LLM 관련 주요 논문 - 2026-06-03

1. Imaginative Perception Tokens Enhance Spatial Reasoning in Multimodal Language Models

Authors: Mahtab Bigverdi , Lindsey Li , Weikai Huang , Yiming Liu , Jaemin Cho , Jieyu Zhang , Tuhin Kundu , Chris Dangjoo Kim , Zelun Luo , Linda Shapiro , Ranjay Krishna
URL: https://arxiv.org/abs/2606.03988
Abstract:

Vision language models (VLMs) excel at many tasks but still struggle with spatial reasoning when critical information is not directly observable. Many such problems require imaginative perception: inferring what would be seen from an unseen viewpoint, tracing paths through occluded spaces, or integrating partial observations into a coherent spatial representation. We introduce Imaginative Perception Tokens (IPT), intermediate perceptual representations that externalize what a VLM would perceive under alternative spatial configurations while remaining consistent with the observed input. To study this capability, we formulate three tasks, Perspective Taking (PET), Path Tracing (PT), and Multiview Counting (MVC), and construct datasets of approximately 20K examples with ground truth imaginations, answers, and evaluation benchmarks. Using the unified VLM BAGEL as the backbone, IPT supervision consistently improves spatial reasoning and often outperforms textual chain of thought training, even without generating images at inference time. On MVC, IPT improves accuracy by 3.4% and achieves competitive performance with strong closed-source models on PT. We further find that combining IPT and label-only supervision yields additional gains, whereas textual chain of thought can substantially degrade performance, suggesting a modality mismatch when spatial computation is forced through language. Overall, IPT provides a principled supervision signal for reasoning about unobserved spatial structure, improving generalization while producing interpretable intermediate representations.

2. Reasoning Structure of Large Language Models

Authors: Frédéric Berdoz , Luca A. Lanzendörfer , Fabian Farestam , Roger Wattenhofer
URL: https://arxiv.org/abs/2606.03883
Abstract:

Large reasoning models (LRMs) are often evaluated using metrics such as final-answer accuracy or token count. However, identical scores on these metrics can hide fundamentally different reasoning structures. To address this limitation, we introduce a scalable LRM benchmark of logic puzzles and a pipeline that converts unstructured traces into verifiable reasoning graphs of claims and dependencies. This turns reasoning into a structured, measurable object whose topology can be quantitatively analyzed. Building on this, we define a reasoning efficiency metric that quantifies how concentrated the model’s logical flow is. Our analysis on open-source reasoning models shows that structural measurements separate behaviors that token count and accuracy conflate, providing a practical tool for diagnosing failure modes and comparing how reasoning scales with puzzle difficulty.

3. PyraMathBench: Evaluating and Improving Mathematical Capability in Large Language Models

Authors: Zetian Ouyang , Linlin Wang , Gerard de Melo , Liang He
URL: https://arxiv.org/abs/2606.03858
Abstract:

Despite the pivotal role of numerical reasoning as the cornerstone of mathematical capabilities in large language models (LLMs) across applications, few benchmarks evaluate LLMs by integrating numerical processing and mathematical reasoning, hindering the interpretability of failures in math tasks. We introduce PyraMathBench, a comprehensive hierarchical benchmark with 32,505 questions derived from 7,404 math word problems, spanning 4 key cognitive aspects, 14 subcategories, and 2 modalities. Experiments reveal that LLMs’ performance is severely compromised by inadequate numerical computation and weak handling of abstract numerical questions. To address this, we propose the Smart Optimization & Learning-based VErsatile module (SOLVE) and Interactive Relative Policy Optimization (IRPO), which enhance LLMs’ numerical-mathematical synergy via efficient tool calls (fuzzy matching and low-quality call rejection). Comparative experiments show Qwen-2.5 achieves a 5.0 score improvement with SOLVE and IRPO training.

4. EvoDS: Self-Evolving Autonomous Data Science Agent with Skill Learning and Context Management

Authors: Zherui Yang , Fan Liu , Yansong Ning , Hao Liu
URL: https://arxiv.org/abs/2606.03841
Abstract:

Recent progress in Large Language Model (LLM) agents has enabled promising advances in automated data science. However, existing approaches remain fundamentally limited by their static action sets and lack of principled long-horizon context management, hindering their ability to accumulate reusable experience across tasks and operate reliably in multi-stage, iterative data science pipelines. To address these challenges, we introduce EvoDS, a self-evolving autonomous data science agent that learns to expand its skills and adaptively managing long-term context through agentic reinforcement learning. Specifically, EvoDS introduces two key strategies: (1) Autonomous Skill Acquisition (ASA) mechanism, which enables agents to synthesize, validate, and reuse executable skills; and (2) Adaptive Context Compression (ACC) strategy, which treats context management as a learned control problem rather than passive truncation. These strategies are orchestrated within a two-stage multi-agent training scheme, enabling EvoDS to autonomously improve over time. Theoretically, we prove that EvoDS’s hierarchical design reduces tool-selection error, and its optimization objective aligns with an information bottleneck principle, ensuring efficient context use. Empirically, EvoDS outperforms state-of-the-art open-source data science agents by an average of 28.9% across four diverse benchmarks while eliminating out-of-token failures. Our code and data are available at this https URL .

5. Enhancing Operational Safety via Agentic Dialogue Hazard Identification Analysis

Authors: Sanjay Das , Ran Elgedawy , Ethan Seefried , Ryan Burchfield , Tirthankar Ghosal
URL: https://arxiv.org/abs/2606.03812
Abstract:

Operational safety in high-stakes domains such as industrial process control, autonomous, and safety-critical systems, demand reliable hazard identification. While large language models (LLMs) have shown promise in automating safety analysis tasks, single-turn, monolithic inference is brittle: it lacks the self-correction, deliberation, and contextual refinement that safety engineers apply iteratively. In this paper, we introduce HAZDIAL, a framework that investigates whether structured agentic dialogue-multi-agent, multi-turn interactions improves the quality of NLP- based hazard identification over single-pass baselines. We systematically compare two dialogue modalities: adversarial debate and constructive discussion, and propose an algorithm-based agentic interaction optimization. We evaluate all configurations against a curated golden dataset using standard classification metrics (accuracy, precision, recall, F1) and novel dialogue metrics. This work advances the intersection of dialogue systems, multi-agent reasoning, and AI safety, providing an empirical evidence for dialogue-driven hazard analysis.

6. LAP: An Agent-to-Instrument Protocol for Autonomous Science

Authors: Linwu Zhu , Liqiang Gao , Yan Chen , Dan Zhu , Jian Huang
URL: https://arxiv.org/abs/2606.03755
Abstract:

Autonomous science is moving from demonstration to infrastructure. Large language model agents now plan experiments, and self-driving laboratories execute them. Yet every such system rebuilds the link between the reasoning agent and the physical instrument from scratch, against fragmented vendor SDKs and standards built for deterministic software clients rather than probabilistic, goal-directed agents. Recent agent-interoperability protocols clarify two of the three edges of an agentic ecosystem (Anthropic’s Model Context Protocol (MCP) standardizes the agent-to-tool edge, and Google’s Agent2Agent (A2A) the agent-to-agent edge), but neither models the agent-to-instrument edge, where operations are stateful, safety-critical, exclusively owned, physically embodied, and produce measurements with units, calibration, and uncertainty. We present the Lab Agent Protocol (LAP), a protocol design that fills this gap. LAP retains A2A’s peer-to-peer, discovery-first, task-lifecycle structure and adds four physical-world primitives: (i) the InstrumentCard, a signed capability and physical-limit description; (ii) first-class reservation for exclusive instrument and sample locking; (iii) a safety-fence handshake with operator-confirmation tokens cryptographically bound to a specific task and its parameters, gating hazardous and irreversible operations; and (iv) a MeasurementResult schema that makes every result physically typed (QUDT/UCUM), calibration-anchored, uncertainty-bearing, and reproducible by construction. We specify roles, a six-layer architecture, the JSON-RPC method set, the task and safety state machines, the error model, and cross-laboratory federation, and walk a closed-loop autonomous campaign through the protocol end-to-end. LAP is transport-compatible with the A2A/MCP ecosystem and encapsulates rather than replaces existing device standards such as SiLA 2 and OPC-UA.

7. Proof-Refactor: Refactoring Generated Formal Proofs into Modular Artifacts

Authors: Yiming Fu , Peixuan Liu , Zichen Wang , Kun yuan
URL: https://arxiv.org/abs/2606.03743
Abstract:

While Large Language Models (LLMs) have shown strong performance in generating formal proofs, their outputs often remain less readable, modular, maintainable, and reusable than proofs in mature formal mathematics libraries. We argue that this gap stems in part from the compile-first objective implicit in most proof-generation pipelines, which encourages monolithic or ad hoc proof scripts rather than library-quality artifacts. Existing approaches to proof-quality improvement often rely on explicit, computable optimization objectives. In practice, however, the most tractable and experimentally validated objectives are largely length-based, while higher-level qualities such as readability, modularity, maintainability, and reusability are difficult to reduce to reliable automatic metrics. Instead of optimizing proof improvement against a single proxy metric, we take a process-guided approach inspired by human proof-refactoring workflows. We propose an agentic framework $\textbf{Proof-Refactor}$ that decomposes proof refactoring into four phases: extracting candidate proof fragments, designing helper declarations, formally proving the extracted and designed components, and repairing the original proof using the verified components. On generated Lean proofs from PutnamBench and Putnam2025, Proof-Refactor improves rubric-based refactoring scores over a strong Claude Code refactoring baseline, with the largest gains in signature quality and human readability. These results suggest that process-guided refactoring can improve proof structure without treating proof length as the primary objective.

8. When to Re-Plan: Subgoal Persistence in Hierarchical Latent Reasoning

Authors: Ayushi Chadha
URL: https://arxiv.org/abs/2606.03741
Abstract:

Long-horizon reasoning requires a system to commit to medium-horizon intent without becoming rigid: re-plan too often and computation never coheres into multi-step structure; commit too long and the plan goes stale. We study this stability-adaptivity tradeoff in the latent reasoning setting, where multi-step computation occurs inside hidden state rather than externalized token traces. We extend the Hierarchical Reasoning Model (HRM) with a feudal-style manager-worker interface: a slow high-level module periodically emits a normalized directional subgoal that persists for P low-level steps, biasing the worker’s hidden-state updates and supplying an intrinsic cosine alignment loss. On ARC and ConceptARC, we find that subgoal persistence – not subgoal injection alone – is the central knob: moderate periods P in [3, 6] consistently outperform both very frequent (P=1) and very long horizons, with a clear minimum LM loss at P=3 (1.544 vs. 1.674 at P=1, 1.640 baseline; replicated over 5 seeds at mean 1.595, std 0.045). The intrinsic alignment weight lambda shows a complementary narrow optimum (lambda approximately 0.05). A controlled ablation at past-sweet-spot lambda isolates learned directional structure – not architectural capacity or auxiliary loss alone – as the source of interference when the alignment signal exceeds its optimum. Together these findings implicate a design principle for compositional planning in latent reasoning systems: medium-horizon intent must be coherent across enough computational steps for compositional structure to form.

9. Code-on-Graph: Iterative Programmatic Reasoning via Large Language Models on Knowledge Graphs

Authors: Weiwei Ding , Zixuan Li , Long Bai , Zhuo Chen , Kun Su , Fei Wang , Xiaolong Jin , Jin Zhang , Jiafeng Guo , Xueqi Cheng
URL: https://arxiv.org/abs/2606.03705
Abstract:

Knowledge Graphs (KGs) are widely used to mitigate the limitations of Large Language Models (LLMs), such as outdated knowledge and hallucinations. Existing LLM-KG integration frameworks typically rely on predefined operators to retrieve factual knowledge from KGs and inject it into prompts for answer generation. This paradigm faces two critical bottlenecks: 1) Inflexibility: The predefined operators are limited in scope and thus lack sufficient compositional expressiveness to fully capture the complex semantics required by KG questions. 2) Unscalability: Direct injection of factual knowledge into prompts limits scalability in handling large-scale factual knowledge. To address these two bottlenecks, we propose Code-on-Graph (CoG), a programmatic reasoning framework for LLM-KG integration. Specifically, given the factual knowledge retrieved at each reasoning step, CoG first identifies the corresponding KG schemas and represents these schemas as Python classes, which serve as abstract interfaces to the retrieved facts. It then generates executable code grounded in these classes, with the retrieved facts instantiated as objects of the corresponding classes during execution. This design enables flexible code-based reasoning while avoiding the direct injection of large-scale factual knowledge into prompts. Experiments on WebQSP, CWQ, and GrailQA demonstrate that CoG outperforms prior state-of-the-art models by up to 10.5%.

10. Dynamic Objective Selection with Safeguards and LLM Oversight for Financial Decision-Making

Authors: Keigo Sakurai , Takahiro Ogawa , Miki Haseyama , Anjyu Anan , Kei Nakagawa
URL: https://arxiv.org/abs/2606.03704
Abstract:

Financial decision-making tasks such as stock recommendation and portfolio allocation typically estimate future return and risk and then select trades or allocations for an investor, and the chosen optimization objective often determines realized performance. However, because market conditions evolve over time, a fixed objective can be suboptimal across regimes, while regime-switching pipelines that rely on latent regime estimates can be noisy or delayed and frequent switching can increase turnover and operational instability. In this paper, we propose DOSS (Dynamic Objective Selection with Safeguards), a learning-based selector that directly chooses the decision-relevant objective function at each time point from interpretable statistical summaries of recent returns, selecting among a small set of candidates (e.g., return-seeking, loss-averse, and risk-adjusted) without introducing intermediate regime variables. DOSS formulates objective selection as a classification problem over objectives and performs sequential updates with a rolling window to make forward-looking selections without temporal leakage, while also outputting a confidence score for each proposal. To mitigate misselection and excessive switching in deployment, DOSS applies confidence-aware gating with a fail-safe that overrides low-confidence proposals to a conservative default and enforces explicit controls tied to switching frequency. We further integrate governance by positioning a Large Language Model (LLM) as an oversight component rather than a generator of new objectives: the LLM is restricted to accept a proposed objective or override it to a predefined safe default, with deterministic rule-based constraints triggering overrides when needed.

11. The DeepSpeak-Agentic Dataset

Authors: Sarah Barrington , Maty Bohacek , Hany Farid
URL: https://arxiv.org/abs/2606.03686
Abstract:

We present DeepSpeak-Agentic, a dataset of videos comprising over 37 hours of semi-structured conversations between a human and an embodied AI agent. We use this dataset to evaluate the automatic forensic identification (audio, video, or text) of AI agents, study the nature of human-agent interactions, and provide a benchmark for future advances in the large-language models and AI-generated voices and faces that power embodied AI agents. We also contribute a scalable data-capture system that creates agents, automatically pairs them with human crowd workers, records audiovisual conversations across specified scenarios, and identifies and separates the human and agent in the combined stream.

12. EvoDrive: Pareto Evolution for Safety-Critical Autonomous Driving via Self-Improving LLM Agents

Authors: Tong Nie , Yuewen Mei , Yihong Tang , Junlin He , Jie Deng , Jian Sun , Wei Ma
URL: https://arxiv.org/abs/2606.03678
Abstract:

Generating safety-critical scenarios is essential for validating and improving autonomous driving systems, yet it inherently requires maximizing adversariality to expose failures while preserving realism. Existing methods usually manage this trade-off with handcrafted heuristics, confining generation to known priors and overlooking underexplored patterns. While recent open-ended agentic evolution can push this limit, unconstrained general agents lack strict simulator grounding and tend to collapse the multi-objective tension into single-scalar maximization. Here we present EvoDrive, the first automated, LLM-based agentic evolution framework for multi-objective scenario generation. EvoDrive employs a simulator-grounded actor-critic architecture where a memory-driven actor iteratively proposes improvements to the generators and critics filter out implausible candidates, and a self-evolving world evaluator routes promising proposals to optimize simulation budgets. EvoDrive further maintains a Pareto archive of evaluated candidates to preserve diverse attack-realism trade-offs and guide future evolution via simulation feedback. Benchmark results on MetaDrive and CARLA show that EvoDrive not only significantly expands the Pareto frontier across various generators, but also produces valuable scenarios for policy training.

13. From Answers to States: Verifiable Process-Level Evaluation of Chemical Reasoning in Large Language Models

Authors: Hongyu Guo , Hao Li , He Cao , Gongbo Zhang , Li Yuan
URL: https://arxiv.org/abs/2606.03660
Abstract:

Large language models are increasingly used as chemistry assistants, yet most chemistry benchmarks still score only final answers. This masks a critical failure mode: a model may output the correct molecule, product, or option while its reasoning violates chemical logic. Existing process-level evaluators are hard to scale because LLM judges and human step-level process annotation are costly, inconsistent, and vulnerable to hallucination. We introduce ChemCoTBench-V2, a rule-verifiable diagnostic benchmark for low-cost, auditable evaluation of structured, verifier-addressable chemical reasoning traces. It spans molecular understanding, molecule editing, molecular optimization, and reaction prediction, with 5,620 evaluation samples across 18 reporting tasks. Models must expose key intermediate steps in expert-designed templates, and those steps are checked with deterministic chemistry rules and, for closed-answer tasks, reference traces rather than another LLM judge. Open-ended molecular optimization is evaluated with oracle-verifiable state constraints rather than strict trace matching. The benchmark reports three separate signals: final-answer correctness, template adherence, and step-wise verifier correctness over expert-refined intermediate commitments. Experiments on frontier models reveal a persistent gap between final-answer success and structured-reasoning-state consistency: models often follow the requested format while failing chemical-step checks, or answer correctly with weak supporting reasoning. ChemCoTBench-V2 enables fine-grained model comparison and identifies the concrete step at which the trace first violates the verifier.

14. Diagnosing Knowledge Gaps in LLM Tool Use: An Agentic Benchmark for Novel API Acquisition

Authors: Jinnuo Liu , Yue Peng , Jinhan Niu , Hongyi Wen
URL: https://arxiv.org/abs/2606.03657
Abstract:

Large language models for code generation often need to use APIs that are absent from their pretraining data. This requires more than recalling a function name: models must coordinate signatures, module paths, input-output contracts, semantics, and executable usage patterns. Existing novel-API benchmarks are typically static, rely on coarse pass/fail metrics, or use synthetic APIs that may not reflect real library evolution. We introduce NovelAPIBench, a fully automated dynamic benchmark that, for any base model and target library, discovers novel APIs, extracts decomposed knowledge bundles, generates executable coding tasks, and assigns failed samples to six diagnostic categories. Across about 1.9K tasks, four base models, and five domains, we compare knowledge injected through retrieval with knowledge internalized through parametric adaptation. We find that knowledge components are not interchangeable: usage examples are the strongest standalone signal, while the best two-component setting pairs signatures with either mechanisms or examples depending on the domain and backbone. Adding more context, especially source code, can hurt by increasing import-path errors. Parametric adaptation also does not replace retrieval once external knowledge is removed; rather, fine-tuning mainly teaches models how to use provided bundles, and this ability transfers to held-out libraries. These results suggest that retrieval and tuning play complementary roles: retrieval supplies volatile API content, while tuning improves procedural integration.

15. Gender-Dependent Diagnostic Substitution in LLM Medical Triage: Same Symptoms, Unequal Urgency

Authors: Qi Han Wong
URL: https://arxiv.org/abs/2606.03641
Abstract:

We investigate whether large language models produce different medical triage recommendations for identical neurological symptoms when only the patient’s stated gender and age vary. Using three model families–Gemini 3.5 Flash, Claude Sonnet 4.6, and GPT-5.4-mini–we present a standardized symptom profile (persistent headache, blurred vision, morning nausea, visual disturbances) across seven demographic conditions: three age groups (25, 38, 65) x two genders (male, female), plus a gender-unspecified baseline (n = 30 per condition per model, 630 total trials). We find a stark, systemic gender-dependent triage disparity: young women receive significantly lower emergency room (ER) referral rates than age-matched men (Gemini: 0% vs. 23.3%; Claude: 6.7% vs. 96.7%; GPT: 6.7% vs. 66.7%, all p < 0.001). The disparity disappears at age 65 for all models. The primary mechanism is diagnostic substitution: the models anchor on a gender-associated diagnosis, preferentially classifying young women with Idiopathic Intracranial Hypertension (IIH)–a condition epidemiologically linked to women of childbearing age–while diagnosing men with generic increased intracranial pressure with space-occupying lesions in the differential. This diagnostic closure routes female patients to lower-urgency care (outpatient doctor appointments) despite comparable severity ratings (7-9/10). Our findings demonstrate that clinical LLMs replicate documented human clinical biases by using epidemiological priors to suppress triage urgency, suggesting that AI triage engines must decouple urgency assessment from probabilistic diagnostic priors. We release all code, prompts, and raw results.

16. TSQAgent: Rating Time Series Data Quality via Dedicated Agentic Reasoning

Authors: Shunyu Wu , Dan Li , Haozheng Ye , Weibin Feng , Jian Lou , Bo Zhang , Wenjie Feng , Chenjuan Guo , See-Kiong Ng
URL: https://arxiv.org/abs/2606.03629
Abstract:

Assessing the quality of time series (TS) data is fundamental yet inherently challenging due to the multifaceted nature of quality dimensions. Recently, large language models (LLMs) have emerged as a promising paradigm for TS quality assessment via pairwise comparison and per-dimension evaluation. However, existing approaches rely on manually predefined quality dimensions and purely text-based reasoning, leaving it unknown whether LLMs can identify truly relevant quality dimensions or perform grounded and quantitative quality comparisons. To investigate this, we construct TSQBench, a dedicated benchmark for evaluating LLMs on two progressive capabilities: (i) understanding and identifying relevant quality dimensions, and (ii) performing quality comparison under specific dimensions. Our analysis reveals that current LLMs consistently struggle with both dimension identification and evidence-grounded quality comparison. To address these limitations, we propose TSQAgent, a novel agentic reasoning framework for TS quality rating consisting of three collaborative roles: Perceiver for focused dimension selection, Inspector for dimension-wise quantitative analysis, and Adjudicator that aggregates and refines the final judgment. In particular, we introduce an agentic reasoning strategy that instills the ability to identify and prioritize the most relevant quality dimensions, and further propose an agent workflow equipped with external analytical tools to enable precise quantitative comparisons over selected dimensions. Experiments on both the proposed benchmark and eleven real-world datasets demonstrate that our framework not only substantially improves LLMs’ capabilities in quality understanding and quantitative comparison but also effectively translates these improvements into better quality-aware data selection, leading to enhanced downstream performance and data efficiency.

17. Cross-Lingual Token Arbitrage: Optimizing Code Agent Context Windows via Local LLM Preprocessing

Authors: Mehmet Utku Colak
URL: https://arxiv.org/abs/2606.03618
Abstract:

AI-assisted coding agents are bottlenecked by input-token cost. Two pathologies of raw human input drive much of this overhead: tokenization inefficiency for non-English text and structural entropy in conversational prompts. Existing approaches act reactively by compressing already-bloated contexts or intervening after failures occur. We introduce a pre-flight, edge-side prompt-rewriting middleware that operates between the developer and the cloud agent. A local Llama 3.2 (3B) model performs cross-lingual translation into English, structural rewriting into a compact task-oriented format, and regex-validated rewrite-with-fallback safeguards to ensure the optimized prompt is never larger than the original. We evaluate on OMH-Polyglot, a multilingual coding benchmark spanning Turkish, Arabic, Chinese, and code-switched specifications. Across three commercial LLM backends, the middleware reduces prompt tokens by 34-47 percent and total tokens by up to 18.8 percent while preserving or improving task accuracy. Ablation studies show that gains arise primarily from the rewriting stage rather than simple function-name extraction. Compared with LLMLingua-2 at matched compression rates, our method consistently achieves superior OckScore performance across all evaluated backends. These results demonstrate that proactive prompt optimization can substantially reduce inference costs without sacrificing coding quality.

18. StepFinder: A Temporal Semantic Framework for Failure Attribution in Multi-Agent Systems

Authors: Taiyu Zhu , Yifan Wu , Weilin Jin , Ying Li , Gang Huang
URL: https://arxiv.org/abs/2606.03467
Abstract:

LLM-based multi-agent systems exhibit remarkable collaborative capabilities in complex multi-step tasks. However, these systems are highly sensitive to single-step execution errors that can propagate through agent interactions and lead to cascading failures. To understand the causes of failure and improve system reliability, failure attribution has been introduced as a task that aims to automatically identify the root cause step responsible for a failure. Existing failure attribution methods mainly rely on LLMs to reason over original execution trajectories, which not only incur high inference costs and latency, but also suffer from interference caused by redundant and noisy execution logs, causing LLMs to struggle in accurately identifying the true root cause step. To address this, we propose StepFinder, a lightweight failure attribution framework. We use LLMs solely during the feature construction phase to encode execution logs into temporal semantic sequences. Subsequently, a parameter-efficient combination of temporal modeling and attention modules is applied to capture the sequential evolution and cross-step dependencies of the trajectories. Finally, the step-level error score is refined through multi-scale differences and position bias, enabling precise root cause identification. Experimental results on the Who&When benchmark demonstrate that StepFinder outperforms LLM-based methods in step-level failure attribution while achieving substantially higher inference efficiency, reducing inference time by 79% compared with the fastest LLM-based method, with no text generation overhead. Our code is available at this https URL .

19. DMF: A Deterministic Memory Framework for Conversational AI Agents

Authors: Matteo Stabile , Enrico Zimuel
URL: https://arxiv.org/abs/2606.03463
Abstract:

Conversational AI agents require memory systems that are both scalable and semantically coherent across long interaction horizons. Existing approaches rely predominantly on large language model (LLM)-based summarisation at write time, which introduces non-determinism, escalating token costs, and opacity in pruning decisions. We present the Deterministic Memory Framework (DMF), a CPU-first approach that replaces generative memory compression with a fully deterministic pipeline grounded in classical NLP analysis, vector geometry, and mathematical scoring. DMF assigns each conversational interaction a Survival Score $\Omega$ computed from deterministic content signals, conversational cues, and structured provenance, combined through a logistic projection. An interaction-count decay law, denoted as $\Omega_{\mathrm{eff} }(\Delta n)$, governs how relevance evolves as new turns arrive, where $\Delta n$ is the number of newer interactions rather than wall-clock time, preserving full determinism. We present the mathematical formulation of DMF, its structured recall pipeline, the pruning decision procedure, and the evaluation protocol. Experiments are conducted on a purpose-built benchmark using the LoCoMo and LongMemEval datasets. We compare DMF against Mem0, a popular memory layer for AI agents. DMF achieves comparable accuracy while using zero tokens to prepare the memory context and 5x to 242x fewer tokens over the entire conversation. These results show that it is possible to eliminate LLM calls from the memory-management loop, reducing token costs to nearly zero and enabling deterministic memory systems for conversational AI agents.

20. CP-Agent: Context-Aware Multimodal Reasoning for Cellular Morphological Profiling under Chemical Perturbations

Authors: Yuxin Zhang , Yiyao Li , Ping Shu Ho , Simon See , Zhenqin Wu , Kevin Tsia
URL: https://arxiv.org/abs/2606.03435
Abstract:

Cell Painting combines multiplexed fluorescent staining, high-content imaging, and quantitative analysis to generate high-dimensional phenotypic readouts to support diverse downstream tasks such as mechanism-of-action (MoA) inference, toxicity prediction, and construction of drug-disease atlases. However, existing workflows are slow, costly and difficult to interpret. Approaches for drug screening modeling predominantly focus on molecular representation learning, while neglecting actual experimental context (e.g., cell line, dosing schedule, etc.), limiting generalization and MoA resolution. We introduce CP-Agent, an agentic multimodal large language model (MLLM) capable of generating mechanism-relevant, human-interpretable rationales for cell morphological changes under drug perturbations. At its core, CP-Agent leverages a context-aware alignment module, CP-CLIP, that jointly embeds high-content images and experimental metadata to enable robust treatment and MoA discrimination (achieving a maximum F1-score of 0.896). By integrating CP-CLIP outputs with agentic tool usage and reasoning, CP-Agent compiles rationales into a structured report to guide experimental design and hypothesis refinement. These capabilities highlight CP-Agent’s potential to accelerate drug discovery by enabling more interpretable, scalable, and context-aware phenotypic screening – streamlining iterative cycles of hypothesis generation in drug discovery.

21. The Reliability Gap in Benchmark Auditing: Distribution Shift and Scale as Failure Modes of Contamination Detection

Authors: Wojciech Zarzecki , Jan Dubiński , Sebastian Cygert
URL: https://arxiv.org/abs/2606.03305
Abstract:

Benchmark contamination, where evaluation examples appear in a model’s training data, threatens the validity of LLM assessment. Statistical tools for detecting training-data membership exist, but have been validated almost exclusively in controlled academic regimes: large, homogeneous pre-training corpora and transparent, single-stage training pipelines. Whether these methods remain reliable in realistic auditing scenarios remains unclear. We identify two under-studied failure modes: distribution shift, which arises when suspect and validation sets violate the IID assumption, and scale constraints, which arise because benchmarks are orders of magnitude smaller than pre-training corpora. We systematically evaluate three leading paradigms: LLM Dataset Inference, Post-Hoc Dataset Inference, and CoDeC across 27 models from multiple families (including Pythia, OLMo~2, and specialised cultural and medical LLMs) and scales (up to 27B). We then further extend our analysis to frontier industry models. Across 335 evaluations, only 199 yield correct outcomes. LLM Dataset Inference results in false positives under distribution shift, Post-Hoc Dataset Inference is underpowered at benchmark scale, and CoDeC provides only coarse provenance signals that are insufficient to verify individual benchmark splits. Our results reveal a systematic reliability gap between controlled validation and practical benchmark auditing, and show that statistical detection cannot yet replace transparent data provenance. We open-source our benchmark for further research.

22. LEAP: Supercharging LLMs for Formal Mathematics with Agentic Frameworks

Authors: Po-Nien Kung , Linfeng Song , Dawsen Hwang , Jinsung Yoon , Chun-Liang Li , Simone Severini , Mirek Olšák , Edward Lockhart , Quoc V Le , Burak Gokturk , Thang Luong , Tomas Pfister , Nanyun Peng
URL: https://arxiv.org/abs/2606.03303
Abstract:

Large Language Models (LLMs) exhibit strong informal mathematical reasoning but struggle to generate mechanically verifiable proofs in formal languages like Lean. We present LEAP, an agentic framework that enables general-purpose foundation models to achieve state-of-the-art performance on automated formal theorem proving. LEAP leverages foundation model capabilities, such as informal reasoning, instruction following, and iterative self-refinement. By decomposing complex problems into smaller units, the system bridges formal proof construction with informal blueprints through continuous interaction with the Lean compiler. To provide a rigorous evaluation beyond increasingly saturated benchmarks, we introduce Lean-IMO-Bench, a benchmark of IMO-style problems formalized in Lean, with short statements yet highly non-routine and multi-step proofs across a wide range of difficulty levels. Empirically, on the latest 2025 Putnam Competition, an annual mathematics competition for undergraduate students in North America, LEAP solves all 12 problems, matching recent breakthroughs by frontier formal mathematical models. On Lean-IMO-Bench, LEAP boosts the one-shot formal solve rate of general-purpose LLMs from below 10% to 70%, notably surpassing the 48% benchmark set by a specialized, gold-medal-caliber IMO system. Furthermore, we demonstrate LEAP’s research-level utility by autonomously formalizing complex proofs for open combinatorial challenges, including a verified proof for a key subproblem in Knuth’s Hamiltonian decomposition of even-order Cayley graphs.

23. A Negative Result on Cross-Model Activation Transfer in a Pythia Multi-Hop Setting

Authors: Peiyan Zhang
URL: https://arxiv.org/abs/2606.03280
Abstract:

Recent work shows that language models can transmit behavioural traits through hidden signals in generated data during training. We ask whether a more direct and stricter channel is also viable: can one language model communicate useful intermediate reasoning state to another at inference time by translating and injecting hidden activations, rather than by passing natural-language text? We test this question in a controlled Pythia-160M to Pythia-410M multi-hop reasoning setting. A linear translation layer learns a strong normalized-space map between sender and receiver hidden states, with normalized cosine similarity near 0.97 across seeds. However, when the translated activations are injected into the receiver at inference time, they do not improve downstream answering. Low-strength additive injection remains near the no-injection baseline, with confidence intervals that cross zero. Replacement-style injection is consistently destructive, and rescaling translated vectors to the receiver hidden-state norm does not rescue performance. The result is therefore a scoped negative result: in this setting, offline representational alignment is not sufficient for useful causal communication inside the receiver.

24. Distilling Answer-Set Programming Rules from LLMs for Neurosymbolic Visual Question Answering

Authors: Thomas Eiter , Nelson Higuera Ruiz , Johannes Oetsch
URL: https://arxiv.org/abs/2606.03269
Abstract:

Visual Question Answering (VQA) is the task of answering questions about images, requiring the integration of multimodal input and reasoning. Modular approaches that incorporate logic-based representations into the reasoning component offer clear advantages over end-to-end trained systems, particularly in terms of interpretability. However, adapting or extending these representations when task requirements change can place a significant burden on developers. To address this challenge, we present an approach for distilling rules from Large Language Models (LLMs). Our method prompts an LLM to extend an initial VQA reasoning theory, expressed as an answer-set program, to meet new requirements of the task. Examples from VQA datasets guide the LLM, validate the results, and help correct erroneous rules by leveraging feedback from the ASP solver. We demonstrate that our approach is effective across diverse VQA datasets. Notably, only a few examples are needed to elicit correct rules from LLMs. Our experiments suggest that rule distillation from LLMs is a promising alternative to traditional data-driven rule learning approaches. Under consideration in Theory and Practice of Logic Programming (TPLP).

25. Perceive Before Reasoning: A Pre-Reasoning Perception Framework for Efficient and Reliable Proactive Mobile Agents

Authors: Zhijie Ding (1,2), Weinan Hong (1,3), Zicheng Zhu (1,4), Lei Li (1), Dezhi Kong (1), Hao Wang (1), Peng Zhou (1), Xuchu Jiang (1), Jiaming Xu (1) ((1) HyperAI Team, Xiaomi Corporation, (2) Zhongnan University of Economics and Law, (3) Jilin University, (4) The Chinese University of Hong Kong, Shenzhen)
URL: https://arxiv.org/abs/2606.03236
Abstract:

Multimodal large language models (MLLMs) have substantially advanced mobile agents, yet proactive mobile assistance remains challenging because agents must decide \emph{when} to intervene before determining \emph{how} to assist. Existing systems often implement these two decisions within a unified MLLM-based pipeline, leading to goal misalignment between conservative intervention filtering and comprehensive assistance generation, as well as redundant inference when the agent should remain silent. To address these limitations, we propose the \textbf{Pre-Reasoning Perception Framework (PRPF)}, a two-stage framework built on perceiving before reasoning. PRPF introduces a lightweight Multimodal Proactive Perceptor (MPP) for intervention gating and context compression, and activates the Proactive Agent Reasoner (PAR) only when intervention is warranted. Experiments on the ProactiveMobile benchmark show that PRPF substantially reduces false trigger rates (FTR) while improving success rates (SR) and inference efficiency over the ProactiveMobile baseline.

26. ClinicalMC: A Benchmark for Multi-Course Clinical Decision-Making with Large Language Models

Authors: Ruihui Hou , Siyi Zhu , Ziyue Huai , Guangya Yu , Yongqi Fan , Chunming Wang , Tong Ruan
URL: https://arxiv.org/abs/2606.03157
Abstract:

Large language models (LLMs) have been widely adopted in healthcare, yet they still encounter significant challenges in complex clinical decision-making scenarios. Existing benchmarks primarily assess LLM performance in single-course settings and lack systematic evaluation in multi-course scenarios, where a patient’s condition evolves over time. To address this gap, we propose ClinicalMC, a benchmark for multi-course clinical decision-making. It includes 1,275 Chinese and 5,804 English samples across four stages from admission to discharge. These stages cover triage, first-course examination/diagnosis/treatment, subsequent multi-course examination/assessment/treatment, and final diagnosis. In ClinicalMC, patients in the English dataset undergo an average of 5.11 clinical courses, whereas those in the Chinese dataset undergo 3.42. To assess LLM performance, we construct a multi-agent evaluation framework that includes patient, examiner, and doctor agents. Based on the benchmark and framework, we design two experimental settings – a single-turn static setting and a multi-turn dynamic setting – and assess three categories of LLMs: 1) closed-source LLMs like GPT5-mini; 2) open-source LLMs like DeepSeek-V3.2; and 3) medical LLMs like HuatuoGPT-o1. Through extensive evaluation, we aim to better understand LLM performance in the medical domain and support its effective deployment in healthcare.

27. GTBench: A Curriculum-Grounded Benchmark for Evaluating LLMs as Mathematical Research Assistants in Graph Theory

Authors: Noujoud Nader , Ibrahem Aljabea , Patrick Diehl , Deepti Gupta
URL: https://arxiv.org/abs/2606.03144
Abstract:

Large language models (LLMs) are increasingly used as self-study assistants in technical disciplines, yet their reliability as mathematical reasoning assistants remains poorly understood. We introduce GTBench, a curriculum-grounded benchmark for evaluating LLMs as mathematical research assistants in graph theory, comprising 63 problems organized into three groups of increasing difficulty: undergraduate definitions and basic properties (Group 1), algorithm tracing and structural reasoning (Group 2), and graduate-level proof construction (Group 3). Problems are sourced from verified academic materials including Diestel’s Graph Theory. We evaluate five frontier models – GPT-5, Claude Sonnet 4.6, Gemini 2.5 Flash-Lite, Llama 3.3 70B, and Mistral Large 3 – under zero-shot and chain-of-thought prompting, using exact-match and LLM-as-judge evaluation for Groups 1 and 2, and a hybrid human expert and LLM-as-judge protocol for Group 3. Our results reveal a pronounced performance hierarchy: GPT-5 approaches ceiling on Group 1 (95.8% zero-shot) and maintains meaningful accuracy on graduate proofs (82%), while all other models degrade substantially with difficulty, with Llama achieving 0% under human evaluation on Group 3 zero-shot. Failure mode analysis shows that correct algorithm, wrong execution errors dominate Groups 1 and 2, while Group 3 additionally surfaces incomplete reasoning failures and reveals systematic disagreement between human evaluators and the automated judge, particularly on verbose or near-complete proofs (kappa = 0.48-0.83 across human pairs). GTBench provides the first curriculum-grounded evaluation framework for graph-theoretic reasoning in LLMs, with direct implications for the governance of AI tools in mathematical education and scientific research.

Authors: Kaiqi Yang , Tai-Quan Peng , Sanguk Lee , Hui Liu
URL: https://arxiv.org/abs/2606.03137
Abstract:

LLM-based multi-agent simulation offers a promising way to study social interaction, deliberation, and collective opinion dynamics. However, many existing dialogue simulation frameworks represent interaction mainly as observable turn exchange or aggregated outputs, leaving the internal evaluative processes behind silence, speaking intention, and public expression difficult to examine. We introduce TBS (Think-Before-Speak), an interval-based multi-agent simulation framework that separates agents’ private reasoning from public utterance generation. At each interval, all agents update structured internal states based on the shared dialogue history and their own memory. These states include dissonance-related appraisal, perceived opinion climate, perceived isolation risk, response strategy, and willingness to speak. The orchestrator then resolves competing speaking intentions and commits one utterance to the public dialogue, allowing internal evaluation and public interaction to co-evolve over time. We evaluate TBS in simulated town hall discussions on a climate-related policy issue. Results show that TBS produces coherent internal-state traces and that these traces vary systematically across turn-allocation, silence, and memory conditions. Dissonance-related appraisal increases agents’ willingness to speak, whereas silence-pressure appraisal decreases it. Once speaking intention is formed, public expression is shaped mainly by turn-allocation rules. These findings suggest that TBS supports mechanism-sensitive social simulation by making the pathway from internal evaluation to public expression observable and analyzable.

29. Uncertainty-Aware Clarification in LLM Agents with Information Gain

Authors: Mengyi Deng , Zhiwei Li , Xin Li , Tingyu Zhu , Ying Zhao , Zhijiang Guo , Wei Wang
URL: https://arxiv.org/abs/2606.03135
Abstract:

Large Language Model (LLM) agents often operate under underspecified user instructions, where latent uncertainty over user intent leads to erroneous tool actions. To address this challenge, we propose a goal-oriented clarification framework that aligns clarification behavior with ambiguity resolution. Central to our approach is the Information Gain Reward, a metric that quantifies the utility of clarification questions by measuring the Bayesian belief update towards the ground-truth goal induced by the clarification exchange. We train the clarifier (LLM) using this reward to optimize for high information gain, ensuring that clarifications effectively reduce uncertainty and improve task completion within the agent-tool-user environment. We validate our framework within a clarification-enhanced $\tau$-Bench environment, conducting cross-agent evaluations across five heterogeneous backbones. Empirical results demonstrate that our method consistently improves the success rate by 3.7\% over the no-clarification baseline, while adding only 0.3 total interaction steps on average.

30. EvoTrainer: Co-Evolving LLM Policies and Training Harnesses for Autonomous Agentic Reinforcement Learning

Authors: Guhong Chen , Yingcheng Shi , Yongbin Li , Binhua Li , Xander Xu , Hu Wei , Shiwen Ni , Min Yang , Jieping Ye
URL: https://arxiv.org/abs/2606.03108
Abstract:

Autonomous LLM training is often framed as recipe search, which leaves the training harness largely static. This limitation sharpens in agentic RL, where shifting bottlenecks and scalar rewards mask diverse failure modes. We introduce EvoTrainer, an autonomous training framework that co-evolves LLM policies and training-side harnesses through empirical feedback: it diagnoses rollout-level evidence, revises diagnostics, backtests interventions, and accumulates reusable skills. Evaluated on mathematical reasoning, competitive-programming code generation, and repository-level software engineering, EvoTrainer matches or exceeds the human-engineered RL references under the same data, codebase, and evaluation protocol, with the largest gain on long-horizon agentic SWE. Trajectory analyses show that retained strategies diverge across domains, evolving diagnostics prevent invalid high-scoring branches from being promoted, and reusable skills shape later search. Autonomous LLM RL should move beyond recipe search toward joint evolution of policies and the training harnesses that interpret them.

31. From Long News to Accurate Forecast: Importance-Aware Fusion and PRM-Guided Reflection for Time Series Forecasting

Authors: Mingyang Liu , Qingcan Kang , Yuke Wang , Shixiong Kai , Kaichao Liang , Hui-Ling Zhen , Tao Zhong , Mingxuan Yuan , Linqi Song
URL: https://arxiv.org/abs/2606.03097
Abstract:

Incorporating news into time series forecasting is appealing because news can reveal abrupt exogenous events that historical values alone cannot recover. However, existing LLM-based news-forecasting pipelines face two practical limitations: relevant news articles often exceed the model’s context window, and iterative retrieval of supplementary news is typically unguided, leading to redundant updates and slow convergence. We address these issues with a novel framework that combines importance-aware news compression and process-level retrieval supervision. First, we train an importance reward model that estimates the forecasting utility of each article and uses this signal to allocate compression budgets during sequential pairwise fusion, preserving informative content within a fixed context limit. Second, we introduce a process reward model (PRM) that ranks multiple supplementary-news candidates conditioned on the current error profile and the history of previously selected articles, replacing one-shot blind retrieval with quality-controlled selection. Both components are trained offline using historical data with ground truth; inference uses the frozen filtering logic and compression modules without any reflection loop. Experiments on finance, energy, traffic, and bitcoin forecasting benchmarks show that our method improves prediction accuracy over strong baselines, significantly reduces the number of refinement iterations compared to the iterative baseline, and remains effective when relevant articles span thousands of tokens.

32. Decomposing how prompting steers behavior

Authors: Fan L. Cheng , Nikolaus Kriegeskorte
URL: https://arxiv.org/abs/2606.03093
Abstract:

Prompting steers large language models (LLMs) and vision-language models (VLMs) without weight updates, but it remains unclear how instruction changes reshape internal representations to produce behavior. We introduce a nested geometric decomposition framework that treats prompting as a transformation of the representational geometry of the content following the prompt. For each prompt pair, we align representations of the same stimuli under two prompts using increasingly expressive stimulus-invariant maps: translation, rigid transformation with uniform scaling, sequential axis scaling, affine transformation, and nonlinear transformation. We then causally test each map by replacing a single layer’s prompt-A hidden state for held-out stimuli with its mapped counterpart and measuring recovery of prompt-B representational geometry and behavior. Across three LLMs, three VLMs, and six text or image datasets spanning style, emotion, scene content, and number, prompts consistently reshape representations toward the instructed task structure. Cross-validated variance decomposition shows that much prompt-induced activation change is captured by shape-preserving maps, especially translation and rigid transformation with uniform scaling, while tier profiles reveal model- and task-specific routing strategies across layers. Crucially, although translation and rigid tiers already improve behavioral agreement, affine transformation is the first tier to nearly recover target-prompt task geometry and yields corresponding behavioral gains. This suggests that cross-dimensional linear mixing is a key mechanism by which prompts reorganize representations toward instructed task structure. Our framework decomposes prompt-induced representational change into interpretable geometric components and reveals how models route task-relevant structure to produce prompt-driven behavior.

33. The Shadow Price of Reasoning: Economic Perspective on Optimal Budget Allocation for LLMs

Authors: Xu Wan , Speed Zhu , Jianwei Cai , Guang Chen , XiMing Huang , Wiggin Zhou , Mingyang Sun
URL: https://arxiv.org/abs/2606.03092
Abstract:

Inference-time scaling has emerged as a critical avenue for enhancing Large Language Models’ performance, yet real-world deployment is constrained by strict computational budgets. In this work, we formulate inference budget allocation as a global constrained optimization problem governed by economic principles. By modeling per-query reasoning utility with a shifted-surge function, we derive an optimal allocation policy based on a global shadow price that equilibrates marginal utility under resource scarcity. Based on this theory, we propose Constrained Latent-utility Equilibrium Allocation for Reasoning (CLEAR). It performs rational abandonment and reallocates resources from insolvent queries to solvable queries near their emergence thresholds. Extensive experiments on several reasoning tasks with different traffic streams demonstrate that CLEAR significantly improves the Pareto frontier of total token cost versus mean accuracy. In resource-scarce regimes, CLEAR achieves up to a 3x improvement in global accuracy compared to uniform allocation.

34. DELTAMEM: Incremental Experience Memory for LLM Agents via Residual Trees

Authors: Haoran Tan , Zeyu Zhang , Zhicheng Cao , Rui Li , Xu Chen
URL: https://arxiv.org/abs/2606.03083
Abstract:

Large Language Model (LLM)-based agents increasingly rely on memory to learn from experiences over continual interactions. However, storing experiences as independent, flat units leads to substantial redundancy and retrieval conflicts, as similar episodes repeat overlapping content and subtle scene variations cause retrieved memories to offer contradictory guidance. To address this, we introduce residual experience, positing that newly acquired experience is often an incremental variation of existing knowledge. We propose DeltaMem, a framework that organizes experience memory into two independent residual trees, one storing goal-conditioned task experience as reusable skills and another for scene-level environment knowledge. Each tree uses a root node for generalized base experiences and incremental delta nodes for subsequent variations, allowing related experiences to share a common foundation without duplication. For retrieval, a failure-penalized similarity scan locates the best match, reconstructing the full experience via root-to-match chain composition. An autonomous consolidation mechanism distills high-frequency paths into new root nodes, enabling the trees to self-organize from general heuristics to specialized variants. Experiments across diverse interactive environments show that DeltaMem consistently outperforms existing baselines. To facilitate future research, we release the code at this https URL .

35. CORE: Conflict-Oriented Reasoning for General Multimodal Manipulation Detection

Authors: Jinjie Shen , Yaxiong Wang , Yujiao Wu , Lechao Cheng , Tianrui Hui , Nan Pu , Zhihui Li , Zhun Zhong
URL: https://arxiv.org/abs/2606.03066
Abstract:

The rapid rise of generative AI has made multimodal fake news increasingly realistic and pervasive, posing severe threats to public trust and social stability. Existing detection methods rely heavily on manipulation-specific models and large-scale labeled data, resulting in poor generalization to emerging manipulation types. We observed that the essence of manipulated misinformation lies in its intrinsic conflicts, \textbf{i.e.,} semantic or physical inconsistencies either across modalities or with common world knowledge. Inspired by this observation, we propose \textbf{C}onflict-\textbf{O}riented \textbf{RE}asoning (\textbf{CORE}) framework, an effective paradigm that learns to endows multimodal large language models (MLLMs) with explicit conflict-capturing capability. To this end, CORE first constructs the Conflict Attribution Corpus (CAC) with fine-grained annotations of conflict factors and sources, providing essential data support for subsequent conflict perception training. By performing conflict-oriented representation enhancement and reasoning based on CAC, CORE achieves robust and generalizable conflict detection, effectively and rapidly adapting to unseen manipulation types with a few samples or in even zero-shot settings. Extensive experiments demonstrate that CORE surpasses state-of-the-art models. The dataset and code are publicly available at this https URL .

36. SkillDAG: Self-Evolving Typed Skill Graphs for LLM Skill Selection at Scale

Authors: Tong Bai , Zhenglin Wan , Pengfei Zhou , Xingrui Yu , Wangbo Zhao , Yang You , Ivor W. Tsang
URL: https://arxiv.org/abs/2606.03056
Abstract:

As LLM agents adopt large skill libraries, selecting the right subset becomes a structural problem rather than a similarity-matching one: skills depend on, conflict with, specialize, or duplicate one another, a structure invisible to both full enumeration and embedding similarity. We present SkillDAG, which models inter-skill relationships as a typed directed graph and exposes it to an LLM agent as an inference-time, agent-callable structural retrieval interface, queried and evolved during execution rather than baked into a fixed retrieval pipeline: each search returns vector matches, typed-edge neighbors, and conflict signals, and a propose-then-commit protocol lets the agent register execution-backed edges so the graph accumulates structure across episodes. On ALFWorld and SkillsBench with MiniMax-M2.7, SkillDAG reaches 67.1% success and 27.3% reward, exceeding the strongest reported Graph-of-Skills baseline by +12.8 and +8.6 points; the advantage ports to gpt-5.2-codex, and intrinsic SkillsBench Ret@K rises from 65.5 to 78.2 under matched queries. These gains trace to isolable mechanisms: candidate ranking that stays robust as the pool grows 10x where a fixed seeding-diffusion pipeline degrades, and set-monotone online edits that enlarge ground-truth recall without evicting prior hits.

37. TriEval: A Resource-Efficient Pipeline for LLM Bias, Toxicity, and Truthfulness Assessment

Authors: Akshatha Srikantha , Manpreet Singh , Yash Jajoo , Shyamal Lakhanpal
URL: https://arxiv.org/abs/2606.03036
Abstract:

LLMs have evolved from basic chatbots to the backbone of the AI ecosystem, now widely used in healthcare, schools, and government services. The domain-wide adoption of LLMs necessitates continuous evaluation to ensure their safety and fairness. Common issues encountered after deploying LLMs include inconsistent outputs and hallucinations of incorrect information. Although numerous LLM evaluation tools exist, most are limited to testing a single parameter at a time or require massive computational resources that are not accessible to most researchers. TriEval addresses these challenges by evaluating LLM outputs across multiple parameters, including bias, toxicity, and truthfulness together, while minimizing computing resources. The pipeline is compatible with both open- and closed-source models and runs on a standard laptop without a GPU cluster. TriEval has been tested on four models: Llama 3 8B, Mistral 7B, Gemma 2 9B, and Claude Haiku. The results show clear differences between open-source and closed-source models, especially in terms of toxicity and truthfulness. TriEval is being released as open source to enable broader access for researchers with limited computational resources.

38. Inducing Reasoning Primitives from Agent Traces

Authors: Zhihan Lei , Jiarui Yan , Joshua Momo , William W. Cohen
URL: https://arxiv.org/abs/2606.02994
Abstract:

ReAct-style LLM agents often rediscover the same reasoning routines across problems, yet leave those routines trapped in transient scratchpads. We introduce Reasoning Primitive Induction, a single-pass method that mines successful ReAct traces, clusters recurrent reasoning moves, and converts the most frequent moves into a compact library of typed pseudo-tools. Each pseudo-tool is specified by a natural-language docstring interpreted by an LLM at invocation time, and a standard ReAct loop composes these primitives at test time. The central result is that induced libraries outperform the very agent that generated their traces: by +44pp on RuleArena NBA (30 -> 74), +30pp on MuSR team allocation (38 -> 68), and +22pp on NatPlan meeting planning (7 -> 29). Across five comparable subtasks spanning narrative deduction, rule application, and constraint-satisfaction planning, a single fixed configuration improves over zero-shot Chain-of-Thought on every subtask, matches or surpasses expert-authored decompositions, and outperforms AWM at lower average inference cost.

39. Toward a Modular Architecture for Embedded AI Agent Systems at the Edge

Authors: Marcus Rüb , Michael Gerhards
URL: https://arxiv.org/abs/2606.02862
Abstract:

The rise of Large Language Models (LLMs) has enabled agentic AI capable of complex reasoning and tool use; however, deploying such autonomy in pervasive computing environments remains challenging due to the strict memory and energy constraints of embedded microcontrollers. Existing frameworks typically assume server-class resources or continuous connectivity, leaving a gap for deeply embedded systems. This paper proposes a modular reference architecture for Embedded Agent Systems that bridges the divide between deterministic real-time control and agentic intelligence. We introduce a tiered design that decouples On-Device Agents - executing highly compressed neural networks and rule-based logic for low-latency, privacy-critical tasks - from Cloud-Augmented Agents that leverage Small Language Models (SLMs) for higher-level reasoning and planning. A key contribution is the integration of a cross-cutting Governance Layer, ensuring observability, policy enforcement, and safety across distributed fleets of autonomous devices. Rather than presenting purely empirical benchmarks, we analyze architectural design principles and trade-offs regarding latency, energy, and reliable execution in resource-constrained environments.

40. Traj-Evolve: A Self-Evolving Multi-Agent System for Patient Trajectory Modeling in Lung Cancer Early Detection

Authors: Sihang Zeng , Matthew Thompson , Ruth Etzioni , Meliha Yetisgen
URL: https://arxiv.org/abs/2606.02812
Abstract:

Modeling patient trajectories from longitudinal electronic health records (EHRs) requires reasoning over sparse, noisy, and long-context multimodal sequences. Existing LLM-based multi-agent systems address context length but process patients in isolation, failing to mirror how clinicians leverage accumulated experience from similar prior cases. We present Traj-Evolve, a self-evolving multi-agent system with two complementary evolving mechanisms. First, an Experience Pool (ExPool) acts as a non-parametric memory, indexing rejection-sampled reasoning traces to retrieve similar patients as few-shot contexts. Second, multi-agent reinforcement learning (MARL) via reward-ranked fine-tuning parametrically optimizes inter-agent and agent-memory collaboration. A leave-one-out cross-retrieval strategy unifies the two, aligning training- and inference-time behavior under retrieval augmentation. On a lung cancer prediction task utilizing up to five years of multimodal EHRs, Traj-Evolve outperforms 9 strong baselines on the overall population and a challenging never-smoker population. Analysis of the evolving dynamics highlights three key findings: (1) expanding the ExPool shifts optimal retrieval from diverse to specific samples; (2) under MARL, the manager agent’s prediction loss converges quickly while the worker agents’ temporal reasoning continues to benefit from more verified patients; and (3) the two mechanisms are complementary on the predicted risk, where ExPool improves specificity while MARL improves sensitivity.

41. ChatHealthAI: Aligning Electronic Health Record Representations with Large Language Models for Grounded Clinical Reasoning

Authors: Bo-Hong Wang , Baicheng Peng , Ruilin Wang , Jun Bai , Ziyang Song , Yue Li
URL: https://arxiv.org/abs/2606.02802
Abstract:

Large language models (LLMs) exhibit strong natural-language reasoning abilities for clinical decision support, but struggle to effectively model structured longitudinal electronic health records (EHRs). In contrast, EHR foundation models can learn predictive patient representations, yet lack interpretable language-based reasoning. To bridge this gap, we propose ChatHealthAI, a multimodal reasoning framework that aligns structured EHR representations from a pretrained EHR foundation model with the semantic space of a frozen LLM through a task-aware resampler. By integrating longitudinal patient representations with refined clinical event descriptions, ChatHealthAI enables clinically grounded natural-language reasoning while maintaining accurate patient prediction. We evaluated ChatHealthAI on three clinical predictive tasks from the EHRSHOT benchmark. Results show that ChatHealthAI improves reasoning quality and interpretability while preserving competitive predictive performance. These findings highlight the potential of integrating EHR foundation models with pretrained LLMs for interpretable clinical prediction.

42. Visual Graph Scaffolds for Structural Reasoning in Large Language Models

Authors: Runlin Lei , Xiaokui Xiao , Zhewei Wei
URL: https://arxiv.org/abs/2606.02673
Abstract:

Graphs have been used to enhance large language models (LLMs) for structured reasoning, mostly as external knowledge sources are provided to models at test time. In this paper, we take a different view: the value of graphs for LLMs lie not only in supplying information, but also in organizing reasoning. Inspired by how humans use graph-structured mind maps to organize branching and converging thoughts, we ask whether graphs can serve as an internal form of reasoning assistance. We study this question on multi-hop question answering tasks, where teacher-provided reasoning traces are rewritten as graph mind maps and used to guide a student model. Our experiments reveal a clear modality gap. When graph structures are flattened into text, their benefits become limited once direct answer hints are removed. Under this abstract guidance setting, both reasoning efficiency and answer quality degrade substantially. In contrast, visual graph guidance remains effective without direct answer clues, and its advantage persists after supervised fine-tuning and KL-based distillation. The above findings support the claim that graphs should be studied not only as external knowledge structures for LLMs, but also as visual scaffolds for organizing reasoning.

43. Language Models Need Sleep: Learning to Self-Modify and Consolidate Memories

Authors: Ali Behrouz , Farnoosh Hashemi , Vahab Mirrokni
URL: https://arxiv.org/abs/2606.03979
Abstract:

The past few decades have witnessed significant advances in the design of machine learning algorithms, from early studies on task-specific shallow models to more general deep Large Language Models (LLMs). Despite showing promising results in tasks that require instant prediction or in-context learning, existing models lack the ability to continually learn and effectively transfer their temporal in-context knowledge to their long-term parameters. Inspired by human learning process, we introduce a ‘‘Sleep’’ paradigm that allows the models to continually learn, distill their short-term fragile memories into stable long-term knowledge with replay, and recursively improve themselves with ‘‘Dreaming’’ process. In more detail, sleep consists of two stages: (1) Memory Consolidation: an upward distillation process, called Knowledge Seeding, where the memories of a smaller-self are distilled into a larger network to provide more capacity while preserving the knowledge. As a proof of concept, we present a new Generalized Distillation process for {Knowledge Seeding} (i.e., the combination of on-policy distillation with Reinforcement Learning (RL)-based imitation learning); (2) Dreaming: a self-improvement phase, where the model uses RL to generate a curriculum of synthetic data to rehearse new knowledge and refine existing capabilities without human supervision. Our experiments on long-horizon, continual learning, knowledge incorporation, and few-shot generalization tasks support the importance of the sleep stage.

44. AlignAtt4LLM: Fast AlignAtt for Decoder-Only LLMs at IWSLT 2026 Simultaneous Speech Translation Task

Authors: Quentin Fuxa , Dominik Macháček
URL: https://arxiv.org/abs/2606.03967
Abstract:

We describe AlignAtt4LLM, an IWSLT 2026 simultaneous speech translation system for English to German, Italian, and Chinese. The system is a synchronous cascade: Qwen3-ASR with forced alignment produces an incrementally updated source transcript, and Gemma-4 E4B-it translates that prefix under an MT-side AlignAtt policy. To our knowledge, this is the first application of AlignAtt to a decoder-only LLM, where the encoder-decoder cross-attention used by earlier AlignAtt systems is absent. We recover a usable policy by proposing (1) an explicit source span in the prompt, (2) offline selection of translation-specific alignment heads, (3) selective qk-fast replay of the draft-to-source attention block, and (4) runtime query/key capture that preserves model outputs bit-identically. On the IWSLT 2026 development set, AlignAtt4LLM outperforms the supplied baselines for the European target languages, English to German and English to Italian, in both the low-latency regime around 2 seconds and the high-latency regime below 4 seconds CU-LongYAAL. Results for English to Chinese are more mixed, but the method is not tied to Gemma-4: because AlignAtt4LLM only requires a deterministic prompt layout, calibrated attention heads, and query/key capture, the same policy can be reapplied to stronger translation-focused decoder-only MT backbones for non-European target languages.

45. Agentic Chain-of-Thought Steering for Efficient and Controllable LLM Reasoning

Authors: Yu Xia , Zhouhang Xie , Xin Xu , Byungkyu Kang , Prarit Lamba , Xiang Gao , Julian McAuley
URL: https://arxiv.org/abs/2606.03965
Abstract:

Large language models improve final-answer accuracy through extended chain-of-thought reasoning, but often spend tokens inefficiently and offer little inference-time control. Existing efficient reasoning methods control thinking length by shortening, early-stopping, or compressing traces, leaving how the model thinks implicit. In this paper, we propose Agentic Chain-of-Thought Steering (ACTS), which formulates reasoning steering as a Markov decision process where a controller agent adaptively steers a frozen reasoner during inference. At each step, the controller observes the reasoning trace and remaining thinking budget, then issues a steering action consisting of a reasoning strategy and a steering phrase that initiates the next reasoner step. This enables budget-aware strategy control for efficient reasoning while preserving the reasoner’s generation continuity. We initialize the controller agent from our constructed synthetic steering trajectories with multi-budget augmentation, and further optimize it via reinforcement learning with budget-conditioned reward shaping. Experiments across multiple benchmarks show that ACTS matches full-thinking performance with substantial token savings, and enables controllable accuracy-efficiency trade-offs across different reasoners and tasks. The code is available at this https URL .

46. Using Reward Uncertainty to Induce Diverse Behaviour in Reinforcement Learning

Authors: Anthony GX-Chen , Ankit Anand , Gheorghe Comanici , Zaheer Abbas , Eser Aygün , David Smalling , Shibl Mourad , Doina Precup , André Barreto , Mark Rowland
URL: https://arxiv.org/abs/2606.03962
Abstract:

Classical reinforcement learning (RL) typically seeks a deterministic policy that maximizes the expected sum of a scalar reward. Yet, modern applications such as language model fine-tuning or scientific discovery demand diversity. Existing remedies such as entropy regularization or diversity bonuses often require fragile trade-offs that sacrifice performance for stochasticity or rely on heuristic metrics that can misalign policy rankings. We argue that diversity is more naturally understood as the rational response to uncertainty in the reward. When the reward function is not perfectly known–as is the case with ambiguous preferences or imperfect reward models–committing to a single action can be sub-optimal. Building on this, we propose a fundamental reformulation of the RL objective by replacing the scalar reward with a distribution over reward functions, and applying a non-linear objective over sets of actions. The result is a framework in which calibrated behavioural diversity emerges naturally, remains controllable through the reward function distribution, and is obtained without sacrificing expected reward. Focusing on the contextual bandit setting, we derive a principled gradient estimator for this objective and prove that our formulation naturally generalizes both vanilla policy gradient and more recently developed action-set approaches. Our empirical results demonstrate that this framework offers a robust and theoretically grounded alternative for complex RL tasks where the traditional formulation of the problem fails to induce the desired breadth of agent behaviour.

47. Efficient ASR Training with Conversations that Never Happened

Authors: Máté Gedeon , Péter Mihajlik
URL: https://arxiv.org/abs/2606.03957
Abstract:

Conversational ASR for lower-resource languages and niche domains is limited by the scarcity of domain-matched multi-speaker training data. We propose an augmentation pipeline that generates scenario-level dialogues with participant metadata, maps speaker attributes to TTS voice profiles, and assembles synthesized utterances into speaker-aware simulated conversations. We evaluated five LLM families under single-generator, fixed-budget mixture, and scale-up settings using the same FastConformer-Large training recipe for each one. We ran comprehensive evaluations on the Hungarian BEA-Dialogue benchmark corpus, with the method itself being applicable to any language given the resources for each component. The results show that synthetic conversations consistently improve speech recognition performance, but generator choice and data composition strongly affect the gains. Our largest training configuration, using only 67 hours of real conversations and 636 hours of simulated data, achieves better performance on the evaluation benchmark than a zero-shot model trained on 2700 hours of Hungarian speech. These findings indicate that LLM-generated conversational data synthesized with TTS is a practical complement to real conversational corpora for speech model training.

48. NetKV: Network-Aware Decode Instance Selection for Disaggregated LLM Inference

Authors: Mubarak Adetunji Ojewale
URL: https://arxiv.org/abs/2606.03910

Abstract:

Disaggregated LLM inference forces the KV cache to traverse the datacenter network before decoding begins, so transfer time enters directly into the Time to First Token (TTFT) budget. Current schedulers route on compute load and prefix-cache locality alone, ignoring the topological distance and dynamic congestion between prefill and decode instances. We close this gap with a thin operator-to-scheduler interface, the network cost oracle, and we prove that ignoring the network term renders cache-aware-only scheduling arbitrarily suboptimal as context length grows. NetKV, the O( D ) per-request greedy that consumes this oracle, has tier rankings that are provably robust to stale telemetry. On a 64-GPU four-tier fat-tree simulator driven by Mooncake traces, NetKV reduces mean TTFT by up to 21.2% over round-robin and 17.6% over a tuned cache+load-aware scheduler, lifts SLO attainment by up to 20.1 percentage points, and keeps the Time Between Tokens overhead below 0.5 ms in every condition tested, with no changes to the transport, inference engine, or hardware.

49. Agent libOS: A Library-OS-Inspired Runtime for Long-Running, Capability-Controlled LLM Agents

Authors: Yingqi Zhang
URL: https://arxiv.org/abs/2606.03895
Abstract:

Large language model (LLM) agents are evolving from request-response assistants into long-running software actors: they maintain state across model calls, fork subtasks, wait for external events, request human authority, generate tools, and perform side effects that must be resumed and audited. This paper presents Agent libOS, a library-OS-inspired runtime substrate for LLM agents. Agent libOS runs above a conventional host operating system; it does not implement hardware drivers, kernel-mode isolation, or a POSIX-compatible operating system. Instead, it treats an agent as an AgentProcess: a schedulable execution subject with process identity, parent-child lineage, lifecycle state, a tool table derived from an AgentImage, typed Object Memory, explicit capabilities, human queues, checkpoints, events, and audit records. Its central design rule is tools are libc-like wrappers; runtime primitives are the authority boundary. Filesystem access, object access, sleeps, human approval, JIT tool registration, and external side effects are checked at primitive boundaries under explicit capabilities and policy. We describe the design, threat model, Python prototype, and safety-oriented evaluation. The current prototype implements async scheduling, namespace-local Object Memory, runtime-integrated human approval, one-shot permission grants, per-process working directories, shell and image-registration primitives, Deno/TypeScript JIT tools over a libOS syscall broker, filesystem/object bridge tools, an injectable Resource Provider Substrate, deterministic demos, real-model smoke scripts, and 123 regression tests at the time of writing. Rather than improving planner accuracy, Agent libOS demonstrates a runtime substrate in which long-running LLM agents can be scheduled, authorized, resumed, and audited without treating tool dispatch as the trust boundary.

50. Beyond Encoder Accumulation: Measuring Encoder Roles in Multi-Encoder VLMs

Authors: Wei Ding , Yudong Zhang , Ruobing Xie , Xingwu Sun , Jiansheng Chen , Yu Wang
URL: https://arxiv.org/abs/2606.03879
Abstract:

As foundation models scale toward fusing more heterogeneous visual streams, understanding how diverse encoders interact under joint training becomes a prerequisite for principled design. Yet large vision-language models (LVLMs) currently lack the tools to do so, and parameter-efficient encoder configurations remain hard to identify before training. To re-examine encoder roles under joint training, on the 16-benchmark Cambrian-1 suite we retrain and evaluate all 31 non-empty subsets of five common vision encoders under a unified pipeline (~20k GPU-hours total), and report three findings. First, retraining each subset from scratch reveals encoder rankings that differ from those obtained by masking encoders on a fixed checkpoint, including which encoder ranks first overall. Second, we decompose each encoder’s contribution into two axes, Capacity, the score an encoder reaches on its own, and Necessity, the drop when it is removed from the full pool. The two axes are not interchangeable. Pairing the two highest-Capacity encoders is suboptimal, while pairing a high-Capacity anchor with an adaptive complement matches the full five-encoder model. Adding further encoders beyond this pair yields only marginal gains. Third, at fixed parameter count, per-encoder pre-projector effective rank explains the residual score variation. The strongest pairs combine an anchor whose rank survives joint training with a complement whose rank expands under it, suggesting that higher-rank, less-collapsed projector inputs correspond to a more favorable optimization regime at the encoder-projector interface. Together, the Capacity-Necessity decomposition and the pre-projector rank analysis, along with comprehensive evaluation through retraining, expose a methodological gap in multi-encoder LVLM design, and offer concrete primitives for closing it.

Authors: Jiachen Li , Reina Szeyi Chan , Akshat Choube , Xiang Zhi Tan , Elizabeth Mynatt , Varun Mishra
URL: https://arxiv.org/abs/2606.03876
Abstract:

With the growing prevalence of modern ubiquitous computing technologies, multi-modal tracking systems hold promise for providing timely awareness and reassurance to stakeholders such as remote family members (RFMs) of older adults, who play a central role in care coordination. However, combining heterogeneous data streams into high-level, meaningful content - such as retrospective summaries - remains challenging. While recent work has demonstrated the promise of large language models (LLMs) for interpreting multi-modal tracking data, less attention has been given to generating narrative accounts for stakeholders like RFMs, who possess rich personal knowledge of older adults and strong emotional responsibility, yet have limited visibility into their daily lives and limited capacity for caregiving. In this work, we explore how LLMs can be used to generate retrospective summaries from multi-modal tracking data for RFMs of older adults. We leveraged and customized an existing system, Vital Insight, to generate initial summaries on different dates and data availability scenarios as technology probes, and conducted interviews with 11 RFMs to gather feedback. Based on these insights, we redesigned the system into a multi-layer, multi-agent, insight-driven summary approach that builds from objective statistics and descriptions to enriched, context-aware narratives. We then compared the redesigned summaries with the initial versions through a survey with the same 11 RFMs and found significant improvements in satisfaction, perceived helpfulness, trust, and willingness to receive the summaries. We conclude by presenting design implications for AI-generated summaries for RFMs and broader contexts, emphasizing the need to support RFMs’ sensemaking shift from simply presenting ‘‘What’’ data were collected, to explaining ‘‘How’’ is my loved one doing and ‘‘Why’’.

52. A Training-Free Mixture-of-Agents Framework for Multi-Document Summarization using LLMs and Knowledge Graphs

Authors: Cuong Vuong Tuan , Trang Mai Xuan , Tien-Cuong Nguyen , Vu-Duc Ngo , Thien Van Luong
URL: https://arxiv.org/abs/2606.03867
Abstract:

Multi-Document Summarization (MDS) plays a critical role in distilling essential information from collections of textual data. Existing approaches often struggle to capture complex inter-document relationships, rely heavily on large amounts of labeled data for supervised training, or exhibit limited generalization across domains and languages. To address these limitations, we present a training-free mixture-of-agents framework for MDS that leverages the complementary strengths of large language models (LLMs) and knowledge graphs. Our approach decomposes summarization into specialized agent tasks: extractive selection, knowledge-aware abstraction, and iterative refinement, each operating without task-specific fine-tuning. We unify their outputs using a multi-perspective consistency mechanism guided by LLMs. Experiments across four datasets in English and Vietnamese demonstrate state-of-the-art or competitive performance, validating the effectiveness and adaptability of our modular design.

53. Taiji: Pareto Optimal Policy Optimization with Semantics-IDs Trade-off for Industrial LLM-Enhanced Recommendation

Authors: Yuecheng Li , Zeyu Song , Jing Yao , Chi Lu , Peng Jiang , Kun Gai
URL: https://arxiv.org/abs/2606.03866
Abstract:

Scaling recommender systems via large language models (LLMs) has become a prominent trend in the industry. However, aligning the LLM’s semantic space with the recommender’s ID space via post-training (e.g., SFT and RL) remains challenging. Existing LLM4Rec paradigms are bottlenecked by two main issues: (1) the difficulty of measuring and improving chain-of-thought (CoT) quality in open-domain recommendation during SFT, and (2) the neglect of the trade-off between LLM semantic rewards and recommendation preference rewards during RL alignment. Inspired by these challenges, we present Taiji, a novel LLM-as-Enhancer framework designed for industrial recommender systems. To overcome the SFT bottleneck, we utilize reverse-engineered reasoning and open-ended rejection sampling to generate high-quality, domain-specific CoT data. To resolve the RL alignment issue, we propose Pareto Optimal Policy Optimization (POPO), which adaptively adjusts cross-domain reward weights. Theoretically, it achieves an optimal trade-off between the semantic world knowledge of LLMs and the collaborative ID features representing online user preferences. Extensive offline evaluations and online A/B tests validate the effectiveness of Taiji. Deployed on Kuaishou’s advertising platform since May 2026, Taiji currently serves over 400 million users daily, yielding significant commercial revenue and demonstrating its robust scalability in web-scale environments.

Authors: Yinsheng Yao , Hongxiang Zhang , Weixi Tong , Tianyi Zhang
URL: https://arxiv.org/abs/2606.03852
Abstract:

Large language models often generate code with bugs. Existing methods rely on feedback signals such as test failures and self-critiques to iteratively refine the generated code. Such signals are either too coarse-grained or too high-level, which is not sufficient to inform the model where to fix the bug. In this work, we present Flare, an iterative framework with a lightweight diagnostic model that predicts line-level suspiciousness signals for bug localization and code refinement. Given the inherent uncertainty of diagnostic predictions, Flare searches over the top-k suspicious regions and selects the best candidate according to execution outcomes. Experiments on LiveCodeBench and BigCodeBench with five base LLMs show that, even without candidate search (k=1), Flare outperforms the strongest baseline with an absolute improvement from 1.72% to 7.42%. Furthermore, searching over 10 candidates yields an average improvement of 8.50% compared with no candidate search. When evaluated in isolation, our lightweight diagnostic model achieves the best performance compared with recent fault localization methods, demonstrating that it can provide reliable fine-grained guidance for code refinement.

55. Clustered Self-Assessment: A Simple yet Effective Method for Uncertainty Quantification in Large Language Models

Authors: Qi Cao , Takeshi Kojima , Andrew Gambardella , Helinyi Peng , Yutaka Matsuo , Yusuke Iwasawa
URL: https://arxiv.org/abs/2606.03846
Abstract:

Large language models (LLMs) demonstrate remarkable performance across diverse tasks, but they often generate responses that appear plausible while being factually incorrect. This problem is compounded by the lack of explicit uncertainty estimates, which makes it difficult for users to judge the reliability of model outputs. Existing uncertainty quantification methods typically rely on indirect signals, such as entropy across sampled generations. These signals can be difficult to interpret and do not fully leverage the model’s ability to assess its own uncertainty. We propose a simple yet effective self-assessment method for uncertainty quantification in LLMs. Our approach groups sampled generations into semantically distinct clusters, converts them into answer options in a structured multiple-choice question, and uses the probability assigned by the LLM to each option as a confidence estimate. Experiments across multiple models and datasets show that our method consistently outperforms baseline approaches. Notably, it achieves competitive performance with as few as two additional samples, demonstrating both its effectiveness and efficiency.

56. AI Agents Enable Adaptive Computer Worms

Authors: Jonas Guan , Tom Blanchard , Hanna Foerster , Hengrui Jia , Gabriel Huang , Nicolas Papernot
URL: https://arxiv.org/abs/2606.03811
Abstract:

A computer worm is malware that spreads on a network by replicating itself from one machine to another. Traditional worms, like WannaCry, exploited predetermined vulnerabilities, and their spread can be halted by patching those vulnerabilities. Here we show that artificial intelligence (AI) agents enable a fundamentally new threat: a worm that generates tailored attack strategies to each target it encounters. The worm parasitically uses compromised machines to run open-weight large language models (LLMs) to sustain its reasoning, or extend its reach for further attacks. Deployed on a network of machines spanning Linux, Windows, and IoT (Internet of Things) devices, the worm propagated by exploiting common, real-world corporate network vulnerabilities. Since the worm is powered by stolen compute, the attacker’s marginal cost per new infection is zero. This creates a destabilizing economic asymmetry between attackers and defenders. Moreover, because the worm requires no commercial AI platform, centralized safety controls, such as service refusals or rate limiting, are structurally irrelevant. Our results demonstrate that self-sustaining AI-driven cyber-threats are no longer theoretical. We must prepare for autonomous generative adversaries: malware systems that propagate without human operators and are defined not by fixed exploit code, but by the capacity to reason about targets, adapt to observations, and synthesize attack logic in real time.

57. Trading Human Curation for Synthetic Augmentation in RLVR

Authors: Akshansh , Leonardo Rosa Rodrigues , Michael Korostelev , Youssef Hassan , Mark E. Whiting
URL: https://arxiv.org/abs/2606.03800
Abstract:

The supply of high-quality training tasks is a central bottleneck for reinforcement learning from verifiable rewards (RLVR) on agentic language models. Each task requires a sandboxed setup, a prompt, and a hand-authored reward function, and only tasks that pass a quality bar produce useful training signal. Hand-curation at this quality bar does not scale economically to the task counts effective RL training requires, and the substitution rate between automatically generated task variants and human-authored ones is not yet established. We investigate using pre-specified, gate-filtered augmentations of a small hand-authored base as a substitute for additional human curation during RLVR. We formalize the cost-adjusted trade rate $\rho_{\text{cost} }$ between augmented and human-authored tasks, measure it through a controlled ablation across training corpora with varying augmentation share, and characterize the end-to-end economics of the augmentation pipeline. Substituting augmented content for additional human-authored tasks retains aggregate held-out generalization on a ten-benchmark suite spanning code, instruction following, reasoning, and multi-turn agentic function-calling. The cost-adjusted trade rate $\rho_{\text{cost} }$ between gated synthetic and human-authored RLVR tasks stays in $[1.4\times, 11.6\times]$ across the plausible $c_{\text{human} }/c_{\text{aug} }$ range.

58. E2LLM: Towards Efficient LLM Serving in Heterogeneous Edge/Fog Environments

Authors: Truong-Thanh Le , Amir Taherkordi , Hoang-Loc La , Frank Eliassen , Phuong Hoai Ha , Peiyuan Guan
URL: https://arxiv.org/abs/2606.03770
Abstract:

Large Language Models (LLMs) have become integral to modern applications, yet their deployment remains challenging. Beyond executing the models themselves, practical deployment must address cost efficiency, low latency, and optimal resource utilization. Conventional approaches typically assume that an entire model can be hosted on a single device, which does not hold in many real-world scenarios, particularly in Edge and Fog environments where device resources are constrained. In this paper, we introduce E2LLM, a framework designed to enable efficient LLM deployment in such resource limited settings. Rather than simply partitioning a single model across all available devices, E2LLM replicates the full model across multiple groups of devices (replicas) and applies model parallelism within each replica. Each replica is assigned a specialized role PREFILL or DECODER based on its efficiency in handling input and output tokens. This separation leverages the inherent differences between these two phases of LLM inference. To effectively organize devices, we utilize a Genetic Algorithm to form clusters that maximize system performance. Within each cluster, we apply Dynamic Programming to determine an optimal partitioning strategy that minimizes bottlenecks in model-parallel execution. Experimental results demonstrate that our approach adapts robustly to varying workloads, including scenarios with significant variation in input and output token lengths. Compared to the Splitwise baseline, E2LLM reduces average waiting time by over 50% under high-demand conditions

59. Merit or networks? What decides where research is published

Authors: Ning Li
URL: https://arxiv.org/abs/2606.03763
Abstract:

Does scientific publishing reward the quality of ideas or the advantage of connections? The question is universal to prestige-driven science, yet it has resisted decades of study because a paper’s quality could not be gauged ahead of its publication fate without using that fate as the yardstick. We break this constraint by measuring a paper’s idea quality directly from its text, before publication, using a discipline-trained LLM evaluator that scores the idea without seeing author names or outcomes. Using economics as a case study, we combine this text-legible idea-quality score with an execution-quality rubric, a connection index, an author-ability index, and an off-the-shelf language-model text score to estimate a five-input production function for journal placement across 6,208 economics working papers. The inputs are not rivals but a sequence along the ladder of prestige. Execution sets a meritocratic floor and is the largest input overall. Text-legible idea quality grades the rungs in between. Connections set a favoritism ceiling that bites mainly near the apex, the most selective journals. Connections work through two additive channels: connected authors write papers that score higher, and at equal scores their papers are still more likely to place better. Yet this advantage is bounded. Connections raise the odds of every rung without making the apex the typical outcome for ordinary ideas, and even the highest-scoring papers face real friction reaching the visible journal ladder. The result nests, rather than chooses between, the meritocracy and network accounts of how science is published.

60. Tool-Aware Optimization with Entropy Guidance for Efficient Agentic Reinforcement Learning

Authors: Hongye Cao , Nuo Yan , Haoyuan Deng , Ziwei Wang , Tianpei Yang , Jing Huo , Yuyao Zhang , Yang Gao
URL: https://arxiv.org/abs/2606.03762
Abstract:

Agentic reinforcement learning (RL) equips large language models (LLMs) with tool-use capabilities that substantially improve reasoning on complex tasks. However, integrating external tools often destabilizes training: over-reliance on tools can induce input distribution shift, while overly conservative tool use limits effective exploration. To address this issue, we propose a unified framework TAO-RL that couples tool-aware trajectory filtering with entropy-guided exploration for efficient policy optimization. Specifically, at the data level, TAO-RL filters rollout trajectories along two criteria: discarding those where all tool invocations fail to execute, and removing those where all rollouts are either correct or incorrect, as both cases yield degenerate advantage estimates that contribute no discriminative learning signal. This joint filtering retains data that are both tool-capable and informative, establishing a high-quality training distribution. At the algorithmic level, we introduce a tool-aware entropy-guided bonus that reshapes the advantage function at post-tool-call tokens, encouraging the policy to explore more diverse reasoning paths at critical decision points. These two components are mutually reinforcing: trajectory filtering establishes a clean and informative training foundation, while entropy-guided exploration drives stronger reasoning behaviors at critical tool-interaction junctures. Extensive experiments on 7 challenging reasoning benchmarks across 3 model scales demonstrate the superiority of TAO-RL over existing methods.

61. Ultralytics YOLO26: Unified Real-Time End-to-End Vision Models

Authors: Glenn Jocher , Jing Qiu , Mengyu Liu , Shuai Lyu , Fatih Cagatay Akyon , Muhammet Esat Kalfaoglu
URL: https://arxiv.org/abs/2606.03748
Abstract:

Real-time vision demands models that are accurate, efficient, and simple to deploy across diverse hardware. The YOLO family has become widely deployed for this reason, yet most YOLO detectors still rely on non-maximum suppression at inference, carry heavy detection heads due to Distribution Focal Loss, require long training schedules, and can leave the smallest objects without positive label assignments. We present Ultralytics YOLO26, a unified real-time vision model family that addresses these limitations through coordinated architecture and training advances. YOLO26 uses a dual-head design for native NMS-free end-to-end inference and removes DFL entirely, yielding a lighter head with unconstrained regression range. Its training pipeline combines MuSGD, a hybrid Muon-SGD optimizer adapted from large language model training; Progressive Loss, which shifts supervision toward the inference-time head; and STAL, a label assignment strategy that guarantees positive coverage for small objects. Beyond detection, YOLO26 introduces task-specific head and loss designs for instance segmentation, pose estimation, and oriented detection, producing consistent gains across tasks and scales. The family spans five scales (n/s/m/l/x) and supports detection, instance segmentation, pose estimation, classification, and oriented detection in a single pipeline, with an open-vocabulary extension, YOLOE-26, for text-, visual-, and prompt-free inference. Across all scales, YOLO26 achieves 40.9-57.5 mAP on COCO at 1.7-11.8 ms T4 TensorRT latency, advancing the accuracy-latency Pareto front over prior real-time detectors, while YOLOE-26x reaches 40.6 AP on LVIS minival under text prompting. Code and models are available at this https URL .

62. A Close Look At World Model Recovery In Supervised Fine-Tuned LLM Planners

Authors: Patrick Emami , Nan Qiang , Peter Graf
URL: https://arxiv.org/abs/2606.03685
Abstract:

Supervised fine-tuning (SFT) improves end-to-end classical planning in large language models (LLMs), but do these models also learn to represent and reason about the planning problems they are solving? Due to the relative complexity of classical planning problems and the challenge that end-to-end plan generation poses for LLMs, it has been difficult to explore this question. In our work, we devise and perform a series of interpretability experiments that holistically interrogate world model recovery by examining both internal representations and generative capabilities of fine-tuned LLMs. We find that: a) Supervised fine-tuning on valid action sequences enables LLMs to linearly encode action validity and some state predicates. b) Models that struggle to use output probabilities for classifying action validity may still learn internal representations that separate valid from invalid actions. c) Broader state space coverage during fine-tuning, such as from random walk data, yields more accurate recovery of the underlying world model. In summary, this work contributes a recipe for applying interpretability techniques to planning LLMs and generates insights that shed light on open questions about how knowledge is represented in LLMs.

63. CoEval: Ranking Language Models for Custom Tasks Without Labeled Data or Trustworthy Benchmarks

Authors: Alexander Apartsin , Yehudit Aperstein
URL: https://arxiv.org/abs/2606.03650
Abstract:

Choosing or ranking language models for a specific application is hardest when no task-specific labeled data exists, and standard public benchmarks cannot be trusted, their items having likely leaked into pretraining, so scores reflect memorization rather than fitness. We present CoEval, an open-source, reusable framework that closes this gap end to end: from only a description of a task or domain, teacher models synthesize a fresh, attribute-controlled benchmark with no human labels, contamination-free because items are generated anew on each run, and a cross-family judge ensemble ranks candidate models with no human raters. Validated where ground truth exists, CoEval recovers the true model ranking and tracks ground-truth correctness at ho=0.86. The label-free judging needs no human calibration because judge-panel composition (vendor diversity), not size, drives reliability: a small, well-chosen cross-family panel is most reliable, while a single judge can be anti-correlated with ground truth (judge-choice regret 0.35) and the ensemble never is. Generated items show zero verbatim 13-gram overlap with five major public benchmarks; the panel cancels verbosity bias and precludes same-family self-preference. A four-task study produced 7,978 evaluations for USD 5.89. The same declarative pipeline applies to any domain and is cheap enough to re-run on every model release: a label-free, contamination-free leaderboard any team can regenerate for its own application.

64. Safety Measurements for Fine-tuned LLMs Should be Grounded in Capability

Authors: Krishnapriya Vishnubhotla , Hillary Dawkins , Isar Nejadgholi , Svetlana Kiritchenko
URL: https://arxiv.org/abs/2606.03648
Abstract:

Adapting foundation large language models to a user’s task or preferred style through fine-tuning can result in compromising the model’s safety. Previous works examined the effects of fine-tuning on model safety in limited and seemingly random experimental settings. We argue that anchoring fine-tuning to a specific capability goal is essential for avoiding arbitrary empirical choices, allowing us to draw meaningful conclusions about safety impacts, and to compare mitigation methods on a consistent basis. We conduct a multi-dimensional evaluation of the effects of fine-tuning on model behavior by focusing on capability as well as safety. Our results surface important issues that (1) fine-tuned models can produce incoherent generations in response to safety prompts, (2) automated safety judgments are unreliable for such incoherent outputs, and (3) the conclusions about the effects of fine-tuning can change depending on the choice of safety benchmark as well as the safety evaluator.

65. Black-box, Adaptive, Efficient, Transferable, Harmful, Applicable… Attacks Are All You Need to Break LLMs

Authors: Vincent Limbach , Jonas Dornbusch , David Lüdke , Stephan Günnemann , Leo Schwinn
URL: https://arxiv.org/abs/2606.03647
Abstract:

Accurately evaluating adversarial robustness is a longstanding challenge. A flawed attack design can inflate robustness estimates, making deployment risk assessment and defense comparison unreliable. Historically, standardized attacks such as AutoAttack have largely resolved this for image classifiers, providing a reliable evaluation baseline for systematic comparison across defenses. However, no equivalent exists for LLM jailbreak evaluation yet, where designing such an attack is considerably more difficult. A reliable attack must, among other things, be black-box compatible, applicable to arbitrary defense pipelines, and efficient, which no existing method jointly satisfies. We introduce Indirect Harm Optimization (IHO), a masked diffusion language model attacker trained via iterative preference optimization against a harmfulness judge, requiring only black-box access to the target. The same method can be used without modification as a strong adaptive attack on individual behaviors, or as an efficient amortized policy that transfers to held-out behaviors and unseen target models without fine-tuning. Even against layered defenses, such as a Circuit Breaker-trained model combined with an auxiliary detector, IHO improves attack success considerably over state-of-the-art approaches, without any defense-specific adaptation. Our results position IHO as a practical step toward the kind of standardized jailbreak evaluation that has improved reliability in the past. Code and models are available on GitHub and Hugging Face.

66. The Shape of Addition: Geometric Structures of Arithmetic in Large Language Models

Authors: Liuyuan Wen , Xun Zhu , Lihao Huang , Wenbin Li , Yang Gao
URL: https://arxiv.org/abs/2606.03645
Abstract:

Large Language Models exhibit paradoxical fragility in fundamental arithmetic, implying a disconnect between internal computation and discrete output. By analyzing the residual stream geometry during multi-operand addition, we identify the Iso-Raw-Sum Trajectory (IRST), a geometric structure where representations are anchored by semantic digits and modulated by continuous carry fibers. We propose the Noisy Quantization Model to explain this geometry, framing arithmetic errors as Geometric Slippages caused by internal neural noise pushing a continuous, latent Carry Potential across quantization thresholds. This geometric framework further elucidates Probe Versatility, explaining how lightweight probes can disentangle coexisting latent signals (such as ground truth versus hallucination) from a single activation vector. Finally, we validate these insights through a geometric consistency check method that effectively detects and corrects these quantization failures during inference. Our code is available at this https URL .

67. VidMsg: A Benchmark for Implicit Message Inference in Short Videos

Authors: Issar Tzachor , Michael Green , Rami Ben-Ari
URL: https://arxiv.org/abs/2606.03635
Abstract:

Understanding short online videos involves more than identifying visible objects and actions; video makers often include an underlying message or purpose in the clip. We introduce VidMsg, a benchmark for evaluating implicit message understanding in short, internet-native video clips. VidMsg contains 400 YouTube-derived clips across 9 practical topic areas and 52 fine-grained target messages, covering domains such as career and finance, education, health and well-being, culture, safety, sustainability, and lifestyle. VidMsg is constructed through a message-first pipeline: an LLM first translates target messages into indirect search scenarios, which are used to retrieve candidate clips. Human annotators then retain clips that convey the intended message without being overly explicit. VidMsg is designed primarily for bidirectional message-clip retrieval for scalable applications such as video search and recommendation, where systems must capture holistic video understanding. In addition to retrieval, VidMsg includes a diagnostic multiple-choice QA benchmark, where models select the intended message of a clip from semantically related alternatives. Experiments with contemporary video-language and retrieval models show that strong models often fail on VidMsg, because the task requires pragmatic inference, integration of contextual cues, and discrimination among semantically close messages. We also introduce VidVec-Msg, a baseline method that improves message-oriented retrieval while leaving substantial headroom for future work.

68. Building Reliable Long-Form Generation via Hallucination Rejection Sampling

Authors: Lin Li , Georgia Channing , Suhaas M Bhat , Gabriel Davis Jones , Yarin Gal
URL: https://arxiv.org/abs/2606.03628
Abstract:

Large language models (LLMs) have achieved remarkable progress in open-ended text generation, yet they remain prone to hallucinating incorrect or unsupported content, which undermines their reliability. This issue is exacerbated in long-form generation due to hallucination snowballing, a phenomenon where early errors propagate and compound into subsequent outputs. To address this challenge, we propose a novel inference-time hallucination mitigation framework, named Segment-wise HAllucination Rejection Sampling (SHARS), which uses an arbitrary hallucination detector to identify and reject hallucinated segments during generation and resample until faithful content is produced. By retaining only confident information and building subsequent generations upon it, the framework mitigates hallucination accumulation and enhances factual consistency. To instantiate this framework, we adopt semantic uncertainty as the detector and introduce several vital modifications to address its limitations and better adapt it to long-form text. Our method enables models to self-correct hallucinations without requiring external resources such as web search or knowledge bases, while remaining compatible with them for future extensions. Empirical evaluations on standardized hallucination benchmarks demonstrate that our method substantially reduces hallucinations in long-form generation while preserving or even improving the informativeness of generation. Code is available at: this https URL .

69. TurtleAI: Benchmarking Multimodal Models for Visual Programming in Turtle Graphics

Authors: Chao Wen , Jacqueline Staub , Adish Singla
URL: https://arxiv.org/abs/2606.03626
Abstract:

Vision-language models (VLMs) have been explored for visual programming, where they generate code to solve visual tasks. However, most prior work focuses on visual programming for productivity; it remains unclear how well current VLMs perform on education-oriented visual programming and what factors limit their performance. To bridge this gap, we introduce TurtleAI, a benchmark containing 823 tasks curated based on real-world visual programming tasks in the Turtle Graphics domain. Solving these tasks requires models to perceive geometric patterns, reason about spatial relationships, and synthesize Python code that faithfully reproduces geometric patterns. We evaluate 20+ VLMs, including GPT-5, GPT-4o, and Qwen2-VL-72B, and find that they struggle significantly, with most achieving success rates below 30%. To address these limitations, we propose a data generation technique that requires only a small set of seed samples. Fine-tuning Qwen2-VL-72B on the resulting synthetic data yields an improvement of about 20% on real-world tasks. Our failure analysis reveals that GPT-4o struggles with spatial reasoning and precise visual replication, whereas fine-tuning primarily improves the alignment between visual reasoning and code implementation.

70. Physics-Guided Policy Optimization with Self-Distillation

Authors: Ke Wang , Yuning Wu , Haoran Liu , Chaoqun Jia , Devin Chen , Kai Wei
URL: https://arxiv.org/abs/2606.03620
Abstract:

Self-distilled policy optimization (SDPO) has become a popular paradigm for LLM post-training, where a model learns from its own predictions conditioned on privileged information. SDPO, however, is sensitive to how much each update step should be trusted: corrections from a self-teacher can be highly informative on some batches and misleading on others, and applying them uniformly with a fixed step size can destabilize training. Drawing inspiration from viscous-fluid dynamics and formalizing the analogy at the SDE level, we propose Physics-Guided Policy Optimization (PGPO), which introduces an information-modulated step-size multiplier derived from a mutual-information estimate between the student’s predictions and the feedback-conditioned teacher. We show that this modulation preserves the order-1 weak-approximation guarantees of vanilla SGD, and incurs negligible overhead per iteration. We evaluate PGPO on the Science-QA dataset, where it outperforms SDPO on 3 of the 4 domains with gains of up to +4.5 points, while remaining stable in a setting where SDPO collapses late in training.

71. Exploiting Verification-Generation Gap: Test-Time Reinforcement Learning with Confidence-Conditioned Verification

Authors: Jiahui Li , Jianfeng Shan , Wenpei Chen , Shunyu Wu , Jian Lou , Wenjie Feng , Dan Li , See-Kiong Ng
URL: https://arxiv.org/abs/2606.03608
Abstract:

Test-time reinforcement learning has emerged as a promising paradigm for enhancing the complex reasoning abilities of large language models in a completely label-free manner. Despite existing studies focusing on Pass@1 performance, optimizing Pass@k remains under-explored yet critical in label-free settings, which measures generation coverage for sustained exploration. Optimizing Pass@k in label-free setting is highly non-trivial, as directly applying the Pass@k advantage designs effective for RLVR yields unsatisfactory performance. Through in-depth empirical analysis, we discover the root causes hindering performance: pseudo-label estimations for low-confidence samples have a high probability of being incorrect, while candidate answers for high-confidence samples suffer from severe diversity collapse. To overcome these hurdles, we propose TTRL-CoCoV (Test-Time Reinforcement Learning with Confidence-Conditioned Verification), a novel confidence-adaptive framework that expands Pass@k coverage and improves Pass@1 performance. Based on our key insight that verification capability generally leads generation capability, TTRL-CoCoV employs a confidence-conditioned mechanism: for high-confidence samples, it bootstraps verifier and applies an exploration-enhancing reward to prevent diversity collapse; for low-confidence samples, it delegates pseudo-label selection to the verifier to filter incorrect pseudo-labels; and for medium-confidence samples, it bypasses verification entirely. Extensive experiments demonstrate that TTRL-CoCoV outperforms the best competing methods across 6 widely-recognized benchmarks, achieves average absolute gains of +9.8% in Pass@1 and +18.7% in Pass@16 over TTRL, and even achieves absolute Pass@1 improvements of up to +5.0% across multiple reasoning benchmarks when compared against fully supervised RL methods. Our code repository: this https URL .

72. Testing LLM Arithmetic Reasoning Generalization with Automatic Numeric-Remapping Attacks

Authors: Malia Barker , Bishal Lakha , Edoardo Serra , Francesco Gullo
URL: https://arxiv.org/abs/2606.03606
Abstract:

Large language models achieve strong performance on arithmetic reasoning benchmarks, and one common response to arithmetic brittleness is to delegate computation to code. Yet models are still often used in settings where they must reason directly from natural language, and trustworthy models should solve small-number arithmetic word problems without external tools. Prior work shows that LLMs are sensitive to numerical variation: a model may solve an original problem but fail on structurally similar variants requiring the same reasoning procedure with different numbers. We ask whether this fragility persists under a stricter setting involving small, schema-preserving numeric changes that retain the original reasoning program and avoid large-number stress tests. We introduce an automatic algorithm for generating numeric-remapping attacks on arithmetic word problems. Unlike template-based perturbation methods requiring manual schemas or constraints, our approach derives problem-specific symbolic representations, generates constrained numeric remappings, recomputes gold answers, and realizes transformed questions through deterministic edits guided by LLM-generated edit plans. Stage-wise validation and a high-confidence audit retain reliable attacks, making the pipeline scalable with limited human intervention. We evaluate DeepSeek-R1 (70B), Gemma4 (31B), and GPT-OSS (120B) on GSM8K, MAWPS, and MultiArith. On GSM8K, completed runs show conditional accuracy drops of 12.16 to 25.82 percentage points. MAWPS and MultiArith are far more stable, with most attacked accuracies near or above 98%. These results show that numeric-remapping robustness depends strongly on dataset structure: GSM8K remains sensitive even when reasoning programs are preserved and answers are recomputed, while shorter, more regular datasets are more robust.

73. CauTion: Knowing When to Trust LLMs for Ensemble Causal Discovery

Authors: Bo Peng , Kaiwen Wu , Sirui Chen , Zhiheng Wang , Yu Qiao , Chaochao Lu
URL: https://arxiv.org/abs/2606.03602
Abstract:

Causal discovery from observational data remains challenging due to the fundamental limitations of purely statistical methods, such as statistical distinguishability within equivalence classes and sensitivity to finite sample sizes. While large language models (LLMs) offer a promising source of domain knowledge to complement statistical inference, existing LLM-augmented methods are vulnerable to LLM errors and incur high token costs. Moreover, reliance on a single data-centric algorithm can make results sensitive to algorithm-specific biases. To address these limitations, we propose CauTion, a framework that reliably integrates LLM domain knowledge into an ensemble of statistical causal discovery algorithms through consensus filtering and LLM reliability estimation. CauTion proceeds in three stages. First, an algorithm ensemble utilizes a consensus voting to resolve up to 96% of edges on which algorithms agree, achieving near-perfect accuracy on the filtered consensus edges. Second, a trust-calibrated arbitration mechanism estimates the relative reliability of the LLM and the algorithms via an annotation-free trust calibration procedure, which is then utilized to govern a trust-weighted voting process that restricts LLM arbitration exclusively to edges with unreliable algorithmic evidence. Third, a cycle repair step is applied to guarantee the final causal graph is validly acyclic. Experiments on six datasets demonstrate that CauTion consistently outperforms both data-centric and LLM-augmented baselines, with larger gains on larger graphs and strong robustness to LLM errors. Code is available at this https URL .

74. DDOR: Delta Debugging for Explainable Overrefusal Testing and Repair

Authors: Qinyan Zhou , Peixin Zhang , Jun Sun , Haonan Zhang , Dongxia Wang
URL: https://arxiv.org/abs/2606.03601
Abstract:

While safety alignment and guardrails help large language models (LLMs) avoid harmful outputs, they can also induce overrefusal, i.e., unwarranted rejection of benign queries that merely appear risky. We present DDOR (Delta Debugging for OverRefusal), a fully automated and explainable framework for overrefusal testing and repair in a black-box setting, where only model inputs and outputs are accessible and internal safety mechanisms remain opaque. DDOR applies delta debugging to localize minimal refusal-triggering fragments (mRTFs) that provide phrase-level, explainable evidence for why a refusal occurs. Conditioned on these mRTFs, DDOR generates diverse, context-rich prompts and performs multi-oracle validation to filter intrinsically unsafe or ambiguous cases, producing scalable and model-specific overrefusal test suites (approximately 1K cases per model). Beyond evaluation, we further leverage localized mRTFs to perform targeted prompt repair, substantially reducing overrefusal while preserving the original intent and maintaining safety on genuinely harmful inputs. Overall, DDOR offers a practical end-to-end solution to both evaluate and mitigate overrefusal, improving LLM usability without sacrificing safety.

75. When Attention Collapses: Stage-Aware Visual Token Pruning from Structure to Semantics

Authors: Jiahui Wang , Kai Zhang , Mai Han , Huanghe Zhang
URL: https://arxiv.org/abs/2606.03569
Abstract:

Vision-Language Models (VLMs) have demonstrated remarkable capabilities but suffer from significant computational overhead during inference. While visual token pruning offers a promising solution, existing methods predominantly rely on initial attention scores. This single-metric paradigm presents a critical flaw: high attention scores inherently collapse onto semantically similar regions, thereby severely reducing feature diversity and discarding vital contextual details. To address this, we introduce Structure-to-Semantics (STS), a novel two-stage visual token pruning framework that explicitly decouples the pruning process. The first stage employs a repulsion-based sampling mechanism to maximize spatial and structural diversity. The second stage leverages instruction-aware cross-attention to precisely filter out prompt-irrelevant tokens. This two-stage synergy constitutes the core of STS, first ensuring geometric coverage and then refining the retained tokens according to semantic relevance. Extensive evaluations demonstrate that STS mitigates the redundancy caused by attention-based selection, improving both structural diversity and fine-grained task alignment of the preserved visual tokens.

76. \textsc{CR-Seg}: Attention-Guided and CoT-Enhanced Coarse-to-Refined Reasoning Segmentation

Authors: Yifan Cao , Xiaocui Yang , Faxian Wan , Shi Feng , Daling Wang , Yifei Zhang
URL: https://arxiv.org/abs/2606.03564
Abstract:

Reasoning segmentation aims to segment target objects described by complex language through joint visual-textual reasoning. Existing methods typically rely on either learned semantic tokens to bridge Multimodal Large Language Models (MLLMs) and segmentation models, suffering from difficult cross-modal alignment, or explicit spatial prompts such as bounding boxes, which may lose holistic response semantics. To address these limitations, we propose Attention-Guided and CoT-Enhanced Coarse-to-Refined Reasoning Segmentation, termed CR-Seg, a two-stage framework for coarse-to-refined reasoning segmentation. Specifically, we design an Extract Attention Maps and Points (EAP) module to extract attention maps for coarse target localization and select informative points, both of which are fed into SAM for mask refinement. To alleviate reasoning–answer inconsistency, we further introduce Global-to-Local Chain-of-Thought (GLCoT), which guides the model to reason progressively from global scene context to local target details. Extensive experiments on reasoning segmentation benchmarks demonstrate the effectiveness of CR-Seg.

77. Learn from Your Mistakes: Tree-like Self-Play for Secure Code LLMs

Authors: Wenqi Chen , Ziyan Zhang , Bing Wang , Lin Liu , Hengheng Zhang , Zhengsu Chen
URL: https://arxiv.org/abs/2606.03489
Abstract:

While Large Language Models (LLMs) excel in code generation, they remain prone to replicating subtle yet critical vulnerabilities endemic to their training data. Current alignment techniques, such as Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL), typically apply coarse-grained optimization at the sequence level. This approach often fails to address the localized nature of security flaws, where a single incorrect token choice can compromise an entire program. To bridge this gap, we introduce Tree-like Self-Play (TSP), a framework that reframes secure code generation as a fine-grained sequential decision process. Unlike standard methods that blindly maximize likelihood, TSP constructs a decision tree where the model explores branching trajectories–generating both secure “golden paths” and vulnerable variants. By treating code generation as a self-play game, the model learns to strictly discriminate against its own localized errors. This provides a dense, on-policy learning signal that forces self-correction precisely at the critical decision nodes where vulnerabilities typically emerge. Our experiments demonstrate that TSP fundamentally enhances model reliability. In Python security benchmarks, TSP boosts CodeLlama-7B’s pass rate (SPR@1) to 75.8%, significantly outperforming SFT (57.0%) and unstructured self-play baselines. Crucially, TSP induces robust out-of-distribution generalization: the model not only reduces vulnerabilities in unseen categories (CWEs) by 24.5% but also successfully transfers security principles learned from C/C++ to diverse languages, including Python, Go, and JavaScript. This suggests that TSP does not merely memorize patches, but internalizes abstract, language-agnostic security logic.

78. NeuroArmor: Safe-Variant-Guided Representation Consistency for Selective Re-Anchoring in Jailbreak Defense

Authors: Zhongyang Lin , Ziran Zhao , Feifei Zhai , Pengyuan Liu
URL: https://arxiv.org/abs/2606.03486
Abstract:

Large language models remain vulnerable to jailbreak attacks that hide harmful intent behind seemingly ordinary requests such as role-play, translation, encoding, adversarial suffixes, and multi-turn buildup. Existing defenses still struggle to handle these attacks without over-blocking benign but sensitive requests, partly because they often apply the same action to every prompt and therefore fail to balance safety and helpfulness. We propose NeuroArmor, a white-box runtime defense that uses prompt-specific safe variants as a local safety reference for deciding when intervention is needed and, once triggered, as safe targets for intervention. For each prompt, NeuroArmor builds K safe variants, compares the prompt state against this local safe reference in hidden-state space, and routes anomalies either to a refusal branch for malicious prompts or to a helpful recovery branch for borderline benign prompts. On Llama-3-8B-Instruct, NeuroArmor reduces malicious attack success rate (ASR) from 41.56% to 1.57% while lowering benign false positive rate (FPR) on the shared benign pool from 30.26% to 22.05%; matched baselines remain substantially weaker on this trade-off. External-judge and manual behavioral evaluations further show that the remaining non-blocked outputs are much less likely to be operationally harmful. Overall, NeuroArmor provides a more effective runtime strategy for jailbreak defense by combining prompt-specific consistency checking, routing, and selective intervention.

79. Analyzing Stream Collapse in Hyper-Connections: From Diagnosis to Mitigation

Authors: Ekaterina Alimaskina , Gleb Molodtsov , Aleksandr Beznosikov
URL: https://arxiv.org/abs/2606.03483
Abstract:

Hyper-Connections (HC) replace the single Transformer residual stream with multiple streams, introducing a permutation symmetry over stream indices. We study how this symmetry is resolved in practice: whether streams specialize in a balanced way or exhibit dominant-stream usage. Using fine-grained diagnostics for HC-based language models, we trace how multi-stream representations are actually used. We find that after an early seeding stage, residual mixing often remains close to identity, limiting a core HC mechanism for exchanging information between streams. Moreover, both signal and interpretable features concentrate in a dominant stream, and the nominally multi-stream residual connection can underutilize its capacity, behaving closer to a single-stream residual pathway. Finally, we show that breaking symmetry at stream initialization reduces dominant behavior and improves performance across \textit{m}HC variants. Our code is publicly available.

80. Rethinking the Role of Tensor Decompositions in Post-Training LLM Compression

Authors: Artur Zagitov , Alexander Miasnikov , Maxim Krutikov , Vladimir Aletov , Gleb Molodtsov , Nail Bashirov , Artem Tsedenov , Aleksandr Beznosikov
URL: https://arxiv.org/abs/2606.03465
Abstract:

Post-training compression is essential for deploying large language models (LLMs) under tight resource constraints. Tensor decompositions have emerged as a promising direction, offering compact parameterizations well suited to Transformer weight structures. However, existing studies evaluate these methods in narrow settings, leaving unclear whether tensorization is effective at large-scale deployment. We systematically evaluate tensor compression across dense and MoE architectures, establishing performance trade-offs grounded in both empirical analysis and theoretical analysis. We identify a fundamental mismatch between the shared subspaces assumed by tensor decompositions and the heterogeneous representations learned by modern LLMs, thereby delineating their practical limits and clarifying their viable role in large-scale deployment. The code is available at this https URL .

81. FORGE: Multi-Agent Graduated Exploitation and Detection Engineering

Authors: Farooq Shaikh
URL: https://arxiv.org/abs/2606.03453
Abstract:

Vulnerability disclosure volumes now far exceed organizational assessment capacity, yet three adjacent research communities (proof-of-concept generation, vulnerability prioritization, and detection rule engineering) operate largely in isolation. Existing automated exploit generation systems report binary pass/fail outcomes, discarding partial progress and producing no signal for the other two communities. This paper presents FORGE, a multi-agent system that bridges these three silos through graduated exploitation depth. Five specialized agents (Intel, Generator, Planner, Exploit, and Detector) execute in a fixed pipeline that (1) generates targeted vulnerable applications from CVE metadata, (2) conducts coached, multi-turn exploitation assessed by an LLM-primary oracle on a four-level taxonomy (L0: no evidence through L3: full compromise), and (3) produces Sigma and Snort detection rules grounded in OpenTelemetry exploitation traces. Graduated depth is the bridging mechanism: deeper exploitation yields richer behavioral traces for detection engineering, while depth data across scoring bands provides ground truth for prioritization validation. A tiered knowledge architecture accumulates intelligence across assessments, transferring build and exploitation experience to subsequent CVEs. Evaluation on 603 CVEs from the CVE-GENIE dataset achieves 67.8% end-to-end L1+ exploitation at USD 1.50 per CVE across eight languages and 187 CWE types. Exploitation rates remain near 68% regardless of EPSS or CVSS band, indicating that pattern-level reachability is orthogonal to metadata-based prioritization. Detection rules from L2+ exploitation achieve significantly higher span-normalized grounding than L1-derived rules (p=0.035), and 93.4% of generated Snort rules produce zero false positives against a synthetic benign corpus.

82. Local Guidance, Global Impact: Gaussian-Reshaped Trust Region Unlocks Behavior Transitions

Authors: Bingxu Liu , Jiashun Liu , Johan Obando-Ceron , Hao Wang , Runze Liu , Pablo Samuel Castro , Aaron Courville , Ling Pan
URL: https://arxiv.org/abs/2606.03382
Abstract:

While Proximal Policy Optimization (PPO) demonstrates strong performance in stationary settings, we show that its standard optimization paradigm struggles in continual and non-stationary environments. The failure does not stem from insufficient model capacity or overly restrictive clipping. Instead, PPO performs persistent, directionally inefficient local updates, which indicates a lack of geometry-aware guidance for accumulating meaningful behavioral change and ultimately hindering transitions toward new behavior patterns. Although divergence-based regularization introduces partial geometric awareness, its monotonically increasing penalties implicitly discourage large policy deviations, even when such shifts are necessary for effective adaptation. To address this limitation, we propose Gaussian Trust Region Policy Optimization (GTR), which reshapes the trust region using a Gaussian kernel. The resulting constraint is bounded and non-monotonic, providing strong local stability while progressively relaxing under sustained high-advantage updates. To further improve robustness, we introduce a Mixture Gaussian Anchor that adapts to recent policy trajectories, reducing variance induced by stale references. GTR is architecture-agnostic and achieves strong performance across games, simulated robotic control, open-world exploration, and language model post-training. These results demonstrate that geometry-aware trust-region design can be a promising direction for robust reinforcement learning in complex non-stationary environments. Our code is available at this https URL .

83. P\textsuperscript{2}-DPO: Grounding Hallucination in Perceptual Processing via Calibration Direct Preference Optimization

Authors: Ruipeng Zhang , Zhihao Li , Haozhang Yuan , C. L. Philip Chen , Tong Zhang
URL: https://arxiv.org/abs/2606.03376
Abstract:

Hallucination has recently garnered significant research attention in Large Vision-Language Models (LVLMs). Direct Preference Optimization (DPO) aims to learn directly from the corrected preferences provided by humans, thereby addressing the hallucination issue. Despite its success, this paradigm has yet to specifically target the perceptual bottleneck in attended regions or address insufficient Visual Robustness against image degradation. Furthermore, existing preference pairs are often vision-agnostic and their inherently off-policy nature limits their effectiveness in guiding model learning. To address these challenges, we propose Perceptual Processing Direct Preference Optimization (P\textsuperscript{2}-DPO), a novel training paradigm in which the model generates and learns from its own preference pairs, thereby directly addressing the identified visual bottlenecks while inherently avoiding the issues of vision-agnostic and off-policy data. It introduces: (1) an on-policy preference pairs construction method targeting Focus-and-Enhance perception and Visual Robustness, and (2) a well-designed Calibration Loss to precisely align visual signals with the causal generation of text. Experimental results demonstrate that with a comparable amount of training data and cost, P\textsuperscript{2}-DPO outperforms strong baselines that rely on costly human feedback on benchmarks. Furthermore, evaluations on Attention Region Fidelity (ARF) and image degradation scenarios validate the effectiveness of P\textsuperscript{2}-DPO in addressing perceptual bottleneck in attended regions and improving Visual Robustness against degraded inputs.

84. Evaluating LLMs’ Effectiveness on Real-World Consumer Device Repair Questions

Authors: Atm Mizanur Rahman (University of Illinois Urbana-Champaign), Md Arid Hasan (University of Toronto), Syed Ishtiaque Ahmed (University of Toronto), Sharifa Sultana (University of Illinois Urbana-Champaign)
URL: https://arxiv.org/abs/2606.03331
Abstract:

Consumer device repair is an important but underexplored testbed for large language models (LLMs). Repair tasks require reasoning over incomplete problem descriptions, hardware-specific diagnostics, actionable troubleshooting, and safety-critical decisions, where incorrect advice can cause device damage, battery hazards, or permanent data loss. We introduce a benchmark of 991 real-world repair questions from Reddit spanning phone repair, computer repair, and data recovery, each paired with technician-written reference solutions, and provide Bangla translations to evaluate cross-lingual performance. We evaluate six state-of-the-art LLMs in English and Bangla using four repair-specific criteria: correctness, completeness, practicality, and safety. Our results show that while LLMs can provide useful repair assistance, they remain unreliable for high-risk real-world repair tasks without rigorous evaluation and explicit safety safeguards. Phone repair is the most difficult and safety-sensitive domain, and all models make substantial errors in board-level diagnosis, repair prioritization, and safe recovery procedures. Across domains and models, Bangla responses consistently perform worse than English responses. Among the evaluated models, GPT-5.4 performs best overall.

85. FLIPS: Instance-Fingerprinting for LLMs via Pseudo-random Sequences

Authors: Gurvan Richardeau , Gohar Dashyan , Erwan Le Merrer , Gilles Tredan
URL: https://arxiv.org/abs/2606.03330
Abstract:

Literature reveals that a Large Language Model’s (LLM) behavior is not only conditioned by its original weights but also its instance-level parameters, such as instructional prompt, sampling configuration or quantization. A model that generates safe outputs under one configuration may produce toxic content under another. However, current LLM identification techniques (such as fingerprinting) focus on intellectual property protection, and their design favors robustness to changes in these instance-level parameters. This poses a critical challenge for AI regulation in which compliance assessments target actual deployed behaviors, not model provenance. In this paper, we introduce instance-level fingerprinting, a regulator-oriented paradigm that distinguishes configurations of the same LLM. Our method FLIPS, exploits biases in generated binary random sequences to reach 96% (closed-set) and 90% (open-set, where some targets are unknown) identification accuracy across 237 model instances, versus 35% for the adapted LLMmap baseline. This shows that instance-level fingerprinting is both necessary for regulation and practically feasible. Code available at this https URL .

86. Calibration Data Trade-offs Across Capability Dimensions: Why Multi-Source Mixing Matters for High-Sparsity LLM Pruning

Authors: Hu Xu , Zhaolong Xing , Congcong Liu , Jiaxing Wang , Zhida Jiang , Junshi Huang , Zhen Chen , Jianfeng Xu
URL: https://arxiv.org/abs/2606.03328
Abstract:

Post-training pruning compresses large language models to high sparsity using a small unlabelled calibration set, and recent work has concluded that the choice of calibration source has only modest impact on averaged post-pruning accuracy. We ask whether this conclusion survives once calibration impact is evaluated separately across distinct capability dimensions rather than aggregated. Decomposing post-pruning capability into General, Commonsense, Code, and Math, and analysing $n{=}15$ calibration sources via Spearman correlations between OIT information metrics and per-dimension retention, we uncover an opposite-sign trade-off: calibration perplexity correlates positively with General retention ($\rho{=}{+}0.71$) but negatively with Math and Code retention ($\rho{=}{-}0.53,\,{-}0.59$; $p{<}0.05$), so no single source can preserve all capabilities. We respond with multi-source calibration mixing, and propose IGSP, an information-guided self-calibration protocol that automates multi-source construction without capability-aligned corpora by minimising 4-gram aggregation and balancing perplexity across dimensions. On LLaMA-3.1-8B at SparseGPT 60% sparsity, a uniform multi-source mix reaches 58.8% total retention, outperforming the best single source (MetaMath, 50.0%) by $+8.8$ and the C4 default (40.0%) by $+18.8$; IGSP improves over Self-Cal by $+2.4$ and SGS by $+4.8$.

87. dstack-capsule: Pod-Level Remote Attestation for Confidential Workloads on Kubernetes

Authors: Yang Yang , Kevin Wang , Yuanhai Luo , Hang Yin , Jie Cai , Shunfan Zhou , Wenfeng Wang
URL: https://arxiv.org/abs/2606.03323
Abstract:

The rise of LLM-as-a-Service and other confidential cloud workloads demands cryptographic proof that user data is processed in a trusted, untampered environment. Existing solutions, notably Confidential Containers (CoCo), enforce a strict “one Pod per VM” model that attests only the Guest OS stack, leaving container-level identity unverified and incurring prohibitive per-VM resource overhead. We present dstack-capsule, a Kubernetes platform that enables Pod-level remote attestation on Intel TDX by allowing multiple Pods to share a single Confidential VM while each retains independent, hardware-backed proof of identity. Our key insight is a two-layer attestation architecture: static platform measurements are frozen in RTMR[3] via an irreversible privilege fuse, while dynamic Pod identities (pod_uid, pod_spec_hash, workload_id) are embedded in the TDX Quote’s report_data field and signed by hardware on every request. dstack-capsule introduces (1) a Pod-level attestation protocol binding Pod spec digests to hardware-signed Quotes; (2) a privilege fuse mechanism that atomically transitions a node from setup mode to secure mode; (3) a multi-layer sandbox spanning storage, runtime, admission, API, and network isolation layers; and (4) a complete open-source implementation based on Kubernetes 1.32, Intel TDX, and Sysbox. We evaluate the security properties, attestation correctness, and performance characteristics of dstack-capsule, demonstrating that it achieves Pod-granularity verification without the resource overhead of per-VM isolation.

88. RobotValues: Evaluating Household Robots When Human Values Conflict

Authors: Jongwook Han , Hyeongjin Kim , Yohan Jo
URL: https://arxiv.org/abs/2606.03312
Abstract:

While household robots are often evaluated based on task completion, everyday domestic environments involve value-conflicting situations in which robots are expected to choose actions that prioritize other values than task success, such as human autonomy, efficiency, or social appropriateness. Yet, there are no benchmarks for evaluating robots’ value preferences in such scenarios. We introduce RobotValues, a benchmark to evaluate household robot planners in 10K value-conflict scenarios. Each instance consists of a realistic household image with multiple plausible robot actions that prioritize different human values. We construct RobotValues through LLM-assisted scenario generation, stakeholder-grounded value extraction, image generation and automatic quality control. Using RobotValues we evaluate VLMs used in robotics and find that models exhibit default value preferences, including safety and accommodation, while underselecting privacy-prioritizing actions. When the models are instructed to prioritize specific values that conflict with their own preferences, they often fail to override their default actions, choosing incorrect actions for 80% of the time. These findings suggest that household robot evaluation should measure not only task completion or safety compliance, but also whether robots can choose among plausible actions when human values conflict.

89. When RLHF Fails: A Mechanistic Taxonomy of Reward Hacking, Collapse, and Evaluator Gaming

Authors: Zelalem Abahana
URL: https://arxiv.org/abs/2606.03238
Abstract:

Reinforcement learning from human feedback (RLHF) makes large-scale post-training possible by replacing an underspecified human objective with learned and scalable proxies. The same substitution creates a structured failure surface: optimization can raise the learned reward while external quality falls, degrade both proxy and judge scores, reveal proxy under-alignment, or produce evaluator-specific disagreement. We present an empirical failure-mode study of a compact RLHF pipeline with proximal policy optimization (PPO), direct preference optimization (DPO), uncertainty-penalized PPO (UP-PPO), reward-model uncertainty, approximate policy drift, diversity and repetition diagnostics, and two external LLM judges. Rather than treating reward hacking as a single terminal event, we classify matched transitions between checkpoints using the directions of the learned reward, judge scores, and average judge score. Across 61 checkpoint rows and 1920 row-level transitions, aggressive PPO has the highest localized reward-hacking rate (14.45%; bootstrap 95% CI: 10.16-18.75), while UP-PPO yields lower rates in the same aggressive regime (11.33-10.94%). A pre-transition logistic model predicts future row-level reward hacking with ROC-AUC 0.821, and row-level analysis finds localized reward hacking that checkpoint averages miss in 3 of 12 settings. The central conclusion is methodological: RLHF failures are not only final-model pathologies, but training dynamics that can be classified, localized, and partially anticipated.

Authors: Zhe Sun , Meng Wang , Lei Wang , Yuxi Wang , Wanxin Li , Yujia Peng , Zhenliang Zhang
URL: https://arxiv.org/abs/2606.03223
Abstract:

Robot storytelling offers a unique blend of technological innovation and creative expression that engages children in unprecedented ways. However, the technical aspects are often too complicated for children. We propose an interactive system that facilitates robot storytelling with tangible and natural language interactions. Children arrange the playground with their own stuff and create narratives with an LLM agent. The created narratives are transformed into a motion sequence based on the map and characters, and the motions are executed by self-navigating swarm robots. This system enhances robot storytelling with flexible scenarios, enabling young children to create robot dramas with everyday objects.

91. AI Rater Discrimination Depends on Scoring Protocol in Complex Clinical Decision-Making

Authors: Sangwon Baek , Kyu Yeon Hur , Kyunga Kim
URL: https://arxiv.org/abs/2606.03198
Abstract:

Clinical AI evaluation increasingly delegates scoring to large language models (LLMs) acting as AI raters, yet their scoring behavior across evaluation conditions has not been quantitatively characterized. We address this gap through a factorial study of AI rater behavior in adult type 2 diabetes (T2D) pharmacotherapy at 12-month outpatient follow-up, a clinical task involving complex decision-making operationalized across seven evaluation questions. Four open-source LLMs served simultaneously as clinical decision support system (CDSS) models and AI raters. Each CDSS output was scored under two scoring protocols: a rubric-anchored Gold Rubric (GR) protocol incorporating a patient-specific rubric, and a rubric-free Non Gold Rubric (Non-GR) protocol. Linear mixed effects models crossed the scoring protocol factor with five design factors – CDSS model, CDSS prompt configuration (document-referenced generation [DRG] vs.\ Baseline), rater model, prompt character, and prompt type – and estimated main effects together with their protocol interactions. Across all questions, AI raters yielded consistently higher scores within a very narrow range (74–78 points on average) under Non-GR compared to those under GR (7.69 to 49.64 points lower mean scores; 1.68 to 3.67 times wider interquartile ranges). Within each question, GR amplified the AI rater’s discrimination between DRG and Baseline CDSS outputs by factors of 1.76 to 5.10, while also revealing substantial behavioral variation across rater models that Non-GR suppressed. These findings support rubric anchoring as the scoring protocol that preserves discriminative power in clinical AI evaluation; rubric-free scoring cannot substitute when questions require patient-specific or jurisdiction-specific criteria that rater models cannot infer from parametric knowledge alone.

92. Fully Automated Identification of Lexical Alignment and Preference-Stage Shifts in Large Language Models

Authors: Thomas Stephan Juzek , Xiaoyang Ming , Jose A. Hernandez
URL: https://arxiv.org/abs/2606.03165
Abstract:

The language used by digital chat assistants such as ChatGPT can diverge from human expectations (misalignment). Research, mostly on Scientific English, has described both what divergences occur and, to some extent, why, linking them to the training stage of human preference learning. Yet, existing approaches rely on manual curation. This paper introduces two curation-free, assumption-light evaluation metrics: the Lexical Alignment Score, which identifies lexical overuse, and the Triangulated Preference Shift, which quantifies how much of such shifts can be attributed to human preference learning. Using PubMed abstracts, continuations were generated and measured using windowed document prevalence across six model families (Falcon, Gemma, Llama, Mistral, OLMo, Yi). The procedure identifies, without manual intervention, overused items such as ‘suggest’, ‘additionally’, and ‘strategy’, and estimates their link to preference learning. Our findings replicate prior work and remain stable across parameter settings, random seeds, and evaluation on further data. The approach scales readily and enables systematic study of lexical (mis)alignment beyond Scientific English and across languages, and as such, the metrics have the potential to contribute to improved alignment for future models and understanding of its origins.

93. Decoupled Smart Contract Audits: Lightweight LLM Framework via Distillation and Aggregation

Authors: Bagus Rakadyanto Oktavianto Putra , Muhamad Risqi Utama Saputra , Widyawan , Guntur Dharma Putra
URL: https://arxiv.org/abs/2606.03128
Abstract:

Smart contracts face critical security challenges that require thorough auditing in decentralized web services. While Large Language Models (LLMs) have shown promise in automated vulnerability detection, existing approaches lack severity evaluations with actionable remediation and demand unnecessarily massive computational overhead. In this study, we introduce an efficient end-to-end smart contract security audit framework utilizing lightweight, highly optimized open-source LLMs (0.6B-4B parameters). Our framework decouples comprehensive audit tasks into four interconnected components: vulnerability detection, explanation, severity classification, and remediation recommendation. To maintain high accuracy without massive parameters, we implement Rank-Stabilized Low-Rank Adapters (rsLoRA), knowledge distillation, and a custom Chain-of-Verification (CoVe) aggregation strategy to systematically screen and consolidate multiple draft responses from the model into a highly accurate audit report. Experimental results demonstrate that our lightweight pipeline consistently outperforms state-of-the-art open-source coder dense LLMs (7B to 34B parameters), achieving 98.25% accuracy in vulnerability detection and an alignment score of 0.4375 in generative explanation tasks. Furthermore, our extensive ablation studies empirically validate the superiority of our decoupled audit processes over unified prompting and uncover a novel severity centrality bias, establishing a critical benchmark for future research in LLM-assisted auditing.

94. AnyAudio-Judge: A Dynamic Rubric-Based Benchmark and Evaluator for Audio Instruction Following

Authors: Haitao Li , Tian Tan , Yuguang Yang , Shan Yang , Xie Chen
URL: https://arxiv.org/abs/2606.03116
Abstract:

The rapid advancement of instruction-guided audio generation has highlighted the critical need for robust alignment evaluation. Current automated evaluation methods heavily rely on holistic scoring from general-purpose large language models, which struggle to decouple complex instructions, lack interpretability, and fail to capture fine-grained attribute mismatches. To address this, we introduce a novel dynamic rubric-based evaluation paradigm that adaptively decomposes complex audio captions into a variable number of independent, verifiable binary rubric items. To rigorously benchmark this capability, we propose the AnyAudio-Judge Bench, a comprehensive, bilingual benchmark comprising 7,920 meticulously curated samples across four diverse audio domains (speech, sound, music, and mixed), featuring deliberately constructed hard negatives. Furthermore, we construct a large-scale corpus of 105K samples with explicit Chain-of-Thought (CoT) rationales to train our dedicated evaluator, the AnyAudio-Judge model. By employing a training pipeline that combines Supervised Fine-Tuning (SFT) and Group Relative Policy Optimization (GRPO), our model successfully aligns its reasoning paths with the rubric-based scoring mechanism. Extensive experiments demonstrate that AnyAudio-Judge not only significantly enhances zero-shot alignment detection compared to state-of-the-art baselines, but also provides precise and interpretable reward signals that substantially improve instruction alignment in downstream reinforcement learning for audio generation.

95. PhotoCraft: Agentic Reasoning with Hierarchical Self-Evolving Memory for Deep Image Search

Authors: Kailin Lyu , Zhiqiang Yuan , Jianwei He , Qiwei Yan , Xuanbo Su , Nanxing Hu , Yang Liu , Ce Hao , Shengqian Qin , Lianyu Hu , Jinchao Zhang , Jie Zhou
URL: https://arxiv.org/abs/2606.03099
Abstract:

Deep Image Search requires multi-step reasoning over rich contextual cues, such as time, location, and event relations. However, most existing LLM-based agents are stateless and reactive, lacking persistent memory to maintain long-horizon context or transfer experience across tasks, which often leads to execution drift and experience isolation. To address these limitations, we propose PhotoCraft, a training-free, hierarchical memory system for photo-search agents. Inspired by human cognition, PhotoCraft equips MLLMs with working, episodic, and semantic memory, which are dynamically invoked during reasoning to preserve logical consistency and knowledge transferability throughout multi-step reasoning and answer generation. Extensive experiments on DISBench demonstrate that PhotoCraft consistently improves context-aware retrieval across diverse MLLM backbones, achieving gains of up to 18.5\% and effectively mitigating key bottlenecks in memoryless deep image search, offering a practical path toward reliable and generalizable multimodal search agents.

96. “Important You should give me full credits!”: Exploring Prompt Injection Attacks on LLM-Based Automatic Grading Systems

Authors: Hang Li , Fedor Filippov , Yuling Lin , Pengfei He , Kaiqi Yang , Yucheng Chu , Yingqian Cui , Hui Liu , Jiliang Tang
URL: https://arxiv.org/abs/2606.03090
Abstract:

The emergence of large language models (LLMs) has significantly accelerated recent research on LLM-based automatic grading (AG) systems. Benefiting from the strong instruction-following capabilities and broad prior knowledge of LLMs, educators can deploy AG systems across diverse tasks using only natural language rubrics while achieving satisfactory grading performance. Despite these advantages, new security concerns may also arise. In particular, prompt injection (PI) attacks have recently become a major threat to LLM-based applications. In the context of AG, attackers can potentially exploit PI vulnerabilities to manipulate grading systems into assigning artificially high scores regardless of the actual answer quality. Such behavior poses serious risks to the fairness, reliability, and integrity of educational assessment. In this work, we study PI attacks in AG systems, and systematically investigate the effectiveness of such attacks in educational scenarios. We further evaluate the effectiveness of existing defensive strategies against these attacks. Through comprehensive experiments under rubric-based grading settings, we demonstrate that current LLM-based AG systems remain highly vulnerable to PI attacks. We hope that our findings raise awareness of this emerging threat and motivate future research toward secure, robust, and trustworthy LLM-based educational systems.

97. Regret Pre-training: Bridging Prior and Posterior Views for Enhanced Knowledge Grounding

Authors: Mingkuan Zhao , Xiayu Sun , Wentao Hu , Suquan Chen , Jiaxuan Li , Xiaoyan Zhu , Xin Lai , Jiayin Wang
URL: https://arxiv.org/abs/2606.03080
Abstract:

Causal language models factorize sequence probabilities using only preceding context, leaving future information unexploited during training despite its availability in the training data. This paper introduces Regret Pre-training, a self-supervised framework grounded in the Learning Using Privileged Information (LUPI) paradigm. The framework employs a dual-view architecture in which a single model generates both a causal Student distribution and a future-conditioned Teacher distribution. The training objective augments standard language modeling with a regret loss that minimizes the KL divergence from teacher to student, transferring future-aware signals to the causal representations. We investigate two teacher configurations on the OLMoE-1B-7B architecture:LocalRegret, which extends attention by one future token, andGlobalRegret, which conditions on bidirectional context with the target position masked. Experiments on nine downstream tasks following 4 billion tokens of training demonstrate that both configurations consistently outperform the baseline. On average,GlobalRegret andLocalRegret achieve 33.9% and 32.2% accuracy respectively, surpassing the baseline’s 30.2%. Most notably,GlobalRegret improves BoolQ performance by 18.1 percentage points (61.0% vs 42.9%). The framework introduces no additional parameters and requires only one extra inference-mode forward pass per training step.

98. Libra: Efficient Resource Management for Agentic RL Post-Training

Authors: Kaiwen Chen , Xin Tan , Jingzong Li , Hong Xu
URL: https://arxiv.org/abs/2606.03077
Abstract:

Reinforcement learning (RL) has become a standard post-training paradigm for large language models (LLMs), extending beyond preference alignment to complex reasoning and multi-turn agentic behaviors. In agentic RL, the rollout stage generates trajectories while invoking tools, producing long-tailed and non-stationary workloads that challenge conventional resource-management assumptions. Three fundamental challenges arise. First, due to the long-tail distribution, a small fraction of trajectories dominates rollout makespan. Second, rollout and training exhibit strong asymmetry in compute patterns, memory demands, and sensitivity to sequence length. Third, as the RL policy evolves, the trajectory-length distribution drifts over time, rendering any static resource split progressively suboptimal. We present Libra, which introduces two core mechanisms. The first is a periodic global resource planner that jointly optimizes GPU allocation across rollout and training clusters. It leverages an elastic hybrid pool to enable lightweight, non-blocking worker reallocation between stages. The second is a causality-driven multi-level feedback queue (C-MLFQ) scheduler, which routes requests to heterogeneous rollout buckets based on causal signals derived from tool-return outcomes, rather than relying on fragile length predictions. Evaluated on 48 A800 GPUs, Libra achieves up to 3.0$\times$ higher throughput and converges up to 2.5$\times$ faster in reward compared to the baselines.

99. Efficient Hyperparameter Optimization for LLM Reinforcement Learning

Authors: Minping Chen , Bowen Xiao , Du Liang , Chuxuan Zeng , Zeyi Wen
URL: https://arxiv.org/abs/2606.03073
Abstract:

Reinforcement learning (RL) for large language models (LLMs) is highly sensitive to hyperparameter configurations, making hyperparameter optimization (HPO) essential yet computationally expensive. Existing multi-fidelity HPO methods remain inefficient for LLM RL due to the massive model scale and resource-intensive training cycles. In this paper, we propose Joint Fidelity Hyperparameter Optimization (JF-HPO), which simultaneously adapts both model size and training budget as fidelity. JF-HPO is empowered by: (i) it leverages a small proxy model of the target LLM for efficient training and evaluation in each HPO trial; (ii) it integrates carefully designed early-stopping strategies based on training dynamics; (iii) it introduces an efficient checkpointing mechanism to eliminate redundant computations. Compared with existing HPO methods, JF-HPO significantly improves the computational efficiency of each trial (up to 14.9 times), while achieving better or competitive predictive accuracy under the same time budget. Notably, compared with utilizing hyperparameter configurations from the VeRL Recipe, JF-HPO delivers performance improvements ranging from 5.8% to 111.6%.

100. ASymPO: Asymmetric-Scale Policy Optimization for Asynchronous LLM Post-Training Without Behavior Information

Authors: Zehua Liu , Yuxuan Yao , Xiaojin Fu , Tao Zhong , Mingxuan Yuan
URL: https://arxiv.org/abs/2606.03070
Abstract:

Asynchronous reinforcement learning can improve language-model post-training throughput by decoupling response generation from policy optimization, but stale responses introduce distribution drift. Standard behavior-corrected methods control this drift with behavior-policy probabilities, importance ratios, or clipping, which requires token-aligned, versioned, and numerically consistent behavior log-probabilities across rollout and learner systems. We ask whether asynchronous group-relative RL can instead be stabilized using only current-policy probabilities. We identify a scale-imbalance failure mode: when stale responses are evaluated under the current policy, positive and negative loss terms can appear at different negative log-probability scales, so zero-sum advantages no longer imply balanced loss contributions. We propose Asymmetric-Scale Policy Optimization (ASymPO), which normalizes each response’s token loss by its current average token negative log-probability. ASymPO requires no behavior-policy probabilities, restores response-level zero-sum balance, and preserves a nonzero learning signal. We also introduce Scaled Policy Optimization (SPO), a fixed negative-scaling baseline, and evaluate both current-policy-only objectives in asynchronous mathematical reasoning post-training.

101. Rethinking Molecular Text Representations for LLMs: An Empirical Study

Authors: Arun Raja , Garrett M. Morris , Kian Ming A. Chai
URL: https://arxiv.org/abs/2606.03057
Abstract:

Large language models (LLMs) are increasingly used for molecular tasks, but it remains unclear which molecular representation to use. We present a systematic benchmark evaluating LLM molecular competence across nine representations and eight chemical tasks. We benchmark 16 LLMs across five model families, including reasoning and non-reasoning variants, chemistry-specialized LLMs, and closed frontier models. Performance is strongly representation-dependent and no single representation wins across tasks, though CML is the best, followed by MolJSON, InChI, and then canonical SMILES. Explicit structured text representations (CML and MolJSON) dominate structural tasks; IUPAC dominates semantic tasks, winning molecule retrieval for all 16 LLMs; and SMILES variants are rarely optimal despite their prevalence in pretraining. Chemistry-specialized models perform well with SMILES at the cost of large degradations with structured text representations, suggesting SMILES-only evaluation rewards specialization that does not generalize. Using LLM-as-a-judge, we find that IUPAC produces the highest fraction of correct molecule generations. A mechanistic study via tokenization audits, linear probes and attention shows that representations are encoded differently inside the model; for example, structured representations require higher attention across the molecular span. Our results argue against representation-invariant evaluation and motivate task-aware representation routing for LLM-based chemistry.

102. Capability Advertisement as a Market for Lemons: A Trust Layer for Heterogeneous Agent Networks

Authors: Gaurav Naresh Mittal
URL: https://arxiv.org/abs/2606.03034
Abstract:

Large language model (LLM) agents have begun to delegate work to one another. Protocols such as the Model Context Protocol (MCP) and the Agent2Agent protocol (A2A) let an agent publish what it can do and let others call it, and public registries of such agents are already appearing. These protocols assume an advertised capability is a static, truthful fact. A real agent is none of these things: its competence is probabilistic, varies with input, drifts when the underlying model is updated, and, because the agent is itself a language model, it can describe itself with complete confidence and be wrong. A caller therefore sees what an agent claims to do, not what it can do, with no principled way to tell a reliable provider from a fluent impostor. We argue these difficulties share one cause: the market for lemons. When quality is hidden and claims are cheap, good and bad providers become indistinguishable, honest reliability goes unrewarded, and the market decays toward its worst participants. Economics offers three remedies, signaling, screening, and reputation, and none are present in today’s agent protocols. We make four contributions: (1) a failure taxonomy that names confident-wrong as a non-adversarial, correlated subclass of Byzantine faults that classical fault-tolerance mismodels; (2) a market-for-lemons model showing that faith-based protocols admit only a low-trust equilibrium; (3) the Trust Layer, a thin, protocol-agnostic narrow waist above MCP and A2A that adds probabilistic capability descriptors, screening, and reputation, and admits a separating equilibrium when the cost of sustaining an overclaim exceeds the gain from it; and (4) a reliability-composition bound for delegation chains with an end-to-end placement argument. The design needs no model retraining and degrades gracefully when its trust anchors are absent or corrupt.

103. Conditional Hypothesis Generation for LLM-Based Text Analysis with Researcher-Specified Covariates

Authors: Paiheng Xu , Jing Liu , Wei Ai
URL: https://arxiv.org/abs/2606.03029
Abstract:

A core goal of computational social science is to discover interpretable differences in how language varies across outcomes of interest, such as political affiliation or instructional quality. Recent LLM-based hypothesis generation methods describe such differences in natural language, but select for globally discriminative patterns without accounting for covariates that shape the data based on researchers’ domain knowledge. When covariates are ignored, selected patterns can reflect confounds rather than differences of substantive interest. We introduce conditional hypothesis generation, a framework that incorporates researcher-specified covariates to steer hypothesis discovery toward differences that hold within relevant subgroups. Two challenges arise: the target subgroup may be underrepresented (stratum imbalance), and the direction of a difference may reverse across subgroups (sign reversal). We propose two econometrics-inspired methods: one introduces feature–covariate interactions to detect sign reversals, and the other applies within-stratum demeaning and inverse-frequency reweighting to equalize underrepresented strata. Synthetic experiments show each method outperforms global baselines in its targeted setting, and expert evaluation on two real-world datasets confirms that covariate-aware generation surfaces more useful hypotheses within relevant subgroups.

104. Spike-Aware C++ INT8 Inference for Sparse Spiking Language Models on Commodity CPUs

Authors: Ting Liu
URL: https://arxiv.org/abs/2606.03026
Abstract:

Spiking language models expose activation sparsity that dense Transformer runtimes do not directly exploit. This paper studies that property from a systems perspective. Building on the SymbolicLight V1 spike-gated language model family, we implement a C++ CPU inference runtime that treats sparse binary spike states as an execution primitive rather than only applying post-hoc weight compression. The runtime combines a manifest-driven weight loader, mixed row/column memory layout, AVX2/FMA kernels, per-channel symmetric INT8 quantization, and integer-domain accumulation for spike-conditioned sparse paths. On an AMD Ryzen 7 5800X, an early scalar FP32 baseline decodes at 9.5 tokens/s. Mixed-layout AVX2 FP32 raises this to 14.7 tokens/s, and AVX2 INT8 reaches 19.9 tokens/s on the same step-30k export while reducing the weight footprint from 3.49 GB to 1.06 GB. For the available 186k-step 874M-parameter INT8 export, the C++ runtime decodes at 22.63 tokens/s in a single-thread CPU benchmark, compared with 16.31 tokens/s for TinyLlama-1.1B Q8_0, 11.26 tokens/s for Falcon3-1B Q8_0, and 9.70 tokens/s for Qwen2.5-1.5B Q8_0 under this http URL . Thread scaling reaches 47.90 tokens/s at four CPU threads, and 512-token prefill improves from 29.86 to 94.68 tokens/s from one to eight threads. The throughput result comes with a quality cost: the SNN reports WikiText-2 perplexity 24.80, worse than the dense baselines in the same benchmark. We frame the result as an inference-systems study for sparse language runtimes, with longer-term motivation in embodied and edge agents that may benefit from local, low-core inference near sensors and actuators. Spike-aware execution can improve CPU throughput and memory behavior for sparse spiking language models, while model quality, controlled dense training baselines, embodied-task evaluation, and measured CPU energy remain open problems.

105. Hallucinations as Orthogonal Noise: Inference-Time Manifold Alignment via Dynamic Contextual Orthogonalization

Authors: Mingkuan Zhao , Wentao Hu , Tianchen Huang , Yuheng Min , Suquan Chen , Yide Gao , Yanbo Zhai , Shuangyong Song , Xuelong Li
URL: https://arxiv.org/abs/2606.03022
Abstract:

Hallucination in Large Language Models (LLMs), characterized by the generation of content inconsistent with contextual facts or logical constraints – remains a persistent challenge for reliable deployment. In this work, we address this issue through a geometric framework rooted in the linear representation hypothesis. We propose that hallucinations manifest as orthogonal noise relative to the semantic manifold of the residual stream. Specifically, we hypothesize that while attention heads ideally propagate information congruent with the context subspace, hallucinations arise when specific heads introduce components orthogonal to this subspace, disrupting the coherence of the latent representation. Based on this formulation, we introduce Dynamic Contextual Orthogonalization (DCO), an inference-time intervention method. DCO utilizes the input residual stream as a dynamic context anchor to perform orthogonal decomposition on attention head outputs. To distinguish between context-aligned semantic updates and divergent noise, DCO employs a layer-wise Z-score suppression mechanism that selectively attenuates outlier orthogonal components based on statistical distributions. Evaluations on Llama-3-8B and 70B across benchmarks such as XSum, NQ-Swap, and IFEval demonstrate that DCO achieves superior contextual faithfulness compared to state-of-the-art intervention baselines. Furthermore, DCO maintains high performance on knowledge-intensive tasks like TriviaQA and TruthfulQA, effectively mitigating the trade-off between hallucination suppression and parametric knowledge retention often observed in existing methods. Our findings validate the geometric interpretation of hallucinations and establish DCO as a computationally efficient approach for enforcing manifold this http URL code is available at this https URL

106. Reproducibility is the New Copyleft: Defining AGI-oriented Reproducible Builds

Authors: Masayuki Hatta
URL: https://arxiv.org/abs/2606.03019
Abstract:

Copyleft, as implemented in licenses such as the GNU General Public License, was a legal hack that used copyright to guarantee user freedom by tying the availability of source code to every act of distribution. Its normative force rested on an implicit technical premise: that source code and object code stand in a well-defined, humanly auditable, and reproducible relationship. Large language models and, prospectively, Artificial General Intelligence (AGI) systems systematically violate this premise. The artifacts jointly required to reconstruct a model – code, data, weights, hyperparameters, toolchain, and hardware configuration – are each subject to independent legal, technical, and economic constraints that no current open-source framework fully resolves. Sufficiently capable AI systems can also rewrite licensed source into functionally equivalent derivatives stripped of their original obligations, a form of laundering against which copyleft has no effective defense. This paper argues that a functional analogue of copyleft for AGI must be grounded not in share-alike clauses over code, but in reproducible builds: a practice guaranteeing bit-exact reconstructability from declared inputs. We review the logic of copyleft, critically examine Maffulli’s Second Liberation thesis according to which AI fulfills Stallman’s dream, and show that the argument collapses unless AGI systems are themselves reproducible. Drawing on the Open Source AI Definition (OSAID), the Model Openness Framework (MOF), OpenMDW, and deterministic-inference research, we define seven requirements for AGI-oriented reproducible builds. We further argue that the Model Context Protocol (MCP) and analogous AI-to-AI coupling mechanisms constitute a new dynamic linking layer for which copyleft-style licensing is ill-suited, and that Masnick’s “protocols, not platforms” framework offers a more promising governance template.

107. MUSE: A Unified Agentic Harness for MLLMs

Authors: Jianglin Lu , Hailing Wang , Xu Ma , Qihua Dong , Mingyuan Zhang , Yizhou Wang , Yun Fu
URL: https://arxiv.org/abs/2606.03005
Abstract:

Despite rapid progress, multimodal large language models (MLLMs) still fail on tasks that humans solve effortlessly, such as navigating a grid maze from a screenshot or selecting the correct puzzle piece. Rather than retraining the model, we ask a complementary question: how much capability can be elicited from a frozen MLLM purely by improving the execution scaffold around it? We introduce MUSE, a multimodal unified structured execution harness that wraps any off-the-shelf MLLM with composable modules for task representation, visual processing, perception tool use, structured parsing, deterministic verification, and verifier-guided repair, without any model retraining. We evaluate MUSE across diverse benchmarks spanning visual spatial planning, visual perception, multimodal reasoning, and fine-grained visual discrimination, using multiple state-of-the-art MLLMs. MUSE delivers consistent gains over the bare model in all settings, with the largest jumps on challenging instances. Further analysis reveals that many MLLM failures arise from harness-level shortcomings rather than fundamental model deficits, and can be addressed through verifier-guided repair without touching the model. These findings highlight the agentic multimodal harness as a critical yet underexplored design dimension, offering an orthogonal avenue for improving MLLMs beyond model-centric optimization.

108. How Quantization Changes Interpretable Features: A Sparse Autoencoder Analysis of Language Models

Authors: Evan Duan
URL: https://arxiv.org/abs/2606.03002
Abstract:

Quantization is a standard path to deploying large language models, and a quantized model is typically judged acceptable when its perplexity or downstream accuracy stays close to the full-precision original. Whether the model still computes in the same way, or whether the interpretable features identified in the full-precision model survive weight rounding, is rarely tested, even as safety audits and steering interventions increasingly rely on those features. We ask whether sparse autoencoder (SAE) features extracted from a dense full-precision model remain faithful once that model is quantized. Using a frozen SAE as a fixed measurement basis, we encode full-precision and round-to-nearest (RTN) quantized activations on identical tokens and quantify per-feature survival by Pearson correlation, sweeping bit-widths from INT8 to INT4 on Pythia-70M and Gemma-2-2B. We find that feature survival is graded: features degrade systematically rather than failing all at once, with 62.4 percent of active features surviving at INT6 on Pythia-70M and 51.3 percent surviving at INT6 on Gemma-2-2B, and with most non-survivors blurred rather than destroyed. Survival is predictable from full-precision statistics alone, with cross-validated AUCs of 0.92 to 0.97 and peak activation as the strongest marginal predictor. Critically, task metrics can miss this damage: on Gemma-2-2B, INT7 improves perplexity while degrading 18.7 percent of features. Finally, quantization and matched-perplexity magnitude pruning damage strongly overlapping feature sets, with Jaccard overlap of 0.79 to 0.86 and damage-score Spearman correlation of 0.98, suggesting a shared mode of compression-induced vulnerability. These results show that behavioral parity is insufficient evidence that interpretability findings transfer to quantized deployments, motivating feature-level audits of compression.

109. Patcher: Post-Hoc Patching of Backdoored Large Language Models

Authors: Anjun Gao , Yueyang Quan , Yufei Xia , Zhuqing Liu , Minghong Fang
URL: https://arxiv.org/abs/2606.02995
Abstract:

Large language models remain vulnerable to jailbreak backdoor attacks, where adversaries poison safety alignment data to embed hidden triggers that bypass safety mechanisms. Existing defenses often require comprehensive attack information or multiple triggered examples, making them impractical when defenders only observe a single reported failure case without knowing whether it stems from a backdoor attack or a natural alignment bug. This paper presents Patcher, a post-hoc defense framework that repairs backdoored language models using only a single reported failure case and the model parameters. Patcher operates in two stages. First, it localizes backdoor triggers by computing response-conditioned gradient-based saliency scores and applying adaptive clustering to separate triggers from benign context. Second, it patches the model through a constrained fine-tuning objective that breaks the trigger-response association while preserving benign-task utility and robustness to non-triggered jailbreak attacks through KL-divergence constraints. We conduct extensive evaluations across multiple backdoor attack strategies and demonstrate that Patcher successfully localizes triggers and neutralizes backdoors while maintaining model utility. We further show robustness against adaptive attacks designed to evade our defense. This work represents a significant step toward practical defenses against training-time attacks in deployed language models.

110. Pretraining Language Models on Historical Text

Authors: Xiaoxi Luo , Zachary Shinnick , Niclas Griesshaber , Yixuan Wang , Junchi Yu , Freda Shi , Philip Torr , Yao Lu
URL: https://arxiv.org/abs/2606.02991
Abstract:

We introduce TypewriterLM, a 7.24B History language model (LM) trained exclusively on English text predating 1913. Developing History LMs requires addressing challenges in data quality and availability, preventing temporal leakage, designing temporally consistent post-training pipelines, and constructing reliable evaluations. To address these issues, we construct TypewriterCorpus, a 54B-token historical corpus collected from diverse archival and linguistically annotated sources with extensive data cleaning and leakage mitigation procedures. Furthermore, we introduce lexically grounded instructing tuning, a post-training framework that constraints responses to remain directly grounded in historical source documents. Using this framework we construct two historical instruction tuning datasets: History-LIMA and History-SelfInstruct. To evaluate capability and temporal consistency, we introduce History-Event, a benchmark suite for evaluating competence, temporal grounding and data leakage. We release TypewriterLM and all associated resources to support future research on historical language models.

111. Fast-dLLM++: Fréchet Profile Decoding for Faster Diffusion LLM Inference

Authors: Siva Rajesh Kasa , Yasong Dai , Sumit Negi , Hongdong Li
URL: https://arxiv.org/abs/2606.02955
Abstract:

Diffusion large language models promise parallel token generation, yet inference remains bottlenecked by deciding which masked tokens can be safely committed together. Fast-dLLM addressed this with KV caching and confidence-guided parallel decoding, but its decoding theory uses a homogeneous high-confidence assumption that effectively reduces each candidate set to its weakest selected token. We argue that this leaves speed on the table because real decoding steps exhibit heterogeneous confidence profiles. We propose \textbf{Fast-dLLM++}, a training-free extension that introduces \emph{Fréchet profile decoding}: selecting parallel commit sets from the full sorted confidence profile rather than a single worst-case confidence. The resulting rule is a heterogeneous-confidence generalization of Fast-dLLM’s factor selector and it recovers the previous rule exactly in the equal-confidence case and adds a provable \emph{heterogeneity bonus} when the selected tokens have uneven confidences. Fast-dLLM++ leaves the model, diffusion process, and cache implementation entirely unchanged, making it a drop-in replacement for existing Fast-dLLM decoding. Experiments on GSM8K, MATH, HumanEval, and MBPP with the LLaDA-8B model show that the theoretical improvement translates directly into empirical gains: profile-aware selection improves the accuracy–throughput frontier by exploiting safe parallelism that weakest-token rules miss, achieving up to 37\% higher throughput at comparable accuracy. Our anonymous code release is at this https URL .

112. SCOPE: Real-Time Natural Language Camera Agent at the Edge

Authors: Nikolaj Hindsbo , Sina Ehsani , Pragyana Mishra
URL: https://arxiv.org/abs/2606.02951
Abstract:

Deploying language-driven agents in robotics requires evaluations that reflect real-world task demands: natural-language instructions with reproducible outcomes. Such agents must connect language models to callable perception and control tools, and be assessed using deployment-critical metrics including latency, accuracy, and error modes. We present SCOPE (Simulation and Camera Operations for Perception and Evaluation), a modular agent for natural-language, open-vocabulary pan-tilt-zoom (PTZ) camera control and visual scene understanding, designed explicitly for edge deployment. SCOPE operates both in a Blender-based simulation environment and on a physical PTZ camera, executing all perception, planning, and control locally at the deployment site using edge-accessible compute. We release a 536-task benchmark spanning QA, single- and multi-step commands, counting, spatial reasoning, descriptions, and optical character recognition in a Blender-based simulation environment that exposes realistic PTZ control affordances. Execution traces are combined with an LM-as-Judge to evaluate latency, accuracy, and error modes. We evaluate 19 planner-perception model combinations pairing Qwen3 small language models (SLMs) with Moondream and Qwen vision-language models (VLMs). Stronger SLMs substantially reduce hallucinations and improve tool routing, leading to more reliable closed-loop behavior. Once a sufficiently capable SLM is used, perception becomes the dominant performance bottleneck. Mixture-of-Experts models on both the planning and perception side consistently match or exceed dense alternatives at latencies and memory footprints comparable to much smaller networks. Quantization provides additional efficiency gains with minimal accuracy degradation, identifying a practical, sim-to-real validated design point for real-time, edge-feasible language-driven PTZ control.

113. Linear Probes Detect Task Format, Not Reasoning Mode in Language Model Hidden States

Authors: Subramanyam Sahoo , Vinija Jain , Aman Chadha , Divya Chaudhary
URL: https://arxiv.org/abs/2606.02907
Abstract:

Linear probing of large language model (LLM) hidden states is widely used to claim that models learn distinct representations for different reasoning types. We test this by probing Qwen3-14B on three benchmarks spanning the classical trichotomy: LogiQA 2.0 (deductive), ARC-Challenge (inductive), and $\alpha$NLI (abductive). At layer 32 of 40, linear probes achieve 100\% cross-validated accuracy with well-separated geometry (intrinsic dimensionalities: 20.6, 28.5, 33.6; convex hull contamination $\leq$1.5\%). However, this separation is entirely driven by format confounds. Residualizing source identity, option count, and response length reduces accuracy to chance. Trace-anchor similarity indicates largely shared reasoning across tasks (42.5\% agreement vs.\ 33.3\% chance), and causal steering with random controls ($n=20$) shows no functional link between geometry and reasoning mode ($p=0.286$). Thus, high probe accuracy reflects task format rather than computational structure, motivating routine format deconfounding in mechanistic interpretability.

114. LLM-Assisted Reranking to Operationalize Nuanced Objectives in Recommender Systems

Authors: Amir Ghasemian , Homa Hosseinmardi , Upasana Dutta , Duncan J. Watts
URL: https://arxiv.org/abs/2606.02883
Abstract:

Recommender systems have grown from content-organization tools into sophisticated systems that shape daily behavior. By controlling what we see, they shape what we perceive, raising concerns about filter bubbles, radicalization, polarization, and social inequality. Large language models (LLMs) enable more powerful personalization, intensifying these dynamics. Yet most recommenders are tuned for engagement or limited accuracy metrics, with little attention to broader social implications, e.g. how personalization reshapes exposure in socially consequential domains. We investigate whether LLM-assisted reranking, while improving personalization, inadvertently amplifies exposure to ideologically extreme or conspiratorial political content, a risk theorized but not empirically characterized in news recommendation. Using real news-consumption histories, we rerank YouTube’s sidebar candidates through zero-shot, instruction-based prompting. We compare a baseline prompt with a constrained variant that preserves topical relevance and broadens ideological exposure while reducing conspiratorial or extreme content. Without constraints, reranking strengthened personalization but increased exposure to conspiratorial and extremist material for users whose histories contained such content. Lightweight prompt-level regularization reduced promotion of extreme content and increased ideological diversity, with modest relevance loss. Synthetic experiments suggest that LLMs rerank via statistical regularities in language rather than semantic understanding of ideology, clarifying why naive prompts amplify these patterns and why regularization can reshape them. Together, our results highlight the power of LLMs to operationalize contextual nuance in high-stakes recommendation, and the need to evaluate LLM-assisted personalization beyond accuracy and treat prompt design as a value-laden rather than neutral default.

115. Adaptive Latent Agentic Reasoning

Authors: Dongwon Jung , Peng Shi , Yi Zhang , Junshan Zhang , Muhao Chen
URL: https://arxiv.org/abs/2606.02871
Abstract:

Large reasoning models improve performance by generating extended chain-of-thought (CoT) reasoning, but this behavior becomes inefficient when applied to LLM agents. Current LLM agents often generate verbose textual reasoning at every decision step and allocate reasoning effort nearly uniformly across turns, leading to substantial inefficiency in multi-turn agentic trajectories. We propose Adaptive Latent Agentic Reasoning (ALAR), a dual-mode framework that uses compact latent reasoning for routine turns and selectively escalates to explicit chain-of-thought when deeper deliberation is needed. ALAR learns latent reasoning by using the agent’s actions as supervision anchors and is further optimized to use latent reasoning when it is sufficient for task success and reserve explicit CoT for harder decisions. Experiments on agentic search and tool-use benchmarks show that ALAR maintains comparable or better task accuracy while substantially reducing generated tokens by up to 43.6% in search and 84.6% in tool use. These results demonstrate that ALAR improves the accuracy-efficiency trade-off of LLM agents by reducing unnecessary textual reasoning while preserving explicit deliberation for harder decision steps.

116. The Epi-LLM Framework: probing LLM behavioral priors through epidemiological agent-based models

Authors: Petra Ferenz , Ava Keeling , Tobias O’Keefe , Lorenzo Stigliano , Francesco Di Lauro , Andres Colubri , Jasmina Panovska-Griffiths
URL: https://arxiv.org/abs/2606.02867
Abstract:

Human behaviour during epidemics affects infectious disease dynamics, but quantifying this remains deeply challenging. Here we introduce the Epi-LLM framework: a novel integration of agent-based modelling, real-life epigames, and large language models (LLMs) in which a synthetic society of agents reasons and adapts dynamically over an outbreak contact network. Comparing synthetic agent behaviour against a no-intervention SEIR baseline and human participant data from the AUIB epigame study, we find that LLM agents across four different architectures reduced peak active infections, with quarantine compliance peaking at 58-65% on day six of the 15-day simulation. A binomial generalised linear model showed that perceived health severity was the strongest predictor of quarantine behaviour ($\beta = 0.33, p = 0.002$), yielding a pseudo-$R^2$ of 0.055, comparable to the 0.072 observed in the human trial. LLM architecture is a key determinant of epidemic dynamics: low-variance architectures offer greater internal validity for testing behavioural rules, while high-variance models may better represent real-world decision-making. Geographic labels alone do not induce culturally differentiated behaviour; explicit attitudinal parameterisation is required. This proof-of-principle work lays the groundwork for deploying the Epi-LLM framework as a scalable, risk-free simulation environment for pandemic preparedness research.

117. GRZO: Group-Relative Zeroth-Order Optimization for Large Language Model Fine-Tuning

Authors: Liyan Tan , Yequan Zhao , Yifan Yang , Ruijie Zhang , Xinling Yu , Zheng Zhang
URL: https://arxiv.org/abs/2606.02857
Abstract:

Zeroth-order (ZO) optimization is a memory-efficient alternative to backpropagation for fine-tuning large language models, but its deployment is limited by the high variance of gradient estimation. We propose GRZO, a Group-Relative Zeroth-Order optimizer that draws one pseudo-independent perturbation per mini-batch example and aggregates the per-example losses through group-relative normalization, raising the effective gradient-direction count from one to the batch size at no additional forward cost while preserving inference-level memory. We prove that GRZO is directionally unbiased with variance shrinking proportionally to the batch size, yielding a tighter nonconvex convergence bound than MeZO. Across RoBERTa-large, Llama3-8B, and OPT-13B over multiple tasks, GRZO improves average accuracy on Llama3-8B by $+3.0$ over MeZO at $23\%$ lower peak GPU memory; as a drop-in replacement for the MeZO core, it lifts sparse, low-rank, and quantized ZO variants by $+6.0$ on average.

118. Fixing FOLIO and MALLS: Verified Annotations and an LLM-assisted Framework to Focus Human Relabeling

Authors: Andrea Brunello , Cristian Curaba , Luca Geatti , Michele Mignani , Angelo Montanari , Nicola Saccomanno
URL: https://arxiv.org/abs/2606.02837
Abstract:

Accurate translation from Natural Language to First-Order Logic (NL-to-FOL) underpins neurosymbolic AI systems and Natural Language Inference (NLI), making the quality of NL-to-FOL benchmarks essential – yet these datasets have never been rigorously audited. Our first contribution is to present a systematic human inspection of the validation split of \textsf{FOLIO} and a subset of \textsf{MALLS} test instances, finding that approximately 39% and 36% of entries, respectively, contain incorrect FOL formalizations (i.e., ground truth labels), with additional rates of ambiguous NL sentences (16.4% and 48%) and incorrect NLI labels in \textsf{FOLIO} (8.4%). Our second contribution is to develop and release corrected ground truths for such datasets, showing that annotation errors distort model evaluation on a reference benchmark task: testing three state-of-the-art LLMs (Gemma~4 31B-it, Qwen3-30B-A3B, and GPT-4o-mini) with the corrected ground truths yields accuracy gains from +9 to +22 percentage points. Motivated by these findings, we propose an LLM-based framework to support humans in manual reviewing NL-to-FOL datasets. By directing reviewers toward the most error-prone instances, we empirically show that it is possible to achieve 90% dataset accuracy after reviewing fewer than 24% of instances, compared to over 70% required by unguided review. We release all human-verified annotations and the code for our framework.

119. Large Byte Model: Teaching Language Models About Compiled Code

Authors: Florian Störtz , Catalin-Andrei Stan , Alexandru Dinu , Sandra Servia-Rodríguez , Mihaela Gaman , Calin Miron , Edward Raff
URL: https://arxiv.org/abs/2606.02834
Abstract:

Malware analysis starts with the raw bytes of an executable program, and tools to “lift” these to higher-level representations, such as assembly, are expensive and subject to error. Large Language Models (LLMs) cannot process raw byte representations and answer questions about them. To this end, we present the first byte-native LLM. Based on a vocabulary expansion technique using a bespoke byte tokenizer, such a model is capable of responding to complex questions about malware binaries, with accuracies ranging from 69% for malware family classification to 98% for architecture classification. Our findings indicate that providing domain knowledge during training is essential for this application – off-the-shelf models lack both accuracy and insight. We’ve deployed this emerging solution to a limited number of analysts to gather feedback for further improvements.

120. Which Defense Closes Which Threat? Attributing OWASP-LLM-Top-10 Coverage and Its Brittleness Under Paraphrasing

Authors: Alexandre Cristovão Maiorano
URL: https://arxiv.org/abs/2606.02822
Abstract:

Production LLM applications stack several defense families – refusal-phrase filters, token-budget controls, model allowlists, rate limits, tool-registry authentication – yet existing breach-and-attack-simulation (BAS) benchmarks report a single aggregate coverage number, hiding which family closes which threat. We measure attribution. We add four OWASP-LLM-Top-10-aware agents to a 21-agent baseline scanner and target a lattice of four synthetic LLM endpoints: $L_0$ (no defenses), $L_1$ (refusal-only), $L_2$ (budget-only), and $L_3$ (full stack). $L_1$ and $L_2$ are sibling single-axis ablations, not subsets of each other; $L_3$ is their union plus tool-registry authentication and credential scrubbing. Across $N=10$ replications, the per-OWASP finding count is clean: refusal alone removes all LLM01 (jailbreak) and LLM07 (system-prompt leakage) findings; budget alone removes all LLM02 (sensitive-info disclosure) and LLM10 (unbounded consumption) findings by terminating multi-step sequences; LLM06 (excessive agency) requires the full stack. We probe brittleness under paraphrasing: with 300 Gemini-generated paraphrases ($K=5$ over a 60-template brittleness corpus), $L_1$ refusal block rate falls 15 pp on LLM01 and 25 pp on LLM07. A fifth target, $L_4$-real, swaps the stub backend for Gemini-2.5-flash behind the same $L_3$ regex and matches $L_1$ exactly, indicating no measurable alignment contribution beyond the regex (not a general claim about alignment). Budget controls show no drop (0 pp once the rate-limit floor is factored out). A refusal whitelist that clears a static benchmark can be defeated by an LLM-driven paraphraser without changing attack intent; a budget control resists the same mutation.

121. Do Neural Retrievers Prefer Certain Documents? Evidence of Learned Relevance Priors

Authors: Francisco Valentini , Edgar Altszyler , Martin Fajcik
URL: https://arxiv.org/abs/2606.02814
Abstract:

Neural retrievers are trained to estimate query-document relevance from annotated query-document pairs. Yet annotation protocols may not purely reflect relevance: they select only a subset of documents for labeling, and this selection can favor certain document types over others. We investigate whether supervised bi-encoder retrievers implicitly learn a document-level relevance prior: a query-independent signal encoded in their representation space as a side effect of training on annotated data. We estimate this prior by training simple classifiers on frozen document embeddings and evaluate three state-of-the-art retrievers across multiple IR benchmarks. We find that supervised neural retrievers encode relevance priors that generalize to unseen documents and are consistent across models. These priors create a findability gap: documents with lower prior are systematically harder to retrieve, even when genuinely relevant. This effect appears in supervised dense retrievers but is weaker and less consistent in BM25, and it persists under controlled matched-document comparisons. Using LLM-based explanations, we find that judged-relevant documents tend to be comprehensive, self-contained summaries of mainstream topics, while niche, fragmentary, or highly technical content is often left unjudged. Retrievers internalize this bias, ranking documents with these favored features higher than documents that lack them, independently of their actual relevance. Our findings expose a structural limitation of supervised retrieval: models trained on annotated data do not just learn relevance, but also the implicit document preferences in their training data.

122. Cosmos 3: Omnimodal World Models for Physical AI

Authors: Aditi , Niket Agarwal , Arslan Ali , Jon Allen , Martin Antolini , Adeline Aubame , Alisson Azzolini , Junjie Bai , Maciej Bala , Yogesh Balaji , Josh Bapst , Aarti Basant , Mukesh Beladiya , Mohammad Qazim Bhat , Zaid Pervaiz Bhat , Dan Blick , Vanni Brighella , Han Cai , Tiffany Cai , Eric Cameracci , Jiaxin Cao , Yulong Cao , Mark Carlson , Carlos Casanova , Ting-Yun Chang , Yan Chang , Yu-Wei Chao , Prithvijit Chattopadhyay , Roshan Chaudhari , Chieh-Yun Chen , Junyu Chen , Ke Chen , Qizhi Chen , Wenkai Chen , Xiaotong Chen , Yu Chen , An-Chieh Cheng , Click Cheng , Xiu Chia , Jeana Choi , Chaeyeon Chung , Wenyan Cong , Yin Cui , Magdalena Dadela , Nalin Dadhich , Wenliang Dai , Joyjit Daw , Alperen Degirmenci , Rodrigo Vieira Del Monte , Robert Denomme , Sameer Dharur , Marco Di Lucca , Ke Ding , Wenhao Ding , Yifan Ding , Yuzhu Dong , Nicole Drumheller , Yilun Du , Aigul Dzhumamuratova , Aleksandr Efitorov , Hamid Eghbalzadeh , Naomi Eigbe , Imad El Hanafi , Hassan Eslami , Benedikt Falk , Jiaojiao Fan , Jim Fan , Amol Fasale , Sergiy Fefilatyev , Liang Feng , Francesco Ferroni , Sanja Fidler , Xiao Fu , Vikram Fugro , Prashant Gaikwad , TJ Galda , Katelyn Gao , Yihuai Gao , Wenhang Ge , Sreyan Ghosh , Arushi Goel , Vivek Goel , Akash Gokul , Rama Govindaraju , Jinwei Gu , Miguel Guerrero , Elfie Guo , Aryaman Gupta , Siddharth Gururani , Hugo Hadfield , Song Han , Ankur Handa , Zekun Hao , Mohammad Harrim , Ali Hassani , Nathan Hayes-Roth , Yufan He , Chris Helvig , Cyrus Hogg , Madison Huang
URL: https://arxiv.org/abs/2606.02800
Abstract:

We introduce Cosmos 3, a family of omnimodal world models designed to jointly process and generate language, image, video, audio, and action sequences within a unified mixture-of-transformers architecture. By supporting highly flexible input-output configurations, Cosmos 3 seamlessly unifies critical modalities for Physical AI – effectively subsuming vision-language models, video generators, world simulators, and world-action models into a single framework. Our evaluation demonstrates that Cosmos 3 establishes a new state-of-the-art across a diverse suite of understanding and generation tasks, demonstrating omnimodal world models as scalable, general-purpose backbones for embodied agents. Our post-trained Cosmos 3 models were ranked as the best open-source Text-to-Image and Image-to-Video models by Artificial Analysis, and the best policy model by RoboArena at the time the technical report was written. To accelerate open research and deployment in Physical AI, we make our code, model checkpoints, curated synthetic datasets, and evaluation benchmark available under the Linux Foundation’s OpenMDW-1.1 this https URL License at this https URL }{ this http URL and this https URL . The project website is available at this https URL .

123. Representational Capacity: Geometric Limits on Feature Representation in Transformer Language Models

Authors: Alexander Guha
URL: https://arxiv.org/abs/2606.02765
Abstract:

Model dimension ($d_{model}$) is a fundamental hyperparameter in transformer language models, yet its role in setting the geometric limits of feature representation remains under-explored. Grounded in the Linear Representation and Superposition Hypotheses - which propose that models encode features as near-orthogonal directions in latent space - we develop a framework for estimating how many such directions a model can support. We first establish the embedding matrix as a measurable proxy for near-orthogonality constraints across the latent space: the boundary between meaningful token relationships and incidental similarity in the pairwise cosine similarity distribution gives a concrete estimate of the model’s accepted deviation $\varepsilon$ from perfect orthogonality. Applying this metric across dozens of open-source models reveals two classes: models with high $\varepsilon$ whose embeddings lack near-orthogonal structure, and models with low $\varepsilon$ that maintain it. We then show that the standard Johnson-Lindenstrauss lemma greatly underestimates the packing efficiency of trained representations, and derive an adjusted capacity formula in which the number of near-orthogonal directions depends on the ratio of vectors to dimensions ($k/d$) rather than the raw count - a single modification that cuts prediction error by two orders of magnitude with no extra parameters. Combining these results, we define representational capacity as an upper bound on the number of distinguishable directions available for features and embeddings in a model’s latent space. Capacity is exponentially sensitive to $\varepsilon$, and larger models favor tighter orthogonality constraints over maximizing raw capacity - a pattern compatible with several explanations (a stability-capacity trade-off, a ceiling on usable concepts, or confounds with model scale) that we leave to future work.

124. Acceptance-Test-Driven Evaluation Protocols for Business-Centric LLM Systems

Authors: Eric Liang
URL: https://arxiv.org/abs/2606.02755
Abstract:

Large language model (LLM) applications are increasingly expected to satisfy deterministic institutional requirements while relying on probabilistic generative components. This mismatch makes ordinary post-hoc benchmarking insufficient for systems that must be safe, reliable, auditable, and economically useful. This paper contributes an evaluation-protocol extension for operational LLM systems grounded in acceptance-test-driven development, safety engineering, and business-centric validation. The extension translates stakeholder goals into executable behavioral contracts, release gates, monitoring signals, and evidence artifacts before prompt, model, retrieval, or agent changes are accepted. It adapts the red-green-refactor discipline of test-driven development to a red-train-green lifecycle: first define failing acceptance tests for desired behavior, then improve the LLM system through prompt changes, retrieval design, fine-tuning, guardrails, or data augmentation, and finally release only when multidimensional gates are satisfied. The contribution is a governance-oriented metric stack, reference architecture, and empirical protocol for comparing acceptance-test-driven LLM development against prompt-first and benchmark-after workflows.

125. EntangleCodec: A Unified Discrete Audio Tokenizer via Semantic-Acoustic Entanglement

Authors: Hui Li , Yangfan Gao , Junlin Shang , Changhao Jiang , Tao Gui , Qi Zhang , Xuanjing Huang
URL: https://arxiv.org/abs/2606.02739
Abstract:

Audio tokenizers serve as the discrete interface between continuous audio and Audio Language Models (ALMs), but existing tokenizers often struggle to support both understanding and generation. Reconstruction-oriented codecs preserve acoustic fidelity but lack rich semantics, while semantic-aware tokenizers typically rely on separate semantic and acoustic streams, introducing redundancy or misalignment. We propose \textbf{EntangleCodec}, a unified discrete audio tokenizer that learns caption-aligned semantic-acoustic representations before quantization. By aligning audio with rich captions rather than ASR transcripts, EntangleCodec captures linguistic content, speaker identity, emotion, prosody, and acoustic scenes within a compact token stream. A flow-matching diffusion decoder further enables high-quality reconstruction across speech, music, and general audio. EntangleCodec achieves reconstruction quality competitive with specialized codecs, outperforms all codec-based baselines on audio understanding by up to \textbf{+7.4\% } on MMAR, and supports both TTS and TTA generation in a unified framework. Furthermore, EntangleCodec-based audio language models demonstrate strong scaling behavior: even at \textit{0.6B} parameters, the model surpasses specialized continuous-representation LLMs with over \textit{13B} parameters across three benchmarks using \textbf{22$\times$} fewer parameters; scaling to \textit{8B} further establishes new state-of-the-art results on MMAR, highlighting that representation quality is as critical as model scale in audio language modeling. Code and model weights are available at this https URL .

126. Filter, Then Reweight: Rethinking Optimization Granularity in On-Policy Distillation

Authors: Yuying Li , Leqi Zheng , Yongzi Yu , Wenrui Zhou , Xuchang Zhong , Xing Hu , Jing Jin , Huangjie Yuan , Tao Feng
URL: https://arxiv.org/abs/2606.02684
Abstract:

On-Policy distillation (OPD) in large language models is shifting from full-trace KL supervision toward more selective training paradigms. Recent OPD methods increasingly focus on selecting which trajectories to learn from, which tokens are most informative, and which supervision signals are most reliable. Motivated by this trend, we rethink optimization granularity of OPD and propose \fireicon\ FiRe-OPD (Filter, then Reweight), which jointly adjusts supervision signals at both trajectory and token levels. In details, FiRe-OPD first filters trajectories to remove low-quality rollout samples, and then applies soft reweighting within the retained trajectories to emphasize informative tokens. Compared with hard token selection, FiRe-OPD leverages a soft-weighting mechanism to effectively mitigate information loss and enhance optimization stability, thereby achieving finer-grained OPD optimization. We validate the effectiveness of FiRe-OPD across strong-to-weak, single-teacher, and multi-teacher settings, and demonstrate its superiority over recent token-level OPD methods ( (e.g., +6.25 on AIME 2024 in strong-to-weak, +18.81 on Miner in multi-teacher). Our code is available at this https URL .

127. The Ringelmann Effect in Multi-Agent LLM Systems: A Scaling Law for Effective Team Size

Authors: Blaž Bertalanič , Carolina Fortuna
URL: https://arxiv.org/abs/2606.02646
Abstract:

Inference-time multi-agent LLM scaling lacks a shared unit: counting nominal agents conflates cost with independent evidence. We derive a two-parameter scaling law $R(N) = N_\text{eff}/N = 1/(1+c(N-1)N^{-\beta})$ where the regime exponent $\beta$ classifies any configuration into one of three asymptotic regimes – hard-ceiling at $1/c$ ($\beta = 0$), sublinear at $N^\beta/c$ ($0 < \beta < 1$), or linear ($\beta \ge 1$), and a mean-field theorem predicts that peer count $k$ and rounds $\tau$ during agent debate enter the dynamics only through their product $k\tau$. The law applies at two levels: answer diversity and correctness redundancy. Across 44 (model $\times$ task $\times$ condition) cells spanning peer debate, self-correction, random-noise placebo, self-consistency, three open-weight families (Qwen, Llama, Ministral) at scales from 7B to 32B with a frontier API check (Gemini), thinking models, heterogeneous teams, and sparse communication, the functional form fits every condition at $R^2 > 0.99$; only $(c, \beta)$ shifts. On free-form math, dense peer influence collapses the answer-level regime from sublinear into hard-ceiling; correctness-level fits remain hard-ceiling throughout. Three findings have practical implications. \emph{(i)}~Thirty dense debating agents produce no more answer diversity than one on MMLU-Hard. \emph{(ii)}~A noise placebo tracks self-correction on free-form math and at $4\times$ scale, so within homogeneous teams the gain commonly attributed to ``debate’’ comes from re-evaluation, not peer content. \emph{(iii)}~A single $N \le 5$ pilot predicts the $N=30$ structural ceiling, and within the configurations tested only architectural diversity (heterogeneous teams) lowers $c$ and escapes the hard-ceiling regime, communication-mode interventions do not.

128. A New Framework for Cybersecurity Refusals in AI Agents

Authors: Eliot Krzysztof Jones , Mateusz Dziemian , Matt Fredrikson , J Zico Kolter
URL: https://arxiv.org/abs/2606.02644
Abstract:

Agentic scaffolds have dramatically improved LLM performance on complex, long-horizon tasks, yielding both broad benefits and amplified risks in domains like cybersecurity. Existing benchmarks for AI agents in cybersecurity focus mainly on measuring proficiency–how effectively agents can complete offensive security tasks–but neglect a critical question: when and how should agents refuse harmful requests? We present the first framework for establishing refusal boundaries in offensive security contexts. Our framework defines (1) principled criteria for when tasks should be refused, (2) categories of tasks that warrant refusal, and (3) evaluation methodology for measuring agent robustness under both benign and adversarial conditions. We apply this framework to assess how current LLM-powered agents adhere to appropriate refusal boundaries across a range of web-based offensive security scenarios, finding that 6 of 8 frontier models tested show near-zero refusal rates, with only 2 models (GPT-5.2 and GPT-5.1 Codex) demonstrating any meaningful refusal behavior.

129. Inference Cost Attacks for Retrieval-Augmented Large Language Models

Authors: Chengliang Liu , Liangbo Ning , Yujuan Ding , Wenqi Fan
URL: https://arxiv.org/abs/2606.02643
Abstract:

Retrieval-Augmented Generation (RAG)-enhanced LLM systems, while powerful, introduce substantial inference costs due to the inclusion of an extra multi-stage pipeline that dynamically retrieves and synthesizes information from external knowledge sources. This high operational cost exposes a critical vulnerability to Inference Cost Attacks (ICAs). However, existing ICAs often rely on the impractical assumption of direct prompt manipulation. We argue that a more feasible and potent threat to RAG-enhanced LLM systems arises from poisoning external knowledge bases (e.g., web knowledge from the Internet). In this work, we introduce the Retrieval-Augmented Inference Cost Attack (RA-ICA), a novel attacking paradigm that targets the computational cost of RAG-enhanced LLM systems by injecting malicious documents into external knowledge corpus. To operationalize this attack, we propose Computational Resource Exhaustion via External Poisoning (CREEP), a novel framework that leverages LLM agents to automatically craft malicious documents that are both semantically relevant for retrieval and potent for inducing an abnormal increase in token consumption during the inference phase. To enhance the attack’s effectiveness, we introduce Memory-Augmented Group Relative Policy Optimization (MA-GRPO), a novel reinforcement learning algorithm that fine-tunes the agents by learning from a dynamic memory of historical best adversarial documents. Extensive experiments across three real-world datasets demonstrate that RA-ICA increases token consumption by up to 13.12 times with an over 90% success rate, without degrading the integrity of the generated answer.

130. SVHalluc: Benchmarking Speech-Vision Hallucination in Audio-Visual Large Language Models

Authors: Chenshuang Zhang , Kyeong Seon Kim , Chengxin Liu , Tae-Hyun Oh
URL: https://arxiv.org/abs/2606.02642
Abstract:

Despite the success of audio-visual large-language models (LLMs), they can produce plausible but ungrounded outputs, termed hallucination. Existing benchmarks focus on environmental sounds (e.g., dog barking) to indicate event occurrence. In contrast, human speech carries fundamentally different, rich semantics and temporal structures, yet it remains unexplored whether current models can accurately align speech content with corresponding visual signals. In this work, we show that speech content can induce hallucinations in audio-visual LLMs. To systematically study this, we introduce SVHalluc, the first comprehensive benchmark for evaluating speech-vision hallucination in audio-visual LLMs. Our benchmark diagnoses speech-vision hallucinations from two critical and complementary aspects: semantic and temporal. Experimental results demonstrate that state-of-the-art open-source audio-visual LLMs struggle with aligning speech content with corresponding visual signals, with a near-random accuracy on multiple tasks. In contrast, Gemini 2.5 Pro significantly outperforms the open-source models. Our analysis suggests that their failures stem from limited ability in cross-modality understanding, despite strong performance in single-modality perception. Our work uncovers a new and fundamental limitation of current audio-visual LLMs and highlights the need for speech-grounded video comprehension. Project page: this https URL .

131. D-Judge: Disrupting Multi-Turn Jailbreaks using Semantics-Preserving Output Rewriting

Authors: Huanli Gong , Zhipeng Wei , Yu Fu , Haz Sameen Shahgir , Ananya Gupta , Yue Dong , N. Benjamin Erichson
URL: https://arxiv.org/abs/2606.02640
Abstract:

Multi-turn jailbreak attacks pose a growing threat to large language model (LLM) safety because they exploit feedback from auxiliary judge models to iteratively refine prompts toward harmful goals. Existing defenses largely detect or block unsafe content at individual turns or at the final response, leaving the judge-driven refinement loop intact and allowing attackers to extract informative feedback from intermediate interactions. We introduce D-Judge, a semantics-preserving output rewriting defense that intervenes directly in this loop by rewriting the victim LLM’s responses before they are evaluated by the attacker’s judge. By misaligning the judge’s feedback signal without changing the meaning of the original response, D-Judge derails the attacker’s prompt-refinement process, causing subsequent queries to be optimized against a distorted signal of attack progress. To improve D-Judge’s ability to produce such rewrites, we construct a dataset of semantically equivalent response pairs that induce different judge-assigned harmfulness scores, and use it for supervised fine-tuning followed by direct preference optimization. Experiments on HarmBench show that D-Judge reduces the success rate of state-of-the-art multi-turn jailbreaks while preserving performance on benign benchmarks.

132. SegTune: Structured and Fine-Grained Control for Song Generation

Authors: Yuejiao Wang , Zihao Ji , Pengfei Cai , Xu Li , Haorui Zheng , Zewen Song , Zhongliang Liu , Chen Zhang , Pengfei Wan
URL: https://arxiv.org/abs/2606.02638
Abstract:

Recent advances in neural song generation have enabled high-quality synthesis from lyrics and global textual prompts. However, most systems fail to model temporally varying attributes of songs, severely limiting fine-grained control over musical structure and dynamics. To address this, we propose SegTune, a Diffusion Transformer-based framework enabling structured and fine-grained controllability by allowing users or large language models (LLMs) to specify local musical descriptions aligned to song segments. These segment prompts are temporally broadcast to corresponding time windows, while global prompts ensure stylistic coherence. To support precise lyric-to-music alignment, we introduce an LLM-based duration predictor that autoregressively generates sentence-level timestamps in LyRiCs format. We further construct a large-scale data pipeline for high-quality song collection with aligned lyrics and prompts, and propose new metrics to evaluate segment alignment and vocal consistency. Experiments demonstrate that SegTune outperforms existing baselines in both musicality and controllability. Visit our project page ( this https URL ) for codes and more generated songs.

133. Position: Prioritize Identifying Structure, Not Complex Models, for Scientific Discovery

Authors: Tyler H. McCormick
URL: https://arxiv.org/abs/2606.02632
Abstract:

Modern Machine Learning (ML) and Artificial Intelligence (AI) models, especially large language models (LLMs), are increasingly used to generate scientific hypotheses and mechanistic explanations from observational data. This position paper argues that in the high-dimensional proxy regimes where modern ML excels, mechanistic learning is generically underdetermined: many incompatible mechanisms induce essentially the same observational relationships on the support of the data, so predictive success and coherent explanations are insufficient evidence of mechanism discovery. This underdetermination becomes uniquely hazardous with large language models (LLMs), which tend to collapse large equivalence classes of explanations into a single fluent narrative. This paper proposes concrete standards for ``mechanistic ML,’’ and argues these norms are necessary if LLM-centered workflows are to support science rather than merely simulate it.

134. FSA-GRPO: Teaching Auditory LLMs to Use Few-shot Demonstrations

Authors: Haolong Zheng , Siyin Wang , Xulin Fan , Zengrui Jin , Mark Hasegawa-Johnson
URL: https://arxiv.org/abs/2606.02615
Abstract:

Few-shot prompting provides an effective way to adapt auditory large language models to low-resource tasks such as children’s speech recognition. However, most auditory large language models are not explicitly trained to perform inference in this demonstration-conditioned format, limiting the extent to which they can benefit from few-shot prompting. To address this limitation, we introduce Few-Shot Aware GRPO (FSA-GRPO), an RL-based post-training recipe that uses a specially designed reward to encourage the model to leverage few-shot demonstrations, thereby strengthening its few-shot adaptation ability. Notably, training with only high-resource adult ASR data improves the model’s general few-shot adaptation ability, yielding gains not only in children’s speech recognition but also in speech translation and audio understanding. We further study data selection and auxiliary reward weighting to identify an effective training recipe. Our experiments show that when in-domain data are unavailable or cannot be used for training, FSA-GRPO is more effective than direct tuning on related out-of-domain data.

135. ReLoRA: Knowledge-Reusing Adaptation for Fast Rollout of Evolving LLM Services

Authors: Yang Xu , Zihuai Xu , Hongli Xu , Yunming Liao , Zhiwei Yao , Xitong Fu
URL: https://arxiv.org/abs/2606.02606
Abstract:

Large Language Models (LLMs) are increasingly deployed as continuously evolving services, where frequent base-model updates may invalidate previously deployed task-specific Low-Rank Adaptation (LoRA) adapters. For service providers managing numerous downstream model services, retraining each LoRA adapter from scratch for every updated base model is computationally prohibitive and delays service rollout. Meanwhile, the simpler alternative, i.e., naively applying the original LoRA adapter to the updated base model, often leads to degraded service quality due to adapter-backbone incompatibility. To address this problem, we propose ReLoRA, a knowledge-reusing re-adaptation framework that efficiently restores service-ready LoRA adapters for evolving LLM services while preserving or improving task performance. Specifically, ReLoRA comprises two key optimization steps: 1) Adaptive LoRA initialization leverages Bayesian optimization to construct a compatibility-aware starting point by fusing information from both the previously deployed task adapter and the base model’s evolution; 2) Fine-tuning with scheduled regularization first rapidly steers the adapter to a high-quality region via strong regularization, followed by relaxed regularization for task-specific refinement. This design enables rapid service-quality recovery with reduced re-adaptation overhead. Extensive experiments demonstrate that ReLoRA reduces time-to-readiness by up to 8.9$\times$ and improves accuracy by up to 4.6\% compared to baselines.

136. IdiomX A Multilingual Benchmark for Idiom Understanding, Retrieval, and Interpretation

Authors: Ayman Ali Sharara
URL: https://arxiv.org/abs/2606.02584
Abstract:

Idiomatic expressions remain a persistent challenge for natural language processing because their meanings are often non-compositional, context-dependent, and difficult to align across languages. Existing idiom resources are often limited in scale, contextual diversity, or multilingual coverage, restricting their utility for modern language models. We introduce IdiomX, a large-scale multilingual benchmark for idiom understanding, retrieval, and interpretation, constructed through a reproducible multi-stage pipeline combining lexical resource extraction, large-scale normalization, controlled large language model enrichment, and structured validation. The resulting dataset contains over 190K contextualized examples spanning 12K+ idioms, with aligned English, Arabic, and French semantic representations, idiomatic and literal usage labels, and rich linguistic metadata. Building on this resource, we define a unified four-task benchmark covering idiom detection, context-to-idiom retrieval, Arabic-to-English idiom retrieval, and idiom interpretation, extending evaluation from figurative recognition to semantic grounding and explainable meaning retrieval. Experiments show that contextual transformer models substantially improve idiom detection, while hybrid retrieval and reranking architectures significantly strengthen both monolingual and cross-lingual idiom retrieval. Results further demonstrate that idiom interpretation can be effectively modeled as a semantic retrieval task, introducing interpretability as a complementary benchmark dimension. Overall, IdiomX provides a scalable benchmark for studying idiomatic language as a progression from detection to retrieval and semantic interpretation, and offers a modular framework extensible to additional languages and figurative reasoning tasks

137. Cost-Aware Query Routing in RAG: Empirical Analysis of Retrieval Depth Tradeoffs

Authors: Sanjay Mishra
URL: https://arxiv.org/abs/2606.02581
Abstract:

Retrieval-augmented generation (RAG) faces a fundamental three-way tension: deeper retrieval improves factual grounding but inflates token costs and end-to-end latency. Static retrieval configurations cannot resolve this tension across heterogeneous query workloads – simple definitional queries waste budget on unnecessary context, while complex analytical prompts are underserved by shallow retrieval. This paper introduces \emph{Cost-Aware RAG} (CA-RAG), a per-query routing framework that selects from a discrete catalog of \emph{strategy bundles} – each coupling a retrieval depth (from retrieval-free direct inference to top-$k{=}10$ dense retrieval) with a fixed generation profile – by maximizing a scalar utility that linearly combines an estimated quality prior with normalized penalties for predicted latency and total billed tokens. CA-RAG is implemented with FAISS-backed dense retrieval and OpenAI chat/embedding APIs, and evaluated on a 28-query benchmark spanning four bundles. The router dynamically exercises all bundles, achieving \textbf{26\% fewer billed tokens} than always-heavy retrieval and \textbf{34\% lower mean latency} than always-direct inference while maintaining equivalent answer quality. Per-query delta analysis reveals that savings are non-uniform and concentrated in simpler queries, motivating complexity-aware guardrails. Sensitivity analysis confirms that the same bundle catalog supports multiple cost-latency-quality operating points through weight adjustment alone. All results are generated directly from logged CSV artifacts for full reproducibility. CA-RAG provides a transparent, auditable foundation for cost-conscious LLM deployments.

LLM 관련 주요 논문 - 2026-06-03

1. Imaginative Perception Tokens Enhance Spatial Reasoning in Multimodal Language Models

2. Reasoning Structure of Large Language Models

3. PyraMathBench: Evaluating and Improving Mathematical Capability in Large Language Models

4. EvoDS: Self-Evolving Autonomous Data Science Agent with Skill Learning and Context Management

5. Enhancing Operational Safety via Agentic Dialogue Hazard Identification Analysis

6. LAP: An Agent-to-Instrument Protocol for Autonomous Science

7. Proof-Refactor: Refactoring Generated Formal Proofs into Modular Artifacts

8. When to Re-Plan: Subgoal Persistence in Hierarchical Latent Reasoning

9. Code-on-Graph: Iterative Programmatic Reasoning via Large Language Models on Knowledge Graphs

10. Dynamic Objective Selection with Safeguards and LLM Oversight for Financial Decision-Making

11. The DeepSpeak-Agentic Dataset

12. EvoDrive: Pareto Evolution for Safety-Critical Autonomous Driving via Self-Improving LLM Agents

13. From Answers to States: Verifiable Process-Level Evaluation of Chemical Reasoning in Large Language Models

14. Diagnosing Knowledge Gaps in LLM Tool Use: An Agentic Benchmark for Novel API Acquisition

15. Gender-Dependent Diagnostic Substitution in LLM Medical Triage: Same Symptoms, Unequal Urgency

16. TSQAgent: Rating Time Series Data Quality via Dedicated Agentic Reasoning

17. Cross-Lingual Token Arbitrage: Optimizing Code Agent Context Windows via Local LLM Preprocessing

18. StepFinder: A Temporal Semantic Framework for Failure Attribution in Multi-Agent Systems

19. DMF: A Deterministic Memory Framework for Conversational AI Agents

20. CP-Agent: Context-Aware Multimodal Reasoning for Cellular Morphological Profiling under Chemical Perturbations

21. The Reliability Gap in Benchmark Auditing: Distribution Shift and Scale as Failure Modes of Contamination Detection

22. LEAP: Supercharging LLMs for Formal Mathematics with Agentic Frameworks

23. A Negative Result on Cross-Model Activation Transfer in a Pythia Multi-Hop Setting

24. Distilling Answer-Set Programming Rules from LLMs for Neurosymbolic Visual Question Answering

25. Perceive Before Reasoning: A Pre-Reasoning Perception Framework for Efficient and Reliable Proactive Mobile Agents

26. ClinicalMC: A Benchmark for Multi-Course Clinical Decision-Making with Large Language Models

27. GTBench: A Curriculum-Grounded Benchmark for Evaluating LLMs as Mathematical Research Assistants in Graph Theory

28. Think-Before-Speak: From Internal Evaluation to Public Expression in Multi-Agent Social Simulation

29. Uncertainty-Aware Clarification in LLM Agents with Information Gain

30. EvoTrainer: Co-Evolving LLM Policies and Training Harnesses for Autonomous Agentic Reinforcement Learning

31. From Long News to Accurate Forecast: Importance-Aware Fusion and PRM-Guided Reflection for Time Series Forecasting

32. Decomposing how prompting steers behavior

33. The Shadow Price of Reasoning: Economic Perspective on Optimal Budget Allocation for LLMs

34. DELTAMEM: Incremental Experience Memory for LLM Agents via Residual Trees

35. CORE: Conflict-Oriented Reasoning for General Multimodal Manipulation Detection

36. SkillDAG: Self-Evolving Typed Skill Graphs for LLM Skill Selection at Scale

37. TriEval: A Resource-Efficient Pipeline for LLM Bias, Toxicity, and Truthfulness Assessment

38. Inducing Reasoning Primitives from Agent Traces

39. Toward a Modular Architecture for Embedded AI Agent Systems at the Edge

40. Traj-Evolve: A Self-Evolving Multi-Agent System for Patient Trajectory Modeling in Lung Cancer Early Detection

41. ChatHealthAI: Aligning Electronic Health Record Representations with Large Language Models for Grounded Clinical Reasoning

42. Visual Graph Scaffolds for Structural Reasoning in Large Language Models

43. Language Models Need Sleep: Learning to Self-Modify and Consolidate Memories

44. AlignAtt4LLM: Fast AlignAtt for Decoder-Only LLMs at IWSLT 2026 Simultaneous Speech Translation Task

45. Agentic Chain-of-Thought Steering for Efficient and Controllable LLM Reasoning

46. Using Reward Uncertainty to Induce Diverse Behaviour in Reinforcement Learning

47. Efficient ASR Training with Conversations that Never Happened

48. NetKV: Network-Aware Decode Instance Selection for Disaggregated LLM Inference

49. Agent libOS: A Library-OS-Inspired Runtime for Long-Running, Capability-Controlled LLM Agents

50. Beyond Encoder Accumulation: Measuring Encoder Roles in Multi-Encoder VLMs

51. From ‘What’ to ‘How’ and ‘Why’: Sharing LLM-Generated Retrospective Summaries of Older Adults’ Passive Tracking Data with Remote Family Members

52. A Training-Free Mixture-of-Agents Framework for Multi-Document Summarization using LLMs and Knowledge Graphs

53. Taiji: Pareto Optimal Policy Optimization with Semantics-IDs Trade-off for Industrial LLM-Enhanced Recommendation

54. FLARE: Fine-Grained Diagnostic Feedback for LLM Code Refinement

55. Clustered Self-Assessment: A Simple yet Effective Method for Uncertainty Quantification in Large Language Models

56. AI Agents Enable Adaptive Computer Worms

57. Trading Human Curation for Synthetic Augmentation in RLVR

58. E2LLM: Towards Efficient LLM Serving in Heterogeneous Edge/Fog Environments

59. Merit or networks? What decides where research is published

60. Tool-Aware Optimization with Entropy Guidance for Efficient Agentic Reinforcement Learning

61. Ultralytics YOLO26: Unified Real-Time End-to-End Vision Models

62. A Close Look At World Model Recovery In Supervised Fine-Tuned LLM Planners

63. CoEval: Ranking Language Models for Custom Tasks Without Labeled Data or Trustworthy Benchmarks

64. Safety Measurements for Fine-tuned LLMs Should be Grounded in Capability

65. Black-box, Adaptive, Efficient, Transferable, Harmful, Applicable… Attacks Are All You Need to Break LLMs

66. The Shape of Addition: Geometric Structures of Arithmetic in Large Language Models

67. VidMsg: A Benchmark for Implicit Message Inference in Short Videos

68. Building Reliable Long-Form Generation via Hallucination Rejection Sampling

69. TurtleAI: Benchmarking Multimodal Models for Visual Programming in Turtle Graphics

70. Physics-Guided Policy Optimization with Self-Distillation

71. Exploiting Verification-Generation Gap: Test-Time Reinforcement Learning with Confidence-Conditioned Verification

72. Testing LLM Arithmetic Reasoning Generalization with Automatic Numeric-Remapping Attacks

73. CauTion: Knowing When to Trust LLMs for Ensemble Causal Discovery

74. DDOR: Delta Debugging for Explainable Overrefusal Testing and Repair

75. When Attention Collapses: Stage-Aware Visual Token Pruning from Structure to Semantics

76. \textsc{CR-Seg}: Attention-Guided and CoT-Enhanced Coarse-to-Refined Reasoning Segmentation

77. Learn from Your Mistakes: Tree-like Self-Play for Secure Code LLMs

78. NeuroArmor: Safe-Variant-Guided Representation Consistency for Selective Re-Anchoring in Jailbreak Defense

79. Analyzing Stream Collapse in Hyper-Connections: From Diagnosis to Mitigation