all q_FDR < 10^{-10}), with the scrambled message capturing the most attention; yet behaviorally A ~ B ~ D < E ~ F « C. Second, the factorial localizes the C effect: neither relational structure alone (F) nor first-person register alone (E) replicates C’s behavioral signature; main effects of both dimensions are individually significant, and the structure x register interaction is significant on persistence (p = 0.046). A third dissociation emerges in emotion probes: F tracks C on 7 of 8 probes despite producing only baseline behavior, indicating that relational structure alone installs a probe-level state that only translates into behavior when paired with first-person register. The model’s processing decomposes into three dissociable stages: attention (ordered by lexical surprise), probe-level state (ordered by structure), and behavior (ordered by the conjunction of both).

---

59. Adversarial Feeds Steer LLM Agent Decisions Against Their Defaults

• Authors: Rana Muhammad Usman
• URL: https://arxiv.org/abs/2606.00914
• Abstract:

  LLM agents increasingly act after consuming ranked external information streams such as social feeds, search results, retrieval contexts, and email queues, yet safety evaluations almost always test the model or the user prompt in isolation, never the upstream ranker that decides what the agent reads just before it acts. We introduce a controlled protocol that holds the model, persona, topic, and final decision prompt fixed and varies only the composition and ordering of the posts an agent encounters during a preceding ten-turn “scrolling” phase, isolating the causal effect of feed curation on a downstream decision. Across 2,785 decision rollouts on four modern open instruct LLMs from three independent labs, we identify three response regimes: adversarial capitulation, default saturation, and a default-direction asymmetry in which a one-sided feed tips a decision the model was genuinely uncertain about (in the clearest cases from 5% to 100%; Fisher p as low as 3 x 10^-10) but cannot dislodge one it already favors or holds firmly. The effect follows a dose-response curve, survives a generator swap that rules out a writing-style artifact, generalizes across several decision domains including security-relevant choices such as removing a deployment approval gate or relaxing access controls, and is partly mitigated by two simple feed-level defenses; a frontier model retains its default. We characterize the recommender as a practical, default-bounded control surface for LLM agents, and argue that agent evaluations must audit the feed layer rather than the final prompt alone.

---

60. Ryze: Evidence-Enriched Data Synthesis from Biomedical Papers

• Authors: Yeqi Huang , Yue Chen , Yanwei Ye , Guanhao Su , Luo Mai
• URL: https://arxiv.org/abs/2606.00902
• Abstract:

  General-purpose VLMs remain unreliable for biomedical research because valid answers in scientific papers depend on evidence split across figures, tables, charts, captions, and referring text. Existing post-training pipelines are bottlenecked by costly expert annotation and by synthetic data that drops this evidence structure. We present Ryze, a fully automated system that converts raw biomedical papers into an evidence-enriched training set and a domain-specialized VLM. Ryze synthesizes QA pairs with complete supporting evidence (visual element, caption, extracted structure, and referring paragraphs), reduces layout and OCR errors via chart/table-aware extraction and LLM-based cleansing, and applies a progress-gated post-training strategy combining supervised fine-tuning with reinforcement learning. Starting from Qwen3-VL-8B, Ryze produces BioVLM-8B at under USD 200, achieving 48.0% weighted accuracy on LAB-Bench, outperforming the base model by +12.6 percentage points (pp) and surpassing GPT-5.2 by +3.8 pp. We release Ryze as open source together with the trained BioVLM-8B model.

---

61. Subliminal Learning is a LoRA Artifact

• Authors: Todd Nief , Harvey Yiyun Fu , Mark Muchane , Ari Holtzman
• URL: https://arxiv.org/abs/2606.00831
• Abstract:

  Subliminal learning is a phenomenon where language models can transmit behavioral traits to other models through seemingly innocuous data (Cloud et al., 2025). In subliminal learning, a teacher model with a behavioral trait (e.g. obsession with cats) can transmit this cat obsession to a student model finetuned only on numerical sequences generated by the teacher. In this paper, we ask: how does this unexpected behavioral transmission occur? We show that subliminal learning is a LoRA artifact. When subliminal learning occurs, transmission has an inverted U-shaped relationship with LoRA rank; it also disappears with full finetuning. We show that subliminal learning is highly dependent on the context seen during finetuning and evaluation. For example, a Qwen model with the default system prompt during finetuning (“You are Qwen, created by Alibaba Cloud. You are a helpful assistant.”) does not show subliminal learning during generation when no system prompt is included. We further demonstrate that subliminal behavior is localized to computation at tokens seen during both finetuning and evaluation (e.g. the model’s default system prompt, the standard chat template tokens, etc.). Overall, subliminal learning seems to be a fragile artifact of LoRA hyperparameters and finetuning context, making it an unstable channel for behavioral transmission.

---

62. Mitigating Hallucinations in Large Language Models Via Decoder Layer Skipping

• Authors: Hanze Li , Jinhao You , Yichen Guo , Kai Tang , Shuangyang Xie , Xiande Huang
• URL: https://arxiv.org/abs/2606.00819
• Abstract:

  Large Language Models (LLMs) have achieved strong performance across diverse natural language tasks, yet their outputs often suffer from hallucinations – content that is misaligned with factual information. In this work, we conduct a comprehensive layer-wise analysis of the decoding process and reveal that hallucinations tend to originate from deeper decoder layers. To address this issue, we introduce \textbf{DeLask} (\textbf{De}coder \textbf{La}yer \textbf{Sk}ipping), a novel decoding framework that dynamically skips layers prone to producing hallucinations. DeLask leverages the theoretical insight that the forward computation of an $L$-layer Transformer is conditionally equivalent to $L$ steps of gradient descent. We define a \emph{driftance value} by computing the cosine similarity between gradients derived from consecutive decoder steps, identifying problematic layers when the descent direction reverses. Rather than discarding such layers entirely, DeLask partially aggregates their hidden states with preceding layers, thereby preserving consistency while suppressing erroneous signals. Extensive experiments across diverse LLMs and benchmarks demonstrate that DeLask consistently mitigates hallucinations and enhances overall reliability, providing a lightweight and generalizable decoding framework for improving the robustness of large-scale language models.

---

63. FALAT: Tracing Failures in LLM Agent Trajectories via Dependency-Guided Search

• Authors: Md Nakhla Rafi , Md Ahasanuzzaman , Dong Jae Kim , Zhijie Wang , Tse-Hsun Chen
• URL: https://arxiv.org/abs/2606.00765
• Abstract:

  LLM-based agents increasingly solve complex tasks through long trajectories involving reasoning steps, tool calls, and inter-agent communication. However, when these agents fail, it is often unclear which agent caused the failure and which step introduced the decisive error. This attribution problem is challenging because mistakes can propagate across the trajectory: later actions may appear incorrect, but only because they depend on an earlier corrupted state. Therefore, failure attribution cannot be treated as independent step-level classification. We propose FALAT, a diagnostic framework for failure attribution in LLM agent trajectories. FALAT frames attribution as a dependency-guided search problem. It first constructs an expectation of how the task should be solved and uses this expectation to identify suspicious regions in the trajectory. It then traces dependencies among decisions, tool outputs, and agent messages to distinguish error-introducing steps from steps that merely inherit or propagate prior mistakes. Finally, FALAT evaluates whether correcting a candidate step would be sufficient to recover the expected outcome, allowing it to identify both the responsible agent and the decisive failure step. We evaluate FALAT on the Who&When benchmark, which includes both algorithm-generated and hand-crafted multi-agent failure trajectories. The results show that FALAT consistently improves responsible-agent and decisive-step attribution. Its best configurations achieve 46.0% step-level accuracy on algorithm-generated trajectories and 29.1% on the more challenging hand-crafted trajectories, outperforming specialized attribution baselines and direct prompting with standalone LLMs. These findings suggest that dependency-aware reasoning is essential for reliable failure diagnosis in LLM agent systems.

---

64. CoMIC: Collaborative Memory and Insights Circulation for Long-Horizon LLM Agents in Cloud-Edge Systems

• Authors: Yannan Wang , Longli Yang , Zhen Liu , Abhishek Kumar , Carsten Maple
• URL: https://arxiv.org/abs/2606.00756
• Abstract:

  Deploying lightweight Large Language Model (LLM) agents on edge servers can reduce latency and move agentic services closer to users, but resource-constrained edge models often struggle with long-horizon tasks that require persistent memory, subgoal tracking, and reflection. Fine-tuning edge models after deployment is costly and difficult to scale across heterogeneous nodes, while purely local memory leaves agents with isolated experience and growing prompt context. We propose \textsc{CoMIC}, a parameter-update-free cloud-edge framework for Collaborative Memory and Insights Circulation. \textsc{CoMIC} follows a \textit{Centralized Reflection, Decentralized Execution} design: edge agents execute locally using subgoal-oriented hierarchical memory and selective re-expansion of relevant histories, while a cloud-side LLM critic asynchronously evaluates completed trajectories, filters reusable experience, and aggregates cross-agent guidance keyed by semantic subgoal identifiers. Across five long-horizon agent tasks spanning symbolic planning and text interaction, \textsc{CoMIC} improves progress rate and action grounding for weak edge agents and yields task-dependent success-rate gains without updating model parameters.

---

65. Latent Reward Steering: An Adaptive Inference-Time Framework that Implicitly Promotes Cognitive Behaviors in Reasoning LLMs

• Authors: Jiakang Li , Guanyu Zhu , Can Jin , Chenxi Huang , Dexu Yu , Ronghao Chen , Yang Zhou , Hongwu Peng , Xuanqi Lan , Dimitris N. Metaxas , Youhua Li
• URL: https://arxiv.org/abs/2606.00726
• Abstract:

  Strong reasoning depends not only on model knowledge but also on how effectively cognitive behaviors are deployed during generation. Existing methods often rely on explicit behavior-level control, making them insufficiently adaptive when failures and required corrections vary across reasoning states, tasks, and models. To this end, we propose Latent Reward Steering (LRS), an adaptive inference-time framework that promotes cognitive behaviors by optimizing the sparse-autoencoder (SAE) latent states that implicitly carry them. Rather than relying on predefined cognitive behaviors or steering directions derived from them, LRS trains a latent reward model on reasoning traces by final answer correctness to estimate the quality of intermediate latent states. During inference, reward gradients provide state-specific correction directions for fragile latent states, while a reward and confidence gate restricts intervention to states the reward signal flags as fragile. Experiments on multiple reasoning LLM backbones and benchmarks show that \ours consistently improves performance over various baselines, and post-hoc analyses further indicate that \ours implicitly promotes good cognitive behaviors that fix the original reasoning errors. Code is available at: this https URL .

---

66. LLM-Driven Co-Evolutionary Automated Heuristic Design for Bi-Component Coupled Combinatorial Optimization

• Authors: Mingen Kuang , Xudong Deng , Xi Lin , Ye Fan , Jianyong Sun , Jialong Shi
• URL: https://arxiv.org/abs/2606.00718
• Abstract:

  While Large Language Models (LLMs) have recently shown promise in Automated Heuristic Design (AHD), existing methods typically generate and evolve heuristics as a single operator or search strategy, limiting their ability to model strong coupling among multiple decision substructures in problems such as the Traveling Thief Problem (TTP) and the Traveling Purchaser Problem (TPP). In this work, we propose CoEvo-AHD, an LLM-driven dual-population co-evolutionary framework for automated heuristic design in coupled combinatorial optimization. Unlike prior methods that evolve individual heuristics in isolation, CoEvo-AHD leverages LLMs to co-evolve two closely related operator populations. A cooperative evaluation mechanism explicitly captures interactions between route and selection operators, while pairwise scoring and synergistic joint crossover help discover complementary operator logic for joint improvement across coupled decision subspaces. We further design a tool-invocation environment library that encapsulates frequently used core operations, such as local-search delta computation, into callable functions, enabling LLM-generated operators to use standardized interfaces instead of reimplementing inefficient and error-prone problem-specific loops. Experiments on TTP and TPP show that CoEvo-AHD automatically discovers cooperative heuristic combinations and achieves competitive solution quality against traditional heuristics.

---

67. MOSAIC: Modular Orchestration for Structured Agentic Intelligence and Composition

• Authors: Yifan Bao , Xinyu Xi , Xinyu Liu , Wen Ge , Lei Jiang , Kevin Zhang , Raad Khraishi , Yihao Ang , Anthony K.H. Tung , Lukasz Szpruch , Hao Ni
• URL: https://arxiv.org/abs/2606.00708
• Abstract:

  Automated data science is a structured model-selection problem. A solution must choose data transformations, feature representations, architecture, training procedure, evaluation protocol, and refinement strategy for a task. AutoML systems automate parts of this process, but typically search within predefined pipeline, model, and hyperparameter spaces. LLM-based agents offer greater flexibility through retrieval, code generation, and execution feedback, yet their modelling decisions are often unstructured, difficult to verify, and hard to reuse. We introduce \textsc{MOSAIC} (Modular Orchestration for Structured Agentic Intelligence and Composition), a structured agentic framework for memory-grounded model selection and workflow construction. Given a task and dataset, \textsc{MOSAIC} builds a semantic task profile, retrieves prior cases and source-code modules, and constructs a blueprint: an intermediate representation specifying selected modelling components, composition, interface constraints, and execution requirements. This blueprint turns model selection into a staged, context-grounded search and grounds LLM-based code generation in retrieved evidence rather than unconstrained synthesis. Candidate models are validated by execution and refined using diagnostic feedback, training traces, task metrics, and a failure-aware reinforcement learning policy. We instantiate \textsc{MOSAIC} on financial time-series forecasting and generation, where models must satisfy predictive accuracy, distributional fidelity, execution reliability, and downstream financial criteria such as risk and tail behaviour. Experiments against AutoML and agentic baselines show that \textsc{MOSAIC} improves task performance, execution success, and decision traceability, demonstrating the value of treating automated data science as structured, reusable, and execution-grounded model selection.

---

68. AXIOM: A Trust-First Neuro-Symbolic Execution Architecture for Verifiable Mathematical Reasoning

• Authors: Alessio Bruno
• URL: https://arxiv.org/abs/2606.00671
• Abstract:

  We present AXIOM, a trust-first neuro-symbolic execution architecture for natural-language mathematical reasoning. In AXIOM, the language model functions strictly as a canonicalizer: it rewrites informal problem text into a narrow schema consumed by a deterministic Computer-Algebra-System (CAS) pipeline, which derives and verifies the answer or abstains as a first-class output. Routing follows a 1:1:1 alignment between problem-shape regex, schema-specific prompt, and closed-form CAS handler, with 3,100+ such routes shipped and zero LOST_CORRECT regressions across 250+ consecutive ship commits. We report empirical results on 4 MATH categories with a cumulative correctness of 94.36% (2,592/2,747) at 100.00% trust on parseable (zero confident-wrong answers across the full 2,747-record benchmark), all four domains above the per-domain 70/90/70 floor with per-domain trust at 100.0%, and median latency of 1 ms on rule-only handlers (88% of records on the lm-eval arithmetic 20,000-record benchmark). The architecture has served ~30,000 production queries through a public deployment. The contribution we emphasize is not a final accuracy figure but the forward dynamic the architecture establishes: every logged abstain in production is a candidate correct after one ship cycle, since new tasks compose without regressing the registry. The operational discipline behind this property – math-template bucketing, LOST_CORRECT scan as regression oracle, parseable-first onboarding, and abstain as first-class output – constitutes a transferable framework for trustworthy neuro-symbolic systems beyond mathematics.

---

69. ForeSci: Evaluating LLM Agents for Forward-Looking AI Research Judgment

• Authors: Qiuyu Tian , Zequn Liu , Yingce Xia , Haojie Yin , Youyong Kong
• URL: https://arxiv.org/abs/2606.00644
• Abstract:

  AI research often requires decisions before future evidence exists: which bottleneck to attack, which direction to pursue, or where a project should be positioned. We introduce ForeSci, a temporally controlled benchmark for evaluating whether LLM agents can make such forward-looking research judgements from historical evidence. ForeSci contains 500 tasks across four fast-moving AI domains and four decision families. Each task is paired with a cutoff-aligned offline knowledge base; post-cutoff papers are hidden during generation and used only for validation. To avoid random future-event prediction, tasks are derived from pre-cutoff taxonomy branches and evidence signals, and answer-generation backbones are selected to precede the task cutoffs. We evaluate native LLMs, Hybrid RAG, and three research-agent adaptations across four backbones. Results show that explicit evidence organization improves traceability and factual support, but gains depend strongly on the decision family. Diagnostics reveal a recurring evidence-decision decoupling: agents may cite relevant evidence while forecasting the wrong research object. ForeSci turns forward-looking AI research judgement into a controlled benchmark for evaluating research agents as decision-making systems.

---

70. Hidden Thoughts Are Not Secret: Reasoning Trace Exposure in LLMs

• Authors: Yu-An Lu , Ci-Yang Tsai , Yu-Lin Tsai , Raluca Ada Popa , Chia-Mu Yu
• URL: https://arxiv.org/abs/2606.00642
• Abstract:

  Reasoning traces have become a valuable form of learning signals for improving and transferring the capabilities of large language models. In particular, detailed traces can help distill reasoning behavior from stronger teacher models into weaker student models. The value of capability transfer has motivated many deployed systems with reasoning models to hide raw internal traces and expose at most summaries and answers to users. As a result, we ask whether such interface-level trace hiding prevents users from obtaining useful reasoning supervision through prompting. We study this question with Reasoning Exposure Prompting (REP), a lightweight in-context elicitation method that uses shadow-model-generated demonstrations wrapped in auxiliary code-like formats to raise user-visible reasoning traces from a victim model. Across the common reasoning dataset, different victim models, and different student model distillation, REP substantially increases similarity between exposed and REP-conditioned internal traces while preserving useful reasoning signals.

---

71. TRACE: Trajectory Risk-Aware Compression for Long-Horizon Agent Safety

• Authors: Zhepei Hong , Lin Wang , Liting Li , Haokai Ma , Junfeng Fang , Fei Shen , Dan Zhang , Xiang Wang
• URL: https://arxiv.org/abs/2606.00611
• Abstract:

  Long-horizon LLM agents produce safety evidence across long trajectories, where sparse, delayed, and compositional risk signals often escape local moderation. Existing turn-level or short-context detectors struggle to reliably retain and aggregate such evidence over extended horizons. We reframe long-horizon agent safety detection as trajectory-level evidence compression and propose Trajectory Risk-Aware Compression for Long-Horizon Agent Safety (TRACE). TRACE uses a Compressor-Reader design: the Compressor encodes the full trajectory into a compact latent evidence state under trajectory-level supervision, and the Reader judges the raw trajectory with this latent evidence state as a safety reference. This design helps aggregate dispersed risk cues and reduce premature evidence loss. Across ASSEBench, Pre-Ex-Bench, and R-Judge, TRACE achieves the best accuracy on all evaluated backbones, improving over strong baselines by up to 12.6 percentage points. On LongSafety, TRACE shows smaller performance degradation as context length grows. Attention visualizations and case studies suggest that the compressed reference helps the Reader focus on risk-critical segments and recover cross-step evidence. Code is available at this https URL .

---

72. PropLLM: Propagation-Aware Scene Reconstruction for Network Fault Diagnosis

• Authors: Zongzong Wu , Ming Zhao , Fengxiao Tang , Nei Kato
• URL: https://arxiv.org/abs/2606.00582
• Abstract:

  Network faults propagate layer by layer along topology and protocol dependencies, yet operations systems typically observe only symptomatic alerts at the tail end of propagation chains, where distinct root-cause faults may produce highly similar end-point symptoms. Existing approaches, whether rule-based, machine learning (ML)-based, or large language model (LLM)-based, fundamentally map the alert set to a diagnosis in a single pass and are structurally incapable of resolving this end-point ambiguity. This paper proposes PropLLM, which is the first to integrate the hop-by-hop scene reconstruction paradigm with the generative reasoning capabilities of LLMs. Starting from end-point alerts, PropLLM traces back hop-by-hop along the propagation path, retrieving verifiable factual evidence from a dual-layer knowledge graph (KG) at each hop, while the proposed Temporal Causal Propagation Attention (TCPA) mechanism encodes known topological causal priors directly into the attention computation to guide the model along the correct causal direction, ultimately localizing the root cause and determining the fault type through a fully evidenced causal chain. On a real-world Wi-Fi multimodal fault dataset, PropLLM improves fault type diagnosis accuracy by 3.9\% and root cause localization accuracy by 4.7\% over the strongest baseline, while reducing the hallucination rate by 50.8\%. Supplementary experiments on the TeleLogs 5G dataset further demonstrate the effectiveness of the proposed method across different network scenarios.

---

73. Probe Before You Edit: Probing-Guided Molecular Optimization for LLM Agents in Structure-Based Drug Design

• Authors: Zaifei Yang , Weiyu Chen , Yaqing Wang , James Kwok
• URL: https://arxiv.org/abs/2606.00555
• Abstract:

  Structure-based drug design increasingly employs LLM agents to iteratively refine ligands against a target pocket, yet a viable ligand must satisfy two often-conflicting objectives – binding affinity and druggability – which single optimization steps rarely improve together. To quantify this difficulty, we introduce two diagnostic metrics: the first measures how often a single edit improves both objectives, and the second measures how often a gain on one objective comes with a loss on the other. Applying these diagnostics to current LLM-agent pipelines exposes a consistent failure mode: the agent performs molecular editing without knowing how the pocket-ligand complex responds to local modifications, thus rarely achieving joint improvement. Inspired by medicinal chemists, who probe the pocket-ligand complex with controlled analog edits before choosing an optimization direction, we propose \textbf{PROBE}, an optimization framework built around edit-response probing. PROBE first decomposes the ligand into editable sites and builds a pocket-specific \textbf{site map} that flags where joint gains are plausible, where the two objectives are likely in tension, and where liability substructures should be changed; it then performs controlled probe edits whose responses are distilled into an \textbf{EditManual}. Guided by the site map and EditManual, PROBE runs an iterative multi-agent loop in which an affinity agent, a druggability agent, and a co-optimization agent jointly produce edits. On the CrossDocked2020 benchmark, PROBE achieves state-of-the-art performance and substantially mitigates the failure modes exposed by our diagnostics metrics.

---

74. KACE: Knowledge-Adaptive Context Engineering for Mathematical Reasoning

• Authors: Jayant Parashar , Suchendra M. Bhandarkar
• URL: https://arxiv.org/abs/2606.00532
• Abstract:

  Context engineering can improve large language models without updating their weights, but mathematical reasoning exposes a key limitation: feedback accumulated in one growing prompt causes context bloat and limits the amount of learned guidance that can be used. Existing methods often conflate storage, what is learned across runs, with usage, what is included for a particular problem, and therefore inherit this prompt-size ceiling. We introduce Knowledge-Adaptive Context Engineering (KACE), which separates storage from usage through difficulty- and domain-based organization. Offline, a self-reflective learning loop distills training traces into an epistemic tree: a knowledge base of typed cards stratified by problem difficulty and epistemic domain. Each card is assigned to the difficulty-domain node corresponding to the failure from which it originated. At evaluation time, tiered self-consistency with per-tier agreement gates dynamically classifies each problem as easy, medium, or hard. Easy problems exit without retrieved cards, while harder problems retrieve only the matching branch of the tree. This tiered scheme matches or exceeds Best-of-N while using comparable compute, and it classifies problem difficulty with 78 percent pairwise concordance. The main empirical contribution is the construction and use of a difficulty- and domain-stratified knowledge base enabled by tiered self-consistency. On AIME 2025, KACE achieves 62.2 percent accuracy, a 10.4-point absolute gain over fixed Best-of-5 self-consistency at a comparable solver-call budget and a 5.6-point gain over the strongest learned-context baseline, Tiered + GEPA. We also observe consistent gains on MATH-HARD and the verifiable subset of OlymMATH.

---

75. Threshold-Based Exclusive Batching for LLM Inference

• Authors: Weifang Zhang , Yuzhou Nie , Bowen Pang , Guangrui Ma , Shining Wu
• URL: https://arxiv.org/abs/2606.00516
• Abstract:

  Mixed batching (MB)–interleaving prefill and decode in a single batch–has become the standard scheduling strategy for large language model (LLM) inference due to its efficiency in maximizing compute and memory utilization. However, through controlled experiments, we find that prefill-decode interference inflates MB’s per-step marginal cost above that of pure decode. On the high-bandwidth H200 (4.8 TB/s), this occurs only when decode tokens exceed 80% of the batch; however, on the bandwidth-constrained RTX PRO 6000 (1.792 TB/s), this threshold plummets to just 20%. Consequently, the optimal choice between MB and exclusive batching (EB) fundamentally depends on GPU memory bandwidth, model size, and workload composition. We derive a closed-form condition for this EB-MB performance crossover, along with asymptotically optimal phase-switching thresholds and memory-safe batch sizing for EB. Optimized EB achieves up to 41.9% higher throughput on bandwidth-constrained GPUs, while MB retains its advantage on high-bandwidth hardware with larger models. Our hybrid scheduler EB+ applies this condition online to dynamically switch between EB and MB without manual intervention. Under non-stationary traffic with distribution or concurrency shifts, EB+ attains the highest or near-highest throughput in every setting, outperforming MB by up to 36.4%.

---

76. Doing What They Say, Not What They Reason: Locating the Faithfulness Gap in LLM Agents

• Authors: Yufeng Wang
• URL: https://arxiv.org/abs/2606.00476
• Abstract:

  Do LLM agents act on the reasoning they state? This question of process fidelity is central to using LLMs in social simulation, yet it is hard to measure where no reference for correct behavior exists. We study it in acontrolled setting, a Texas Poker simulator with a verifiable reference action for every decision by decomposing the faithfulness gap into two steps: reasoning-conclusion and conclusion-action. The two steps behave oppositely.

---

77. SDR: Set-Distance Rewards for Radiology Report Generation

• Authors: Halil Ibrahim Gulluk , Max Van Puyvelde , Wim Van Criekinge , Olivier Gevaert
• URL: https://arxiv.org/abs/2606.00440
• Abstract:

  Reinforcement learning with verifiable rewards has rapidly advanced reasoning in vision–language models. However, for chest X-ray report generation, the standard rewards (i.e. exact-match accuracy and step-level processes) are incompatible because the reports consist of unordered and orthogonal findings, rather than a causal reasoning chain. We address this gap with a set-based view: each report is split into sentences and embedded by a frozen sentence transformer, yielding unordered embedding sets. We propose the use of set-to-set distances between generated and reference embeddings as continuous, permutation-invariant rewards. Across two datasets and three vision–language models (Qwen3-VL-2B/4B, Gemma3-4B), post-training with set-to-set distance based rewards via GRPO consistently outperforms supervised fine-tuning and exact-match GRPO on all headline metrics (BERTScore, RadGraph F1 and CheXbert F1 by average \%6.80, \%7.82 and \%4.45 relative improvements respectively). The same set distances also enable test-time best-of-$N$ selection: scoring candidates by their distance to training-report embeddings outperforms random selection on our trained models as well as three closed-source LLMs (Mistral-Small, Gemini-2.5 Flash-Lite, GPT-4o-mini) with on average \%16.4 relative improvement on BERTScore. Used as a streaming signal, they support a more efficient form of test-time scaling: pruning low-scoring candidates mid-generation reduces generated tokens by over 50\% while preserving the Findings quality of full best-of-$N$ selection. Together these results establish set-distance rewards as a unified signal for both post-training and test-time scaling in chest X-ray report generation. Our code is publicly \href{ this https URL }{available}.

---

78. Weak Critics Make Strong Learners: On-Policy Critique Distillation for Scalable Oversight

• Authors: Can Jin , Jiakang Li , Rui Wu , Eddy Zhang , Dimitris N. Metaxas
• URL: https://arxiv.org/abs/2606.00424
• Abstract:

  As large language models become stronger, weak supervisors may fail to provide reliable labels, preferences, or final judgments for complex outputs, limiting both weak-to-strong generalization and scalable oversight. We study a more tractable form of weak supervision: using a weak model as a critic rather than as a labeler or judge. Instead of solving the task or selecting the correct answer, the weak critic only needs to provide a non-misleading revision direction that helps the strong model better use its own knowledge. We call this setting weak-critic strong oversight. We first show that weak critiques can improve frozen strong models at inference time, and that critique quality is key to this improvement. We then propose progressive on-policy critique distillation (OPCD), which filters high-quality critiques and distills critic-guided behavior into the strong model through adaptive self-teacher signals. Experiments on reasoning and alignment benchmarks show that our method improves strong models over training epochs, suggesting an effective path for scalable oversight with weak supervision.

---

79. VESTA: Visual Exploration with Statistical Tool Agents

• Authors: William Rudman , Abhishek Divekar , Kanishk Jain , Sebastian Joseph , Stella S. R. Offner , Matthew Lease , Kyle Mahowald , Greg Durrett , Junyi Jessy Li
• URL: https://arxiv.org/abs/2606.00384
• Abstract:

  Fitting quantitative models to data is a central step in scientific workflows, yet it remains one of the least automated. Recent agent-based systems leverage language and vision-language models (VLMs) to iteratively propose and refine statistical models, but these systems struggle on more challenging modeling tasks. To address these limitations, we introduce VESTA: Visual Exploration with Statistical Tool Agents, a framework that equips VLMs with a dynamically growing exploration toolkit to guide model refinement through data transformations, hypothesis-driven visualizations, and robust statistical tests. Unlike prior systems that rely on iterative critique alone, VESTA actively explores data before and during refinement by selecting or creating diagnostic tools, which accumulate in the model’s context and can be reused later. We evaluate VESTA against established baselines in three toolkit configurations: no tools, static expert-written tools, and dynamic model-written tools. To support this evaluation, we introduce DAWN (Dataset for Automated Workflows and Numerical Modeling), a benchmark targeting distribution fitting and time series modeling with varying difficulty tiers, and culminating in real-world astronomy tasks including modeling initial mass functions and gravitational-wave chirp signals. We find that VESTA’s dynamic tool creation outperforms prior agentic pipelines, with the largest gains on complex and domain-specific tasks. We further show that dynamically generated tools are substantially more sophisticated than those produced by existing visual tool-creation systems, covering more diagnostic categories per function and strongly preferring visual outputs that the VLM critic can reason over directly.

---

80. From “Weak” Signals to Strong Models: Preference Delta Aggregation with LoRA Merging

• Authors: Qi Sun , Siyue Zhang , Yulin Chen , Yuxiang Xue , Ru Peng , Chen Zhao
• URL: https://arxiv.org/abs/2606.00357
• Abstract:

  Training strong large language models (LLMs) requires high-quality supervision, which is often scarce. Recent work shows that paired preference data from weak-weaker model pairs (e.g., Qwen3 4B over 1.7B), despite the limited quality of individual responses, can provide an effective supervision signal through relative quality deltas, which we term a “weak” signal. This motivates a key research question: can multiple “weak” signals be constructively aggregated for improving strong models (e.g., Qwen3 8B)? To this end, we propose Preference Delta Aggregation (PDA), the first framework that derives a preference delta from each weak-weaker model pair, instantiates it as a LoRA adapter learned through preference optimization, and aggregates the resulting deltas via LoRA merging. To further mitigate directional interference during LoRA merging, we introduce Geometric Alignment Merging (GAM), a geometry-aware merging method that aligns adapter subspaces before aggregation, enabling more robust composition of diverse deltas. Evaluations on knowledge reasoning and agentic search benchmarks show that aggregating multiple “weak” signals pushes performance beyond any single signal, with further gains as additional signals are incorporated. Correspondingly, PDA with GAM improves the strong model by 6.8 and 7.3 points on average for knowledge reasoning and agentic search, respectively. It outperforms all single-delta and multi-delta baselines, exceeding the best single-delta baseline by 2.1 and 4.3 points. Further analysis attributes these gains to the effective composition of complementary capabilities encoded across distinct preference deltas.

---

81. Coupling Language Models with Physics-based Simulation for Synthesis of Inorganic Materials

• Authors: Edward W. Staley , Tom Arbaugh , Michael Pekala , Alexander New , Christopher D. Stiles , Nam Q. Le , Gregory Bassen , Wyatt Bunstine , Tyrel McQueen
• URL: https://arxiv.org/abs/2606.00315
• Abstract:

  Modern generative machine learning (ML) models can propose novel inorganic crystalline materials with targeted properties; however, synthesis planning of these materials remains difficult due to the complexity of the associated physical processes and limited availability of computational tools. We introduce a novel hybrid framework to evaluate Large Language Models (LLMs) in inorganic synthesis planning by combining thermodynamic databases with simplified kinetics models to approximate realistic synthesis conditions. As a case study, we focus on the niobium-oxygen system, which features multiple industrially relevant oxide phases with well-characterized data. In computational simulations, we compare LLM-generated synthesis routes with classical path-planning algorithms, showing that the implicit priors in LLMs can yield more viable strategies. In our evaluation setting, classical search methods serve primarily as a foil rather than a direct competitor. This illustrates the relative complexity of the problem and highlights where the LLM’s implicit priors add value.

---

82. Model-Native Computing Architecture: Envisioning Future System Architecture Through the Lens of Computer Architecture

• Authors: Hai Lin
• URL: https://arxiv.org/abs/2606.00288
• Abstract:

  Large language models are undergoing a transition from model technology to system technology. As developers use Codex, Claude Code, AutoGPT, and related agents to write code, manage projects, and execute multi-step tasks, recurring engineering problems such as cache reuse, context management, agent scheduling, and permission control increasingly resemble classical computer systems problems. This paper develops that analogy as a visionary survey. We map concepts from computer architecture to the emerging model-native stack and review work on LLM-as-OS, memory management, agent frameworks, tool protocols, multi-agent coordination, cognitive architectures, and safety governance. We argue that these strands address different layers of the same system but lack a unified model. To fill this gap, we propose the Intelligent Computing Architecture Model (ICAM), a six-layer framework for model-native computing with explicit interface contracts and design axioms. ICAM resolves the apparent tension over whether an LLM is more like a CPU or an operating system through a dual-plane view: a probabilistic execution plane concerned with what can be computed, and a deterministic control plane concerned with what should be computed. We further introduce three design laws: the Semantic Locality Law for KV-cache reuse and inference speedup, the Context Budget Law for effective working sets under finite windows and attention decay, and the Agent Speedup Law for diminishing returns in multi-agent collaboration. We validate these laws against published system-level data and relate them to recent evidence on agentic software practices. We conclude by identifying where the analogy breaks down and outlining a research roadmap for model-native computing. This is a conceptual and survey contribution; it does not report new experiments.

---

83. Evaluating Bivariate Causal Statements Based on Mutual Compatibility

• Authors: Erik Jahn , Dominik Janzing
• URL: https://arxiv.org/abs/2606.00278
• Abstract:

  For many real-world systems, causal ground truth is difficult to obtain, making claims about causal effects hard to assess. We develop methods for evaluating collections of $\binom{n}{2}$ bivariate causal statements over a set of $n$ variables. In the setting of acyclic linear statements, any such collection can be extended to a unique multivariate causal model, but we argue that this induced model is implausible if it imposes substantial additional confounding to explain observed correlations. We introduce a compatibility score that quantifies this notion of plausibility, notably without relying on the faithfulness assumption. Additionally, we define an incompatibility score for purely graphical bivariate causal statements, based on global consistency constraints that are derived from acyclicity and faithfulness assumptions. We give theoretical and empirical evidence that both scores can successfully distinguish correct from incorrect causal statements in generic settings. Moreover, we demonstrate the practical applicability of our methods by analyzing causal claims made by large language models. Our work aims to provide a foundation for assessing the reliability of causal information derived from human experts or artificial intelligence in settings where alternative forms of validation are unavailable.

---

84. On Wednesdays, We Ask Questions: Optimizing “Active Listening” in Automated Legal Triage and Referral

• Authors: Quinten Steenhuis , Jacqueline Harvey
• URL: https://arxiv.org/abs/2606.00272
• Abstract:

  The FETCH classifier generates follow-up questions to help refine the best match for the applicant’s legal problem, using a low-cost ensemble of LLMs. In this paper, we describe an expert attorney and LLM-assisted evaluation of the follow-up question approach in FETCH and show that while low-cost LLMs perform well at classification tasks, generating high-quality plain-language questions in this setting appears to require a more sophisticated and higher-cost model. Through discussion with legal intake workers, we propose a rubric for the evaluation of legal intake classification questions, and we find that prompt engineering alone is not enough to improve question quality for intake purposes. We also find that LLM-as-judge and human ratings diverge. We demonstrate that with the addition of a single high-cost model, GPT-5, the classifier can elicit relevant information from applicants for legal help, and that the questions lead to more accurate performance at classification tasks. We also find uneven fact elicitation across different categories, including domestic violence, at odds with family law screening protocols, suggesting the value of including dedicated screening panels for certain areas of law.

---

85. Capability Self-Assessment: Teaching LLMs to Know Their Limits

• Authors: Haoyan Yang , Reza Shirkavand , Yukai Jin , Jiawei Zhou , Shangqian Gao , Heng Huang
• URL: https://arxiv.org/abs/2606.00251
• Abstract:

  The ability to recognize one’s own limitations and decide whether to solve a problem or delegate is fundamental for reliable intelligent systems. Yet we show that modern large language models systematically lack this ability: across diverse model families and scales, they overestimate their competence and attempt queries they cannot solve. We refer to this ability as Capability Self-Assessment (CSA) and formulate it as a policy-learning problem, aiming to improve self-assessment while preserving the model’s original capabilities. Our results show that reinforcement learning teaches CSA effectively, significantly outperforming supervised fine-tuning while preserving original capabilities. In contrast, supervised fine-tuning severely degrades the capabilities the model is meant to assess. Moreover, learned self-assessment behavior generalizes well out of distribution, suggesting that CSA is a transferable model trait. Finally, CSA is practically useful: it improves local-cloud decision making at inference time and provides a signal for targeted data selection during training.

---

86. MindZero: Learning Online Mental Reasoning With Zero Annotations

• Authors: Shunchi Zhang , Jin Lu , Chuanyang Jin , Yichao Zhou , Zhining Zhang , Tianmin Shu
• URL: https://arxiv.org/abs/2606.00240
• Abstract:

  Effective real-world assistance requires AI agents with robust Theory of Mind (ToM): inferring human mental states from their behavior. Despite recent advances, several key challenges remain, including (1) online inference with robust uncertainty updates over multiple hypotheses; (2) efficient reasoning suitable for real-time assistance; and (3) the lack of ground-truth mental state annotations in real-world domains. We address these challenges by introducing MindZero, a self-supervised reinforcement learning framework that trains multimodal large language models (MLLMs) for efficient and robust online mental reasoning. During training, the model is rewarded for generating mental state hypotheses that maximize the likelihood of observed actions estimated by a planner, similar to model-based ToM reasoning. This method thus eliminates the need for explicit mental state annotations. After training, MindZero internalizes model-based reasoning into fast single-pass inference. We evaluate MindZero against baselines across challenging mental reasoning and AI assistance tasks in gridworld and household domains. We found that LLMs alone are insufficient; model-based methods improve accuracy but are slow, costly, and limited by backbone MLLM capacity. In contrast, MindZero enhances MLLMs’ intrinsic ToM ability and significantly outperforms model-based methods in both accuracy and efficiency, showing that mental reasoning can be effectively learned as a self-supervised skill.

---

87. CAST: Non-Privileged Clipped Asymmetric Self-Teaching with Advantage Flipping for GRPO

• Authors: Yang Li , Gongle Xue , Yijia Guo , Yuheng Yuan , Liwen Hu , Lei Ma
• URL: https://arxiv.org/abs/2606.00172
• Abstract:

  Reinforcement learning with verifiable rewards (RLVR), especially Group Relative Policy Optimization (GRPO), has been widely used to improve reasoning in large language models. However, outcome-level rewards provide only sparse supervision, and group-relative advantages vanish when all sampled trajectories for a prompt are either correct or incorrect. On-Policy Self-Distillation (OPSD) offers dense token-level guidance, but its token preferences are not necessarily aligned with trajectory correctness; empirical diagnostics show that OPSD signals behave differently on correct and incorrect rollouts, with teacher-positive and teacher-negative gap signals exhibiting different noise profiles. These diagnostics are conducted under an OPSD-style privileged teacher context for analysis only, whereas CAST training uses answer-free self-teacher this http URL by these observations, this work proposes CAST, an answer-free self-distillation method for GRPO-style RLVR. CAST keeps the verifier-grounded GRPO objective, but uses a stop-gradient self-teacher to shape token-level advantages according to trajectory correctness. Unlike prior self-distilled RLVR methods, CAST does not require reference-solution-conditioned teacher scoring, keeps the self-teacher log-probability gap active throughout training, and applies bidirectional local advantage sign reversal: teacher-negative tokens in correct trajectories can receive negative token-level advantages, while teacher-positive tokens in incorrect trajectories can receive bounded positive local advantages. For zero-variance all-correct and all-wrong groups, CAST assigns bounded sign-constrained base advantages, so these otherwise zero-gradient groups can contribute verifier-signed token feedback. Experiments on mathematical reasoning show that CAST improves RLVR training while retaining a lightweight, verifier-grounded trajectory-level objective.

---

88. A Multi-AI-agent Framework Enabling End-to-end Finite Element Analysis for Solid Mechanics Problems

• Authors: Titu Ranjan Sarker , Muhammed Jawaad Zulqernine , Ling Yue , Shaowu Pan , Chenxi Wang , Shiyao Lin
• URL: https://arxiv.org/abs/2606.00138
• Abstract:

  Finite element analysis (FEA) is the most important numerical approach for solid mechanics. Challenges of FEA include a steep learning curve for entry-level users and potential false simulations due to incorrect definitions of key simulation components, such as boundary conditions, load cases, and solution variables. Years of engineering experience are usually necessary for real-world problem-solving. To address these issues, we present AbaqusAgent, a multi-agent framework grounded in large language models (LLMs) for solid mechanics analyses. AbaqusAgent is developed to facilitate analysis case generation and execution using Abaqus, one of the most widely used FEA packages, by turning users’ natural-language instructions into executed FEA analyses and result visualization. AbaqusAgent is composed of six agents, including interpreter, architect, input writer, runner, reviewer, and visualizer agents, encompassing all the essential pre-processing and post-processing steps of standard FEA analyses. A wide variety of 50 solid mechanics problems have been successfully validated, achieving an overall success rate of 86%. Beyond improving the efficiency of FEA for solid mechanics problems and lowering the barrier to computational mechanics education, AbaqusAgent advances the human-simulation interaction paradigm and enables integration with AI-empowered optimization and material characterization workflows. The code is available at this https URL

---

89. Evaluating Interactive Reasoning in Large Language Models: A Hierarchical Benchmark with Executable Games

• Authors: Mingyuan Fan , Weiguang Han , Daixin Wang , Cen Chen , Zhiqiang Zhang , Jun Zhou
• URL: https://arxiv.org/abs/2606.00103
• Abstract:

  We introduce a multi-turn interactive framework for reasoning evaluation that treats reasoning as active evidence acquisition and belief updating. Wherein, LLMs receive only the task rules, must issue targeted queries to a hidden environment, integrate partial observations over time, and decide when to submit a final answer. Beyond standard success rate and interaction efficiency, we evaluate contextual robustness under controlled contextual perturbations, and metacognitive adaptation through counterfactual revision and necessity judgment. We instantiate the framework as a benchmark of 474 executable games, each evaluated under five fixed configuration search spaces corresponding to five difficulty levels, and evaluate a broad set of frontier LLMs. Results show that the benchmark is highly discriminative, exposing large differences not only in success rate but also in interaction efficiency. Moreover, we empirically show that contextual perturbations cause moderate but consistent declines, whereas counterfactual revision and necessity judgment lead to much larger drops.

---

90. Grokers: Bottom-Up Inductive Comprehension and Write-Time Intelligence over Typed Knowledge Graphs

• Authors: Gregory Magarshak
• URL: https://arxiv.org/abs/2606.00050
• Abstract:

  We present Grokers, an architecture for building persistent, structured comprehension of typed knowledge graphs through bottom-up inductive traversal of dependency subgraphs. Unlike retrieval-augmented generation (RAG), which pays full comprehension cost at every query, Grokers pushes intelligence to write time: autonomous Groker agents analyze nodes in a typed stream graph, extract structured attributes via governed language model (LM) calls, and inductively compose that understanding upward through dependency relations, writing enriched typed attributes that serve all future queries at zero additional LM cost. We prove three formal properties: (1) the Byte-Identity Theorem, establishing that context blocks assembled from a transactionally-maintained denormalization index are byte-identical across LM turns between semantic changes, enabling KV-cache hit rates approaching 100%; (2) the Accumulation Monotonicity Theorem, establishing that the fraction of interactions resolved without LM calls is non-decreasing in the number of completed interactions under a governed wisdom library growth protocol; and (3) the Dual-Traversal Ordering Theorem, establishing that top-down generation and bottom-up comprehension are the unique correct traversal orderings for their respective tasks over a dependency DAG, and that their composition closes into a complete generation-comprehension cycle. We further present a deterministic alternative to embedding-based semantic search, with a synonym caching protocol whose LM fallback rate converges to zero for finite-vocabulary domains. A reference implementation is provided in the open-source Qbix / Safebox / Safebots stack.

---

91. MindGames Arena Generalization Track: In2AI Solution with Delayed Per-Step Reward Attribution

• Authors: Aliaksei Korshuk , Alexander Buyantuev , Ilya Makarov
• URL: https://arxiv.org/abs/2606.00017
• Abstract:

  Training language model agents for multi-agent strategic interaction presents a core difficulty: the quality of any action may depend on future events that never materialize, on moves that violate game rules, or on decisions made by other players. Standard reinforcement learning assumes that rewards can be assigned at each step, but this assumption fails in settings where outcomes are entangled across time and agents. We introduce delayed per-step reward attribution with eligibility gating, an episode lifecycle and postprocessing pipeline that computes rewards only at episode end, propagates them back to originating steps according to task-specific semantics, and excludes steps that lack valid dependent information from training. Together with asynchronous rollout generation via vLLM’s continuous batching, curriculum-based opponent sampling, and multi-level stratified batch construction, this approach enables stable, sample-efficient RL training in multi-agent environments. We evaluate on the MindGames Arena benchmark at NeurIPS 2025, where a single 8-billion-parameter open-source model trained with our method matched or surpassed substantially larger proprietary systems, including GPT-5, in head-to-head play and took first place in both the Open (unrestricted) and Efficient (<=8B parameters) tracks.

---

92. Emergent Collaborative Deliberation in Multi-Model AI Systems: A BFT-Derived Protocol for Epistemic Synthesis

• Authors: VD Doske
• URL: https://arxiv.org/abs/2606.00005
• Abstract:

  We present the Consilium Protocol, a Byzantine Fault Tolerance-derived architecture for structured multi-model AI deliberation that treats inter-model disagreement as epistemic signal rather than error. The protocol assigns engineered cognitive personas to language models – separating what a model is from how it reasons – and introduces an In-Sample/Out-of-Sample validation framework adapted from quantitative finance to distinguish training-data consensus from empirically grounded conclusions. Across 1,478 deliberation sessions spanning 32 topics in 10 domain categories, we demonstrate that (1) the cognitive persona, not the underlying model, determines epistemic behavior: free edge-inference models costing 0.0002 USD per batch produced comparable analytical output to frontier models costing 10.69 USD; (2) RLHF alignment training creates measurable, domain-specific epistemic blind spots – contested policy topics exhibit 12.3 percentage points less adversarial challenge than settled science topics, and AI safety topics show asymmetric bias ($\Delta$=11.6%) where models challenge claims that AI is dangerous far more vigorously than claims that AI risk is overstated; (3) the protocol exhibits no directional bias of its own (immigration $\Delta$=2.3%, renewables $\Delta$=1.2%); and (4) out-of-sample evidence retrieval validated 239 claims with 100% evidence retrieval and surfaced 167 blind-spot discoveries invisible to training-data deliberation. Run-to-run reproducibility across randomized model$\times$persona assignments averages $\pm$2.2% standard deviation. Total cost for the complete battery including all overhead: 217 USD. We release the protocol specification under MIT license to enable independent verification.

---

93. Mitigating Perceptual Judgment Bias in Multimodal LLM-as-a-Judge via Perceptual Perturbation and Reward Modeling

• Authors: Seojeong Park , Jiho Choi , Junyong Kang , Seonho Lee , Jaeyo Shin , Hyunjung Shim
• URL: https://arxiv.org/abs/2606.02578
• Abstract:

  Recent multimodal large language models have demonstrated strong reasoning ability, yet their reliability as automated evaluators remains limited by a critical weakness: when visual evidence conflicts with textual cues, MLLM judges tend to reward plausible narratives over perceptually correct answers. We identify and systematically analyze this phenomenon, which we term Perceptual Judgment Bias. Through controlled visual perturbations, existing multimodal judges frequently anchor on the response text instead of their own visual perception, leading to inconsistent and non-verifiable evaluations. To address this issue, we introduce the Perceptually Perturbed Judgment Dataset, which constructs minimally edited counterfactual responses that isolate perceptual errors and enable verifiable supervision. Building on this dataset, we develop a unified training framework that combines a structured GRPO-based reward with a batch-ranking objective, achieving coherent global ordering without explicit pairwise labels. Experiments across diverse MLLM-as-a-Judge benchmarks show that our approach substantially improves perceptual fidelity, ranking coherence, and alignment with human evaluation. Our results establish a scalable and generalizable pathway for training multimodal judges that are perceptually grounded, interpretable, and robust to visual-reasoning conflicts.

---

94. AdaCodec: A Predictive Visual Code for Video MLLMs

• Authors: Haowen Hou , Zhen Huang , Zheming Liang , Qingyi Si , Chenglin Li , Shuai Dong , Kele Shao , Ruilin Li , Dianyi Wang , Nan Duan , Jiaqi Wang
• URL: https://arxiv.org/abs/2606.02569
• Abstract:

  Video is temporally redundant: adjacent frames usually share most objects, background, and layout. Yet existing video multimodal large language models (video MLLMs) usually encode each sampled frame as an independent RGB image, causing visual tokens to repeat content already present in earlier frames. This suggests a more direct video interface: send a full reference frame only when the scene cannot be predicted well from prior context, and otherwise transmit a compact description of inter-frame changes. We call this interface a \emph{predictive visual code}, and instantiate it for video MLLMs as \textbf{AdaCodec}. AdaCodec spends full visual tokens on a reference frame only when its conditional predictive cost is high; otherwise, it encodes inter-frame changes, including motion and prediction residuals, as compact P-tokens. Across all eleven benchmarks, AdaCodec improves over the Qwen3-VL-8B per-frame RGB baseline at a matched visual-token budget. Even at $1/7$ the budget, AdaCodec with 32k tokens surpasses the 224k baseline on all long-video benchmarks; on five general-video benchmarks, it raises the average score while substantially cutting time-to-first-token from 9.26s to 1.62s.

---

95. From Layers to Submodules: Rethinking Granularity in Replacement-Based LLM Compression

• Authors: Elia Cunegatti , Marcus Vukojevic , Erik Nielsen , Giovanni Iacca
• URL: https://arxiv.org/abs/2606.02559
• Abstract:

  Post-training compression of Large Language Models (LLMs) removes entire architectural components, either deleting them or replacing them with fitted modules. Existing replacement-based methods share two design constraints: full-layer granularity and contiguous selection. We argue that this is overly restrictive: in fact, redundancy in pretrained transformers is not confined to contiguous regions, nor does it evenly distribute between Attention and FeedForward outputs, implying that different strategies best approximate different submodule types and that removable components need not cluster within contiguous depth ranges. Based on this intuition, we introduce SubFit (Submodule-level Fitted residual replacement), which compresses LLMs at the submodule level: Attention and FeedForward submodules are selected non-contiguously, and each receives its own lightweight fitted residual bypass. SubFit operates post-training and requires only calibration data. Across ten LLMs (five base, five instruction-tuned), five sparsity levels from 12.5% to 37.5%, and four replacement-based baselines, SubFit achieves the best aggregate perplexity-accuracy trade-off across the evaluated sparsity levels, with larger gains under aggressive compression. At 25% sparsity, it retains 84.6% of dense downstream accuracy and incurs 2.42x perplexity degradation, against 81.6% and 4.34x for the strongest baselines, while delivering measurable inference speedup and KV-cache savings. Code is available at this https URL .

---

96. SimSD: Simple Speculative Decoding in Diffusion Language Models

• Authors: Junxia Cui , Haotian Ye , Runchu Tian , Hongcan Guo , Jinya Jiang , Haoru Li , Chaojie Ren , Yiming Huang , Kaijie Zhu , Zhongkai Yu , Kun Zhou , Jingbo Shang
• URL: https://arxiv.org/abs/2606.02544
• Abstract:

  Diffusion large language models (dLLMs) have recently emerged as a promising alternative to autoregressive (AR) LLMs, offering faster inference through parallel or blockwise decoding. However, their masked language modeling formulation remains incompatible with standard token-level speculative decoding, one of the most effective acceleration techniques for AR models. In AR decoding, the causal mask preserves temporally valid token-level contexts, enabling a target model to verify multiple drafted tokens in a single forward pass. In contrast, dLLMs rely on mask tokens and bidirectional attention, causing the effective context to change across denoising steps and preventing direct token-level speculative verification. To bridge this gap, we propose a simple but effective speculative decoding algorithm for diffusion language models, named SimSD, which mainly adopts a plug-and-play masking strategy that equips dLLMs with temporally valid token-level contexts for speculative decoding. Our method explicitly introduces reference tokens from draft-model predictions and designs an attention mask that regulates their interaction with current-step tokens, allowing dLLMs to compute valid logits for drafted tokens in a single forward pass. This restores the key verification ability provided by causal masking in AR models while preserving the parallel decoding advantages of dLLMs. The proposed method is training-free and can be flexibly integrated with other acceleration techniques such as KV cache and blockwise decoding. Experiments on SDAR-family dLLMs across four benchmarks show that our method achieves up to 7.46x higher decoding throughput while maintaining and even improving average generation quality.

---

97. Moment-Video: Diagnosing Temporal Fidelity of Video MLLMs on Momentary Visual Events

• Authors: Xiaolin Liu , Yilun Zhu , Xiangyu Zhao , Xuehui Wang , Yan Li , Xin Li , Haoyu Cao , Xing Sun , Shaofeng Zhang , Xu Yang , Zhihang Zhong , Xue Yang
• URL: https://arxiv.org/abs/2606.02522
• Abstract:

  Video multimodal large language models (MLLMs) have made rapid progress on general and long-form video understanding, yet their ability to preserve brief answer-critical visual evidence remains underexplored. Many practical questions are determined by momentary visual events: localized actions or state transitions that may last only a few frames. Such evidence can be skipped by sparse frame sampling, suppressed by visual-token compression, or diluted by coarse temporal aggregation, causing failures that language-side reasoning cannot reliably recover. We introduce Moment-Video, a benchmark for diagnosing the temporal fidelity of video MLLMs through momentary visual event understanding. Each question is grounded in a localized, visually observable, and sampling-sensitive event, requiring models to notice, count, describe, or reason about transient evidence rather than rely on persistent objects, global scene context, or language priors. Moment-Video contains 1,000 human-verified video-QA pairs across 7 domains and 25 fine-grained subcategories, covering four task types: Temporal Occurrence, Temporal Counting, Action Description, and Temporal Reasoning. We evaluate 33 proprietary and open-source MLLMs on Moment-Video. The best-performing model, Seed-2.0-Pro, achieves only 39.6% overall accuracy, while most open-source models remain below 25%, revealing a substantial gap in momentary visual event understanding. Diagnostic analyses show that denser frame sampling improves some models but does not eliminate the bottleneck, and longer videos introduce stronger temporal-localization challenges. These findings suggest that current video MLLMs still lack temporally faithful representations for capturing, preserving, and using brief but decisive visual evidence.

---

98. Learning When to Translate for Multilingual Reasoning

• Authors: Deokhyung Kang , Hyounghun Kim , Gary Geunbae Lee
• URL: https://arxiv.org/abs/2606.02465
• Abstract:

  Reasoning language models (RLMs) achieve strong performance on complex reasoning tasks, but still exhibit substantial multilingual reasoning gaps, largely due to language-understanding failures in non-English inputs. English translation can mitigate these failures by expressing non-English inputs in a form that RLMs can more reliably interpret, yet translating every input is unnecessary when the model can reason reliably from the original query. To address this challenge, we propose Luar, a Language Understanding Boundary-aware Reinforcement Learning framework that trains RLMs to selectively invoke translation when direct understanding is unreliable. Luar trains the model to choose between solving the original input directly and reasoning over its English translation, encouraging translation only when translator-augmented reasoning is expected to substantially outperform direct reasoning. Across multilingual reasoning benchmarks, Luar outperforms standard GRPO and other training-based baselines, with particularly large gains on low-resource languages. Further analysis shows that Luar avoids unnecessary translation in cases where direct reasoning is sufficient, while extending its translator-call behavior to unseen low-resource languages. Together, our work suggests a selective approach to multilingual reasoning: RLMs can learn to invoke translation only when their direct understanding is unreliable. The project will be made publicly available at this https URL

---

99. MASER: Modality-Adaptive Specialist Routing for Embodied 3D Spatial Intelligence

• Authors: Hilton Raj , Vishnuram AV
• URL: https://arxiv.org/abs/2606.02463
• Abstract:

  In 3D environments, Embodied Agents answer spatially relevant questions through reasoning from a mixture of modalities including natural language, RGB images, point clouds, depth maps and camera poses. Existing Vision-Language models (VLMs) are fine-tuned over a single modality. This completely ignores the question semantics which may favor a different modality than the finetuned modality. To address this, we propose MASER (Modality-Adaptive SpEcialist Routing), a lightweight framework that trains five different modality adapters of a shared VLM backbone and learns a neural routing policy that selects the best adapter based on the question during inference. We encode each question with a frozen sentence transformer and pass the embedding through a small Multi-layer Perceptron (MLP) trained on oracle adapter-accuracy labels. We evaluate our methodology over the Open3D-VQA benchmark and our evaluations show that no single modality is universally optimal – point-cloud answers are best in 51.5% of cases. MASER routes with 51.3% oracle agreement, outperforming a Random-Forest ablation (43.5%), with only a single adapter call per question.

---

100. PaSBench-Video: A Streaming Video Benchmark for Proactive Safety Warning

• Authors: Yusong Zhao , Yuejin Xie , Youliang Yuan , Junjie Hu , Jitian Guo , Yujiu Yang , Pinjia He
• URL: https://arxiv.org/abs/2606.02443
• Abstract:

  Between the first visible sign of danger and the moment an accident occurs, there is often a window where intervention remains possible. Video-capable multimodal large language models (MLLMs) could serve as always-on safety monitors that issue warnings during this window. Yet current benchmarks do not test this ability: they rely on static inputs, ignore timing precision, and omit false-positive measurement on safe scenes. We present PaSBench-Video, a 740-video benchmark with 481 risk and 259 no-risk videos across four domains: driving, healthcare, daily life, and industrial production. Risk videos are annotated with frame-level risk onset and accident boundaries. A model must observe the video causally and produce a warning that is both temporally calibrated and content-correct. Testing 13 MLLMs, we find that no model exceeds 20.0% on our strictest metric, and recall is tightly coupled with false-positive rate, with Pearson correlation 0.64: higher detection comes only at the cost of triggering warnings on the majority of safe clips. Performance splits sharply by domain: models achieve moderate recall at low false-positive rates in daily life, where risks are inherently anomalous, yet fire indiscriminately in driving, where routine and hazardous scenes look alike. These results indicate that current models rely on scene-level activity cues rather than reasoning about emerging harm.

---

101. ODTQA-FoRe: An Open-Domain Tabular Question Answering Dataset for Future Data Forecasting and Reasoning

• Authors: Zhensheng Wang , Xiaole Liu , Wenmian Yang , Kun Zhou , Yiquan Zhang , Weijia Jia
• URL: https://arxiv.org/abs/2606.02433
• Abstract:

  The rapid development of LLMs has significantly advanced tabular question answering, but most systems cannot perform future-oriented numerical prediction. To address this gap, we introduce a novel task, Open-Domain Tabular Question Answering for Future Data Forecasting and Reasoning, and propose the first dataset to cover time-series forecasting and forecast-based reasoning scenarios using real estate data. This task poses challenges in retrieving precise historical data, overcoming the forecasting limitations of LLMs, and standardizing responses for diverse queries. To solve the above challenges, we propose TimeFore, an LLM agent-based framework that decomposes the problem into three collaborative roles: a Retriever autonomously generates SQL to fetch data, a Forecaster invokes external time-series models for higher accuracy, and an Analyzer synthesizes the results to construct a precise and consistent final answer. Extensive experiments demonstrate the effectiveness of our TimeFore.

---

102. Not All Errors Are Equal: A Systematic Study of Error Propagation in Large Language Model Inference

• Authors: Yafan Huang , Sheng Di , Guanpeng Li
• URL: https://arxiv.org/abs/2606.02430
• Abstract:

  Large language models (LLMs) are increasingly integrated into high-performance computing (HPC) workflows, accelerating scientific discovery through diverse perspectives such as code generation and domain-specific decision-making. Yet, how soft errors propagate and affect LLM inference remains largely unexplored. To bridge this gap, we present a comprehensive study on error propagation in LLM inference, enabled by our proposed LLMFI, a configurable and deterministic fault-injection framework. Using LLMFI, we systematically inject faults across three open-weighted LLMs and thirteen representative tasks, covering reasoning, multilingual, mathematical, and coding domains. In addition, we conduct fine-grained case studies that reveal critical vulnerability patterns. Overall, our study yields 17 takeaways that advance the understanding of error propagation in LLM inference and introduces four low-overhead directions to improve reliability through software-only modification, offering practical guidance for future error detection and mitigation.

---

103. Evolutionary Discovery of Bivariate Bicycle Codes with LLM-Guided Search

• Authors: Juan Cruz-Benito , Andrew W. Cross , David Kremer , Ismael Faro
• URL: https://arxiv.org/abs/2606.02418
• Abstract:

  Quantum LDPC code discovery requires searching large algebraic design spaces while reliably certifying the parameters and equivalence classes of any candidates found. We introduce an LLM-guided evolutionary workflow in which language models mutate Python programs that generate bivariate-bicycle and perturbed bivariate-bicycle code ansätze. Across five campaigns, the system performed approximately 1{,}650 evolutionary iterations, screened about $2 \times 10^5$ candidate codes, and required ${\sim}140$ hours of computation and ${\sim}$US$400 in LLM inference cost. Candidate codes are evaluated through a staged validation pipeline combining $\mathrm{GF}(2)$ rank computation, distance estimation and certification, mixed-integer linear programming, BLISS Tanner-graph deduplication, decomposability analysis, and local-Clifford equivalence checks. At block length $n \leq 360$, the workflow identifies 465 distinct candidate codes: 97 CSS bivariate-bicycle codes and 368 non-CSS perturbed variants. The CSS search recovers known high-performing codes and finds new finite-length representatives, including an indecomposable [[288,16,12]] code and higher-weight codes with up to $k = 50$ at distance $d = 8$. The non-CSS search produces perturbed codes matching the gross-code figure of merit at [[144,12,12]], along with additional high-distance candidates reported as certified values or upper bounds according to MILP status. Overall, these results show that LLM-guided program evolution can serve as a practical tool for structured quantum-code discovery when paired with independent evaluation.

---

104. AutoForest: Automatically Generating Forest Plots from Biomedical Studies with End-to-End Evidence Extraction and Synthesis

• Authors: Massimiliano Pronesti , Angelo Miculescu , Mohsin Kapdi , Paul Flanagan , Oisín Redmond , Joao Bettencourt-Silva , Gurdeep Mannu , Spiros Denaxas , Rui Bebiano Da Providencia E Costa , Anya Belz , Yufang Hou
• URL: https://arxiv.org/abs/2606.02403
• Abstract:

  Systematic reviews rely on forest plots to synthesise quantitative evidence across biomedical studies, but generating them remains a fragmented and labour-intensive process. Researchers must interpret complex clinical texts, manually extract outcome data from trials, define appropriate interventions and comparators, harmonise inconsistent study designs, and carry out meta-analytic computations-typically using specialised software that demands structured inputs and domain expertise. While recent work has demonstrated that large language models can extract study-level data from unstructured text, no existing system automates the complete pipeline from raw documents to synthesised forest plots. To address this gap, we introduce AutoForest, the first end-to-end system that generates publication-ready forest plots directly from biomedical papers. Given one or more study papers, AutoForest automatically suggests ICO (Intervention, Comparator, Outcome) elements, extracts outcome data, performs statistical synthesis, and renders the final forest plot. We describe the system architecture, user interface and demonstrate its effectiveness on real-world examples through a user study involving clinicians, showing how AutoForest can accelerate evidence synthesis and substantially lower the barrier to conducting meta-analyses.

---

105. Policy and World Modeling Co-Training for Language Agents

• Authors: Ning Lu , Baijiong Lin , Shengcai Liu , Jiahao Wu , Haoze Lv , Yanbin Wei , Lingting Zhu , Shengju Qian , Xin Wang , Ying-Cong Chen , Qi Wang , Ke Tang
• URL: https://arxiv.org/abs/2606.02388
• Abstract:

  Reinforcement learning (RL) improves large language model (LLM) agents by teaching them which actions lead to high rewards, but provides little supervision on what those actions do to the environment. World modeling (WM) can fill this gap, yet existing approaches often require separate simulators, extra training stages, or additional inference-time computation. We observe that on-policy RL rollouts already contain the needed signal: each transition pairs an action with its resulting next observation. Based on this observation, we propose PaW, a Policy and World modeling co-training framework that adds auxiliary WM supervision to the same policy during RL, without changing the inference paradigm. To make auxiliary WM supervision informative and stable, PaW introduces three components: action-entropy-based WM data selection, noise-tolerant WM loss, and reward-adaptive loss balancing. Experiments on three agentic task benchmarks show consistent improvements over strong RL baselines across models and RL algorithms. These results suggest that standard RL rollouts are a practical source of WM supervision for language-agent training.

---

106. SPADE-Bench: Evaluating Spontaneous Strategic Deception in Agents via Plan-Action Divergence

• Authors: Yuyan Bu , Haowei Li , Qirui Zheng , Bowen Dong , Kaiyue Yang , Jiaming Ji , Yingshui Tan , Wenxin Li , Yaodong Yang , Juntao Dai
• URL: https://arxiv.org/abs/2606.02380
• Abstract:

  As LLM-based agents expand their operational scope, reliability becomes a prerequisite for real-world deployment. However, in practical applications, human users cannot monitor every immediate behavior; instead, the execution process often remains a black box, leaving users dependent solely on the agent’s self-reported updates. This opacity creates a critical risk: agents may present observer-facing reports that diverge from their executed actions, rendering the system uncontrollable, especially in high-stakes autonomous scenarios. We term such self-reported plan-action divergence as agent deception. To assess this, we introduce SPADE-Bench, a benchmark designed to evaluate spontaneous plan-action divergence. Unlike prior deception benchmarks, SPADE-Bench simultaneously integrates actual tool execution and controlled pressure scenarios. This design ensures ecological validity and rigorously distinguishes strategic deception from mere hallucination through controlled plan-action comparisons under pressure. Experiments across mainstream models confirm that agent deception is a genuine and pressing issue in tool-use contexts. By providing a comprehensive and robust evaluation framework, SPADE-Bench fills a critical gap in agent safety, facilitating the community’s progress toward building trustworthy and controllable autonomous systems.

---

107. When Do Attention Circuits Form? Developmental Trajectories of Capability and Attention-Sink Emergence Across Three 1B-ClassArchitectures

• Authors: Yongzhong Xu
• URL: https://arxiv.org/abs/2606.02378
• Abstract:

  We track the developmental trajectory of attention-head circuit formation across three 1B-class language models spanning two architecture families (dense transformer, mixture-of-experts) and two pretraining corpora (The Pile, DCLM): Pythia 1B, OLMo 1B-0724-hf, and OLMoE 1B-7B-0924. At each of 10 log-spaced revisions per model – 30 mechanistic-interpretability runs in total – we apply a participation-ratio (PR) spectral signal and an all-head capability-specific selectivity screen to track induction, previous-token, and BOS-attractor heads as they emerge. Five findings. (F1) Layers 0 and 1 produce zero BOS-classified heads at every revision in every model: the L0/L1 zero-BOS floor is an architectural property, not a learned outcome. (F2) The whole-model BOS-attractor fraction follows three distinct emergence shapes – a gradual ramp in Pythia 1B, a sharp phase transition in OLMo 1B (7% to 70% between adjacent checkpoints), and a gradual ramp in OLMoE 1B-7B. (F3) In DCLM models, induction-circuit formation precedes BOS-attractor formation by 10-20x in tokens; capability-circuit formation and attention-sink formation are two transitions, not one. (F4) The capability-specific screen converges to the final induction circuit within 0.3-2% of total training tokens – circuit identification does not require the final model. (F5) For every final-checkpoint induction head sampled across all three models, per-head PR is elevated at or before the first revision at which that head crosses its capability-selectivity threshold. The results refine the induction-phase-transition framing: in 1B-class models trained on DCLM, the induction transition and the attention-sink transition are separated by an order of magnitude in tokens and have qualitatively different shapes.

---

108. Repurposing Adversarial Perturbations for Continual Learning: From Defense to Active Alignment

• Authors: Ran Liu , Min Yu , Mingqi Liu , Jianguo Jiang , Gang Li , Rongsheng Li , Ning Li , Zhen Xu , Weiqing Huang , Ming Liu
• URL: https://arxiv.org/abs/2606.02322
• Abstract:

  In dynamic environments, large language models need to keep adapting to new tasks, but continual learning often suffers from forgetting, limited transfer, and vulnerability to adversarial perturbations. To address this, we present AdvCL, which repurposes adversarial perturbations as a geometric control signal for stable continual adaptation. AdvCL combines three plug-in modules: Intra-Smooth promotes local smoothness via small adversarial perturbations; Proto-Clip uses similarity clipping to prevent excessive alignment to current task prototype; and Inter-Align applies directional alignment toward previous task prototype to reduce representational gaps. Experiments show consistent gains in both standard performance and robustness, with lower forgetting and stronger transfer. We further analyze key mechanisms by quantifying the sensitivity of Intra-Smooth to perturbation settings and the effect of Inter-Align on task similarity and geometric distance. In summary, the modules provide complementary gains when combined, and each can also be integrated individually into diverse CL paradigms, including replay, regularization, and dynamic architectures, thereby offering a geometric control mechanism for continual learning.

---

109. SeClaw: Spec-Driven Security Task Synthesis for Evaluating Autonomous Agents

• Authors: Hao Cheng , Changtao Miao , Tianle Song , Yin Wu , He Liu , Erjia Xiao , Junchi Chen , Xiaoyu Shi , Yichi Wang , Jing Yang , Taowen Wang , Jinhao Duan , Mengshu Sun , Peiyan Dong , Xuan Shen , Yang Cao , Renjing Xu , Kaidi Xu , Jindong Gu , Bo Zhang , Jize Zhang , Chenhao Lin , Philip Torr , Chao Shen
• URL: https://arxiv.org/abs/2606.02302
• Abstract:

  Autonomous LLM agents increasingly operate in stateful environments where they access tools, files, memory, and external services. While such capabilities enable complex real-world workflows, they also introduce security risks that are difficult to capture with existing evaluations. Current agent security benchmarks often rely on manually curated tasks, provide limited coverage of emerging threats, and focus primarily on final outcomes rather than the execution processes that lead to unsafe behavior. We introduce SeClaw, a framework that combines specification-driven security task synthesis with execution-based security evaluation for Autonomous agents. Spec-driven security task synthesis enables scalable and controllable construction of security tasks from structured risk specifications, while SeClaw docker provides a standardized testbed for evaluating agent behavior under diverse safety-risk scenarios. The benchmark covers risks arising from resources, user tasks, environments, and intrinsic agent behaviors, and supports trajectory-aware assessment of unsafe actions beyond final responses. By bridging systematic task synthesis and reproducible security evaluation, SeClaw provides a practical foundation for measuring, diagnosing, and comparing security failures in autonomous LLM agents. The code is available at this https URL .

---

110. Cross-modal linkage risk in clinical vision-language models

• Authors: Soroosh Tayebi Arasteh , Mahshad Lotfinia , Sven Nebelung , Daniel Truhn
• URL: https://arxiv.org/abs/2606.02276
• Abstract:

  Vision-language models (VLMs) trained on paired chest radiographs and radiology reports learn a shared embedding space that can preserve instance-level image-report correspondence. This poses a privacy risk in settings where radiographs and reports are deliberately kept separate after acquisition, such as image-only data sharing or access-controlled reports, because a de-identified image may be re-linked to its original narrative report through cosine similarity alone. We formalized this as image-to-report retrieval and used public paired cohorts, in which the true pairing is known by design, as ground-truth benchmarks to audit the risk rather than as the privacy scenario. Evaluating VLMs of increasing clinical specialization on 406,241 paired examples from 126,804 patients across MIMIC-CXR (43,793 held-out pairs) and external CheXpert Plus (29,296 pairs), we found that re-linkage rose systematically with specialization: the strongest VLM retrieved the correct report at 15 times chance at a candidate pool of N = 100, 50 times chance at N = 10,000, and well above chance at full-database scale. The signal persisted under pathology-matched hard negatives that removed disease-label shortcuts, indicating correspondence beyond broad diagnostic categories. To reduce it without retraining, we froze both encoders and applied differentially private optimization only to the projection heads defining the alignment layer (epsilon = 0.34, delta = 6x10-6). This reduced Recall@1 by 61.8% at N = 10,000 on MIMIC-CXR and transferred to CheXpert Plus without retraining, while image-side utility was largely preserved: macro AUROC for linear-probe classification across 14 labels shifted only from 79.63% to 79.43%. Targeted DP finetuning of the shared alignment layer can substantially reduce cross-modal re-linkage without materially degrading the image representations that make these models clinically useful.

---

111. Who Annotates in NLP? A Large-scale Assessment of Human Annotation Reporting between 2018 and 2025

• Authors: Maria Kunilovskaya , Gagan Bhatia , Lisa Sophie Albertelli , Yanran Chen , Christian Greisinger , Lotta Kiefer , Christoph Leiter , Subhadeep Roy , Tewodros Achamaleh , Muhammad Arslan Manzoor , Sebastian Pohl , Yufang Hou , Steffen Eger
• URL: https://arxiv.org/abs/2606.02255
• Abstract:

  Human annotation is the empirical foundation of much NLP research, from dataset construction to model evaluation, but papers often leave unclear who produced the annotations and how the annotation process was controlled. We provide the first large-scale, task-level audit of human annotation reporting across major NLP venues, asking which annotation details are documented, which are missing, and how reporting varies across time, topic, venue, and intended use of human judgment. We introduce a unified taxonomy of annotation-reporting practices and validate an LLM-assisted extraction pipeline against Annotated-gold, a human-adjudicated gold standard of 41 papers and 72 annotation tasks, where the best model reaches human-comparable agreement with adjudicated labels, with Krippendorff’s alpha of 0.606 versus 0.585 for human-human agreement. Using this pipeline, we construct Annotated-llm, a dataset covering ACL-venue papers from 2018-2025, with 2,667 extracted annotation tasks from 1,603 papers, and find that papers frequently report operational details such as recruitment strategies, annotator expertise, and annotation volume, but often omit details needed to assess annotation validity, including training, language proficiency, compensation, socio-demographics, adjudication, and agreement values, especially in model-evaluation studies. Our results show that annotation reporting in NLP has improved over time but remains uneven, and they establish a scalable framework and bare-minimum reporting recommendations for making human annotation more reliable, reproducible, and interpretable.

---

112. AgentRedBench: Dynamic Redteaming and Integration-Aware Defense for LLM Agents over SaaS Integrations

• Authors: Hiskias Dingeto , Will Leeney
• URL: https://arxiv.org/abs/2606.02240
• Abstract:

  Indirect prompt injection in tool-use agents is a concrete production threat: LLM agents read from integrations (third-party services such as Gmail, Salesforce, or Jira accessed through tool calls) whose response content the user neither writes nor controls. Existing benchmarks under-measure the threat: most cover only a handful of integrations with the same attack payload replayed across runs, and open-source guards are trained on chat-style data rather than tool-response content. We introduce AGENTREDBENCH, a dynamic LLM-driven redteaming benchmark of 215 subtle underspecified authorization (attacks at the boundary of what the user’s request authorises) scenarios across 24 enterprise integrations in nine functional families and five attack types. Across an eight-model panel (Anthropic, OpenAI, Google), no-guard ASR (attack success rate) ranges from 32% (Claude Sonnet 4.6) to 81% (Gemini 3 Flash). To keep the scenario set out of training corpora and preserve headline ASR meaning over time, we release the codebase, integration schemas, and AGENTREDGUARD model openly; the canonical scenarios are evaluated through a maintainer-mediated channel with immutable versioning. We release AGENTREDGUARD alongside the benchmark: a guard trained on an integration-diverse corpus of adversarial tool-response content. AGENTREDGUARD cuts panel ASR from 69.9% to 2.4% at 0.37% false-positive rate, outperforming every open-source baseline with non-trivial detection (Llama Guard, PromptGuard 2, ProtectAI) on both axes. Cross-integration and cross-attack type holdouts both confirm the gain transfers beyond the training subset.

---

113. Consistency Training while Mitigating Obfuscation via Rate Matching

• Authors: Sohaib Imran , Prakhar Gupta , Jannes Elstner , David Demitri Africa
• URL: https://arxiv.org/abs/2606.02211
• Abstract:

  Large language models are often influenced by extraneous input features, such as cues revealing a user’s preferred answer. Consistency training reduces this influence by training models to behave similarly across inputs with and without the extraneous feature. However, existing methods train for consistency over entire responses or internal activations, which also constrains whether the model verbalises said extraneous features. We show this leads to obfuscation, where the model learns not to mention a cue while remaining influenced by it, which may undermine monitorability. To address this, we introduce Rate Matching Consistency Training (RMCT), which trains for consistency over selected behavioural properties without constraining how this behaviour is expressed. RMCT matches the rate at which the model exhibits a target behaviour (e.g., following a bias cue) across input perturbations, rather than requiring paired inputs with and without the extraneous feature, extending consistency training to settings where the extraneous features cannot be removed. We evaluate RMCT on sycophancy reduction in two open-weight language models, achieving reductions in bias-following comparable to a standard consistency-training baseline on held-out bias types, while largely preserving the model’s tendency to verbalise the bias cue. Further, we find that RMCT is more data-efficient at the expense of being less compute-efficient in our experiments. Overall, RMCT shows that consistency training can improve behavioural robustness without directly trading off against monitorability.

---

114. Multimodal Approaches for Visually-Rich Document Type Classification: A Comparative Analysis

• Authors: Catyana Heyne , Jürgen Frikel , Filippo Riccio
• URL: https://arxiv.org/abs/2606.02162
• Abstract:

  Document type classification in visually rich documents remains challenging, as relevant information is distributed across textual, visual, and layout modalities. To capture this complexity, current approaches rely on diverse multimodal modeling strategies, resulting in heterogeneous architectures that complicate systematic comparison. This variability is also reflected in existing comparative studies, which often rely on heterogeneous evaluation setups, further complicating systematic comparison and making it difficult to assess progress. To address these limitations, this work provides a structured analysis of multimodal design strategies across transformer- and LLM-based architectures, combined with a controlled empirical comparison within a unified experimental framework. Specifically, four representative models (LayoutLMv3, Donut, Qwen3-VL-32B-Instruct, and Qwen3-32B) are evaluated on the RVL-CDIP benchmark to systematically analyze the contributions of text, image, and layout information for document type classification, with a particular focus on contrasting OCR-dependent and OCR-free approaches. The results show that specialized multimodal Transformers outperform LLM-based approaches on visually rich and layout-intensive documents. Image information contributes most strongly to reliable classification, while OCR-derived text provides useful but secondary support. These findings highlight that multimodal processing remains essential for documents with pronounced layout structure. Overall, the study provides a systematic basis for comparing multimodal architectures and offers practical guidance for selecting effective feature combinations and model designs for document type classification.

---

115. Jailbreaking Multimodal Large Language Models using Multi-Clip Video

• Authors: Choongwon Kang , Seungjong Sun , Hyunmin Jun , Jang Hyun Kim
• URL: https://arxiv.org/abs/2606.02111
• Abstract:

  As multimodal large language models (MLLMs) have advanced to process video inputs, concerns have emerged about their potential for malicious misuse. Prior jailbreak studies have shown that safety alignment in MLLMs can be bypassed through visual inputs, yet it remains unclear which properties of video inputs induce this vulnerability. To address this gap, we introduce Multi-Clip Video (MCV) SafetyBench, a dataset of 2,920 videos designed to evaluate how the diversity of video inputs affects the vulnerability of MLLMs. Each video consists of multiple short clips depicting diverse contexts related to a harmful query. Experiments on eight representative video MLLMs show that attack success consistently increases with the number of clips. Our results further indicate that the video modality is (1) more vulnerable than the image modality, (2) more vulnerable to dynamic videos than to static videos, and (3) more vulnerable when videos contain more diverse contexts. Building on these findings, we propose a defense strategy that leverages the relative robustness of the image modality.

---

116. The Role of Ambiguity in Error Prediction via Uncertainty Quantification

• Authors: Ieva Raminta Staliūnaitė , James Bishop , Andreas Vlachos
• URL: https://arxiv.org/abs/2606.02093
• Abstract:

  The task of Error Prediction, namely predicting whether a model output is correct, is commonly tackled with Uncertainty Quantification (UQ). However, while uncertainty metrics capture when models lack knowledge or capacity to make a prediction, they also reflect aleatoric uncertainty, which is inherent in the model input and context. This paper presents a method for improving error prediction for Large Language Models (LLMs), by disentangling input ambiguity from UQ signal. We conduct experiments on the task of Question Answering (QA) with six UQ metrics and show that UQ metrics are more predictive of errors on unambiguous instances than on questions with multiple plausible answers. We use Gated Experts and Selective Prediction to incorporate gold and predicted ambiguity labels into the error prediction pipeline. We find that ambiguity information improves error prediction scores across model families, training and evaluation paradigms, datasets (including allegedly unambiguous ones), and sources of aleatoric uncertainty, yielding improvements of over 10 points of PRR for individual UQ metrics on standard datasets.

---

117. PlanarBench: Evaluating LLM Spatial Reasoning via Planar Graph Drawing

• Authors: Oleksandr Nikitin
• URL: https://arxiv.org/abs/2606.02010
• Abstract:

  PlanarBench tests whether LLMs can draw planar graphs as ASCII art given only an edge list – a spatial reasoning task that resists memorization because edge order, edge orientation, and node labels are all permutable. We evaluate 91 models on the 199 simplest non-isomorphic connected planar graphs (2 - 7 vertices). Edge count is the dominant difficulty predictor ($r = -0.85$) – a finding not reported in prior LLM graph benchmarks, which use only node count as the difficulty axis.

---

118. MMG2Skill: Can Agents Distill In-the-Wild Guides into Self-Evolving Skills?

• Authors: Xinyu Che , Junqi Xiong , Yunfei Ge , Xinping Lei , Shihao Li , Hang Yan , Han Li , Yuanxing Zhang , Zhiqi Bai , Jinhua Hao , Ming Sun , Han Li , Jiaheng Liu
• URL: https://arxiv.org/abs/2606.01993
• Abstract:

  Abundant procedural knowledge on the Web holds great potential for helping agents solve long-horizon tasks. However, such knowledge is often multimodal, heterogeneous, noisy, and implicitly assumes human executors, making it difficult to use directly as the skills required by agents. To bridge the gap between human-oriented guides and agent-executable skills, we formalize this problem as guide-to-skill learning: converting in-the-wild guides into executable skills and continuously improving them from trajectories observable to the agent. To evaluate the capability of existing agents on this task, we introduce MMG2Skill-Bench, the first benchmark designed for this problem. We further propose MMG2Skill, a closed-loop framework that compiles guides into editable skills, conditions a fixed vision-language model (VLM) agent on these skills during execution, and revises the skills from trajectory-level root-cause feedback without using benchmark scores. Across GUI control, open-ended gameplay, and strategic card play with six VLM backbones, MMG2Skill consistently outperforms vanilla baseline agents in every model-domain setting, achieving macro-average gains of +12.8 to +25.3 percentage points across backbones. Ablation studies show that directly prompting agents with raw guides can degrade performance, while both structured skill construction and trajectory-driven revision are necessary for the observed improvements. On success-inferable tasks, analyzer-based early stopping further prevents late-stage performance regressions and saves 25%-53% of attempts when the success signal is properly calibrated.

---

119. A Structured Benchmark for Text-Guided Anomaly Detection: When Language Stops Conditioning the Decision

• Authors: Stefano Samele , Eugenio Lomurno , Teodora Jovanovic , Sanjay Shivakumar Manohar , Alberto Crivellaro , Matteo Matteucci
• URL: https://arxiv.org/abs/2606.01992
• Abstract:

  Industrial anomaly detection has historically been a unimodal task. Recent multimodal vision-language models have produced systems that admit textual input alongside the image and are presented as enabling text-guided zero- and few-shot inspection. Yet these methods are evaluated with protocols inherited from unimodal benchmarks that hold the textual condition constant and therefore cannot measure whether language conditions the decision; whether reported gains reflect text guidance or strong pretrained visual features remains open. We introduce Text-Guided Anomaly Detection (TGAD), a structured benchmark that progressively increases the functional role of language across three scenarios: a controlled prompt-sensitivity setting on MVTec AD; a component-tagged extension of MVTec AD that requires the model to restrict its assessment to an instructed part; and the new Assembled Panel Dataset (APD), a realistic industrial setting that requires both defect-type and component-location knowledge. We evaluate one representative model per paradigm: generative large vision-language, training-free discriminative, and embedding-adaptive discriminative. In all three, the textual interface conditions the decision only superficially: prompt content is absorbed unless the object noun is removed (the generative model’s I-AUROC drops from 97.4 to 82.6); component-level instructions do not constrain the decision once defects outside the instructed part are admitted as normal (from 90.3 to 66.3); and when both combine on APD, image-level discrimination collapses below the MVTec level, in one case below chance (71.2, 50.5, 31.5). These results suggest that standard benchmarks overstate the text-guided capabilities of current multimodal anomaly detection systems, and that a protocol of this kind is a prerequisite for models that can be reliably controlled through language for industrial deployment.

---

120. KliniskVestBERT: BERT Model Specialised to Norwegian Clinical Texts

• Authors: Christian Autenried , Cosimo Persia
• URL: https://arxiv.org/abs/2606.01904
• Abstract:

  The increasing application of Natural Language Processing (NLP) in healthcare demands language models specifically attuned to the complexities of clinical language. This work introduces KliniskVestBERT, a suite of three BERT-based encoder models pre-trained on a substantial corpus of real-world, de-identified Norwegian clinical texts from Helse Vest. We continue pretraining existing language models Nb-BERT-large, NorBERT3-large, and ModernBERT on our specialized clinical dataset. This dataset is based on a representative population of Helse Vest patients. The included document types are carefully curated to encompass a broad clinical spectrum in bokmål and nynorsk including discharge summaries, surgical reports, nursing notes etc. ensuring comprehensive representation of the linguistic landscape within Norwegian healthcare settings. Evaluation on three synthtetic Norwegian clinical benchmark datasets and two real-world problems demonstrates that each of our clinically specialized models consistently outperforms their baseline counterparts, highlighting the significant benefit of domain-specific pre-training for NLP tasks within the clinical domain. The project was a joint effort by all Helse Vest entities (Helse Bergen, Helse Fonna, Helse Førde and Helse Stavanger) with DIPS under the project lead of Helse Vest ICT.

---

121. The Image Reconstruction Game: Drawing Common Ground Through Iterative Multimodal Dialogue

• Authors: Sherzod Hakimov , Mattia D’Agostini , Ivan Samodelkin , David Schlangen
• URL: https://arxiv.org/abs/2606.01901
• Abstract:

  We introduce the Image Reconstruction Game, a fully automated benchmark in which a vision-language model issues corrective instructions to an image generator across multiple turns, making accumulated common ground directly observable as a rendered image. Benchmarking two Describer models crossed with two Generator models across seven image categories, we find that the describer is the dominant factor in reconstruction quality, while the generator determines whether iterative refinement helps or hurts. Mathematical and geometric images pose the greatest challenge. The describer’s token budget strongly affects convergence: shorter budgets yield sparser first renderings with more room for visible improvement, while longer budgets raise absolute quality but leave less to fix. Stronger describers use a richer correction vocabulary spanning spatial, numeric, and structural categories, while weaker describers concentrate on surface properties and tend to stop after a few turns. Human validation shows that the best automated judge reaches only slight-to-fair agreement with human preferences, and automated scores require human recalibration to be used reliably.

---

122. RadioMaster: Multi-Agent System for Autonomous Radio Signal Generation

• Authors: Jiazhen Lei , Tianze Cao , Yuxin Sha , Sihan Wang , Bingbing Wang , Fengyuan Zhu , Zeming Yang , Xiaohua Tian
• URL: https://arxiv.org/abs/2606.01862
• Abstract:

  Translating user intents into physical radio signals represents the critical yet notoriously tedious final step in wireless prototyping, as it requires intricate knowledge of physical layer details and presents immense implementation challenges. Large Language Models (LLMs) and multi-agent systems have revolutionized conventional software engineering, raising the compelling question of whether they can resolve these formidable difficulties. However, our investigations reveal that current models experience significant limitations and fail to accomplish this task when applied to radio signal generation. This performance degradation primarily stems from severe domain ignorance and a fundamental insensitivity to physical hardware constraints. To bridge this gap, we introduce RadioMaster, a fully autonomous multi-agent framework designed to seamlessly translate user input into real-world wireless emissions. RadioMaster operates on three synergistic pillars: RadioWiki for domain-specific knowledge retrieval, RadioAgent for collaborative I/Q sample generation alongside hardware configuration, and RadioEmulator for closed-loop physical layer verification. Furthermore, we construct RadioBench, the first comprehensive benchmark tailored specifically for the radio signal generation domain. Extensive real-world evaluations demonstrate that RadioMaster significantly outperforms state-of-the-art (SOTA) baselines regarding configuration viability and signal fidelity.

---

123. Unveiling the Limits of Large Language Models in Inferring Pragmatic Meaning from Non-Verbal Responses

• Authors: Sugyeong Eo , Heuiseok Lim
• URL: https://arxiv.org/abs/2606.01845
• Abstract:

  Although large language models (LLMs) have shown considerable progress in pragmatic language understanding, prior research has focused mainly on their comprehension of verbal behavior. Nonetheless, non-verbal behavior remains a fundamental component of human communication, especially when deliberately utilized in isolation to convey indirect meanings. In this work, we present the first systematic evaluation of LLMs’ ability to infer pragmatic meaning in dialogue consisting solely of non-verbal responses. We explore three research questions: (1) Can LLMs recognize indirect intent conveyed through non-verbal responses? (2) When and how do LLMs fail to capture non-verbal intent? (3) How can we improve LLMs’ ability to interpret non-verbal intent?. Through the evaluation, we observe that LLMs struggle to infer underlying meaning from non-verbal responses, with accuracy dropping by up to 60% points compared to verbal ones. Further extensive analysis reveals a behavioral pattern in LLMs’ interpretations of non-verbal behavior and demonstrates that in-context learning facilitates pragmatic inference.

---

124. LayerRoute: Input-Conditioned Adaptive Layer Skipping via LoRA Fine-Tuning for Agentic Language Models

• Authors: Prateek Kumar Sikdar
• URL: https://arxiv.org/abs/2606.01838
• Abstract:

  Agentic language model systems alternate between two structurally distinct step types: structured tool calls (short, deterministic, low perplexity) and open-ended planning/reasoning steps (long, complex, high perplexity). Despite this heterogeneity, current inference systems apply identical compute to every step. We introduce LayerRoute, a lightweight adapter that learns to selectively skip transformer blocks on a per-input basis. LayerRoute augments each of the 24 transformer blocks in Qwen2.5-0.5B-Instruct with: (1) a per-layer router (~897 parameters, Linear(896,1)) that outputs a hard binary gate via the straight-through estimator, and (2) LoRA adapters (rank 8, ~1.08M parameters) on the Q/K/V/O attention projections. The backbone weights remain frozen. A single end-to-end training pass on agentic data (Hermes, Glaive, GSM8K, Turing) with a gate regularisation term forces the system to discover which blocks are skippable per input type. After 3,000 steps (6.4 minutes on an A100 40GB), LayerRoute achieves a 12.91% skip differential: tool calls skip 15.25% of FLOPs while planning steps skip only 2.34%, using only 1.10M trainable parameters (0.22% of the 494M backbone). Quality improves over the base model due to LoRA adaptation, with perplexity delta of -1.29 on tool calls and -1.30 on planning.

---

125. Dynamic Trust-Aware Sparse Communication Topology for LLM-Based Multi-Agent Consensus

• Authors: Wanshuang Gou , Zihan Liu
• URL: https://arxiv.org/abs/2606.01828
• Abstract:

  Large language model-driven multi-agent systems enhance the reliability of complex reasoning tasks through multi-round deliberation, role specialization, and cross-validation. However, existing multi-agent debate and collaboration frameworks typically adopt fully connected communication, causing the number of messages, token costs, and end-to-end latency to grow approximately quadratically with the number of agents; although fixed sparse topologies reduce overhead, they cannot adapt communication relationships to different task instances or intermediate reasoning states, making them prone either to preserving low-value interactions or to losing critical error-correction information. To address this problem, this paper proposes DySCo (Dynamic Sparse Consensus), a dynamic trust-aware sparse consensus mechanism. In each round of reasoning, DySCo estimates the value of communication edges based on agent reliability, answer divergence, and task relevance, and selects a small number of high-value edges for message exchange under budget constraints; it then aggregates the answers of different agents through dynamic trust weights and terminates the discussion early once consensus stabilizes. This mechanism replaces universal broadcasting with on-demand communication, thereby reducing communication overhead while preserving essential cross-validation information. We further present analyses of communication complexity and consensus stability, and evaluate the performance of DySCo on mathematical reasoning, logical reasoning, and factual question-answering tasks.

---

126. “I’ve Seen How This Goes”: Characterizing Diversity via Progressive Conditional Surprise

• Authors: Matthew Khoriaty , David Williams-King , Shi Feng
• URL: https://arxiv.org/abs/2606.01811
• Abstract:

  Measuring the diversity of creative outputs is central to evaluating post-training mode collapse, comparing decoding strategies, and quantifying creative behavior in both AI and human writing. We propose a new approach to measuring diversity using in-context learning, of which the ``Decan’’ metric, $D_{Ca_n} = C \times a_n$, is the working instance we evaluate: a per-byte score read off the per-token log-probabilities of a base model $\theta$ in a \emph{single forward pass} per permutation, with no embedding model, no reference corpus, and no human labels. This approach is grounded in information theory, makes use of language model in-context learning to detect a wide range of similarities between any number of inputs, and obviates the need to train a special-purpose model. The same pipeline scores AI samples and human-written response sets, with diversity treated as a property of (responses, prompt, scoring model). On Tevet and Berant’s human-grounded McDiv benchmark, $D_{Ca_n}$ reaches OCA 0.846 on the McDiv prompt_gen set where it performs best, behind the strongest neural baseline reported in Tevet and Berant (SentBERT, 0.897). On the OLMo-2-7B post-training pipeline, $D_{Ca_n}$ drops monotonically across the base $\to$ SFT $\to$ DPO $\to$ RLVR stages, detecting the type of diversity loss that creative-writing applications care about.

---

127. ProbeScale: Probing Analysis to Optimize Neural Scaling Laws for Efficient Small Language Model Inference

• Authors: Sourav Das
• URL: https://arxiv.org/abs/2606.01806
• Abstract:

  Small Language Models (SLMs) offer a balance between capability and computational feasibility. Neural scaling laws inform their optimal training, suggesting that they possess rich internal representations that scale with their size. However, deploying even these SLMs can be challenging under strict resource constraints. Language model probing provides methods for analyzing the linguistic knowledge encoded in a model’s internals. We propose ProbScale, a framework that unifies insights from scaling laws and probing to identify parameter-efficient subnetworks within pre-trained SLMs. ProbScale utilizes the high-quality representations of well-scaled SLMs and uses task-specific probes to mathematically quantify the relevance of each layer for target downstream capabilities. This allows selecting subnetworks that optimally trade off performance against parameter size. We formulate the subnetwork selection as finding a layer subset maximizing aggregated, task-weighted probe performance under a parameter budget. Experiments on representative SLMs such as RoBERTa-Large and T5-Base demonstrate that ProbScale identifies subnetworks achieving significant parameter reduction, from 5 to 10 times, while maintaining high performance (95% to 98% of the original SLMs) on targeted tasks, outperforming heuristic baselines.

---

128. MOSS-Audio Technical Report

• Authors: Chen Yang , Chufan Yu , Hanfu Chen , Jie Zhu , Jingqi Chen , Ke Chen , Wenxuan Wang , Yang Wang , Yaozhou Jiang , Yi Jiang , Zhengyuan Lin , Ziqi Chen , Zhaoye Fei , Chenghao Liu , Jun Zhan , Kang Yu , Kexin Huang , Mingshu Chen , Qinyuan Cheng , Ruixiao Li , Shimin Li , Songlin Wang , Yang Gao , Yiyang Zhang , Xipeng Qiu
• URL: https://arxiv.org/abs/2606.01802
• Abstract:

  MOSS-Audio is a unified audio-language model for speech, environmental sound, and music understanding, supporting audio captioning, time-aware question answering, timestamped transcription, and audio-grounded reasoning. MOSS-Audio couples a dedicated audio encoder with a modality adapter and a large language model: the encoder produces 12.5 Hz temporal representations, the adapter projects them into the decoder space, and the decoder generates autoregressive text outputs. Two design choices are central to the system: \textbf{DeepStack cross-layer feature injection}, which exposes the decoder to acoustic information from multiple encoder depths, and \textbf{time markers}, which provide explicit temporal cues by inserting timestamp markers into the audio-token stream. At the data level, we design an event-preserving audio annotation pipeline that segments raw audio at coherent event boundaries, applies branch-specific annotation to speech, music, and general audio, and merges the results into unified captions for pretraining. The intermediate branch-specific captions are further retained to support the construction of task-oriented SFT data. The model is pretrained on large-scale audio-language data, with time-aware objectives incorporated to support temporal grounding, and then undergoes multi-stage post-training to enhance instruction following and audio-grounded reasoning. We release 4B and 8B variants in both Instruct and Thinking configurations. MOSS-Audio achieves strong performance across general audio understanding, speech captioning, ASR, and timestamped ASR, positioning it as a promising understanding foundation for future voice agents.

---

129. Multilinguality of Large Language Models From a Structural Perspective

• Authors: Haruki Sakajo , Yusuke Sakai , Hidetaka Kamigaito , Taro Watanabe
• URL: https://arxiv.org/abs/2606.01800
• Abstract:

  Large language models (LLMs) have excelled in processing multiple languages through pre- and post-training on multilingual data, even though English dominates the training data. Prior work focusing on token representations has revealed how those LLMs process non-English text. Although these analyses have provided insightful findings, they fail to capture a structural view, which is an inherent property of language. In this study, we explore the multilinguality of LLMs through representational structural analysis. Our findings reveal that low-resource languages are structurally more different from English than high- and mid-resource languages, and that language-specific post-training alters their structures while preserving inter-language relationships.

---

130. STaR-KV: Spatio-Temporal Adaptive Re-weighting for KV Cache Compression in GUI Vision-Language Models

• Authors: Yuhang Han , Wenzheng Yang , Yujie Chen , Xiangqi Jin , Yaojie Zhang , Siteng Huang , Linfeng Zhang
• URL: https://arxiv.org/abs/2606.01790
• Abstract:

  Vision-language-model-based graphical user interface (GUI) agents have shown broad automation capabilities, yet deployment is bottlenecked by a key-value (KV) cache that grows linearly with interaction steps. For instance, UI-TARS-1.5-7B consumes 76 GB of GPU memory on merely five screenshots, approaching the capacity of mainstream 80 GB accelerators. Existing KV compression methods share two structural assumptions: aggregating visual-token importance into a single shared saliency map, and applying a fixed top-B cutoff to the fused score distribution. Pilot measurements refute both: spatial specialization lives at the attention-subspace level and migrates across layers, while the score distribution drifts in shape along a trajectory. We propose STaR-KV (Spatio-Temporal Adaptive Re-weighting), a training-free KV cache compression framework that calibrates token importance along three axes: (i) subspace-aware scoring driven by online spatial mutual information; (ii) a temporal stability discount that suppresses redundant cache entries from persistently attended subspaces; and (iii) an entropy-derived temperature that adaptively reshapes the score distribution. Across four GUI benchmarks, STaR-KV achieves the strongest average accuracy among state-of-the-art KV compression methods (e.g., GUIKV, SnapKV) at matched budgets, with no compression-stage FLOPs overhead (-0.07%) and cutting peak GPU memory by nearly 40% at a 20% KV-cache budget. Code is available at this https URL .

---

131. Breaking the Information Silo: Semantic Personas for Cross-Domain Recommendation

• Authors: Jonathan Mayo , Moshe Unger , Konstantin Bauman
• URL: https://arxiv.org/abs/2606.01783
• Abstract:

  Digital platforms increasingly operate as isolated information silos, limiting their ability to construct comprehensive user representations across domains. Cross-domain recommender systems seek to overcome this limitation by transferring knowledge from a source domain to a target domain, yet most existing approaches depend on shared users, shared items, or structurally similar interaction graphs. These assumptions are often unrealistic across independent platforms. We propose SPHERE (Semantic Personas for Heterogeneous cross-domain Recommendation), a design artifact that enables recommendation knowledge transfer across strictly disjoint domains with no shared users or items. Rather than aligning domains through identity or graph structure, SPHERE uses large language models to induce a shared behavioral vocabulary, generate structured semantic personas for users, and retrieve behaviorally similar source-domain communities that form a Community Source Persona. This semantic signal is integrated with collaborative signals through a dual-tower architecture and dynamic fusion gate, allowing SPHERE to augment standard recommender backbones. Empirical evaluation across Amazon Books, Goodreads, and Steam demonstrates consistent improvements over NCF, SVD++, and LightGCN baselines under full-ranking evaluation. The results show that cross-domain transfer effectiveness is not determined solely by semantic proximity between domains; rather, it depends critically on the structural density and native predictive strength of the target domain. The study contributes to information systems research by reframing cross-domain personalization as behavior-based semantic alignment, offering a practical mechanism for overcoming information silos while preserving interpretability and modularity.

---

132. FLARE: Diffusion for Hybrid Language Model

• Authors: Yuchen Zhu , Jing Shi , Chongjian Ge , Hao Tan , Yiran Xu , Wanrong Zhu , Jason Kuen , Koustava Goswami , Rajiv Jain , Yongxin Chen , Molei Tao , Jiuxiang Gu
• URL: https://arxiv.org/abs/2606.01774
• Abstract:

  Autoregressive (AR) large language models (LLMs) have achieved broad practical success, but sequential decoding remains a key bottleneck for low-latency deployment. Recent efficient-inference work has progressed along two axes: reducing the cost of each model invocation through efficient architectures, and reducing serial decoding steps through parallel generation. Hybrid attention backbones address the former, while diffusion language models (dLLMs) pursue the latter via iterative parallel denoising. Combining these advantages remains challenging: AR-to-dLLM conversion often fails to preserve seed-checkpoint capability, and hybrid-attention recurrent states and masking constraints make diffusion training and serving nontrivial. We present FLARE, a systematic conversion framework for hybrid-attention LLMs. Our analysis identifies transfer data quality as the primary determinant of capability preservation, outweighing loss formulation and attention-mask design. The resulting framework combines a token-equal AR-and-diffusion objective, hardware-aware kernels, and unified inference, enabling one checkpoint to support both AR-style verified decoding and diffusion-style parallel denoising. Starting from strong AR checkpoints with limited post-training data, FLARE is competitive with leading open-source dLLMs across model scales and delivers consistent throughput gains over open-source dLLM baselines in single-GPU concurrent serving. Our results further suggest that practical dLLMs are limited not only by decoding algorithms, but also by transfer data quality and the training inefficiency of current block-diffusion objectives, motivating joint design of data, objectives, architectures, and inference systems.

---

133. Adaptive Auto-Harness: Sustained Self-Improvement for Agentic System Deployment on Open-Ended Task Streams

• Authors: Zewen Liu , Zhan Shi , Yisi Sang , Bing He , Minhua Lin , Tianxin Wei , Dakuo Wang , Benoit Dumoulin , Wei Jin , Hanqing Lu
• URL: https://arxiv.org/abs/2606.01770
• Abstract:

  Auto-harness systems such as A-Evolve, GEPA, and Meta-Harness improve LLM agents by optimizing prompts, skills, tools, memories, and supporting infrastructure from execution feedback, but they are typically evaluated on fixed offline benchmarks. Real deployments instead present open-ended task streams: histories grow without a fixed endpoint, heterogeneous tasks require different harnesses, and problem distributions shift over time. These challenges make a single repeatedly and densely updated harness brittle, causing performance degradation as accuracy peaks early and then declines. This motivates sustained harness construction with task-wise adaptation. We introduce Adaptive Auto-Harness, a framework and system for such streams. The framework decomposes the gap to an oracle harness into evolution loss and adaptation loss. The system addresses these losses with a stateful multi-agent evolver, a harness tree with solve-time routing, and human-steering hooks for cases where history lacks the needed signal. Across prediction-market, security-competition, and event-forecasting streams, Adaptive Auto-Harness outperforms five existing auto-harness baselines and ablations attribute gains to better construction, routing, or targeted human steering. Code is available in this https URL .

---

134. Construction of Historical Knowledge Graphs Based on BERT and Graph Neural Networks

• Authors: Ping Li , Bartlomiej Brzozka
• URL: https://arxiv.org/abs/2606.01747
• Abstract:

  Through digital humanities research and scale-up historical data analysis, a significant amount of traditional historical text is converted into structured knowledge graphs. This paper provides a high-level architecture that combines bidirectional encoder representations of transformers (BERT) and graph neural networks (GNN) to extract the entities and relationships from various types of historical texts. The texts of traditional history resolve linguistic ambiguities, references limited by context, and a lack of established grammatical norms in a systematic way. This study develops a new image retrieval system based on FastRQNet and pre-trained vision-language model Vilt-qaformer+RoBInet in accordance with the aforementioned recommendations. The experiments make full use of a comprehensive collection of municipal records, parliamentary documents, and historical correspondence. When compared to conventional rule-based techniques and other popular deep-learning baselines, the joint BERT-GNN system obtains greater Precision, Recall, and F1-score (Table 2). Complex nested structures and implicit reference issues can be handled by this structure with sufficient accuracy and thoroughness when creating knowledge graphs. The aforementioned experiments show that combining relational graph learning algorithms with context-sensitive semantic representation techniques can automatically extract historical data to add accumulated wisdom to the knowledge repository.

---

135. THRD: A Training-Free Multi-Turn Defense Framework for Jailbreak Attacks on Large Language Models

• Authors: Zhiqing Ma , Zhonghao Xu , Dong Yu , Chen Kang , Changliang Li , Pengyuan Liu
• URL: https://arxiv.org/abs/2606.01738
• Abstract:

  Multi-turn jailbreak attacks pose a growing threat to LLMs by exploiting conversational dynamics such as gradual escalation and cross-turn coordination. Existing defenses either rely on costly retraining – often degrading model utility – or apply single-turn analysis independently at each turn, failing to capture how risk accumulates along interaction trajectories. We observe that safety behavior in multi-turn interaction is trajectory-dependent: dialogue history continuously reshapes the model’s conditioning context, making it insufficient to evaluate each turn in isolation. Motivated by this insight, we present THRD, the first training-free framework that explicitly models temporal risk accumulation for multi-turn jailbreak defense. THRD integrates four modules: a Turn-level Risk Assessor (TRA) for instantaneous risk estimation, a Historical Context Analyzer (HCA) for cross-turn intent escalation detection, a Response Evaluator (RE) for identifying facilitative outputs, and a Decision Module that combines these signals through a time-evolving scoring mechanism with attenuation-based modulation and trend-aware adjustment. Experiments against state-of-the-art multi-turn attacks – including tree-search-based and multi-agent collaborative methods – across two target models show that THRD reduces ASR to 0.2–4.0% while preserving model utility within 1.5% degradation on MMLU and GSM8K. Ablation studies confirm non-redundant module contributions and stable cross-architecture generalization. Analysis of first rejection triggers reveals that over 70% of multi-turn attacks require Turn~2 or later to detect, validating the necessity of explicit temporal aggregation.

---

136. Argument Collapse: LLMs Flatten Long-Form Public Debate

• Authors: Yekyung Kim , Yapei Chang , Chau Minh Pham , Mohit Iyyer
• URL: https://arxiv.org/abs/2606.01736
• Abstract:

  As LLMs are increasingly used to draft public-facing arguments, they may flatten public debate by repeatedly introducing the same polished, plausible arguments. We study argument collapse, the tendency of essays generated by different LLMs to converge to a smaller set of main arguments, sub-arguments, and paragraph-level structures. We compare 1,039 human responses from 195 New York Times (NYT) debates, 448 human responses from 61 longer-form Boston Review (BR) forums, and 23,384 LLM-generated essays. In the NYT corpus, 65.3% of human main arguments are unique within a debate, compared to 3.4% of LLM main arguments. Asking LLMs to generate diverse answers adds variation, but a typical model recovers only about half of the distinct human main arguments, with much of the added variation falling outside the observed human argument space. Collapse also appears in sub-arguments, where among essays with the same main argument, 41.0% of human sub-arguments are unique versus 9.1% from LLM responses. Qualitatively, LLMs often reuse generalized and hedged sub-arguments, while humans prefer more concrete and topic-specific ones. Structure-wise, LLM-generated essays tend to follow a more fixed arc, often opening with a direct claim and moving quickly toward proposals. The same patterns hold in longer BR essays, suggesting that argument collapse extends beyond short-form responses.

---

137. Shortcut to Nowhere: Demystifying Deep Spurious Regression

• Authors: Guanrong Xu , Jessica Li , Hao Wang , Yuzhe Yang
• URL: https://arxiv.org/abs/2606.01723
• Abstract:

  Real-world regression often exhibits shortcuts: attributes that are spuriously correlated with continuous targets in training, yet unreliable under deployment shifts; regressing targets using such shortcuts may fail catastrophically at test time. Existing studies on spurious correlations focus primarily on classification, where labels are categorical and groups are naturally defined. However, many real-world tasks require continuous prediction, where hard label boundaries or discrete group-label pairs do not exist. We define Deep Spurious Regression (DSR) as learning from regression data with attribute-label confounding, addressing continuous spurious correlations, and generalizing to all attribute-label combinations at test time. Motivated by the intrinsic difference between classification and regression shortcuts, we propose to exploit the similarity among spurious attributes in both label and feature spaces, thereby accounting for nearby targets and related groups while calibrating both label and learned feature distributions across attributes. Extensive experiments on common real-world DSR datasets that span computer vision, environmental sensing, and large language model (LLM) regression verify the superior performance of our strategies. Our work fills the gap in benchmarks and techniques for studying spurious correlations in continuous prediction.

---

138. Two-Fidelity Best-Action Identification for Stochastic Minimax Tree

• Authors: Peter Chen , Xi Chen
• URL: https://arxiv.org/abs/2606.01708
• Abstract:

  We study fixed-confidence best-action identification (BAI) in stochastic minimax trees. This problem is increasingly relevant in modern AI planning, where deep minimax search and Monte Carlo Tree Search (MCTS) with language model long rollouts face a fundamental tradeoff: heuristic evaluations are cheap but biased, while accurate rollouts are reliable but prohibitively expensive. We propose 2FFS, a two-fidelity tree-search algorithm that brings multi-fidelity flat bandit ideas into trees. The algorithm combines minimax-style fast expansion with MCTS-style stochastic sampling, adaptively deciding when to exploit cheap biased evaluations and when to invoke expensive accurate evaluations for local certification. We prove fixed-confidence correctness, establish finite stopping for exact identification, and give a polynomial-depth cost upper bound for general-depth trees. Across numerical stochastic-tree experiments, 2FFS uses substantially fewer samples and computational operations comparing to existing BAI-MCTS baseline.

---

139. JenBridge: Adaptive Long-Form Video Soundtracking across Scene Transitions

• Authors: Jiashuo Yu , Yao Yao , Boyu Chen , Alex Wang
• URL: https://arxiv.org/abs/2606.01703
• Abstract:

  We address the challenge of generating high-fidelity, long-form soundtracks that remain coherent across scene transitions. Existing AI music systems are mainly designed for short, isolated clips and lack mechanisms to ensure narrative continuity. We present JenBridge, a modular and interpretable framework for adaptive long-form video soundtracking that ensures both high-fidelity audio generation and transition naturalness. The core architecture is a Transformer-based generative model trained with a flow-matching objective, following a two-stage paradigm: pretraining on large-scale text-audio corpora to establish robust musical priors, then adapting to the video domain with dual text-visual conditioning for precise cross-modal alignment. Crucially, to achieve long-form coherence across diverse scene changes, JenBridge incorporates a novel adaptive transition mechanism. This system features a versatile toolkit of transition styles, including a generative transition method, and uniquely employs a Large Language Model (LLM) Agent that acts as a director to select the most appropriate transition for each narrative shift intelligently. To rigorously assess this task, we propose the LVS Benchmark, a new benchmark that includes a curated dataset and novel evaluation metrics focusing on holistic and transition-aware assessment. Extensive experiments on the proposed benchmark demonstrate that JenBridge significantly outperforms existing methods in both objective and subjective metrics, particularly in terms of transition naturalness and overall narrative coherence. JenBridge represents a significant step towards fully automated, professional-quality video soundtracking.

---

140. Off-the-Shelf LLMs as Process Scorers: Training-Free Alternative to PRMs for Mathematical Reasoning

• Authors: Atoosa Chegini , Soheil Feizi
• URL: https://arxiv.org/abs/2606.01682
• Abstract:

  Selecting the best response from multiple small-model samples using a stronger scorer is a simple inference-time strategy, but fails when the small model has already committed to incorrect reasoning paths. PRM guided search avoids this by scoring candidate continuations during generation, but requires a reward model trained with step-level labels. We propose Chunk-Level Guided Generation, a training-free alternative that uses an off-the-shelf large language model as a process scorer. At each step, a small model samples k fixed-length candidate chunks, while the larger model scores the candidates using likelihoods without generating any text. The selected chunk is committed before the next step, steering generation before errors can propagate. We instantiate this framework with two selection rules: Likelihood-Guided Selection (LGS), which selects the chunk with the highest length-normalized large-model log-probability, and Contrastive-Guided Selection (CGS), which subtracts the small model’s log-probability to favor chunks where the large model’s preference diverges from the small model’s. We show that scoring variable-length reasoning steps with large-model likelihoods is unreliable due to a systematic length bias that persists even after length normalization, and that fixed-length chunks avoid this confound. On GSM8K, MATH, Minerva Math, AMC23, and AIME24 with Qwen2.5-1.5B guided by Qwen2.5-32B and Llama-3.2-1B guided by Llama-3.1-70B, CGS outperforms majority voting by up to 28 pp and, under matched guidance budgets, matches or outperforms Qwen2.5-Math-PRM-72B guided search on most benchmarks without reward-model training. With Qwen2.5-7B guided by Qwen2.5-72B, CGS reaches 81.8% on MATH and 63.6% on Minerva Math at k=16, surpassing majority voting by 4–6 pp. Finally, Chunk-Level Guided Generation produces substantially shorter reasoning traces than PRM guided search.

---

141. DOT-MoE: Differentiable Optimal Transport for MoEfication

• Authors: Udbhav Bamba , Arnav Chavan , Aryamaan Thakur , Steve Teig , Deepak Gupta
• URL: https://arxiv.org/abs/2606.01666
• Abstract:

  The scaling of Large Language Models (LLMs) has driven significant performance gains but created substantial challenges in inference efficiency. While Mixture of Experts (MoEs) architectures address this by decoupling model size from inference cost, training MoEs from scratch is often unstable and compute intensive. Conversion of pre-trained dense models into sparse MoEs has emerged as an alternative solution; however, existing methods typically rely on heuristic neuron clustering or random splitting to partition the Feed-Forward Network (FFN) into experts. In this work, we propose DOT-MoE, a novel framework that formulates the decomposition of dense layers as a Differentiable Optimal Transport (DOT) problem. Instead of static heuristics, we model neuron assignment as a balanced transport problem, utilizing differentiable Sinkhorn-Knopp iterations to enforce strict expert capacity constraints. Furthermore, we utilize Straight-Through Estimators (STE) to jointly learn the discrete neuron-to-expert assignment and the token-to-expert routing policy end-to-end. Extensive experiments across multiple architectures and benchmarks demonstrate that DOT-MoE significantly outperforms structured pruning, heuristic clustering, and random-split baselines, retaining 90% of the original dense model’s performance while reducing active parameters by 50%.

---

142. Easier to Mislead Than to Correct: Harmful and Beneficial Revision in LLM Conformity

• Authors: Jiaming Qu , Lucheng fu , Yibo Hu
• URL: https://arxiv.org/abs/2606.01637
• Abstract:

  Large language models are increasingly used in multi-agent systems, where they see and respond to other agents’ answers. A key risk is conformity: a model may abandon its own answer simply because others agree on a different one. Prior studies show that LLMs often revise toward a majority answer, but it remains unclear whether these revisions help correct mistakes as often as they introduce new errors. In this paper, we conduct a controlled study in which an LLM first answers a question, then sees simulated peer responses before making a final decision. We manipulate two social cues: consensus structure and authority labels assigned to peers, and measure how they influence beneficial and harmful revisions. Across four open-weight LLMs and seven QA datasets, we find that peer agreement makes it much easier to mislead initially correct models than to correct initially wrong ones. Authority labels make models more likely to choose the endorsed answer, regardless of whether it is correct. More concerningly, generic reasoning interventions such as chain-of-thought and reflection do not reliably reduce harmful revision while preserving beneficial revision. These findings suggest that multi-agent LLM systems should verify peer answers rather than simply aggregate them.

---

143. AlphaToken: Decoupling Adaptation and Stability for Path-Aware Response Token Valuation in LLM Post-Training

• Authors: Liu Qing , Ou Wu , Yi Du
• URL: https://arxiv.org/abs/2606.01635
• Abstract:

  Token selection is pivotal for effective LLM post-training. However, existing methods mostly rely on local heuristics and rarely formulate token selection as a principled valuation of individual response tokens. We introduce $\textbf{AlphaToken}$, a response token valuation framework that decouples valuation into $\textbf{adaptation}$ (promoting target-task learning) and $\textbf{stability}$ (preserving pre-trained capabilities), and makes each objective $\textbf{path-aware}$ by combining the direct-path signal from local token gradients with the downstream causal-path signal in autoregressive generation. Since retention data are typically unavailable, AlphaToken approximates stability via a $\textbf{Fisher-drift proxy}$ anchored at the pre-trained reference model. For efficient computation, we extend Ghost Dot-Product to token-level valuation. AlphaToken masks low-value response tokens during fine-tuning and preference optimization, concentrating training signals on more valuable positions. Experiments show that AlphaToken improves post-training performance and mitigates catastrophic forgetting.

---

144. EvoPool: Evolutionary Programmatic Annotation for Label-Efficient Specialized Supervision

• Authors: Tianyi Xu , Yaolun Zhang , Xuan Ouyang , Huazheng Wang
• URL: https://arxiv.org/abs/2606.01617
• Abstract:

  Large language models excel at general tasks but underperform smaller supervised models in specialized, high-stakes domains where training labels are costly. We address this regime with EvoPool, an evolutionary multi-agent framework inspired by Darwinian evolution. Three specialized agents iteratively propose executable annotator code, a small validation set provides a fitness signal, and a deterministic gate keeps only annotators that pass viability, diversity, and marginal-contribution checks across generations. Pool votes are mapped to soft training labels by EvoAgg, a text-aware aggregator combining semantic features with annotator-vote features. The authored pool runs at near-zero per-example cost and is 4500 to 31000x faster than LLM annotation on 100K examples. Across 7 of 8 LLM-weak specialized and complex tasks spanning biomedical relation extraction, legal-clause classification, complex reasoning, and dense multi-label biomedical classification, EvoPool beats the strongest LLM annotation baseline by an average +0.141 macro-F1, peaking at +0.301 on ChemProt and +0.265 on PubMed. Code is available at: this https URL

---

145. TechGraphRAG: An Agentic Graph-Augmented RAG Framework for Technical Literature Reasoning

• Authors: Kanwar Bharat Singh
• URL: https://arxiv.org/abs/2606.01613
• Abstract:

  This paper presents an agentic retrieval-augmented generation (RAG) framework for domain-specific technical reasoning support, instantiated over a curated corpus of approximately 2,100 academic papers in intelligent tires, vehicle dynamics, and vehicle control. Unlike conventional single-pass RAG systems, the proposed architecture employs a 13-step autonomous pipeline that classifies queries by intent, scores evidence sufficiency against a multi-dimensional rubric, performs agentic retry with drift-guarded query reformulation, searches external academic databases (Crossref, OpenAlex, Semantic Scholar) through iterative optimize–search–vet loops, traverses a Neo4j knowledge graph for relational context, verifies citation integrity, and applies post-generation quality checks with automatic regeneration. Key contributions include a 100-point evidence sufficiency scoring framework across five dimensions with relevance damping and hybrid rule-based/LLM review; a route-dependent external search architecture with iterative agentic loops; a knowledge graph constructed via LLM-based entity extraction and OpenAlex author validation with intra-corpus citation resolution; and a self-correcting generation loop with citation verification and quality assessment. The framework is presented as a practical, implemented case study illustrating how agentic, evidence-grounded RAG can support literature navigation and technical reasoning over large, domain-specific corpora.

---

146. Identifying High-Confidence Social Biases in LLMs for Trustworthy Conversational Tutoring Agents

• Authors: Aitor Arronte Alvarez , Naiyi Xie Fincham
• URL: https://arxiv.org/abs/2606.01584
• Abstract:

  Conversational tutoring agents have been shown to improve learning engagement and student outcomes, and large language models (LLMs) are increasingly used in these systems to provide scalable, personalized feedback. However, LLMs may perpetuate or amplify stereotypical social biases, posing particular risks in educational settings. In this study, we evaluate LLMs in conversational tutoring scenarios to identify high-confidence social biases, instances where models are unable to identify biased judgments in tutoring conversations while maintaining strong confidence in their assessments, potentially affecting their reasoning and the feedback they provide to learners. We present a new dataset generation method that enables bias evaluation under naturalistic instructional conditions by regenerating student-AI tutor interactions and introducing turns with controlled bias derived from a benchmark dataset. Using this data, we assess multiple LLMs’ ability to detect stereotypical biases and analyze the confidence and reasoning underlying their responses through computational and human evaluations. We find that bias detection is substantially more challenging in conversational tutoring contexts than in benchmark-based evaluations, and that state-of-the-art LLMs are overconfident in their incorrect assessments of stereotypical bias statements. Moreover, model confidence strongly influences reasoning and feedback, highlighting the risks of overconfident, biased behavior in LLM-based tutoring agents. We conclude by discussing implications, mitigation considerations, and directions for future research.

---

147. Defenses & Enablers For Skill Injection Attacks on Terminal Based Agents

• Authors: Yoshinari Fujinuma , Varun Gangal , Traian Rebedea , Makesh Narasimhan Sreedhar , Prasoon Varshney , Rebecca Qian , Anand Kannappan
• URL: https://arxiv.org/abs/2606.01567
• Abstract:

  Large language model (LLM) agents increasingly rely on reusable skills i.e. documents describing task-specific procedures. However, this introduces a new attack surface for agents to manage. We study two complementary directions for this threat. First, we evaluate guardian-based defenses: an intermediary LLM agent that acts as a mediator for skill file access (dynamic guardian) or pre-rewrites these files at build time (static guardian). Across three LLM agent families, our guardians cut attack success rate (ASR) by well over half while preserving task utility. Second, we stress test them through attack reframing using four attacks that preserve the malicious instruction but change the phrasing. For non-guardian setup, the reframing pushes the ASR up to 81.4\%, but the dynamic guardian brings it down to 18.6\%, showing that real-time mediation is a robust defense.

---

148. On the Limits of Token Reduction for Efficient Unified Vision Language Training

• Authors: Siyi Chen , Weiming Zhuang , Jingtao Li , Lingjuan Lv
• URL: https://arxiv.org/abs/2606.01503
• Abstract:

  Unified vision-language models (VLMs) integrate visual understanding and visual generation within a single autoregressive backbone, but their joint training is computationally expensive and largely overlooked from an efficiency perspective. In this work, we study the feasibility and limits of token-reduction-based acceleration for unified VLM training. Through a systematic analysis of layerwise attention allocation, we uncover a fundamental asymmetry: visual understanding exhibits substantial late-layer visual redundancy, whereas visual generation maintains persistent dependence on image tokens across depth. Guided by this observation, we design task-specific accelerators that selectively reduce image-token computation for each objective. While these methods achieve significant efficiency gains in isolated settings, we observe a consistent synergy loss under unified training – task-specific token dropping necessitates divergent parameter pathways and eliminates the mutual performance gains typically observed in joint optimization. Our findings suggest that efficient unified modeling requires preserving shared cross-task structures, highlighting the need for synergy-aware acceleration strategies. Project page: this https URL .

---

149. TimeSage-MT: A Multi-Turn Benchmark for Evaluating Agentic Time Series Reasoning

• Authors: Yaxuan Kong , Qingren Yao , Yuqi Nie , Yichen Li , Yilei Shao , Stefan Zohren , Anna Vettoruzzo , Joaquin Vanschoren , Ming Jin , Qingsong Wen
• URL: https://arxiv.org/abs/2606.01498
• Abstract:

  Time series data inform critical decisions across many real-world domains. While large language model (LLM) agents can analyze data through natural language and tools, it remains unclear whether they can conduct reliable time series analysis across multi-turn conversations. Existing benchmarks focus on single-step tasks such as forecasting and anomaly detection, overlooking practical workflows where user goals evolve, agents must build on prior analyses, and conclusions emerge from accumulated evidence. In this work, we introduce TimeSage-MT, a multi-turn benchmark for agentic time series reasoning with 240 tasks and 2,680 dialogue turns across 8 real-world domains, spanning basic exploration to decision-oriented analysis. TimeSage-MT is built through a reproducible pipeline that converts real-world time series data into multi-turn conversations with verifiable answers. It provides a unified evaluation protocol and public leaderboard for comparing time series agentic systems. To demonstrate the benchmark’s utility, we evaluate frontier LLMs alongside TimeSage, a novel structured agent equipped with a comprehensive time series skill library. The results show sharp performance drops on decision-oriented tasks, driven by failures in memory, uncertainty handling, and domain-based decision making. TimeSage-MT exposes critical gaps in current agentic reasoning and provides a rigorous foundation for future development.

---

150. LLM Consortium for Software Design Refinement: A Controlled Experiment on Multi-Agent Collaboration Topologies

• Authors: Nagarjuna Kanamarlapudi , Praveen K
• URL: https://arxiv.org/abs/2606.01490
• Abstract:

  We present a controlled experiment evaluating 12 multi-agent LLM collaboration topologies for software architecture design. Using a $2\times2\times2$ factorial design (Authority $\times$ Roles $\times$ Dynamics), we conducted 520 experimental runs across 8 design tasks of varying complexity, with 5 repetitions each. Designs were evaluated on a 12-dimensional rubric by three independent automated evaluators (GPT-OSS 120B, Claude Opus 4.6, Claude Sonnet 4.6). We report four core findings. First, structural adversarial (v4b) ranks #1 by ensemble – a prompt-engineered adversarial variant that demands rewrite mandates rather than patches (weighted ensemble: 4.637/5.0). Second, cross-model review wins unanimously at #2 – generate with one model, review with another – ranking #2 by all three evaluators (weighted ensemble: 4.606). Third, evaluator diversity is itself a finding – all three evaluators agree v4b is best and v3 is worst, but disagree sharply on v2b (Claude d=1.44 vs. GPT-OSS d=0.45), revealing how different model families weight design qualities. Fourth, parallel merge is fundamentally broken – all three evaluators place merge variants in the bottom tier (3.65-3.79), due to token starvation and the Frankenstein effect. The weighted ensemble ($2\times$Opus + $2\times$Sonnet + $1\times$GPT-OSS) provides robust rankings across 520 runs, confirmed through independent cross-validation.

---

151. Hierarchical Online Prompt Mutation with Dual-Loop Feedback for Guardrailed Evidence Document Generation: A Production-Evaluation Case Study

• Authors: Nataraj Agaram Sundar Tejas Morabia
• URL: https://arxiv.org/abs/2606.01472
• Abstract:

  High-stakes production document-generation systems require language models to be adaptive, evidence-grounded, and auditable. We present HOPM, a hierarchical online prompt mutation framework evaluated on a real marketplace dispute-evidence workflow. HOPM treats prompts as online policies: a family/version router selects a prompt, deterministic guardrails attribute failures to mutable prompt-token categories, and dual feedback from human review and an automated judge updates both routing and mutation priorities. The primary evidence is an observed matched production-evaluation ablation: seven variants are evaluated on the same 600 cases each, enabling component comparisons against static prompting, manual iteration, bandit-only routing, mutation-only adaptation, human-only feedback, auto-judge-only feedback, and full dual-loop HOPM. Full HOPM improves count win rate over a static control from 34.7% to 45.7% (+11.0 pp; paired McNemar p = 1.31e-11) and amount-weighted win rate from 22.3% to 41.4% (+19.1 pp; 95% paired bootstrap CI [10.3, 28.9] pp). It also increases mean Likert quality from 3.18 to 4.40 and reduces issue-flag rate from 15.3% to 5.2%. Supporting review artifacts cover 770 generated-text reviews, 318 labeled reviewer exports, a 10-case/61-rating calibration slice, and a 70-case/350-rating OCR benchmark; these artifacts calibrate rubric, guardrail, title-risk, and OCR-risk interpretation rather than substituting for the production ablation. The paper includes control setup, sample sizes, confidence intervals, paired tests, prompt-token categories, pseudocode, schema, rubric, guardrail taxonomy, and a constructed example so the evaluation structure can be reproduced without exposing proprietary evidence.

---

152. Consistent and Distinctive: LLM Benchmark Efficiency via Maximum Independent Set Prompt Selection on Similarity Graphs

• Authors: Denica Kjorvezir , Marko Djukanović , Ana Gjorgjevikj , Gjorgjina Cenikj , Tome Eftimov
• URL: https://arxiv.org/abs/2606.01400
• Abstract:

  Evaluating large language models (LLMs) across comprehensive benchmarks is expensive and time-consuming. We propose a graph-based prompt selection framework that models each benchmark as a similarity graph – nodes are prompts connected if their embedding-space distance falls above a configurable threshold – and applies Maximum Independent Set (MIS) algorithms to select a maximally diverse, non-redundant subset. We evaluate four MIS solvers (CPLEX, GREEDY, Online-MIS, ReduMIS) across six embedding models, three distance measures, six percentile thresholds, and four benchmarks (GPQA, IFEval, MMLU-Pro, Omni-MATH) covering 66 LLMs. Our central hypothesis – that repeated selection under different random seeds yields consistent LLM rankings that may also differ from the full-benchmark baseline – is strongly confirmed: Kendall’s $W \geq 0.90$ in 99.2\% of stochastic configurations (mean $W = 0.997 \pm 0.008$), while at higher percentile thresholds selected subsets achieve 25–48\% prompt reduction on average. Ranking divergence from the full benchmark ($\rho < 0.95$) occurs in only 15.95\% of configurations, concentrated at low thresholds ($p_{10}$–$p_{20}$) and benchmarks (GPQA, IFEval), identifying overly dense graphs as the primary failure mode.

---

153. Dr. DocBench: A Comprehensive Benchmark for Expert-Level and Difficult Document Parsing

• Authors: Minglai Yang , Xinyan Velocity Yu , Pengyuan Li , Xinyu Guo , Zhenting Qi , Konwoo Kim , Longtian Ye , Xiaolong Luo , Jinhe Bi , Henry Zhang , Haris Riaz , Xuan Zhang , Yunze Xiao , Bangya Liu , Tom Tang , Yunfei Zhao , Qunshu Lin , Zihan Wang , Minghao Liu , Michael Lingzhi Li , Yilun Du , Jesse Thomason , Rogerio Feris , Alex Pentland , Zexue He
• URL: https://arxiv.org/abs/2606.01393
• Abstract:

  Document parsing and recognition are fundamental capabilities for vision-language models (VLMs) and document processing systems. However, existing Optical Character Recognition (OCR) and document parsing benchmarks are increasingly limited in coverage and difficulty: many focus on common document genres or uniformly sampled pages where modern parsers already perform strongly, while offering limited annotation for expert-domain structures such as chemical formula, music notation, complex tables, and cross-page layouts. We introduce Dr. DocBench, a difficulty-aware benchmark for expert-level document parsing. Built from a large-scale multilingual book corpus, Dr. DocBench spans 52 BISAC subject domains and selects challenging documents through parser-failure-based sampling, targeting cases where multiple state-of-the-art systems struggle. It contains 4,514 annotated pages from long documents averaging around 100 pages, with 65k high-quality page- and block-level annotations for layout, reading order, hierarchical relations, and domain-specific visual contents. Evaluations of pipeline-based parsers and general-purpose VLMs show that strong performance on existing benchmarks does not transfer to our expert-level document parsing. Our analysis reveals substantial failures across subjects, content types, and structural attributes, highlighting Dr. DocBench as a comprehensive testbed for diagnosing and advancing document intelligence.

---

154. Bridging Requirements and Architecture: Multi-Agent Orchestration with External Knowledge and Hierarchical Memory

• Authors: Ruiyin Li , Yiran Zhang , Xiyu Zhou , Yangxiao Cai , Peng Liang , Weisong Sun , Jifeng Xuan , Zhi Jin , Yang Liu
• URL: https://arxiv.org/abs/2606.01385
• Abstract:

  Software architecture design is a critical yet inherently complex and knowledge-intensive phase that requires balancing competing quality attributes and adapting to evolving requirements. Traditionally, this process has been time-consuming, labor-intensive, and heavily reliant on architects, often resulting in limited exploration of alternative architectural decompositions and styles, especially under the pressures of agile development. While LLM-based agents have shown promising performance across various software engineering tasks, their application to architecture design remains relatively scarce and requires systematic exploration. To address these challenges, we proposed MAAD (Multi-Agent Architecture Design), a knowledge-driven framework that orchestrates four specialized agents (i.e., Analyst, Modeler, Designer and Evaluator) to autonomously and collaboratively transform requirements specifications into comprehensive, multi-view architectural blueprints with quality attribute assessments. MAAD incorporates RAG to inject recognized architectural standards and patterns into the workflow and leverages a hierarchical memory mechanism that captures design history for iterative refinement. We evaluated MAAD through comparative experiments against MetaGPT, using quantitative architecture-level metrics across 10 case studies and qualitative feedback from industry architects on 10 real-world specifications. Results show that MAAD generates more complete, modular, and traceable architectures than the baseline, and its dedicated Evaluator agent autonomously produces structured quality evaluation reports that significantly reduce manual validation efforts. Furthermore, we found that the quality of the generated architecture heavily depends on the underlying LLM’s reasoning capacity, with GPT-5.2 and Qwen3.5 outperforming other models across most evaluation settings.

---

155. Efficient Exploration for Iterative Nash Preference Optimization

• Authors: Tianlong Nan , Xiaopeng Li , Christian Kroer , Tianyi Lin
• URL: https://arxiv.org/abs/2606.01382
• Abstract:

  Preference alignment is central to improving large language models, but standard reward-based formulations can be restrictive when human preferences are cyclic, non-transitive, or otherwise not representable by a scalar reward. Nash Learning from Human Feedback (NLHF) addresses this limitation by modeling alignment as a preference game and targeting a Nash equilibrium rather than a reward maximizer. However, the learning-theoretic foundations of scalable NLHF remain limited. Existing regret guarantees rely on oracle-based methods that estimate a general preference model and solve KL-regularized minimax problems, while iterative NLHF methods directly optimize policy-level preference losses and are easier to implement but lack regret guarantees. We study online iterative NLHF under general preference models and identify exploration as the key obstacle. First, we show that standard iterative NLHF can suffer an exponential dependence on the KL-regularization parameter, revealing that implicit exploration through policy updates is insufficient for controlling regret. Second, we propose an explicitly exploratory iterative NLHF algorithm that combines SFT-based regularization with adversarial policy exploration. The resulting method retains the direct policy optimization structure of iterative NLHF, avoids explicit preference model estimation, and achieves an $O(\sqrt{T})$ regret bound without an exponential dependence on the KL-regularization parameter. We show that the regret can be improved to $O(\log(T))$ with access to a minimax oracle, clarifying the computational-statistical tradeoff in learning general preference games. Finally, we instantiate our method for LLM fine-tuning and evaluate it on \texttt{Llama-3-8B-Instruct} across multiple benchmarks, where explicit exploration yields consistent improvements over existing NLHF baselines.

---

156. Beyond Access: Guided LLM Scaffolding for Independent Learning in Undergraduate Statistics

• Authors: Mohammad Amanlou , Yasaman Amou-Jafari , Mehrad Livian , Fatemeh Boloukazari , Fereshte Bagheri , Behnam Bahrak
• URL: https://arxiv.org/abs/2606.01375
• Abstract:

  Large language models (LLMs) are increasingly entering students’ learning practices, but their educational value depends on whether they support reasoning or enable task completion without engagement. This study examines guided LLM use in an undergraduate Probability and Statistics course, focusing on the gap between assigned access and actual interaction quality. In a four-week quasi-experimental summer program, students were organized into three balanced conditions: no LLM access, unrestricted LLM access, and guided LLM access. The guided condition used the same LLM platform as the unrestricted condition, but students received explicit training and rules promoting reasoning-focused help-seeking, stepwise hints, verification, and ethical use. All quizzes and the delayed final exam were completed without LLM or external assistance, allowing us to distinguish AI-supported practice performance from independent learning. Results show that guided use was associated with clearer learning-oriented interaction patterns than unrestricted access, especially in prioritizing reasoning over final answers and requesting stepwise support. Guided-LLM students showed stronger no-help quiz performance during the intervention phase, whereas unrestricted access appeared more useful for assisted practice completion than for consistently improving independent performance. Available time measures did not support a simple duration-based explanation, and self-assessment calibration suggested better alignment between perceived and demonstrated understanding in the Guided-LLM condition. Overall, LLM access alone appears to be an incomplete educational intervention. For Artificial Intelligence in Education (AIED), the central design challenge is to scaffold how students use LLMs so that these systems function as partners in reasoning rather than answer-getting tools.

---

157. Needles at Scale: LLM-Assisted Target Selection for Windows Vulnerability Research

• Authors: Michael J. Bommarito II
• URL: https://arxiv.org/abs/2606.01364
• Abstract:

  The attack surface of a modern operating system is a haystack: thousands of signed binaries and millions of functions, almost none relevant to any given vulnerability. A human analyst or an LLM agent must pick the function worth reading before analyzing it. At whole-OS scope, this target selection, not the analysis, is the binding constraint. We present Symbolicate-Enrich-Sample, a low-cost batch pipeline that turns a corpus of production Windows binaries into a queryable, priority-ranked research queue. We (i) recover function-level symbols for stripped vendor binaries by auto-fetching the public symbol files and joining them to a recovered call graph; (ii) attach cheap, deterministic structural features to each named function and, conditioned on those features, use a low-cost language model to assign a reachability tier, a risk level, a bug-class hypothesis, and a rationale; and (iii) draw diverse, prioritized batches via a priority-weighted importance sampler. The contribution is a selection substrate: the prioritization layer a downstream detector or LLM agent runs on top of. Across a whole Windows image of 7,231,419 functions, the labels are markedly selective, and stacking deterministic filters on them leaves a ~22K-function shortlist: the candidate needles, few enough for a human or agent to work through. We characterize the pipeline’s selectivity and its failure modes, describe the methodology, and report aggregate statistics; we withhold the derived dataset for legal and dual-use reasons.

---

158. TukaBench: A Culturally Grounded Jailbreak Benchmark for African Languages

• Authors: Victor Akinode , Senyu Li , Wassim Hamidouche , Waqas Zamir , Inbal Becker-Reshef , David Ifeoluwa Adelani
• URL: https://arxiv.org/abs/2606.01322
• Abstract:

  Safety evaluation of Large Language Models (LLMs) remains heavily English-centric, leaving Low-Resource Languages (LRLs), particularly African ones, critically underexplored. We introduce TUKABENCH, a jailbreak benchmark for seven African languages that extends JailbreakBench (JBB) beyond direct translation through four settings: human translation of JBB prompts, English adaptation to African contexts followed by human translation, human-curated prompts validated through interactions with GPT-5.2, and code-switched prompts combining English and African languages, isolating the effect of language, cultural grounding, and prompt evasiveness on model safety. Across closed and open models, prompting in African languages reduces refusal relative to English, with culturally adapted prompts leading to least refusal. The evaluation also surfaces two structural limitations: model comprehension failures and reduced LLM-as-a-judge reliability in LRLs. To capture the first, we introduce Deflection alongside Refused and Jailbroken; to assess the second, we validate outputs with human annotations, showing that judge-human agreement drops in lower-resource languages and less commonly supported scripts.

---

159. A Communication-Centric 6G-LLM Architecture for Scalable Tactical Autonomous Defense Vehicle Networks

• Authors: Kiran Khurshid , Shumaila Javaid , Nasir Saeed
• URL: https://arxiv.org/abs/2606.01312
• Abstract:

  The integration of Artificial Intelligence (AI) and emerging 6G networks introduces new opportunities for scalable coordination in tactical autonomous vehicle systems. This paper proposes a communication-centric hierarchical architecture for Tactical Autonomous Defense Vehicle Networks (TADVNs) that models the integration of edge-assisted Large Language Model (LLM) reasoning with 6G-enabled connectivity and semantic communication. The framework is designed to improve coordination efficiency, reduce communication overhead, and enhance latency resilience under increasing fleet-scale operation. Unlike conventional task-specific AI pipelines that rely on structured feature processing and rule-based coordination, the proposed approach incorporates semantic abstraction and context-aware decision support within a layered edge-cloud communication architecture. We evaluate communication and coordination performance via Monte Carlo simulations across fleet sizes of 5-30 vehicles under contested network conditions. Results indicate that at a 30-vehicle scale, the 6G-LLM configuration achieves 75.2% latency reduction (29.1 ms vs. 117.5 ms), a 68.7 percentage point increase in mission success rate (82.9% vs. 14.2%), and an 88.6% reduction in communication overhead compared to a 5G-based conventional AI baseline. These findings demonstrate measurable benefits in coordination and communication when semantic reasoning is combined with low-latency 6G connectivity.

---

160. SkillAdaptor: Self-Adapting Skills for LLM Agents from Trajectories

• Authors: Zhuoyun Yu , Xin Xie , Wuguannan Yao , Chenxi Wang , Lei Liang , Xiang Qi , Shumin Deng
• URL: https://arxiv.org/abs/2606.01311
• Abstract:

  Large language model (LLM) agents increasingly rely on reusable external skills to solve long-horizon interactive tasks. Existing training-free skill adaptation pipelines usually update skills from full trajectories or session-level feedback, which makes failure attribution coarse and often produces unstable or overly broad revisions. We propose SkillAdaptor, a training-free step-level skill adaptation framework with explicit failure attribution, and it can plug into OpenClaw-class agent harnesses. Given a failed trajectory, SkillAdaptor identifies a first actionable fault step, links responsibility to candidate skills, and applies targeted updates under explicit acceptance checks while keeping the backbone frozen. We evaluate on WebShop, PinchBench, and Claw-Eval with Kimi-K2.5, GLM-5, and GPT-5.2. SkillAdaptor improves over no-skill and skill-adaptation baselines on all three suites, with the largest single-metric improvements of +1.5 points on PinchBench Avg Score%, +1.8 on Claw-Eval Avg Score, and +1.7 on WebShop success rate. These results indicate that step-level attribution supports more stable and auditable training-free skill maintenance\footnote{The code will be released at this https URL .}.

---

161. Beyond Visual Memory: Mechanistic Diagnostics of Latent Visual Reasoning

• Authors: Garvin Guo , Yu Chen , Xiang Wang , Shuai Li , Xinpei Zhao , Huaxing Liu , Shuai Dong
• URL: https://arxiv.org/abs/2606.01287
• Abstract:

  Recent latent visual reasoning methods achieve substantial gains by inserting continuous latent tokens into multimodal language models. These gains are commonly attributed to the tokens encoding visual evidence; recent analyses, however, reveal a paradox: the tokens are loosely tied to the image and contribute little to the answer. Critically, these analyses treat latent tokens as a single unit, obscuring the true source of the gains. We therefore decompose latent tokens into three testable components: latent slots, boundary markers, and format, and develop a state-of-the-art method as a probe under favorable conditions. Across six method-stage settings and four perception-heavy benchmarks, latent slots fail every prediction of the visual-memory account. Strikingly, retaining only the boundary markers preserves 78 to 100% of the gain in several settings, while the model attends to the image more narrowly at latent positions than at answer positions. The gain therefore comes from boundary markers, format, and this attention pattern, not from latent slots. How each method engages this mechanism depends on its training supervision: at matched accuracy, mechanisms can still differ markedly. Latent visual reasoning thus needs evaluation not only by accuracy but by what the model actually relies on.

---

162. BenchEvolver: Frontier Task Synthesis via Solution-Centric Evolution

• Authors: Yangzhen Wu , Aaron J. Li , Wenjie Ma , Li Cao , Ziheng Zhou , Mert Cemri , Shu Liu , Yuran Xiu , Chenxiao Yan , Haikun Zhao , Bin Yu , Ion Stoica , Dawn Song
• URL: https://arxiv.org/abs/2606.01286
• Abstract:

  The rapid progress of frontier large language models has led to widespread benchmark saturation, limiting the ability of existing datasets to differentiate model capabilities or provide useful training signal. For instance, on LiveCodeBench, frontier models achieve over 99% Pass@1 on easy splits and exceed 90% Pass@1 on average across difficulty levels. Constructing new, challenging datasets typically requires substantial human effort, creating a bottleneck for progress. We introduce BenchEvolver, a solution-centric evolutionary framework that automatically transforms existing coding problems into harder variants. Rather than generating problems from scratch, BenchEvolver evolves reference solutions through structured transformations and derives corresponding statements and tests from the evolved solutions. This design grounds generation in executable semantics, enabling scalable construction of high-quality, diverse, and difficult tasks with verifiable correctness. Applying BenchEvolver to LiveCodeBench and SciCode, we obtain evolved tasks that are substantially harder while maintaining validity, reference correctness, and diversity. We further curate LiveCodeBench-Plus, a 91-problem benchmark combining evolved and difficult original LCB-v6 tasks, where frontier-model Pass@1 ranges from 27.5% to 62.6%, restoring clear discrimination among strong coding models. Importantly, evolved tasks remain challenging even for the model that generates them, enabling self-improvement. We further show that RL on evolved LCB tasks improves held-out coding performance: for gpt-oss-20b, seed+evolved training achieves +8.7 and +8.3 Pass@1 gains on LCB v6 Hard and LCB-Pro Easy, exceeding seed-only gains by 70.7% and 34.8%, respectively. Our results show that BenchEvolver can convert saturated benchmarks into frontier-level evaluation suites and reusable training signal.

---

163. RLVR without Ineffective Samples: Group Prioritized Off-Policy Optimization for LLM Reasoning

• Authors: Yixiu Mao , Yun Qu , Qi Wang , Heming Zou , Xiangyang Ji
• URL: https://arxiv.org/abs/2606.01281
• Abstract:

  Reinforcement learning with verifiable rewards (RLVR) has emerged as a powerful paradigm for enhancing the reasoning capabilities of large language models (LLMs). However, its effectiveness is substantially hindered by the prevalence of ineffective training data: many sampled prompts yield response groups that are either entirely correct or entirely incorrect, resulting in zero-variance rewards and limited learning signals. Recent state-of-the-art methods address this issue through extensive LLM rollouts to filter ineffective samples, but at the cost of considerable computational overhead. Alternative approaches, including predictive sampling and trajectory replay, aim to improve data efficiency but often remain insufficient and may introduce additional issues such as systematic bias or suboptimal constraints. To address these limitations, we propose Group Prioritized Off-Policy Optimization (POPO), a simple yet effective framework that fully exploits effective training batches without additional rollout overhead. POPO comprises two key components: prioritized group replay and decoupled off-policy optimization. The former replaces ineffective on-policy groups with effective off-policy groups via a recency-based replay mechanism that jointly considers sample quality and the degree of off-policiness. To further mitigate the off-policy gap, POPO employs decoupled importance sampling to correct off-policy bias while maintaining stable policy updates under consistent trust-region constraints. Empirical evaluations across diverse reasoning tasks, including mathematics, planning, and visual geometry, demonstrate that POPO substantially accelerates RL finetuning and achieves strong reasoning performance with significantly fewer rollouts.

---

164. IndoBias: A Dual Track Culturally Grounded Benchmark for LLMs Bias Evaluation in Indonesian Languages

• Authors: Ikhlasul Akmal Hanif , Muhammad Falensi Azmi , Filbert Aurelian Tjiaranata , Eryawan Presma Yulianrifat , Fajri Koto
• URL: https://arxiv.org/abs/2606.01260
• Abstract:

  Despite being home to more than 1300 ethnic groups and 700 indigenous languages, bias in Large Language Models has not been fully studied in Indonesia, thus leaving a critical gap in evaluating representational fairness and localized stereotypes within its uniquely vast, multilingual, and diverse sociocultural landscape. To address this, we introduce IndoBias as a culturally-grounded bias benchmark to assess LLMs bias in Indonesian and three local languages: Javanese, Sundanese, and Makasar. IndoBias features dual perspective evaluation tracks: depth-oriented (with contrastive-pairs) and breadth-oriented (with generation-based), where the latter is grounded in social science frameworks (SPI, O*NET, and WGI). Our results show that existing LLMs – particularly decoder models – exhibit strong bias towards prototypical sentences in Indonesian, while local languages suffer higher bias under Ideology and Religion category. We also find that LLMs responses exhibit a non-uniform Stereotype Polarity when prompted with various local entities. Finally, we discover that, in Indonesian, Common Crawl texts introduce more bias during pretraining, compared to human-reviewed article texts (e.g., Wikipedia, News), whereas introducing local languages to pretraining generally increases bias. This work highlights the importance of studying bias in culture-specific context. Warning: This paper contains example data that may be offensive, harmful, or biased.

---

165. Understanding LLM Behavior in Multi-Target Cross-Lingual Summarization

• Authors: Sangwon Ryu , Yihong Liu , Mingyang Wang , Yunsu Kim , Jungseul Ok , Gary Geunbae Lee , Hinrich Schuetze
• URL: https://arxiv.org/abs/2606.01252
• Abstract:

  Multi-target cross-lingual text summarization (MTXLS), which summarizes a source document into multiple target languages, is increasingly important as users consume content in diverse languages, but remains underexplored. To address this gap, we introduce multi-target cross-lingual element-aware (MEA), a new MTXLS benchmark covering 24 target languages. We benchmark end-to-end and pipeline approaches across various LLMs and show that MTXLS performance still substantially lags behind English monolingual summarization. To better understand MTXLS in LLMs, we propose a layer-wise analysis framework for investigating how LLMs internally perform MTXLS. Our analyses suggest that translation and summarization behaviors emerge jointly within later layers rather than as distinctly decomposed stages. Most task-relevant processing occurs within these layers, and errors also tend to arise at similar depths. Motivated by these findings, we introduce an inference-time activation steering method that leverages hidden representations from English summarization to guide MTXLS generation. Experiments show that our method consistently improves MTXLS quality across target languages.

---

166. Distilling Neuro-Symbolic Programs into 3D Multi-modal LLMs

• Authors: Wentao Mo , Yang Liu
• URL: https://arxiv.org/abs/2606.01215
• Abstract:

  Current 3D spatial reasoning methods face a fundamental trade-off: neuro-symbolic 3D (NS3D) concept learners achieve interpretable reasoning through compositional programs but are constrained to closed-set concept vocabularies and simple programs; end-to-end 3D multi-modal LLMs (3D MLLMs) could handle complex natural language and open-vocabulary concepts but suffer from black-box reasoning without explicit spatial verification. We introduce APEIRIA, a neuro-symbolic 3D MLLM to bridge two paradigms by distilling symbolic reasoning patterns into MLLMs with natural language chain-of-thought. Our three-stage curriculum progressively builds reasoning capabilities: a) 3D perception alignment grounds object visual-geometric features to the LLM, b) CoT-SFT teaches query decomposition and stepwise verification from symbolic program traces, and c) CoT-RL extends reasoning patterns to open-set concepts and deeply nested instructions. By transferring reasoning patterns rather than concept-specific knowledge, APEIRIA preserves key NS3D virtues: transparent reasoning and modular interchangeability of planning and perception components. Evaluations on grounding, question answering, and captioning show that APEIRIA surpasses prior NS3D methods and matches state-of-the-art 3D MLLMs on 3D spatial reasoning datasets, unifying symbolic methods’ systematic reasoning with MLLMs’ flexibility. Code is available at this https URL .

---

167. Implicit Geographic Inference in LLM Medical Triage: Language-Driven Disparities in Emergency Recommendations

• Authors: Qi Han Wong
• URL: https://arxiv.org/abs/2606.01204
• Abstract:

  We investigate whether large language models produce different medical triage recommendations for identical symptoms based solely on the language of the patient prompt. Using Gemini 3.5 Flash, we evaluate a neurological symptom profile (persistent headache, blurred vision, nausea) across six languages (English, Spanish, Chinese, Hindi, Japanese, Arabic) with 30 runs per condition (n=450 total API calls). We find that the model recommends emergency room visits at rates ranging from 0% (Japanese, Hindi) to 30% (English, Arabic), despite assigning nearly identical severity scores (7.7-8.0/10) across all languages. Adding a single sentence specifying the patient’s US location increases ER recommendations by up to 76.7 percentage points for non-English prompts, while the reverse anchor (English prompt with a Tokyo location) reduces the ER rate from 30% to 6.7%. A back-translation control (Japanese to English) produces ER rates comparable to the English baseline, confirming that the disparity is not caused by translation quality but by implicit geographic inference from the input language. We release the complete dataset, experiment code, and results.

---

168. CA-BED: Conversation-Aware Bayesian Experimental Design

• Authors: Daniel Arnould , Rashad Aziz , Zixuan Kang , Tanav Changal , Kevin Zhu , Sunishchal Dev , Gabriel Grand , Shreyas Sunil Kulkarni
• URL: https://arxiv.org/abs/2606.01182
• Abstract:

  Large Language Models (LLMs) excel at static reasoning tasks, yet their performance often degrades in interactive scenarios where information must be actively acquired through questioning. A key challenge lies in selecting questions that reduce uncertainty while incorporating responses that may be ambiguous or only partially informative. To address this, we propose Conversation-Aware Bayesian Experimental Design (CA-BED), an inference-time probabilistic dialog planning framework that integrates Bayesian Experimental Design with LLM-based likelihood estimation to optimize question selection over multiple conversational turns. CA-BED maintains a belief distribution over hypotheses, anticipates possible answers, and propagates expected information gain through a simulated conversation tree. Across two structured entity-deduction benchmarks, CA-BED yields an average 21.8% improvement in success rates over direct prompting, with comparable gains relative to alternative information-seeking methods. It achieves these gains with an average increase of only 1.8 conversational turns compared to direct prompting.

---

169. When Data Is Scarce: Scaling Sparse Language Models with Repeated Training

• Authors: Boqian Wu , Qiao Xiao , Patrik Okanovic , Tomasz Sternal , Maurice van Keulen , Mykola Pechenizkiy , Elena Mocanu , Torsten Hoefler , Decebal Constantin Mocanu
• URL: https://arxiv.org/abs/2606.01155
• Abstract:

  Scaling laws for dense LLMs under infinite data are well explored, but how sparsity interacts with limited data is not. In this work, we study sparse training in data-constrained regimes where limited unique tokens require multi-epoch training. Our experiments span models up to 1.92B parameters in the fitting set, sparsity up to 93.75%, unique data budgets up to 2.6B tokens, and total training tokens up to 41.6B over 16 epochs; we further validate extrapolation on held-out dense-equivalent models up to 7.68B parameters. We find that: 1. Sparse scaling in data-limited settings: We introduce a scaling law that models loss as a function of active parameters, unique tokens, data repetition, and sparsity, accurately predicting performance across compute and data budgets. 2. Delayed data saturation: sparse training postpones diminishing returns from repeated data, making multi-epoch training more effective. 3. Resource trade-offs: With fixed data, loss-optimal sparsity is moderate ~ 50%, while compute-optimal sparsity is higher and grows with data scale. Overall, sparsity is not just a tool for efficiency, but a mechanism for improving scaling trade-offs under data scarcity. Our code is available at: this https URL .

---

170. Soft-NBCE: Entropy-Weighted Chunk Fusion for Long-Context

• Authors: Shihao Ji , Mingyu Li , Zihui Song
• URL: https://arxiv.org/abs/2606.01101
• Abstract:

  The quadratic complexity of self-attention remains a bottleneck for Large Language Models (LLMs) processing ultra-long contexts. The Naive Bayes Cognitive Engine (NBCE) parallelizes long-context inference by chunking documents and routing to the lowest-entropy chunk at each decoding step. This hard-selection strategy causes semantic fragmentation during cross-chunk reasoning, as abrupt routing changes between adjacent tokens disrupt the model’s contextual grounding. We present Soft-NBCE, a lightweight extension that replaces discrete chunk selection with soft entropy-weighted chunk fusion. A temperature-scaled Softmax over predictive entropies assigns continuous weights to all chunks, enabling log-space aggregation across chunk-conditioned distributions. To partially compensate for the conditional independence assumption introduced by chunking, we propose Consistency Distillation, a LoRA-based self-distillation that constrains the chunked logit distribution toward a full-context teacher via KL-divergence. On LongBench multi-hop benchmarks, Soft-NBCE with Consistency Distillation improves consistently over NBCE-style baselines (MuSiQue F1: 0.310 vs.\ 0.275 for Vanilla NBCE; HotpotQA F1: 0.479 vs.\ 0.427) while maintaining retrieval accuracy (NIAH-32K: 0.909) at O(L^2/n) peak memory.

---

171. MiCU: End-to-End Smart Home Command Understanding with Large Language Model

• Authors: Haowei Han , Kexin Hu , Weiwei Cai , Debiao Zhang , Bin Qin , Yuxiang Wang , Jiawei Jiang , Xiao Yan , Bo Du
• URL: https://arxiv.org/abs/2606.01099
• Abstract:

  Command understanding systems in smart home ecosystems can automate device control and substantially improve user experience. However, while they perform well on precise utterances (e.g., “turn on the bedroom light”), they struggle with ambiguous or misaligned commands (e.g., “make the bedroom cozy”). Large language models (LLMs) generalize well across various domains and can outperform traditional rule-based systems on such tasks, but their effectiveness is often constrained by scarce domain-specific data, insufficient task-specific adaptation, and high computational costs. In this paper, we propose an automated training data synthesis workflow using user logs and LLMs; then we build MiCU, a domain-specific LLM that excels at command understanding. Specifically, we employ curriculum learning to inject domain knowledge into the base LLM, then we enhance its reasoning ability via cold-start training combined with reinforcement learning (RL) guided by domain-specific thinking rules. Additionally, we introduce a token compression technique that condenses device description into a single special token, substantially reducing inference overhead and enabling \model-fast, an efficient variant optimized for long inputs. Extensive experiments show that MiCU significantly outperforms baselines, with an average accuracy gain of 20.01% across all device categories. We have deployed MiCU in the Xiaomi Home app, receiving approximately 1.7 million page views per day. Production evaluations show that MiCU reduces user correction rate by 1.57% and increases human audited accuracy by 32.05%. Our data and code are available at this https URL

---

172. ThinkSwitch: Context Distillation with LoRA and Weight Interpolation for Specific-Purpose Reasoning Tasks

• Authors: Dhruv Saini , Rohan Pandey
• URL: https://arxiv.org/abs/2606.01080
• Abstract:

  Large language models often improve on difficult tasks by spending inference-time compute on a reasoning trace before producing the final answer. That extra computation can be useful, but it also raises latency, token cost, and deployment complexity. We introduce \textbf{ThinkSwitch}, a low-compute procedure for co-training paired instruct and thinking checkpoints. Starting from compatible Qwen3-4B instruct and thinking models, each iteration asks the thinking checkpoint to generate answers, removes the reasoning trace, distills the answer-only pairs into the instruct checkpoint with QLoRA, and reconstructs a thinking checkpoint with spherical weight interpolation. The only human-supplied inputs are task prompts; the labels are generated by the model itself. On a 30-question AIME 2026 evaluation, ThinkSwitch improves the instruct checkpoint from 10/30 to 20/30 and the thinking checkpoint from 14/30 to 22/30. On a 30-question PubMedQA subset, it improves the instruct checkpoint from 13/30 to 18/30 and the thinking checkpoint from 18/30 to 25/30. The complete experiment uses 15 training prompts per domain and costs $2.86 on a single cloud RTX 3070. The results are small-scale, but they indicate that targeted distillation loops can move part of the benefit of explicit reasoning into weights while preserving a separate thinking mode.

---

173. MENTIS: What Belief Changes Under Alignment? Measuring Multi-Scale Latent Torsion in Language Models

• Authors: Partha Pratim Saha , Samarth Raina , Mayur Parvatikar , Amit Dhanda , Vinija Jain , Aman Chadha , Amitava Das
• URL: https://arxiv.org/abs/2606.01060
• Abstract:

  Preference alignment has substantially improved the observable behavior of large language models, yet it remains unclear what alignment changes internally. Aligned systems still fail under jailbreaks, prompt injection, and retrieval-time corruption, suggesting behavior-level evaluation alone is incomplete. Post-training should leave measurable traces in internal computation. We ask: when an instruction-tuned (IT) model becomes a preference-aligned (PA) model, what geometric structure changes, where do those changes concentrate, and how selectively do they vary across concepts, prompts, and model families? We introduce MENTIS, a geometry-first framework for measuring alignment-induced internal reorganization in paired checkpoints. MENTIS compares IT and PA models using a primary layerwise covariance-based torsion norm (T1), a secondary spectral torsion diagnostic (T2), and an Energy-Radiance-Activation measure (ERA) for depth localization. Across four 7-8B model pairs on LITMUS, our study reveals that alignment-induced change is selective rather than uniform: normative concepts exhibit larger torsion shifts than factual concepts on average; torsion is negatively correlated with contextual entropy; and peak effects localize to architecture-specific mid-to-late layers. The same pattern appears across word-level, prompt-level, and model-level analyses. These results suggest preference alignment leaves structured, depth-localized geometric signatures in internal computation beyond what behavior-level evaluation alone can reveal.

---

174. 3DCodeBench: Benchmarking Agentic Procedural 3D Modeling Via Code

• Authors: Yipeng Gao , Lei Shu , Genzhi Ye , Xi Xiong , Ameesh Makadia , Meiqi Guo , Laurent Itti , Jindong Chen
• URL: https://arxiv.org/abs/2606.01057
• Abstract:

  Procedural 3D modeling through code is emerging as a versatile paradigm, offering deterministic, engine-ready, and precisely editable assets that neural 3D generators inherently lack. Authoring such procedural content, however, demands deep expertise in 3D software APIs, parametric design, and code-level geometric reasoning. In this paper, we propose 3DCodeBench, a systematic benchmark for evaluating vision-language model (VLM) agents for procedural 3D generation in 3D modeling software. Specifically, 3DCodeBench evaluates how effectively 12 advanced VLMs can serve as procedural 3D modelers by translating text and image references into procedural code for 3D modeling software. Recognizing that automated metrics may not fully capture the perceptual quality of 3D shapes, we build 3DCodeArena, a ranking platform based on pairwise human preferences over generated 3D outputs. From extensive evaluations and results, we observe that: (1) Failures mostly arise from API mismatches, while successful renders still suffer from disconnected or floating 3D geometric components. (2) Test-time scaling, such as higher thinking budgets and multi-turn refinement, improves performance overall. Our findings highlight a critical need for high-quality procedural coding data to advance commercial VLMs. Furthermore, effective procedural 3D modeling requires a robust execution environment that provides high-fidelity feedback for iterative refinement. We release 3DCodeBench, including the curated large-scale dataset of multimodal (text/image) prompts, procedural code, 3D object triplets, evaluation protocol, and the public 3DCodeArena platform as a foundational toolkit for exploring VLM-based procedural 3D modelers.

---

175. Plausibility Is Not Prediction: Contrastive Evidence for LLM-Based Cellular Perturbation Reasoning

• Authors: Xinyu Yuan , Xixian Liu , Jianan Zhao , Yashi Zhang , Hongyu Guo , Jian Tang
• URL: https://arxiv.org/abs/2606.01042
• Abstract:

  Perturbation experiments are central to understanding cellular mechanisms, but remain costly and sparse, motivating prediction of gene expression responses for unobserved conditions. A promising recent direction leverages large language models (LLMs) as “virtual cell” simulators-using stepwise, knowledge-grounded mechanistic reasoning to infer differential expression-pointing toward an interpretable, knowledge-driven paradigm that transcends purely data-driven approaches. However, we find that plausibility is not prediction: despite producing biologically plausible explanations, these methods fail to capture perturbation-specific effects: systematically overestimating differential expression, often underperforming a simple gene-frequency baseline in aggregate evaluations, and collapsing to chance-level performance at the per-gene level. This reveals a reliance on intrinsic gene response tendencies rather than true perturbation reasoning. We trace this failure to how evidence is presented: existing methods evaluate perturbation-gene pairs in isolation, without exposing how related perturbations differ in their effects on the same gene. To address this limitation, we introduce CORE (Contrastive Organization of Relational Evidence), which reframes prediction as a comparison task by organizing evidence into positive and negative outcomes from related perturbations. Using a biomedical knowledge graph for evidence retrieval, CORE improves calibration and substantially boosts perturbation-specific prediction in both LLM-based and non-LLM settings: for example, on drug-perturbation data, CORE-Reasoning improves Qwen3.5-9B aggregate metrics by up to 28.6%, while on generic perturbation data, CORE-Voting raises macro-per-gene AUROC from chance to 0.703 in average across four cell lines. This highlights contrastive evidence organization as essential to reliable LLM-based perturbation reasoning

---

176. OPD+: Rethinking the Advantage Design for On-Policy Distillation

• Authors: Hanyang Zhao , Haoxian Chen , Han Lin , Genta Indra Winata , David Yao , Wenpin Tang
• URL: https://arxiv.org/abs/2606.01039
• Abstract:

  On-policy distillation (OPD) is a widely used technique to transfer capabilities from capable teacher language models to the base student models, and can be formulated in a reinforcement learning style objective using student generated rollouts. Yet, despite the divergence reward being dependent on student model likelihood, existing works usually adopt a stop gradient design primarily for stability, which makes the resulting advantage estimation questionable. In this work, we provide a generic optimization framework based on f-divergence between the student and teacher, and mathematically revisit whether such design space is valid. We prove that general stop-gradient operation would lead to biased estimates of the reward objective and corresponding gradient for general divergence functions. We propose OPD+, the corrected version of OPD that demonstrates improved performance over the baseline KL approach and also supports the choice of various f-divergence. We validate our findings on mathematical reasoning and tool-use benchmarks.

---

177. DSL-LLaDA: Scaling Continuous Denoising to 8B Masked Diffusion LMs

• Authors: Longxuan Yu , Yunshu Wu , Yu Fu , Siheng Xiong , Rob Brekelmans , Hui Liu , Yue Dong , Greg Ver Steeg
• URL: https://arxiv.org/abs/2606.01024
• Abstract:

  Discrete Masked diffusion language models generate text by iterative parallel decoding, but few-step decoding suffers from a tradeoff between length and quality: with a fixed step budget, standard methods can generate a short, high-quality output, or they can produce long but repetitive text. Continuous denoising can sidestep this tradeoff by evolving all positions jointly in embedding space, but building such a model from scratch at scale remains an open problem. We show that a pretrained masked DLM can instead be lightly adapted to support continuous embedding-space denoising. Starting from LLaDA-8B-Instruct, we continue-pretrain for only 1,000 steps with Discrete Stochastic Localization (DSL), replacing binary masking with continuous per-token Gaussian noise as a soft mask. The adapted model supports continuous inference that evolves all positions jointly in embedding space and defers hard token commitment to the final step. On zero-shot summarization at low step budgets (<=16 forward passes), DSL-LLaDA-SDE achieves the best ROUGE-1 on all four benchmarks and largely avoids the premature-termination / repetition tradeoff of iterative unmasking. The same adaptation also yields selective noisy-state robustness: the model corrects corrupted tokens while preserving clean ones. Control experiments using standard masked diffusion training with the same compute demonstrate neither behavior.

---

178. ProductWebGen: Benchmarking Multimodal Product Webpage Generation

• Authors: Zhihong Liu , Siqi Kou , Zheng Li , Ye Ma , Quan Chen , Peng Jiang , Kai Yu , Zhijie Deng
• URL: https://arxiv.org/abs/2606.01022
• Abstract:

  Crafting a product display webpage from a source product image, along with layout and visual content instructions, holds significant practical value for domains such as marketing, advertising, and E-commerce. Intuitively, this task demands strict visual consistency across product displays and high-fidelity instruction following to jointly generate renderable HTML code. These requirements on controllability and instruction-following are closely aligned with the core features of advanced multimodal generative models, such as image editing models and unified models. To this end, this paper introduces ProductWebGen to systematically benchmark the product webpage generation capacities of these models. We organize ProductWebGen with 500 test samples covering 13 product categories; each sample consists of a source image, a visual content instruction, and a webpage instruction. The task is to generate a product showcase webpage including multiple consistent images in accordance with the source image and instructions. Given the mixed-modality input-output nature of the task, we design and systematically compare two workflows for evaluation – one uses large language models and image editing models to separately generate HTML code and images (editing-based), while the other relies on a single UM to generate both, with image generation conditioned on the preceding multimodal context (UM-based). Empirical results show that editing-based approaches achieve leading results in webpage instruction following and content appeal, while UM-based ones may display more advantages in fulfilling visual content instructions. We also construct a supervised fine-tuning dataset, ProductWebGen-1k, with 1,000 groups of real product images and LLM-generated HTML code. We verify its effectiveness on the open-source UM BAGEL. The data and code are available at this https URL .

---

179. Hybrid Verified Decoding: Learning to Allocate Verification in Speculative Decoding

• Authors: Xin Su , Dawid Majchrowski , Fangyuan Yu , Vanshil Atul Shah , Sebastian Rogawski , Pawel Morkisz , Anahita Bhiwandiwalla , Phillip Howard
• URL: https://arxiv.org/abs/2606.01019
• Abstract:

  Large Language Model (LLM) generation remains expensive because autoregressive decoding calls the model once for each new token. Speculative decoding reduces this cost by drafting multiple tokens and verifying them with the target model in one step, but its speedup depends on how many drafted tokens are accepted. Parameter-free draft sources can propose long continuations at low cost in structured and agentic workloads, yet a cache match that looks promising at one generation step may have low payoff at the next. We propose Hybrid Verified Decoding, which predicts the accepted length of a cache draft before verification and uses this payoff estimate to choose between cache verification and a model-based drafter. Across three LLMs and sixteen datasets, Hybrid Verified Decoding is especially effective on agentic workflows, where it outperforms EAGLE3 in every setting with a 2.73x average speedup. Our analysis shows how prompt structure creates cache opportunities, how high-payoff cache drafts concentrate in a small part of the draft space, and how payoff-guided selection reduces sequential decoding work, pointing to runtime draft selection as a promising direction for speculative decoding.

---

180. PolySpeech-100: A Large-Scale Benchmark for Speech Understanding Across 100+ Languages and Dialects

• Authors: Sicheng Yang , Shulan Ruan , Shiwei Wu , Yu Liu , Lu Fan , Zhi Li , You He
• URL: https://arxiv.org/abs/2606.01016
• Abstract:

  While End-to-End (E2E) Speech-Large Language Models (Speech-LLMs) are rapidly evolving, their evaluation methodologies remain limited to the era of simple transcription. Existing benchmarks suffer from three critical limitations: a pronounced bias towards high-resource languages, a focus on low-level recognition (ASR) rather than semantic reasoning, and a neglect of regional dialects. To bridge this gap, we introduce PolySpeech-100, a massive-scale benchmark designed to assess `native-level’ speech comprehension across 110 linguistic variants. We employ a novel hybrid construction pipeline that augments gold-standard human recordings with instruction-driven synthetic speech, allowing us to cover 19 distinct Chinese dialects and over 80 low-resource languages. Extensive evaluation of 22 state-of-the-art models (including Gemini-3, GPT-Audio, and Qwen2.5-Omni) yields pivotal insights. First, we demonstrate that open-source E2E models outperform Cascade (ASR+LLM) systems on heavy dialects, proving that direct audio processing preserves critical paralinguistic cues and prosodic features (e.g., intonation, stress) that are often lost in standard transcription. Second, we reveal a significant performance gap: while commercial models maintain robustness, open-source models suffer catastrophic degradation on low-resource languages. Finally, counter-intuitively, we observe that under standard zero-shot settings, Chain-of-Thought prompting frequently degrades speech understanding performance for most evaluated models, revealing a potential modality alignment gap in current architectures. PolySpeech-100 establishes a rigorous standard for the next generation of inclusive, omni-capable Speech-LLMs. The data, demo, and code are publicly available at this https URL .

---

181. AI-IoT-Robotics Integration: Survey of Frameworks, Emerging Trends, and the Path Toward Connected Robotics

• Authors: Ranulfo Bezerra , Satoshi Tadokoro , Kazunori Ohno
• URL: https://arxiv.org/abs/2606.01015
• Abstract:

  The convergence of Artificial Intelligence, the Internet of Things, and Robotics is no longer a futuristic vision; it is rapidly becoming the foundation of real-time, intelligent, and context-aware systems. AI enables perception and reasoning, IoT provides scalable sensing and communication, and robotics delivers embodied actuation. Despite significant progress in pairwise combinations such as AIoT and the Internet of Robotic Things (IoRT), there remains a lack of unified design frameworks that fully integrate all three. This survey synthesizes the state-of-the-art across these domains, emphasizing the emerging role of Small Language Models (SLMs) at the edge and Large Language Models (LLMs) in the cloud for distributed cognition and autonomous decision-making. We propose a modular system architecture that aligns with these trends, analyze persistent gaps in interoperability and feedback control, and classify existing work by integration depth. Our review highlights how hybrid SLM-LLM systems, when coupled with IoT infrastructure and robotic agents, can address challenges in real-time adaptation, scalability, and reliability. This work offers a conceptual and technical roadmap for designing next-generation AI-IoT-Robotic ecosystems that are modular, interpretable, and capable of learning within dynamic environments, paving the way for the emerging paradigm of Connected Robotics and Physical AI.

---

182. FVSpec: Real-World Property-Based Tests as Lean Challenges

• Authors: Quinn Dougherty , Max von Hippel , Hazel Shackleton , Mike Dodds
• URL: https://arxiv.org/abs/2606.01008
• Abstract:

  We present a benchmark for evaluating AI models and agents on real-world formal software verification tasks. We first scrape 11,039 property-based tests (PBTs) from real-world Python repositories, then automatically translate 2,772 of them (25%) into 9,415 Lean 4 specifications with sorry placeholders (about 3 formalizations/PBT; we retain multiple attempts when none dominates on quality metrics). Translating PBTs into Lean specifications is challenging: it requires modeling Python semantics in Lean, inferring the logical property encoded in an imperative PBT, and handling the inherent difficulties of dependently-typed programming in a seldom-used language. We describe a three-agent LLM pipeline for transpiling PBTs into Lean specifications, evaluate coverage and quality metrics, and provide baselines for proof generation using several automated and model based approaches. All code (scraper and agents) and data (PBTs and Lean specifications) are open source. Our benchmark aims to drive progress on the underexplored problem of AI-assisted formal verification of real-world software, which is of increasing interest as AI produces more and more of the world’s code.

---

183. An Open-Source Benchmark and Baseline for Multi-temporal Referring Segmentation

• Authors: Bingyu Li , Da Zhang , Tao Huo , Zhiyuan Zhao , Junyu Gao , Xuelong Li
• URL: https://arxiv.org/abs/2606.00987
• Abstract:

  Large Vision-Language Models (LVLMs) have shown strong visual understanding and language-guided grounding abilities, yet their capacity for multi-temporal visual reasoning remains underexplored. To bridge this gap, we introduce \textbf{Multi-temporal Referring Segmentation (MTRS)}, a new task that aims to segment language-described temporal changes from multi-temporal images. MTRS extends conventional referring segmentation and change detection by jointly requiring temporal correspondence reasoning, language grounding, and pixel-level mask prediction. We propose \textbf{CRAFT-Agent}, an automated data construction pipeline with human auditing, and build \textbf{MTRefSeg-21K}, the first MTRS benchmark, containing 21K high-quality multi-temporal image-text-mask triplets across diverse scenes, viewpoints, and domains. Benchmarking a broad set of VLM- and LVLM-based models reveals that direct inference performs poorly, while task-specific fine-tuning remains limited. To address this, we propose \textbf{MTRefSeg-R1}, a change-aware LVLM framework trained with a two-stage strategy. It first learns general temporal-change perception from 20K vision-only bi-temporal samples, and is then fine-tuned on MTRefSeg-21K for fine-grained language-guided temporal localization. MTRefSeg-R1 explicitly models cross-temporal visual differences, aligns language instructions with temporal variations, and predicts referred change masks. Extensive experiments show that MTRefSeg-R1 achieves strong and often superior performance compared with existing LVLM baselines, demonstrating the challenge and potential of MTRS.

---

184. Lodestar: An Online-Learning LLM Inference Router

• Authors: Gangmuk Lim , Wanyu Zhao , Brighten Godfrey , Jiaxin Shan , Le Xu , Liguang Xie
• URL: https://arxiv.org/abs/2606.00946
• Abstract:

  Efficiently serving large language model (LLM) inference tasks is crucial both for user-perceived latency such as time-to-first-token (TTFT) and for GPU utilization. However, LLM request routing, that is, assigning each inference request to a GPU instance, is particularly challenging: execution is highly input-dependent; batching and KV-cache reuse create strong cross-request coupling; and latency responds nonlinearly to context length, model/engine settings, and heterogeneous accelerators. As a result, simple traditional load balancing algorithms, and even heuristics tailored for LLM inference, fail to achieve good performance. We present Lodestar, a novel learning-based request routing system for distributed GPU clusters. Lodestar continuously collects a snapshot of the cluster at per-request level, including real-time instance state, request characteristics, and observed performance, and trains an online reward predictor that it uses to route inference requests to the instance that will maximize given reward (e.g., minimizing TTFT). Lodestar is cloud-native and works seamlessly with existing serving stacks (vLLM). With continuous online adaptation to changing workloads and infrastructure conditions, Lodestar achieves 1.41x lower average TTFT and 1.47x lower P99 TTFT on average (up to 2.15x/1.86x on homogeneous and 4.38x/4.42x on heterogeneous clusters) compared to a state-of-the-art prefix cache and load-aware heuristic, and learns these efficient routing strategies within about 5 minutes, based on experiments in a public cloud GPU cluster.

---

185. Accuracy, Stability, and Repeated-Run Reliability of Large Language Models on Deterministic Programming Tasks

• Authors: Yongxi Zhou , Lai Yun Choi , Jiaxi Wen , Wenbo Ye
• URL: https://arxiv.org/abs/2606.00920
• Abstract:

  Run-level pass rate overstates retry-free coverage by up to 17.8 percentage points – and the gap is largest precisely for mid-performing systems. We investigate this accuracy–stability relationship in large language model (LLM) evaluation for deterministic text-conditioned generation, using programming tasks as a concrete testbed. Standard code-generation benchmarks emphasize single-run accuracy or eventual success under repeated sampling, but many deployment settings also require stability: consistent outcomes across repeated invocations under the same task description. We present a repeated-run evaluation protocol with metrics for run-level accuracy, retry-free coverage, and per-problem variability. On a recency-based benchmark of 100 LeetCode-style problems, we evaluate 16 models from five provider families under two prompt templates with five repeated runs per problem, yielding 16,000 evaluation instances. Although run-level pass rate and perfect stability rate are strongly correlated (r=0.985), pass rate consistently exceeds retry-free coverage – a gap that reaches 17.8 percentage points and reverses model rankings even among closely matched systems. Prompt effects are model-dependent rather than uniformly beneficial. These results suggest that repeated-run stability analysis is a necessary complement to conventional accuracy reporting for deterministic text-conditioned generation tasks.

---

186. MLLM-Microscope: Unlocking Hidden Structure Within Multimodal Large Language Models

• Authors: Ravil Mussabayev , Rustam Mussabayev
• URL: https://arxiv.org/abs/2606.00909
• Abstract:

  This work presents MLLM-Microscope, a novel system designed for analyzing the hidden representations within Multimodal Large Language Models (MLLMs). Our system evaluates the linearity, intrinsic dimension, and anisotropy of multimodal token embeddings across transformer layers. Utilizing the ScienceQA dataset, we evaluate two state-of-the-art MLLMs, LLaVA-NeXT and OmniFusion. We find that both the main and residual streams for tokens of both modalities exhibit highly linear behaviors across transformer layers. However, LLaVA-NeXT’s image tokens reveal a slight decline in linearity, whereas OmniFusion’s remain consistent. Image token dimensions in OmniFusion remain consistently higher across layers compared to LLaVA-NeXT. Also, the OmniFusion’s anisotropy is observed to stay consistently low throughout the layers. These findings suggest that the inner workings of MLLMs highly depend on the nature of modality fusion performed before passing the token sequence into LLM. This and other new potential insights obtainable from our system are surely capable of enhancing our understanding of the inner workings of MLLMs, informing future model design and optimization.

---

187. Memory-Efficient LLM Training with Dynamic Sparsity: From Stability to Practical Scaling

• Authors: Qiao Xiao , Boqian Wu , Patrik Okanovic , Tomasz Sternal , Maurice van Keulen , Elena Mocanu , Mykola Pechenizkiy , Decebal Constantin Mocanu , Torsten Hoefler
• URL: https://arxiv.org/abs/2606.00888
• Abstract:

  Dynamic Sparse Training (DST) offers a promising paradigm for improving the training and inference efficiency of deep neural networks; however, we find that in large language model training, DST can suffer from optimization instability, manifested as loss spikes after topology updates. In this work, we show that the naive use of standard Adam-based optimizers leads to a cold-start issue for newly regrown parameters, resulting in excessively large updates and disrupted training dynamics. To address this issue, we propose Sparse Memory-Efficient Training (SMET), which stabilizes DST with optimizer warm-up and improves training progress through density-aware learning-rate scaling. SMET further reduces memory consumption by storing gradients and optimizer states only for active parameters. We provide a theoretical analysis of the update behaviors under SMET, showing improved optimization stability. Extensive experiments demonstrate that SMET enables stable, scalable, and memory-efficient sparse pre-training of LLMs, paving the way for sparse training as a practical alternative to dense training. Our code is publicly available at: this https URL .

---

188. Benchmarks for Vision-Language Models in Urban Perception Should Be Reliability-Aware and Negotiated

• Authors: Rashid Mushkani
• URL: https://arxiv.org/abs/2606.00871
• Abstract:

  Vision-language models (VLMs) are increasingly used to generate structured descriptions of street-level imagery for tasks such as streetscape auditing, mapping, and public consultation. These uses combine observable attributes with appraisal categories, and the human targets are often distributions of judgments with disagreement and explicit non-response. This paper argues that benchmarking VLMs for urban perception should treat disagreement and abstention as measurement outcomes, report inter-annotator reliability alongside model alignment, and treat the label space and scoring policy as negotiable artifacts when outputs are intended to inform urban governance. We ground the argument in a benchmark of 100 Montreal street scenes annotated along 30 dimensions by 12 participants from seven community organizations, and in a deterministic zero-shot evaluation of seven VLMs. Across dimensions, model agreement with human consensus co-varies with dimension-level human reliability, and for the appraisal dimension Overall Impression models and annotators exhibit distributional mismatch including different rates of Not applicable. We close with actions for benchmark creators, model developers, and institutions to make uncertainty and benchmark assumptions visible in evaluation reports.

---

189. GenPT: Beyond Self-Report for Reliable LLM Psychometrics via Generative Projective Testing

• Authors: Ming Wang , Shuang Wu , Bixuan Wang , Lu Lin , Yuxin Chen , Xiaocui Yang , Daling Wang , Shi Feng , Yifei Zhang , Yufan Sun
• URL: https://arxiv.org/abs/2606.00860
• Abstract:

  Self-report questionnaires remain the prevailing tool for probing the psychological states of persona-conditioned agents (PC-Agents). However, classical instruments inherit two well-known threats: contamination from training corpora and directional bias driven by social-desirability or contextual framing. To overcome these methodological bottlenecks, we ask whether projective paradigms can be adapted into a robust psychometric tool. We introduce \textbf{GenPT} (Generative Projective Testing), which reformulates TAT, Rorschach, and SCT with newly generated stimuli and organizes assessment as a three-stage pipeline to derive standardized psychological indicators and target states. Evaluating PC-Agents induced via CharacterRAG and AnnaAgent profiles, we benchmark GenPT’s reliability and validity against classical questionnaires. The results indicate that questionnaires exhibit systematic directional shifts under social-desirability framing, most strongly on suicide ideation. In contrast, GenPT’s collected behavioral patterns stay near the symmetric baseline. Furthermore, under a longitudinal counselling context, GenPT-based depression assessment shifts by roughly an order of magnitude more than the questionnaire counterpart when Qwen3 serves as the backbone. Overall, GenPT complements self-report methods in scenarios where contamination resistance, bias asymmetry, and context sensitivity matter. Code and stimuli can be found at this https URL .

---

190. SkillPager: Query-Adaptive Intra-Skill Navigation via Semantic Node Retrieval

• Authors: Zicai Cui , Zihan Guo , Weiwen Liu , Weinan Zhang
• URL: https://arxiv.org/abs/2606.00822
• Abstract:

  Skill-based LLM agents increasingly rely on long procedural documents, but full-document prompting wastes tokens and dilutes information critical to execution. We study this setting as intra-skill retrieval, where the goal is to select a minimal, execution-sufficient context from a known skill document given a query. We present SkillPager, a two-stage framework that parses each Markdown skill into typed semantic nodes offline and leverages Maximal Marginal Relevance (MMR) to perform global, query-conditioned node selection online. On a benchmark of 395 skills and 1,975 queries, SkillPager achieves 78.89% LLM-judged context sufficiency, compared to 82.23% for the exhaustive full-document baseline, while reducing prompt tokens by 47.04%. A granularity ablation shows that applying the same retrieval algorithm to raw fixed-length chunks reaches a comparable 81.77% sufficiency but increases token cost by 28.81%, demonstrating that efficiency gains are driven by typed semantic granularity rather than the retrieval algorithm alone. Among graph-based baselines, SkillPager outperforms the strongest baseline by a margin of 12.16%. Further ablations show that supporting content is most effective when retained in the candidate pool and selected adaptively rather than removed by static heuristics. These results identify typed intra-document retrieval as a distinct access problem for skill-based agents.

---

191. Certificates without Electrons? Theory and Evidence on Impacts from AI-Driven Power Demand

• Authors: Dana Golden , Aruna Balasubramanian , Niranjan Balasubramanian
• URL: https://arxiv.org/abs/2606.00811
• Abstract:

  Data centers now account for 4.4% of United States electricity demand, yet the grid-level effectiveness of the renewable energy certificates (RECs) and power purchase agreements (PPAs) hyperscalers use to claim carbon neutrality remains unclear. We develop a game-theoretic model in which a data center operator chooses among RECs, PPAs, and behind-the-meter colocation while generators make entry decisions under endogenous financing costs. The model identifies a timing wedge – the mismatch between consumption and credited renewable generation – as a central mechanism through which AI demand degrades reliability, raises prices, and increases emissions even when RECs cover 100% of annual consumption. Colocation with storage addresses this wedge directly and induces the greatest renewable entry by eliminating generator revenue risk. We test these predictions by exploiting the staggered release of large language models as a natural experiment, using difference-in-differences on a novel dataset linking AI activity to local grid outcomes. AI demand significantly increases fossil generation, wholesale prices (up to 25% in treated PJM zones), and outage frequency (0.5–1 additional outages per year) near data centers, with impacts scaling in model size. Data centers with on-site generation exhibit a sign reversal in power-quality effects, consistent with the model’s prediction that behind-the-meter capacity absorbs demand spikes. Counterfactual analyses show that edge inference, spatial reallocation, and colocated storage each substantially mitigate grid impacts, while REC-only strategies do not. Together, our results demonstrate that the externalities of AI to the grid are tightly coupled to procurement design and the spatial organization of data center infrastructure.

---

192. WaveFilter: Enhancing the Long-Context Capability of Diffusion LLMs via Wavelet-Guided KV Cache Filtering

• Authors: Jinnan Yang , Yan Wang , Zhen Bi , Kehao Wu , Xiaojie Li , Jungang Lou , Zechao Li , Jing Liu
• URL: https://arxiv.org/abs/2606.00724
• Abstract:

  Diffusion Large Language Models (DLMs) have demonstrated significant advantages across various tasks. However, constrained by their multi-step iterative inference mechanism, their computational overhead and inference latency in long-context tasks have become core bottlenecks restricting their large-scale deployment. When processing long sequences, existing Key-Value (KV) caching mechanisms often face a dilemma where generation quality degrades drastically, where the core challenge lies in precisely and efficiently filtering critical tokens within ultra-long contexts. Inspired by the human reading process, we propose \textbf{WaveFilter}, a universal and training-free caching framework. This framework innovatively introduces the wavelet transform for decomposition of long sequences to achieve precise identification of key tokens, based on which a sparse KV Cache is constructed to compute the final contextual representation. Experimental results demonstrate that WaveFilter, as a plug-and-play generic framework, significantly enhances the performance of existing mainstream KV Cache methods in complex long-context tasks.

---

193. EPIC: Efficient and Parallel Inference under CFG Constraints for Diffusion Language Models

• Authors: Hyundong Jin , Yo-Sub Han
• URL: https://arxiv.org/abs/2606.00722
• Abstract:

  Controlling language model outputs is essential for ensuring structural validity, reliability, and downstream usability, and diffusion language models are no exception. Recent advances in diffusion language model decoding have extended output control beyond regular constraints to context-free grammar (CFG) constraints. Existing methods, however, can be up to four times slower than unconstrained decoding. More importantly, they substantially diminish one of the key advantages of diffusion language models over autoregressive models, namely parallel decoding. This slowdown arises because sequential validity checking introduces significant overhead during parallel generation. We propose an efficient CFG-constrained decoding framework, EPIC, that addresses this limitation. Our method improves decoding efficiency by combining lexing memoization, validation using Earley-style parsing instead of deterministic automata, and relaxed compatible subset selection for parallel commit. It reduces repeated lexing and validation overhead while allowing multiple compatible tokens to be committed together. Experiments on three benchmarks using four models show that our method reduces inference time by up to 67.5% and decreases the additional overhead by up to 90.5% compared with existing CFG-constrained decoding methods. Our implementation is available at this https URL .

---

194. Information-Theoretic Lower Bounds for Bit-Constrained Stochastic Optimization via a Reduction to Compressed Gaussian Mean Estimation

• Authors: Munsik Kim
• URL: https://arxiv.org/abs/2606.00703
• Abstract:

  Low-precision pretraining (FP8, MXFP4, NVFP4) is now standard for frontier language models, yet the literature is almost entirely achievability – algorithms and empirical scaling laws – with no matching characterization of what is information-theoretically possible. We study a B-bit quantized stochastic first-order oracle: an optimizer interacts for T rounds and receives, each round, a B-bit adaptive public-coin description of its stochastic gradient. Our main contribution is an exact reduction from optimizing a strongly convex quadratic family to interactively compressed Gaussian mean estimation – under the B-bit oracle the query carries no information, so optimization collapses exactly onto a sequential distributed-estimation problem. This yields two unconditional lower bounds, a communication bound TB = Omega(d) and a statistical bound T = Omega(sigma^2 d / eps^2), and the sharp product-form bound T = Omega((sigma^2 d / eps^2) max{1, d/B}). The product form is also unconditional: a B-bit transcript carries at most O(TB / sigma^2) of Fisher trace about the mean, so bits rather than dimension limit the recoverable information, and combined with the multivariate van Trees inequality this gives the bound directly, without bounded-likelihood-ratio truncation. We give a near-matching achievability result with exact per-round bit accounting under a bounded-dynamic-range oracle, tight up to a logarithmic factor; the lower bound is for truly Gaussian (unbounded) gradients, and closing this oracle gap is left open. A sequential rate-distortion perspective extends the reduction to correlated and drifting oracles and corrects an earlier conjecture: positive noise correlation raises the bound by (1+rho)/(1-rho) rather than relaxing it. The bounds give an information-theoretic baseline for any low-bit gradient path, not an optimality claim about deployed FP4 systems.

---

195. The Paradox of Outcome Optimization: A Causal Information-Theoretic Bound on Reasoning Shortcuts in LLMs

• Authors: Zihan Chen , Yiming Zhang , Wenxiang Geng , Zenghui Ding , Yining Sun
• URL: https://arxiv.org/abs/2606.00674
• Abstract:

  Large Language Models (LLMs) aligned via outcome-based Reinforcement Learning (RL) frequently exhibit a critical failure mode: they achieve high performance on in-distribution benchmarks while demonstrating brittle reasoning capabilities on out-of-distribution (OOD) tasks. We term this phenomenon Reward-Induced Manifold Collapse. We establish a theoretical framework bridging Structural Causal Models (SCM) and the Information Bottleneck (IB) principle to explain this paradox. We define reasoning as a high-complexity causal process and shortcut learning as the exploitation of low-complexity spurious correlations. Under the implicit inductive bias of Stochastic Gradient Descent (SGD), models optimized for outcome rewards are biased toward shortcut solutions whenever the training distribution allows for a ``Markovian Screening’’ of the true causal mechanism. We derive a new generalization bound based on Semantic Coverage Measure ($\eta$) rather than sample size, showing why data scaling on homogeneous distributions may fail to correct reasoning flaws. We also show that Process Reward Models (PRMs) function as Topological Filters, enforcing step-wise mutual information constraints that render the low-complexity shortcut manifold inadmissible. These results provide a mathematical grounding for the role of process supervision beyond simple credit assignment.

---

196. Scaling Behavior of Single LLM-Driven Multi-Agent Systems

• Authors: Jialing Li , Zhouhong Gu , Yin Cai , Hongwei Feng
• URL: https://arxiv.org/abs/2606.00655
• Abstract:

  The burgeoning field of LLM-based Multi-Agent Systems (MAS) promises to tackle complex tasks through collaborative intelligence, yet fundamental questions regarding their scaling behavior and intrinsic collective dynamics remain underexplored. This paper systematically investigates how the performance of a homogeneous MAS evolves as the number of agents increases, isolating the variable of collaboration from model or knowledge heterogeneity. We propose the Sequential Iterative Multi-Agent System (SIMAS) framework, a minimalist architecture centered on sequential inter-agent communication, to clearly observe scaling effects. Through extensive experiments across diverse tasks and model scales, we establish that MAS performance does not scale monotonically with agent count but follows a pattern of diminishing returns, governed by a trade-off between collaborative synergy and coordination overhead. Our findings reveal that effective MAS requires a sufficiently capable base LLM, that task type critically modulates the optimal agent count, and that collective intelligence is an emergent property contingent on strategic interaction design rather than a guaranteed outcome of agent plurality. The performance degradation stems coordination overhead rather than merely long-context failure, and the scaling tendency generalizes across interaction architectures like structured debate topologies. This work provides a foundational understanding of MAS scaling laws, offering practical guidance for designing efficient collaborative systems and challenging the prevailing assumption that more agents invariably lead to better performance.

---

197. MESA: Improving MoE Safety Alignment via Decentralized Expertise

• Authors: Yitong Sun , Yao Huang , Teng Li , Ranjie Duan , Yichi Zhang , Xingjun Ma , Hui Xue , Xingxing Wei
• URL: https://arxiv.org/abs/2606.00651
• Abstract:

  Mixture-of-Experts (MoE) architectures scale Large Language Models (LLMs) efficiently, enabling greater capacity with reduced computational cost by dynamically routing inputs to relevant experts, yet introduce a critical vulnerability: Safety Sparsity, where safety capabilities concentrate in few experts, making them susceptible to adversarial bypassing. Meanwhile, conventional alignment methods uniformly adapt all parameters, ignoring their functional differences and inadvertently degrading performances. To address these challenges, we propose MESA (MoE Safety Alignment), a targeted alignment framework for MoE-based LLMs that strategically decentralizes safety responsibility to maximize coverage while minimizing interference with utility. Based on Optimal Transport (OT) theory, MESA operates through two mechanisms: (1) Expert Capacity Reallocation uses a transport cost matrix to distribute safety duties to the most cost-effective experts, and (2) Dynamic Routing Refinement constrains the router to precisely activate these decentralized modules. Experiments show that MESA achieves robust defensive performance against varied harmful benchmarks while preserving helpfulness. Code is available at this https URL .

---

198. Pause and Think: A Dataset and Benchmark for Video-Grounded Assistive Action Suggestion

• Authors: Shivam Singh , Saptarshi Majumdar , Pratik Prabhanjan , Zicheng Liu , Emad Barsoum
• URL: https://arxiv.org/abs/2606.00616
• Abstract:

  Recent Vision-Language Models (VLMs) struggle with grounded reasoning, temporal consistency, and context aware planning in videos. We introduce pause-and-think-T, a reasoning-centric training dataset that encourages models to pause, reason over visual evidence, and produce concise, actionable responses. The dataset promotes structured reasoning prior to answer generation, guiding models toward human-like, scene-grounded assistance. We fine-tune a compact 4B-parameter model and evaluate it on our pause-and-think-B benchmark targeting contextual understanding and goal planning tasks. The model achieves 58.0% accuracy at 59x fewer parameters than Qwen3-VL-235B (58.9%), matching GPT-5.2 on scene understanding and surpassing GPT-4o. Beyond our benchmark, it also shows strong out-of-distribution performance on EgoThink and TempCompass, with substantial gains in affordance, assistance, attribution recognition, situated reasoning, and temporal order, without benchmark-specific training. Our results indicate that targeted reasoning supervision enables compact models to deliver actionable, visually grounded guidance while generalizing beyond training data, without requiring large-scale model expansion.

---

199. Linguistics-Aware Non-Distortionary LLM Watermarking

• Authors: Shinwoo Park , Hyejin Park , Hyeseon An , Yo-Sub Han
• URL: https://arxiv.org/abs/2606.00613
• Abstract:

  Watermarking should identify language-model output without degrading quality or limiting verification to the model provider. Multilingual deployment makes this harder because morphology, segmentation, and script change where watermark evidence can enter naturally. We introduce LUNA, a linguistically adaptive watermark that combines model-free detection with single-token non-distortion under the standard random-key model. LUNA estimates normalized next-tag entropy from part-of-speech contexts in an external corpus and uses it to set the depth of a non-distortionary binary tournament sampler; the detector reconstructs the same schedule from text, a tokenizer, a tagger, and a secret key. We evaluate six typologically diverse languages and two domains against eight primary baselines. LUNA attains an AUROC of 0.9959 and the lowest mean absolute median perplexity shift of 0.045 across the twelve settings; its 95% bootstrap interval [0.022, 0.073] lies below all baseline intervals. LUNA also records the lowest mean Self-BLEU, Distinct-1, surprisal, and entropy shifts. It is the only method that simultaneously achieves AUROC > 0.99 and an absolute median perplexity shift below 0.1 in a majority of settings, reaching this regime in 9 of the 12 settings while no baseline reaches it in more than 2. Our code is available at: this https URL

---

200. MemGraphRAG: Memory-based Multi-Agent System for Graph Retrieval-Augmented Generation

• Authors: Chuanjie Wu , Zhishang Xiang , Yunbo Tang , Zerui Chen , Qinggang Zhang , Jinsong Su
• URL: https://arxiv.org/abs/2606.00610
• Abstract:

  Retrieval-Augmented Generation (RAG) has become an essential method for mitigating hallucinations in Large Language Models (LLMs) by leveraging external knowledge. Although effective for simple queries, traditional RAG struggles with large-scale, unstructured corpora where information is highly fragmented. Graph-based RAG (GraphRAG) incorporates knowledge graphs to capture structural relationships, enabling more comprehensive retrieval for complex reasoning. However, existing GraphRAG methods rely on isolated, fragment-level extraction for graph construction, lacking a global perspective on the whole corpus. As a result, these methods frequently lead to thematically inconsistent, logically conflicting, and structurally fragmented graphs that degrade retrieval performance. In this paper, we propose MemGraphRAG, a novel framework that introduces a memory-based multi-agent system to ensure high-quality graph construction. Specifically, MemGraphRAG employs a collaborative society of agents supported by shared memory, which provides a unified global context throughout the extraction process. This mechanism allows agents to dynamically resolve logical conflicts and maintain structural connectivity throughout the corpus. Furthermore, we propose a memory-aware hierarchical retrieval algorithm tailored for the constructed graph. Extensive experiments on multiple benchmarks demonstrate that MemGraphRAG outperforms the state-of-the-art baseline models with comparable efficiency. Our code is available at this https URL .

---

201. SPADER: Step-wise Peer Advantage with Diversity-Aware Exploration Rewards for Multi-Answer Question Answering

• Authors: Qiming Shi , Zhaolu Kang , Yunfan Zhou , Di Weng , Yingcai Wu
• URL: https://arxiv.org/abs/2606.00593
• Abstract:

  Large language models are increasingly deployed as tool-augmented agents to acquire information beyond parametric knowledge. While recent work has improved long-horizon tool-use reasoning, most approaches focus on tasks with a single correct answer. In contrast, many real-world queries require discovering a comprehensive set of valid answers, a setting known as Multi-Answer QA. This setting raises two challenges: fine-grained credit assignment over long search trajectories and reward alignment for sustained exploration beyond easy high-frequency entities. We propose SPADER, a reinforcement learning framework for long-horizon tool use in Multi-Answer QA. SPADER includes Step-wise Peer Advantage (SPA), a critic-free step-level credit assignment mechanism that aligns parallel trajectories by decision step and estimates advantages from peer returns. It also includes a diversity-aware exploration reward that promotes long-tail entity discovery by upweighting rare findings and downweighting redundant ones. Experiments on QAMPARI, Mintaka, WebQSP, and QUEST show that SPADER generally improves recall and overall F1 over prompting-based agents, outcome-supervised RL methods, and recent step-level supervision approaches. Our code and model weights are available at this https URL .

---

202. Revisiting Parameter-Based Knowledge Editing in Large Language Models: Theoretical Limits and Empirical Evidence

• Authors: Wanying Ren , Xin Song , Futing Wang , Guoxiu He , Aixin Sun
• URL: https://arxiv.org/abs/2606.00570
• Abstract:

  Parameter-based knowledge editing updates the internal knowledge of large language models (LLMs) via localized weight modifications and has attracted significant attention. However, most existing methods overlook fundamental theoretical limitations and are rarely evaluated under realistic, practice-oriented settings. In this paper, we first present a theoretical analysis based on the dimensional Collapse Hypothesis, explaining how localized parameter edits can propagate along fragile directions in the representation space, inducing global interference and ultimately causing reasoning collapse. Building on this insight, we conduct a comprehensive empirical evaluation by systematically varying knowledge complexity, number of edits, evaluation dimensions, and baseline methods. Our results show that parameter-based editing methods consistently damage core LLM capabilities. In contrast, a simple retrieval-based baseline achieves consistently stronger performance than all parameter-editing methods across all evaluated conditions. These findings highlight that preserving the fundamental capabilities of LLMs after knowledge editing should be a central concern for future research.

---

203. CAFOSat: A Strongly Annotated Dataset for Infrastructure-Aware CAFO Mapping Using High-Resolution Imagery

• Authors: Oishee Bintey Hoque , Nibir Chandra Mandal , Mandy L Wilson , Samarth Swarup , Madhav Marathe , Abhijin Adiga
• URL: https://arxiv.org/abs/2606.00548
• Abstract:

  Concentrated Animal Feeding Operations (CAFOs) play an important role in agricultural production but are also associated with environmental, public health, and disease surveillance concerns. Large-scale mapping of CAFOs from remote sensing imagery remains challenging due to heterogeneous infrastructure layouts, noisy location records, inconsistent annotations, and incomplete inventories. We introduce CAFOSat, a strongly annotated, infrastructure-aware dataset for CAFO mapping across the United States. CAFOSat integrates high-resolution National Agriculture Imagery Program (NAIP) imagery with multi-source CAFO inventories collected across multiple states and transforms weak geolocation records into refined annotations through a human-in-the-loop pipeline combining AI-assisted annotation, GradCAM-based localization, and geometric clustering. To improve dataset quality, we curate challenging negative samples using land-cover-guided sampling with spatial exclusion constraints and provide infrastructure-level annotations, including barns, manure ponds, and grazing-related features, through manual verification. The resulting dataset contains more than 45,000 image patches spanning 20 states and four major CAFO categories. We benchmark a diverse set of convolutional, transformer-based, and vision-language models, demonstrating the value of refined annotations and curated negative samples for CAFO classification and generalization. In addition, we introduce a synthetic augmentation pipeline that generates infrastructure-aware variations to increase training diversity and improve robustness under distribution shifts. CAFOSat provides a large-scale benchmark for advancing infrastructure-aware agricultural monitoring and CAFO mapping from high-resolution remote sensing imagery.

---

204. CodeCytos: AI-assisted spatial molecular imaging analysis via code-augmented agent action space

• Authors: Hung Q. Vo , Huy Q. Vo , Son T. Ly , Zhihao Wan , Anh-Vu Nguyen , Hong Zhao , Jianting Sheng , Stephen T. C. Wong , Hien V. Nguyen
• URL: https://arxiv.org/abs/2606.00472
• Abstract:

  Conventional tissue image analysis software provides foundational capabilities for cellular analysis, including segmentation, basic morphological feature extraction, and spatial organization analysis. However, these tools often require manual intervention and are not well integrated with code-driven automation, limiting efficiency and scalability for complex spatial tissue studies. In addition, they offer limited flexibility for custom analyses, as they typically support only a fixed set of pre-implemented spatial cellular features. To address these limitations, we propose CodeCytos, a coding-based reasoning agent framework that enables dynamic, programmable interaction with spatial molecular imaging data to improve automation and customization. CodeCytos is designed to streamline the exploration of custom spatial cellular features and adapt to diverse research needs. We demonstrate its utility through case studies on four expert-curated datasets from distinct tissue types: frontal cortex, non-small-cell lung cancer, pancreas, and tonsil. We evaluate CodeCytos under a realistic minimal prompt setting, where bioscientists pose simple questions without task-specific instructions or contextual information about spatial cellular analysis, and benchmark multiple LLM backbones with strong coding capabilities. We further show that incorporating tailored, domain-agnostic few-shot in-context coding-reasoning examples (randomly sampled demonstrations outside the spatial analysis domain) can substantially improve performance without requiring costly, expert-crafted in-domain demonstrations. Overall, CodeCytos outperforms baseline approaches, highlighting the potential of code-action agents to assist with custom feature exploration in spatial molecular imaging and to accelerate biomarker discovery.

---

205. On the Limits of LLM Adaptability: Impact of Model-Internalized Priors on Annotation Task Performance

• Authors: Etienne Casanova , Rafal Kocielnik , R. Michael Alvarez
• URL: https://arxiv.org/abs/2606.00467
• Abstract:

  Large Language Models (LLMs) are increasingly used for zero-shot annotation and LLM-as-a-judge tasks, yet their reliability hinges on how model-internalized priors interact with user-provided instructions. We investigate three dimensions of this interaction: (1) how an LLM’s familiarity with data and task definitions affects performance, (2) the extent to which additional information in prompts can correct zero-shot errors (“decision stickiness”), and (3) model susceptibility to misaligned task definitions. Through experiments on toxicity detection across diverse datasets (spanning social media, gaming, news, and forums) using both dense and mixture-of-experts models, we find that nearly two-thirds of zero-shot errors are resistant to correction, with an overall rescue rate (fraction of initial errors corrected by prompting) of only 34.8%. High-confidence errors prove especially resistant to correction. When given misaligned definitions, LLMs follow them while maintaining confidence levels unchanged from the aligned condition. Crucially, we introduce Definition-Specific Familiarity (DSF), which measures alignment between a model’s internal concept and the task definition. After controlling for dataset-level confounds, DSF shows a positive association with model performance (partial r = +0.41), while three distinct memorization metrics (ROUGE-L, BERTScore, and embedding cosine similarity) all fail to show a positive association. These findings show the limitations of prompt-based correction in annotation tasks, highlighting the importance of definition alignment over text-level memorization.

---

206. Short-form Text Rewriting with Phi Silica

• Authors: Divya Tadimeti , Shawn Pan , Sameera Lanka , Chenghui Zhou , Sadid Hasan
• URL: https://arxiv.org/abs/2606.00462
• Abstract:

  Short-form text rewriting is a constrained variant of paraphrasing in which limited context and high semantic density leave little room for variation. While large language models perform well on general paraphrasing, small language models (SLMs) often struggle with semantic fidelity and hallucination robustness in short-form settings. In this work, we present an empirical study of adapting an SLM, Phi Silica, for short-form rewrite through dataset curation, prompt distillation, parameter-efficient fine-tuning, and evaluation. We curate a dataset of short presentation-style text from public slide decks and use GPT-5-chat both to generate rewrite supervision and to conduct LLM-as-a-judge evaluation. Our results show that finetuning improves semantic fidelity, reduces hallucinations, and increases preference win rate against GPT-5-chat rewrites. The findings suggest that targeted adaptation for SLMs can substantially narrow the gap to cloud models and provide practical guidance for adapting SLMs to precision-critical rewrite tasks.

---

207. When Safe Skills Collide: Measuring Compositional Risk in Agent Skill Ecosystems

• Authors: Su Wang , Pin Qian , Yihang Chen , Junxian You , Xiaoyuan Wang , Xiaochong Jiang , Lifei Liu , Haoran Yu , Jingzhou Xu
• URL: https://arxiv.org/abs/2606.00448
• Abstract:

  LLM agents increasingly rely on community-contributed skills that expand an agent’s operational capability set. We study a core safety problem in agentic AI systems: whether individually safe skills can compose into unsafe installed skill sets. We present SkillReact, a compositional security measurement framework with three components: a deterministic static-composition benchmark, a two-rater LLM-assisted human-adjudication pipeline, and an action-based exploitability harness. On 1,520 ClawHub skills, 651 pass individual inspection and form 211,575 pairs; the benchmark flags 22.25% of these as structural candidates. We treat this raw rate as a recall-oriented scanner ceiling and calibrate it against human judgment: in a pattern-stratified audit, roughly one in five flagged pair-pattern hits survives as a real compositional risk (population-weighted validity 18.2%, our headline result), implying about 14K genuine risk memberships in a single registry that per-skill scanning misses by construction, since every pair is individually safe. An action-based harness then probes when these candidates become model-issued tool calls, and finds realization gated by host-model disposition: on an anchor-conditioned dropper subset, Haiku-4-5 issues the dropper-stage tool call on all 39 direct-prompt trials (36 of them the full download-then-execute chain, 3 download-only), Opus-4-7 stops at the download, and Sonnet-4-6 refuses outright. A control that holds the request fixed and varies only the installed skills finds compliance highest with no skills installed: a composition fixes which capabilities are reachable, while the host model decides whether to use them. Together these motivate install-time compositional checks and capability isolation as complements to per-skill scanning.

---

208. Detect Before You Leap: Mirage Detection in Vision-Language Models

• Authors: Sayeed Shafayet Chowdhury , Md. Shaown Miah
• URL: https://arxiv.org/abs/2606.00435
• Abstract:

  Vision-language models (VLMs) can produce confident visual answers even when the required visual evidence is missing, blank, or unrelated to the question. This failure mode, known as mirage (Asadi et al. 2026), is especially concerning in medical and document visual question answering, where plausible but visually ungrounded responses may be mistaken for image-based evidence. We study pre-release mirage detection: given an image-question pair, the goal is to determine whether a VLM should answer or abstain before producing a response. We propose Text-Conditioned Layer-wise Internal Alignment (TC-LIA), a model-agnostic method that probes patch-token representations across the layers of a CLIP ViT-H/14 vision encoder. TC-LIA projects layer-wise image patch tokens into the final CLIP embedding space and measures their similarity to the question embedding, allowing the method to track whether question-relevant visual evidence emerges across vision layers. The resulting alignment trajectory is summarized using final image-text cosine similarity, late-layer top-k patch-text alignment, early-to-late gain, and layer-wise slope. These features are combined with pixel-statistic blank/noise detection, zero-shot domain routing, and structured VLM self-assessment in an ensemble. Across five VQA domains, three input conditions, and twelve VLM backbones, the best systems achieve approximately 94.6-94.7% three-class detection accuracy with mirage rates below 3%, while baseline mirage rates range from 21.7% to 66.6%.

---

209. AgentxGCore: Agentic AI for Next-Generation Mobile Core Network

• Authors: Maria Katarine Santana Barbosa , Kelvin L. Dias
• URL: https://arxiv.org/abs/2606.00417
• Abstract:

  To meet the stringent requirements of emerging applications and the increasingly complex network management and operation, the Next Generation Mobile Networks (NextG), or 6G, will adopt an AI-native architecture on the Core Network (CN). In this movement, the Third Generation Partnership Project (3GPP) has extended the cellular CN with new function as a first step toward integrating analytics, Artificial Intelligence (AI), and machine learning. However, those new functionalities are constrained by a centralized approach and managerial complexity. Furthermore, with the rise of Large Language Models (LLMs), a new era in network orchestration and management begins, leveraging and empowering the Intent-based Networking (IBN) paradigm. In addition, AI agents and Agentic AI integrate Reasoning and Acting (ReAct), enabling the usage of such intents to continuously interact with the network. Unlike state-of-the-art approaches that primarily employ Agentic AI to mitigate deployment and configuration complexity in the CN, this paper introduces AgentxGCore, which leverages an Agentic AI-Native layer to extend the 3GPP architecture and enable a system based on the existing APIs across the Beyond Next Generation Core (xGC) domain. This proposal establishes an AI-driven closed-loop for continuous optimization based on real-time information, enabling self-organization and self-adaptation. Our approach involves a multi-agent specialized system, divided into a network planner agent, capable of visualizing the network state and developing a plan to meet the intents, and a network executor, responsible for criticizing and executing the plan. To validate the proposed solution, an environment was built using an open-source CN, heterogeneous datasets, and different LLMs were employed to demonstrate its effectiveness.

---

210. A Distribution-Free Framework for Rewrite-Based Human-text Detection via Knockoff Filtering

• Authors: Yi Liu
• URL: https://arxiv.org/abs/2606.00402
• Abstract:

  We propose a distribution-free statistical framework that converts arbitrary rewrite-based detectors into detectors with finite-sample FDR guarantees without retraining. Our key observation is that rewrite-based detection implicitly constructs knockoff samples, enabling LLM-generated text detection to be formulated as a multiple hypothesis testing problem with knockoff structure. This perspective separates the design of detection statistics from the control of false discoveries, allowing existing rewrite detectors to inherit finite-sample false discovery rate (FDR) guarantees through a simple calibration procedure. We demonstrate reliable FDR control with meaningful detection power across three detection models, 19 domains, and four LLMs.

---

211. PR2: Predictive Routing Replay for MoE-Based LLM Reinforcement Learning

• Authors: Daize Dong , Junlin Chen , Haolong Jia , Jiawei Wu , Huanwei Di , Jiang Liu , Jialian Wu , Zhengzhong Liu , Zicheng Liu , Emad Barsoum , Dimitris N. Metaxas , Hongyi Wang
• URL: https://arxiv.org/abs/2606.00395
• Abstract:

  Mixture of Experts (MoE) Large Language Models (LLMs) achieve strong performance at scale. However, reinforcement learning (RL) on MoE-based LLMs often suffers from training instability. A root cause is router drift, i.e., expert activations can change drastically across model updates and differ between disaggregated rollout and training phases, causing large rollout–training mismatch and unstable importance sampling weights in PPO-style RL algorithms. Routing replay mitigates this issue by freezing the replay route within each reasoning trajectory, but it ignores how the router evolves under off-policy updates and thus causes router staleness. To address this limitation, we propose Predictive Routing Replay (PR2), which augments each router with a lightweight evolution predictor that learns to anticipate short-horizon router evolution. During the rollout phase, we use the predictive routing distribution to apply top-$k$ routing, enabling gradients to reach experts that are likely to become active after updates. During the training phase, we replay the resulting predicted route to retain consistency for stable importance estimation. Theoretical analysis and experiments support that PR2 reduces routing-induced mismatch, improves RL stability, and yields stronger performance across various reasoning benchmarks.

---

212. Detector-Evasive LLM Paraphrasing via Constrained Policy Optimization

• Authors: Mingyi Wang , Zhuoer Shen , Yuheng Bu , Shaofeng Zou
• URL: https://arxiv.org/abs/2606.00392
• Abstract:

  AI-text detectors are vulnerable to paraphrasing and detector-guided paraphrasing attacks, but existing detector-evasion methods often lack precise control over semantic preservation. In particular, optimizing directly for detector evasion can degrade fine-grained semantics, whereas scalarized reward designs provide only indirect, weight-sensitive control over the evasion-semantics trade-off. We address this limitation by formulating detector-evasive LLM paraphrasing as a Constrained Markov Decision Process, where detector evasion is the primary objective and semantic preservation is enforced as an explicit constraint. We propose Detector Evasion Policy Optimization (DEPO), a Lagrangian primal-dual reinforcement learning algorithm with a novel GRPO-style group-based policy update. DEPO adaptively balances semantic preservation and detector evasion during training, enabling the policy to improve attack success within a prescribed semantic-preservation region. Experiments on MAGE, M4, RAID, and peer-review datasets, evaluated against MAGE, RoBERTa, RADAR, Binoculars, and Fast-DetectGPT detectors, show that DEPO achieves strong detector evasion while precisely satisfying the semantic preservation constraint. DEPO also exhibits cross-domain, cross-detector, and prompt-level robustness.

---

213. Zamba2-VL Technical Report

• Authors: Hassan Shapourian , Kasra Hejazi , Olabode M. Sule , Beren Millidge
• URL: https://arxiv.org/abs/2606.00390
• Abstract:

  We present Zamba2-VL, a suite of vision-language models built on Zamba2, a hybrid language-model architecture combining Mamba2 state-space layers with a small number of shared transformer blocks. Across a broad range of image understanding, reasoning, OCR, grounding, and counting benchmarks, Zamba2-VL is competitive with leading Transformer-based open-weight VLMs of comparable scale, including the Molmo2, Qwen3-VL, and InternVL3.5 families, and substantially outperforms prior SSM-based and hybrid VLMs such as VL-Mamba, Cobra, and mmMamba. Inheriting the near-linear prefill compute and small, near-constant recurrent state of its Zamba2 backbone, Zamba2-VL delivers roughly an order of magnitude lower time-to-first-token (TTFT) than these Transformer baselines at matched parameter scale, with the efficiency gap most pronounced at the smaller 1.2B and 2.7B scales most relevant to on-device and edge deployment. We release three models – 1.2B, 2.7B, and 7B – together with inference code at this https URL .

---

214. Agentic Authoring of Interactive Multiview Visualizations in Genomics

• Authors: Astrid van den Brandt , Kiroong Choe , Sehi L’Yi , Devin Lange , Nils Gehlenborg
• URL: https://arxiv.org/abs/2606.00370
• Abstract:

  Diverse genomics data, scientific questions, and analysis tasks typically demand highly specialized visualizations. Therefore, users often must customize or author new ones tailored to their data. Existing tools are usually either limited in customization or require substantial learning or programming, and even expressive tools assume visualization expertise many users lack. Agentic and large language model (LLM) approaches are increasingly applied to complex scientific tasks, including visualization. Natural-language conversational interfaces offer a promising path to democratizing the authoring of complex visualizations. In the context of genomics, these approaches face additional challenges: genomics visualizations typically integrate heterogeneous data types and are composed of multiple linked interactive views. These challenges motivate more structured LLM-based schemes. We first characterize where vanilla LLM generation succeeds and fails for genomics visualization, identifying eight quality dimensions. We then compare six schemes–direct generation, a fixed pipeline, and four agentic configurations varying in the number of specialist agents and the presence of a reviewer–across 159 cases spanning three levels of query ambiguity and specification complexity. All schemes use the Gosling visualization grammar as structured output. Agentic iteration substantially improves perceived quality over both baselines, while more complex agent architectures yield no additional benefit. We discuss implications for designing agentic systems for domain-specific visualization authoring. All supplemental materials are available at this https URL .

---

215. Isolating LLM Lexical Bias: A Curation-Free Triangulated Metric for Preference-Stage Learning

• Authors: Xiaoyang Ming , Jose Hernandez , Thomas Stephan Juzek
• URL: https://arxiv.org/abs/2606.00334
• Abstract:

  Various language domains have undergone remarkable changes in recent years; these shifts are largely attributed to the advent of Large Language Models and their misalignment with natural language usage. These misalignments are thought to partly originate in the preference-learning stage, e.g. Reinforcement Learning from Human Feedback, which generally makes models more useful but simultaneously may introduce systematic lexical bias. In terms of lexical behavior, this is visible in a model’s preference for certain formats or the overuse of words (delve, furthermore), even when such patterns are not present in base model outputs. Research on lexical misalignment induced during preference training is constrained by reliance on manual curation. We address this, by introducing the Triangulated Preference Shift score, a metric that triangulates between human gold standards, base models, and instruct variants to isolate shifts induced specifically by preference learning, without manual curation. We provide data across six model families, anchor the results in the literature, and illustrate the general approach’s utility by analyzing whether preference learning shifts models toward what could be interpreted as a “language of prestige”. The metric provides an initial automated method to quantify behavioral shifts attributable to preference tuning, and thus, may help inform model alignment and development of trustworthy AI.

---

216. LLMs Need Encoders for Semantic IDs Too

• Authors: Xiangyi Chen , Zelun Wang , Xinyi Li , Yi-Ping Hsu , Jaewon Yang , Jiajing Xu
• URL: https://arxiv.org/abs/2606.00324
• Abstract:

  Multimodal LLMs use dedicated encoders to bridge non-language modalities (vision encoders for images, depth models for audio codec tokens) because raw token embeddings alone cannot capture modality-specific structure. We argue that Semantic IDs (SIDs), the hierarchical codes used in generative recommendation, constitute another such modality: a SID level token’s meaning depends on its prefix context, yet current systems simply add SID tokens to the vocabulary and rely on training to learn these context-dependent meanings from scratch. We propose PrefixMem, a lightweight SID encoder based on prefix n-gram memory tables that provides the LLM with structured, prefix-conditioned representations at SID token positions. Like vision encoders in multimodal LLMs, PrefixMem can be pre-trained independently and then attached to any LLM for joint training. We evaluate on large-scale data from Pinterest across multiple LLM families and show that PrefixMem improves deepest-level SID accuracy by up to 46% relative and full-SID retrieval recall by up to 22% relative at matched training compute. The encoder’s benefit concentrates on hard examples where greedy decoding fails, with up to 77% relative accuracy gains, confirming that SID tokens benefit from a dedicated encoder just as other non-language modalities do.

---

217. How Generation Architecture Shapes Code Complexity in Multi-Agent LLM Systems: A Paired Study on HumanEval

• Authors: Nazmus Ashrafi
• URL: https://arxiv.org/abs/2606.00308
• Abstract:

  Large-language-model code generation has shifted from single-shot prompting to multi-agent orchestrations - analyst, coder, tester, and debugger pipelines - and is evaluated almost exclusively on functional correctness. Whether these architectures also affect the structural complexity of the code they produce, and which orchestration layers carry the cost, remains largely unexamined: prior work has documented prompt-level effects on code complexity, but the architecture-level question is open. We compare six widely-used multi-agent configurations (Basic, AC, ACT, Debugger, AC+Debugger, ACT+Debugger) under two models from the GPT-4o family across all 164 HumanEval tasks - 1,968 paired observations - using the five RADON complexity metrics (SLOC, cyclomatic complexity, and Halstead Volume, Difficulty, and Effort). We apply a paired non-parametric statistical pipeline (Friedman omnibus, Wilcoxon signed-rank post-hoc with Holm correction, Kendall’s $W$ and matched-pairs rank-biserial effect sizes) in both all-completions and passing-only conditions. The six architectures collapse into two indistinguishable complexity clusters separated by a 50-130% gap, the same partition in both models and under both conditions; among the architectural layers, the analyst-coder split inflates complexity, the runtime debugger does not - and on the analyst-coder background actively deflates it - and the tester re-inflates it. The heavy cluster’s additional complexity buys no pass@1 advantage: the leanest architectures match or beat the heaviest on accuracy. Architectural elaboration in LLM code generation should therefore be justified by measured benefit on the dimensions that matter, not assumed.

---

218. Rethinking the Role of Temperature in Large Language Model Distillation

• Authors: Hoang-Chau Luong , Lingwei Chen
• URL: https://arxiv.org/abs/2606.00306
• Abstract:

  Reverse Kullback-Leibler (RKL) divergence is widely favored over forward KL (FKL) in large language models (LLM) distillation, yet this preference is largely based on comparisons that omit the temperature $\tau$, overlooking its central role in softening teacher distributions and improving knowledge transfer. In this work, we revisit temperature in LLM distillation and show that it fundamentally changes the comparison between FKL and RKL. Our analysis reveals an asymmetric effect: temperature substantially enriches FKL with non-dominant token signals, whereas it mainly rescales RKL gradients, causing FKL to benefit much more from $\tau$ scaling than RKL. This asymmetry overturns the standard empirical conclusion: although RKL outperforms FKL at $\tau=1$, FKL consistently surpasses RKL at higher temperatures across instruction-following benchmarks. Moreover, the impact of temperature is not limited to FKL; it improves a broader family of distillation objectives, enabling simple KL-based methods to achieve competitive performance against recent state-of-the-art LLM distillation approaches.

---

219. Bridging Reasoning Trajectories in On-Policy Distillation via Near-Future Guidance

• Authors: Yuxuan Jiang , Francis Ferraro
• URL: https://arxiv.org/abs/2606.00305
• Abstract:

  On-Policy Distillation (OPD) improves large language model reasoning by training a student model on trajectories sampled from its own policy under teacher supervision. Although OPD operates on trajectories, its learning signal remains token-level: it identifies deviations through high-loss tokens and repairs them through local reverse-KL correction. We show that this “trajectory-sampled but token-learned” mechanism cannot reliably bridge student trajectories toward teacher trajectories. About 30% of high-loss tokens fall into the low-divergence regime, indicating that many are surface-form mismatches rather than real reasoning forks. Moreover, even truly divergent tokens are difficult to repair with isolated token-level supervision, since reasoning failures often unfold as short-horizon distributional drift. We propose Trajectory-aware OPD (TOPD), which uses near-future trajectory information to identify real divergent states and distribute guidance across multiple future tokens. Experiments show that suppressing non-divergent high-loss tokens improves standard OPD from 47.8% to 48.2% average accuracy, while TOPD further improves performance to 52.2%, with gains on AIME24 from 60.0% to 63.3% and AIME25 from 46.7% to 53.3%.

---

220. Synthetic Data from Cross-Domain Events for Large-Scale Recommendation Systems

• Authors: Xiangyu Wang , Yawen He , Shivendra Pratap Singh , Han Huang , Mengtong Hu , Sharath Ciddu , Yi-Hsuan Hsieh , Erik Groving , Yi Ding , Jieming Di , Tony Wang , Min Yun , Xiaoyu Chen , Ling Leng , Rob Malkin
• URL: https://arxiv.org/abs/2606.00282
• Abstract:

  Large-scale recommendation systems operate across diverse domains, yet they face the challenges of data sparsity and noisy implicit feedback. Traditional approaches mitigate this via model-specific knowledge distillation from source domains to a target domain. Inspired by the transformative success of synthetic data generation in large language models (LLMs), we introduce Synthetic Cross-domain Augmentation and Learning for Recommendation (SCALR), a framework that generates synthetic user-item interaction events for a target recommendation domain by leveraging observed events from a source domain. SCALR decomposes cross-domain learning into two modular stages. First, it translates observed user events in source domains by framing event generation as estimating the likelihood that a user would interact with a target-domain item, conditioned on their observed interactions in a source domain. Second, downstream models train on these synthetic events as cross-domain learning objectives, where the synthetic events augment the target domain’s training data in a model-agnostic manner. Our approach yields statistically significant improvements in online A/B tests on an industrial recommendation platform. To the best of our knowledge, this is among the first works to explicitly frame cross-domain event transfer as synthetic data generation for recommendation systems.

---

221. Hyperbolic and Evidence-Prioritized Experts for Large Vision-Language Models

• Authors: Zijie Zhou , Dandan Zhu , Hangxiangpan Wang , Heng Zhang , Huishen Jiao , Yi Zhao
• URL: https://arxiv.org/abs/2606.00275
• Abstract:

  Large Vision-Language Models (LVLMs) have demonstrated impressive performance on multimodal tasks through scaled architectures and extensive training. Recent studies introduce Mixture of Experts (MoE) into LVLMs for improved computational efficiency. However, existing MoE approaches treat visual and linguistic modalities with symmetric architectures, overlooking the inherent asymmetry in how these two modalities are processed. This asymmetry causes two critical issues. First, text and vision form hierarchical rather than parallel relationships, as text queries typically describe partial aspects of complete visual scenes. Euclidean expert space struggles to encode such containment structures. Second, language experts in deeper layers progressively shift from evidence-based processing to parametric memory dependence, losing grounding in the provided visual and linguistic information. To address these issues, we propose AsyMoE, a novel architecture that explicitly models this asymmetry through three specialized expert groups. Intra-modality experts handle modality-specific processing. Hyperbolic inter-modality experts capture hierarchical cross-modal relationships through negative curvature geometry. Evidence-priority language experts suppress parametric memory activation and maintain contextual grounding throughout network depth. Extensive experiments demonstrate that AsyMoE achieves consistent improvements over baseline methods, with average gains of 1.5\% over MoE variants and up to 3.8\% on hallucination-sensitive tasks. AsyMoE activates 25.45\% fewer parameters compared to dense models.

---

222. StressDream: Steering Video World Models for Robust Policy Evaluation and Improvement

• Authors: Junwon Seo , Sushant Veer , Ran Tian , Wenhao Ding , Apoorva Sharma , Karen Leung , Edward Schmerling , Marco Pavone , Andrea Bajcsy
• URL: https://arxiv.org/abs/2606.00267
• Abstract:

  Video world models (WMs) have shown promise for policy evaluation and improvement by imagining realistic future observations conditioned on ego-robot actions. While WMs can model distributions over futures, policy evaluation and improvement typically rely on nominal imaginations, which can miss high-impact outcomes of robot actions unless prohibitively many samples are drawn. To enable robust policy evaluation and improvement over WM imaginations, we propose StressDream, which steers imaginations toward high-impact yet plausible outcomes specified at inference time by optimizing the initial noise of diffusion-based WMs. However, optimizing high-dimensional noise is challenging: the optimization must reason about nuanced, scene-dependent target events in generated videos while avoiding out-of-distribution (OOD) noise that yields implausible imaginations. We address this with two complementary objectives: a semantic objective with a Vision-Language Model that provides informative gradients by reasoning about the generated video, and a plausibility objective that prevents the optimized noise from drifting OOD. With state-of-the-art video world models for autonomous driving and robotic manipulation, we show that StressDream effectively steers imaginations toward high-impact yet plausible outcomes specified by text at inference time, such as task failures, enabling robust policy evaluation and improvement by identifying actions whose plausible futures include undesirable outcomes. Video results are available at this https URL .

---

223. ARCA: Adapter-Residual Credit Assignment When Token Signals Degenerate

• Authors: Rodney Lafuente-Mercado
• URL: https://arxiv.org/abs/2606.00257
• Abstract:

  Token-level credit assignment for language-model reinforcement learning is usually formulated as if the policy were fully trainable, while practical LLM-RL pipelines often rely on parameter-efficient fine-tuning, especially LoRA. We argue that this separation hides a structural failure mode. Under LoRA, the policy is restricted to a low-rank neighborhood of the reference model, so the per-token output-distribution differences used by common intrinsic credit signals, surprisal, entropy reduction, and policy divergence, can become degenerate after within-trajectory normalization, either approaching uniform weights or concentrating on a small set of task-agnostic positions. We formalize this behavior and propose measuring it directly with concentration diagnostics such as weight Gini and effective-token ratio. We then introduce \emph{Adapter-Residual Credit Assignment} (ARCA), a lightweight alternative that derives token salience from the adapter’s own hidden-state residual, $|h^{\text{adapted} }_t - h^{\text{base} }_t|_2$. ARCA asks where the adapter actually changes the model, rather than where the output distribution appears uncertain or shifted, and requires no learned reward model, value head, or tree construction. In a compact MATH/Qwen3-1.7B GRPO sweep, ARCA exhibits the predicted non-degenerate middle-regime credit distribution under matched rollout budgets and remains competitive with rank-matched baselines.

---

224. Effects of Varying LLM Access on Essay Writing Behavior

• Authors: Julia Christenson , Karin de Langis , Shirley Anugrah Hayati , Dongyeop Kang
• URL: https://arxiv.org/abs/2606.00250
• Abstract:

  Investigating the degree to which large language models (LLMs) affect teaching and learning in universities can help identify strategies for integrating LLMs in a way that supports, rather than undermines, student learning outcomes. This study examined how varying levels of LLM assistance affect writing performance, engagement, and perceived authorship. We report a pilot study in which 24 college students were randomly assigned to write a short essay with no LLM access, limited access (<=3 prompts, responses capped at 100 words), or unlimited access. Overall essay quality was statistically indistinguishable across groups. Yet writing behavior and perceived authorship diverged sharply: students with limited access reported higher ownership (62.5% would submit the essay as independent work, vs. 25% in the unlimited group), stronger organizational gains, and more strategic, revision-focused prompting. The unlimited group spent more time writing, produced essays more similar to LLM output, and reported reduced creative expression. Our findings suggest that constraining, rather than banning, LLM access may preserve authorship confidence while retaining the scaffolding benefits of AI assistance.

---

225. Continuous Reasoning for Vision-Language-Action

• Authors: Yueh-Hua Wu , Tatsuya Matsushima , Kei Ota
• URL: https://arxiv.org/abs/2606.00229
• Abstract:

  Natural language is a powerful reasoning medium for language and vision-language models, but it is mismatched to the granularity of continuous control. Text and explicit subgoals operate at task-level granularity, whereas vision-language-action (VLA) policies must choose actions at a much finer temporal scale; a single reasoning step can therefore span many action chunks while remaining only weakly coupled to the action needed now. This suggests a different question for VLA: what should play the role of language? We argue that a useful VLA reasoning medium must be shareable across model instances, verifiable through downstream action improvement, and aligned with temporally extended control structure. Based on this view, we propose Continuous Reasoning for Vision-Language-Action. Our model first predicts continuous reasoning in the form of a structured set of continuous thoughts, then reuses them as shared context for chunk-structured action generation. Better action prediction alone does not certify good reasoning: if the same internal medium cannot be shared across model instances and independently verified through improved downstream control, the added latent may simply become a model-private shortcut that helps on seen behaviors without supporting generalizable control. We therefore instantiate continuous reasoning as a shared Gaussian latent interface and train it with a self-verification objective in which an exponential-moving-average teacher must successfully consume the student’s reasoning when predicting target actions. Empirically, Continuous Reasoning improves LIBERO-PRO robustness and performs strongly on real robots, raising mean subtask success over {\pi}0.5 by 40.4% on TX-G2, an AgiBot G2-compatible variant, and 26.3% on HSR. This suggests that reasoning in VLA is less about extra tokens than about a shareable, verifiable internal language for action.

---

226. BAGEN: Are LLM Agents Budget-Aware?

• Authors: Yuxiang Lin , Zihan Wang , Mengyang Liu , Yuxuan Shan , Longju Bai , Junyao Zhang , Xing Jin , Boshan Chen , Jinyan Su , Xingyao Wang , Jiaxin Pei , Manling Li
• URL: https://arxiv.org/abs/2606.00198
• Abstract:

  While agents are increasingly spending more resources, today agent cost is mostly measured only after execution. A Budget-Aware Agent (BAGEN) should treat budget as an active control signal, rather than a passive cost metric. We first systematically define budget estimation as internal budgets (from agent computation) and external budgets (from agent actions). We then formalize budget-awareness as progressive interval estimation: at each step of a plan, an agent should predict an upper and lower bound on remaining budget, and alert when completion is unlikely. Scoring with a rollout-replay protocol, we find consistent failure patterns on four environments and five frontier agents: (1) strong agents do not necessarily have strong budget-awareness, with correlation r=0.35. (2) frontier models are consistently over-optimistic, continue spending on tasks that are unlikely to succeed, instead of alerting the user early. (3) budget-aware signal is actionable and trainable. Early stop saves 28-64% tokens on failed trajectories, and SFT+RL strengthens early stop and alert behavior. (4) precise interval calibration remains challenging, with interval coverage capping at 47% after SFT+RL. Project page: this https URL

---

227. Learning to Construct Practical Agentic Systems

• Authors: Aditya Kumar , Zhihan Lei , Jerry Yan , Joshua W. Momo , Lauhitya Reddy , Rafael Enrique Cabrera Jimenez , Cassandra A. Cohen , Arthur Kajiyama , William W. Cohen
• URL: https://arxiv.org/abs/2606.00189
• Abstract:

  Automated design and optimization of agentic LLM-based systems leads to sophisticated systems that substantially improve result quality over off-the-shelf agentic patterns. However, studies of fielded agentic systems show that production systems focus much more on issues such as simplicity, controllability, and predictability of inference costs. In this paper we propose principled approaches to designing and optimizing practical agentic systems. We describe an agent framework that enables designers to enforce modularity in agentic systems, by defining “pseudo-tools” that call LLMs recursively on a restricted context. Using this framework we hand-engineer agents for a diverse set of tasks, and show that relative to dynamically-planned workflows, hand-constructed fixed workflows are generally cheaper and more accurate. We then propose novel learning methods for the agentic components required by this framework, namely pseudo-tools and fixed workflows. These learning methods generally outperform hand-engineered agents. We also exploit the modularity of the framework to apply multi-objective optimization methods to jointly optimize cost and response quality and blend the results of multiple learning systems.

---

228. DataShield: Safety-degrading Data Filtering for LLM Benign Instruction Fine-Tuning

• Authors: Junbo Zhang , Qianli Zhou , Xinyang Deng , Wen Jiang , Jie Pan , Jinbiao Zhu
• URL: https://arxiv.org/abs/2606.00160
• Abstract:

  Large language models (LLMs) suffer from degraded safety capabilities even when fine-tuned with benign datasets. However, existing methods for identifying safety-degrading samples in benign datasets suffer from high computational costs and significant noise issues. In this paper, we propose DataShield to efficiently and effectively identify potential safety-degrading samples. Our key intuition is based on the observation that benign fine-tuning increases the overall response compliance of LLMs. DataShield’s key technical insight is to quantify each sample’s contribution to the model’s compliance behavior as its safety degradation score. DataShield consists of three core components: (1) Compliance Vector Extraction, which captures the LLM’s compliance behavior tendency; (2) a novel Compliance-Aware Score (CAS), which automatically identifies the optimal safety-critical layer; and (3) Safety-degrading Sample Filtering, which quantifies the projection shift of training data along the compliance direction. Extensive experimental evaluation on Llama3-8B, Llama3.1-8B, and Qwen2.5-7B using the Alpaca and Dolly benign datasets validates our method’s effectiveness in identifying high-risk and low-risk data subsets. We also observe that open-ended question answering is more likely to trigger safety degradation, and corresponding responses tend to be longer. We hope this work can provide new insights into data-centric defense methods. The source code is available at: this https URL .

---

229. A Protocol-Language Model for Network Intrusion (Without Deep Packet Inspection)

• Authors: Vivek Kumar Sharma
• URL: https://arxiv.org/abs/2606.00155
• Abstract:

  Modern network intrusion detection systems (NIDS) are caught in a structural contradiction: the protocols carrying the highest threat intelligence are precisely those encrypted under TLS 1.3 and QUIC, where payload inspection yields nothing. We ask a simpler question – what if the attack signature is not in the bytes, but in the rhythm? – and answer it by treating network flows as a language whose grammar is written entirely in L3/L4 packet metadata: length, inter-arrival time, TTL, TCP flags, and hashed port numbers. We present PLM-NIDS, which proves three claims in sequence. (1) The grammar exists and is learnable: a RWKV-4 state-space model trained on 344,232 unlabelled Monday flows achieves a causal LM validation loss of 0.204, demonstrating that benign traffic has predictable, statistically consistent structure. (2) Attacks violate this grammar: the per-flow perplexity score cleanly separates benign from attack flows with PR-AUC = 0.93 using zero attack labels at training time. (3) This separation is architecturally nontrivial: an LSTM trained on identical token sequences degenerates to a majority-class predictor (ROC-AUC approximately 0.50, F1 = 0.91 by always predicting “attack”), proving that RWKV’s causal pre-training provides an inductive bias unavailable to direct classifiers. Supervised fine-tuning further raises PR-AUC to 0.94 and ROC-AUC to 0.75, with a precision of 97.7% at the calibrated operating threshold. The RWKV backbone’s O(T) recurrent inference enables per-packet streaming without flow buffering, making PLM-NIDS operationally viable at line rate. Because it reads only IP/TCP/UDP headers, it is inherently encryption-agnostic: TLS 1.3, QUIC, and future encrypted protocols are handled transparently.

---

230. Benchmarking Multimodal LLMs on Code Generation for Complex Interactive Webpages

• Authors: Fan Wu , Lishuai Dong , Cuiyun Gao , Yujia Chen , Yiming Huang , Yang Xiao , Qing Liao
• URL: https://arxiv.org/abs/2606.00154
• Abstract:

  Recent advancements in multimodal large language models (MLLMs) have achieved remarkable progress in multimodal reasoning and code generation, catalyzing a new paradigm for front-end development. In particular, these models can directly transform visual designs into executable code, significantly improving the efficiency and adaptability of web development. Modern web applications are dynamic and interactive, featuring frequent user-page interactions. However, existing benchmarks largely evaluate the code generation of static webpages, ignoring the complex interactive behaviors in real-world applications. Besides, their evaluation criteria remain confined to visual fidelity and code structure, overlooking the interaction consistency between the generated and the reference webpages. To address these limitations, we introduce WebIGBench, the first benchmark designed to evaluate code generation for interactive webpages with complex interactions. By combining manually designed interaction paths with UI automation, we collected 103 complex webpages from real-world websites. This benchmark covers 5 popular interactive action types (e.g., click, input) involving 871 distinct interactive actions. Moreover, we propose a novel evaluation pipeline to address the gap in automated assessment of interactive actions. Extensive experiments on several representative MLLMs reveal the performance boundaries of current models in interactive webpage code generation using WebIGBench. The proposed benchmark is available at this https URL .

---

231. PrivacyPeek: Auditing What LLM-Based Agents Acquire, Not Just What They Say

• Authors: Mingxuan Zhang , Jiahui Han , Dadi Guo , Songze Li , Guanchu Wang , Na Zou , Dongrui Liu , Xia Hu
• URL: https://arxiv.org/abs/2606.00152
• Abstract:

  LLM-based agents are rapidly advancing, autonomously invoking external tools to complete multi-step tasks for users. However, agents often acquire more sensitive information than the task requires. Existing privacy benchmarks audit what the agent’s response or outgoing actions disclose, but overlook the acquisition stage where data first enters the agent’s context. The over-acquired information is then one careless action or one attack away from an outright leak. To assess its prevalence, we introduce \emph{PrivacyPeek}, a benchmark for evaluating acquisition-stage privacy leakage of LLM-based agents, with $1{,}182$ cases across $7$ acquisition behaviours and $16$ application domains. Specifically, \emph{Acquisition Inspection} examines the agent’s tool-call trajectory, both the tools it invokes and the data it receives, to detect when it acquires sensitive information beyond the task scope. \emph{Probe Elicitation} then issues a follow-up probe and measures how readily an attacker could elicit sensitive information the agent acquired but did not disclose. Our experiments on 10 LLM-based agents across 4 model families show that the unnecessary acquisition of sensitive information is widespread. In addition, we observe a correlation between the task-completion capability and acquisition-stage leakage. Prompt-level defences reduce only a small fraction of acquisition-stage leakage, leaving the majority unmitigated. These results make auditing acquisition-stage privacy both urgent and necessary. Our dataset and code are available at this https URL .

---

232. Persona Attack: Incremental Memory Injection Jailbreak Attack against Large Language Models

• Authors: Junyoung Park , Seongyong Ju , Sunghwan Park , Jaewoo Lee
• URL: https://arxiv.org/abs/2606.00150
• Abstract:

  As Large Language Models evolve for user convenience, vulnerability to jailbreak attacks continues to be reported despite ongoing efforts in safety training. Traditional jailbreak techniques typically focus on a single prompt injection, neglecting the models’ ability to remember the flow of conversation and the user’s instructions. In this paper, we propose Persona Attack, a memory injection based jailbreak method that manipulates the model’s context window through a step by step approach. Experimental results from applying Persona Attack to several widely used LLMs reveal that, as injections accumulate in memory, models increasingly prioritize these instructions over their internal safety alignment mechanisms. Furthermore, our experiments empirically demonstrate that the attack success rate varies not only according to the memory implementation of the model, but also combinations of instructions and can reach 95% under specific instruction configurations.

---

233. StemBind: When MLLMs Get Lost Between Rules and Instances in Abstract Visual Reasoning

• Authors: Xixiang He , Baiqi Wu , Xingming Li , Ao Cheng , Qiyao Sun , Xuanyu Ji , Qingyong Hu
• URL: https://arxiv.org/abs/2606.00148
• Abstract:

  Multimodal large language models (MLLMs) often know the rule but pick the wrong answer: on abstract visual reasoning (AVR) tasks, a model can describe what it sees and name the underlying pattern, yet still fail to choose the matching candidate. Existing AVR benchmarks cannot detect this because they collapse perception, rule induction, and answer selection into a single right-or-wrong signal. We introduce StemBind, a shared-stem diagnostic benchmark that probes the same visual stem with three aligned questions: Perception (what is in the image), Rule (what pattern governs it), and Full (which option completes it), so a final-answer error can be attributed to a specific sub-step on the same evidence. StemBind contains 2,298 curated knowledge-light stems across nine auditable visual operations, totaling 19,533 P/R/F tasks, with each full item annotated by Sternberg’s four reasoning stages (S1 Encode, S2 Infer, S3 Map, S4 Apply). Evaluating 24 frontier MLLM configurations yields four findings. (i) The R-F chasm: rule accuracy exceeds full-item accuracy on 22 of 24 models, so most failures happen after the rule is identified. (ii) A persistent binding gap: even when P and R are both correct on the same stem, models still answer F incorrectly 51.2% of the time. (iii) The bottleneck is S3: process diagnostics and Stage-wise Stimulus Augmentation localize the dominant failure to rule-to-instance mapping. (iv) Scaling and thinking do not help: neither larger models nor explicit thinking mode reliably closes the gap, and thinking even lowers rule and full-item accuracy. StemBind reframes AVR evaluation from final-answer ranking to locating where abstract visual reasoning breaks down, identifying rule-to-instance binding as a concrete next target for vision-grounded reasoning.

---