LLM 관련 주요 논문 - 2026-06-09

1. SearchSwarm: Towards Delegation Intelligence in Agentic LLMs for Long-Horizon Deep Research

Authors: Pu Ning , Quan Chen , Kun Tao , Xinyu Tang , Tianshu Wang , Qianggang Cao , Xinyu Kong , Zujie Wen , Zhiqiang Zhang , Jun Zhou
URL: https://arxiv.org/abs/2606.09730
Abstract:

Large language models are increasingly expected to handle complex, long-horizon real-world tasks whose context demands can grow without bound, yet model context windows remain inherently finite. Recent work explores a paradigm where a main agent decomposes tasks and dispatches subtasks to subagents, which execute and return only summarized results, conserving the main agent’s context budget. However, performing this well requires delegation intelligence: the ability to decompose complex tasks, determine when and what to delegate, and integrate returned results into the ongoing workflow. Training data for this capability is scarce in naturally occurring text, and to our knowledge, how to synthesize such data and train models to acquire this capability remains largely unexplored in the open-source community. To bridge this gap, we present a preliminary exploration targeting deep research, a representative long-horizon agent task. Specifically, we design a harness that guides the model toward high-quality task decomposition and delegation, while constraining subagents to return results properly to support the main agent’s workflow. The harness-guided trajectories naturally encode correct delegation decisions, which we use as supervised fine-tuning data to internalize delegation intelligence into model weights. Our resulting model, SearchSwarm-30B-A3B, achieves 68.1 on BrowseComp and 73.3 on BrowseComp-ZH, the best results among all models of comparable scale. We will release our harness, model weights, and training data to facilitate future research.

2. Beyond Probabilistic Similarity: Structural, Temporal, and Causal Limitations of Retrieval-Augmented Generation in the Legal Domain

Authors: Hudson de Martim
URL: https://arxiv.org/abs/2606.09724
Abstract:

Retrieval-Augmented Generation (RAG) has become a standard architectural response to unreliability in legal AI, yet high-profile failures, including fabricated citations submitted to courts and anachronistic legal content presented as current, continue to appear across jurisdictions. We argue that these failures are not residual confabulations to be eliminated by scaling language models, but symptoms of an architectural mismatch between probabilistic retrieval and the hierarchical, temporal, and institutional structure of legal knowledge. We develop the argument in three moves. First, we articulate the ontological commitment of legal knowledge as a triad of properties derivable from classical legal theory: hierarchical and mereological structure, diachronic dynamism under operational closure, and causal traceability of institutional provenance grounded in the duty of justification. Second, we identify three corresponding pathologies of retrieval (mereological blindness, diachronic blindness, and causal opacity), each developed with an operational definition, a failure mechanism, a canonical example, and detection criteria for diagnostic use. Third, we review the state of the art through this lens, showing that existing approaches address these requirements unevenly and do not yet compose into a paradigm that treats them as co-constitutive. From this analysis we derive four architectural commitments that characterize the deterministic-by-design direction for legal retrieval: ontological primacy, event reification, bitemporal correctness, and deterministic interaction protocols. The framework concerns quaestio juris (which norms apply and in what state) rather than the downstream tasks that act on identified norms, and addresses legislative and constitutional retrieval primarily, with interpretive time as an explicit extension.

3. (Auto)formalization is supposed to be easy: Trellis process semantics for spelling out rigorous proofs

Authors: Wesley Pegden
URL: https://arxiv.org/abs/2606.09674
Abstract:

We present Trellis: an autoformalization system that leverages LLM agents in a deterministically constrained workflow to enforce incremental progress in Lean autoformalization tasks through iterative refinement of natural language proofs. Our approach is motivated by the common mathematician’s notion of what it means to have a rigorous proof in the first place: namely, that it would be routine to elaborate any part of the proof in further detail. The result is a system which aims to achieve reliable autoformalization on a modest budget and with generalist agents, with specialization to autoformalization coming not from any task-specific agent training but instead from a meaning-of-rigor inspired workflow enforced by process semantics. We link to an end-to-end Lean formalization of a recent Ramsey theory breakthrough produced by the process.

4. Correlation Is Not Enough: Embedding Human Metadata for Individual Causal Discovery

Authors: Suraj Biswas , Saurabh Gupta , Pritam Mukherjee
URL: https://arxiv.org/abs/2606.09672
Abstract:

Ask a pretrained biomedical language model whether “cortisol 28 ug/dL” and “stock-market volatility” are related, and it returns a cosine similarity of 0.83 on a scale where 1.0 means identical. The two share no mechanism. This is not a corner case: every off-the-shelf biomedical encoder we tested (BioBERT, PubMedBERT, BioM-ELECTRA) scores unrelated cross-domain pairs between 0.76 and 0.92 when the answer should be near zero. Accuracy on cross-domain discrimination is 0%. Retrieval systems survive this, because a language model downstream filters the noise. A Large Behavioural Model (LBM), a foundation model whose subject is a person rather than a sentence, does not: it reasons over a graph of a user’s life and treats embedding proximity as evidence that two events are causally linked. False proximity writes a false causal edge, and everything downstream inherits the error. Here, embedding geometry is not a tuning knob; it is correctness. We report the fix. A contrastive pass over 72,034 pairs raises PubMedBERT BIOSSES correlation from 0.633 to 0.828 and within-vs-across-domain separation from 1.05x to 1.63x. A second pass, BODHI, mines hard negatives from edges absent in a biomedical knowledge graph and lifts separation to 2.30x and the discrimination gap to +0.392, at a 4.5% BIOSSES cost. On an Intel Xeon 6737P with AMX, OpenVINO cuts single-query latency from 1367 ms to 10 ms (133x) and reaches 555 sentences/sec. One finding contradicts standard advice: FP16 beats INT8 on this silicon at every serving batch size, and we explain why. The same model on a no-AMX Ice Lake instance runs 13-27x slower. We release the benchmark suite, training corpora, the BODHI generator, and the OpenVINO scripts.

5. SpatialWorld: Benchmarking Interactive Spatial Reasoning of Multimodal Agents in Real-World Tasks

Authors: Hongcheng Gao , Hailong Qu , Jingyi Tang , Jiahao Wang , Zihao Huang , Hengkang Qiao , Shihong Huang , Junming Yang , Yi Li , Hongyixuan Yuan , Wenjie Li , Bohan Zeng , Wenbo Li , Bo Wang , Jianhui Liu , Olive Huang , Haoyang Huang , Wentao Zhang , Guoqing Huang , Nan Duan , Yinpeng Dong
URL: https://arxiv.org/abs/2606.09669
Abstract:

Spatial reasoning is a foundational capability for multimodal large language models (MLLMs) to perceive and operate within the physical world. However, existing benchmarks predominantly rely on passive evaluation (e.g., static VQA) or simulator-specific pipelines, failing to assess general interactive spatial understanding. We introduce SpatialWorld, a unified benchmark designed specifically for evaluating the interactive spatial understanding of multimodal agents in complex real-world tasks. Integrating eight heterogeneous simulation backends under a shared, simulator-agnostic protocol, SpatialWorld features 760 human-annotated tasks across diverse domains (e.g., household routines, travel, social collaboration). Agents must solve tasks under vision-only partial observability, actively gathering egocentric visual evidence and expressing decisions via a unified, text-based action interface native to MLLMs. For reliable evaluation, each task includes a human-validated initial state, a reference trajectory, and a terminal-state verifier. Evaluating 15 advanced agents reveals that robust spatial task solving remains challenging: the strongest model, GPT-5, achieves an average task success rate (TSR) of only 17.4%, while the leading open-source model, Qwen-3.5, reaches 14.1%. Further analysis exposes a clear mismatch between task success and execution efficiency, alongside substantial domain-specific performance variations. These bottlenecks in active exploration and long-horizon planning position SpatialWorld as a rigorous testbed for future spatial agents.

6. Optical Reasoning: Rethinking Images as an Expressive Reasoning Medium Beyond Text

Authors: Yutong Bian , Dongjie Cheng , Heming Xia , Yongqi Li , Wenjie Li
URL: https://arxiv.org/abs/2606.09585
Abstract:

Chain-of-Thought (CoT) improves the performance of Large Language Models (LLMs) and has been extended to Multimodal Large Language Models (MLLMs). More recent work further moves from text-based multimodal reasoning toward interleaved-modal reasoning, where intermediate steps can incorporate both textual rationales and visual evidence. In this work, we propose a bolder and more ambitious idea: could images alone serve as the reasoning medium for both language and multimodal tasks? To explore this, we propose optical reasoning, which treats images as a standalone reasoning medium. We instantiate this concept with two variants: typographic-based optical reasoning, which optimizes visual layouts for compact rationale rendering, and graphical-based optical reasoning, which composes text and graphical elements into structured visual rationales. Across mathematical, scientific, and interleaved-modal reasoning benchmarks, optical reasoning can match or even exceed traditional text reasoning while reducing reasoning tokens by an average of 28.57% on language tasks and 16% on multimodal tasks, achieving 1.96 times the token efficiency of text reasoning. These results show that images can effectively and efficiently encode rationales while providing a unified visual canvas for reasoning.

7. TABVERSE: Benchmarking Cross-Format Table Understanding in LLMs and VLMs

Authors: Momina Ahsan , Sarfraz Ahmad , Ming Shan Hee , Roy Ka-Wei Lee , Preslav Nakov
URL: https://arxiv.org/abs/2606.09578
Abstract:

Large Language Models (LLMs) and Vision-Language Models (VLMs) are increasingly evaluated on table reasoning tasks, but the role of table representation remains under-explored. In practice, the same table content may appear in different structural formats, such as HTML, Markdown, and LaTeX, or as rendered images. However, existing evaluations often let content, format, layout, and modality vary together, making it difficult to isolate representation effects. We introduce TABVERSE, a controlled multimodal table benchmark that aligns the same table content across multiple structural formats and rendered images, with question category and difficulty tags. This design enables systematic evaluation of representation effects while holding table content fixed. We evaluate LLMs and VLMs across three tasks: Question Answering (QA), Structural Understanding Capability (SUC), and Structure Reconstruction (SR). Our results show that representation choice substantially affects table understanding. Models generally perform better with structured text than with rendered images, but the size of this gap depends on the task, model, and format. HTML is often the most robust text format, while row-sensitive structural tasks and syntactically usable LaTeX reconstruction remain challenging. These findings show that table representation is a key factor in reliable table evaluation.

8. PRISM: Recovering Instruction Sets from Language Model Activations

Authors: Gilad Gressel , Rahul Pankajakshan , Julia Diament , Efim Hudis , Krishnashree Achuthan , Yisroel Mirsky
URL: https://arxiv.org/abs/2606.09563
Abstract:

As LLMs are deployed as agents, reliable monitoring requires knowing not only what they output, but which instructions are steering their behavior. This is difficult when models infer unintended subgoals, follow contextual cues, or are influenced by prompt injections and hidden objectives. While activation-to-language methods suggest that hidden states can reveal natural-language information, existing approaches are not designed to recover the full set of simultaneous instructions, constraints, prohibitions, and subgoals active in agentic settings. We formalize this problem as instruction set retrieval and introduce PRISM, an activation-conditioned interpreter that decodes hidden states from a frozen target model into a faithful bullet list of active instructions. Unlike prior activation-to-language methods, PRISM is trained to recover instruction sets directly, using judge-guided GRPO to reward covered instructions and penalize unsupported ones. Across benign, constrained, prompt-injection, and hidden-objective settings, PRISM outperforms activation-to-language baselines, especially on security-relevant objectives.

9. AI Scientists Are Only as Good as Their Evidence: A Stratified Ablation of Proprietary Data and Reasoning Skills in Drug-Asset Valuation

Authors: Yinan Wang
URL: https://arxiv.org/abs/2606.09556
Abstract:

AI Scientist agents are often evaluated as if capability were mainly a function of model quality, prompting, or reasoning scaffolds. We test a different hypothesis in drug-asset valuation: for knowledge-intensive scientific decisions, the limiting factor is often the evidence substrate the agent can access. We run a controlled three-arm ablation on a production valuation agent: A is a plain web-only LLM analyst, B adds public structured tools plus a 14-dimension valuation playbook, verifier, objectivity policy and red-team, and C adds the proprietary Noah AI corpus of curated pipeline, trial and deal intelligence. Across a 13-asset stratified benchmark, B improves calibration and audit discipline: tier-in-range accuracy rises from 0.80 to 0.89 and objectivity from 3.16 to 3.30. But B does not remove the factual ceiling. Under capability-superset accounting, A and B recover only 0.25 and 0.38 of the curated gold competitive record, while C recovers 0.96; on the curated long-tail subset, C reaches 0.93 vs. 0.26/0.30. Raw blind-panel decision quality is similar for A and B (7.01 vs. 6.96), so we introduce completeness-aware decision utility: informed decision-quality = decision-quality x gold-coverage. On this metric, C reaches 7.43 vs. 1.76/2.57 for A/B. Even a perfect non-proprietary-data report would be capped at 3.83 by B’s coverage. The result is not that reasoning scaffolds are unimportant; they improve calibration and discipline. Rather, proprietary evidence sets the upper bound of what the AI Scientist can know and therefore decide.

10. From Rigid to Dynamic: Entropy-Guided Adaptive Inference for Long-Context LLMs

Authors: Zhanchao Xu , Haoyang Li , Qingfa Xiao , Fei Teng , Chen Jason Zhang , Lei Chen , Qing Li
URL: https://arxiv.org/abs/2606.09508
Abstract:

Existing sparse attention and KV cache compression methods for long-context LLM inference typically apply fixed sparsity patterns or uniform budgets across all attention heads, overlooking the substantial variation in attention behavior among heads and contexts. We observe two distinct entropy patterns among attention heads: Rigid Heads, whose entropy stays near zero across input segments, and Dynamic Heads, whose entropy fluctuates significantly. Crucially, the distribution of these types is context-dependent and cannot be predetermined offline. We therefore propose EntropyInfer, a training-free framework that uses attention entropy to adaptively allocate compute at the granularity of individual heads and segments during prefilling. For decoding, we introduce a latent KV cache compression scheme that leverages generated output tokens, rather than prefill tokens alone, to identify and retain the most critical cache entries. Extensive experiments on Llama, Qwen and openPangu model series show that EntropyInfer consistently outperforms baselines including SnapKV, AdaKV, and CritiPrefill, achieving up to 2.39$\times$ end-to-end speedup beyond 100k tokens with minimal quality degradation compared to full attention. The code is released in this https URL .

11. Deterministic Integrity Gates for LLM-Assisted Clinical Manuscript Preparation: An Auditable Biomedical Informatics Architecture

Authors: Yoojin Nam , Jinhoon Jeong , Namkug Kim
URL: https://arxiv.org/abs/2606.09500
Abstract:

Objective. Large language models (LLMs) increasingly draft clinical research manuscripts, but their fluency can hide fabricated citations, numbers that drift from source tables, and unmet reporting-guideline items. Existing tools generate text without verifying it, and self-critique inherits the blind spots that produce confident fabrication. We describe an architecture that pairs generation with verification. Methods. The design rests on three principles: decompose the workflow into self-contained skills, gate every stage transition with halt-on-failure, and resolve each integrity question with the cheapest sufficient mechanism – a deterministic, re-executable check where one suffices, and a prose-level probe only where interpretation is unavoidable. This determinism-where-possible split, organized as an integrity-gate taxonomy, is the core contribution. It is realized as MedSci Skills, an open-source toolkit of 43 skills coordinated by one orchestrator, whose deterministic tier comprises 21 standard-library detectors. We evaluate it on three reproducible public-dataset pipelines (STARD, PRISMA, STROBE) and a seeded-defect ablation. Results. Across the three pipelines every content-hash manifest verified clean and the gates surfaced real defects. On 27 identical injected defects the deterministic gates detected all 27 with no false positives on the matched clean fixtures, whereas a generic single-prompt LLM reviewer detected 11, its misses concentrated in generated-code, bibliography-internal, and style defects the prose does not expose. Conclusion. Determinism-where-possible verification yields an auditable, re-executable trail that exposes the evidence a human needs to check an LLM-assisted manuscript – feasibility and reproducibility evidence, not a claim of human-competitive quality, which a separate blinded study addresses. MedSci Skills is MIT-licensed and archived (v3.8.0).

12. LLM-Orchestrated Conformance Checking in Stroke Care Without Computer-Interpretable Guidelines

Authors: Giorgio Leonardi , Stefania Montani , Manuel Striani , Alessandro Canessa , Delfina Ferrandi
URL: https://arxiv.org/abs/2606.09489
Abstract:

Objective: Conformance checking in healthcare seeks to assess whether patient care pathways adhere to clinical guidelines. However, its practical application often depends on the availability of formal, machine-interpretable representations of guidelines, such as Computer-Interpretable Guidelines (CIGs), which are seldom available in real-world clinical settings. Methods: This work introduces a modular framework based on the orchestration of Large Language Models (LLMs) to support medical conformance checking directly from unstructured clinical and guideline texts, without requiring predefined CIGs. The proposed architecture integrates multiple LLMs and supporting components to extract patient traces from clinical discharge letters, identify normative rules from textual clinical guidelines, translate these rules into executable scripts, and compute a Trace Conformance Indicator to quantify compliance within the event log. Results: The framework was implemented and evaluated in the stroke care domain at the neurological ward of Alessandria Hospital. Hundreds of patient traces were automatically extracted from hospital data and assessed against 50 rules derived from the reference guideline. The analysis showed that more than 86\% of the available traces were conformant. Conclusion: The results demonstrate the feasibility of using orchestrated LLMs for practical healthcare conformance analysis. At the same time, the study provides evidence of a high level of adherence to stroke care guidelines at Alessandria Hospital.

13. AliyunConsoleAgent: Training Web Agents in Real-World Cloud Environments via Distillation and Reinforcement Learning

Authors: Bojie Rong , Zheyu Shen , Qiaoping Wang , Pengfei Kang , Yang Xu , Yawen Wei , Hanyu Wu , Zhi Zhao , Leihao Pei , Linquan Jiang
URL: https://arxiv.org/abs/2606.09447
Abstract:

We present AliyunConsoleAgent, a web agent framework for automated documentation verification in real-world cloud consoles. Major cloud platforms encompass hundreds of products with rapid feature iteration, causing console UIs to frequently diverge from their corresponding documentation. Verifying that documented procedures accurately reflect the current console and can be executed end-to-end demands an estimated 4 million recurring inspections annually, yet manual coverage remains below 1%. While agent systems built on frontier proprietary models achieve high success rates, their prohibitive cost and data privacy constraints preclude large-scale deployment. We propose a two-stage training paradigm: supervised fine-tuning (SFT) on distilled frontier-model trajectories, followed by reinforcement learning using Group Relative Policy Optimization (GRPO) and a dual-channel outcome reward model in real cloud environments. To support large-scale RL training, we construct a high-determinism rollout system featuring Terraform-based resource pre-provisioning and LLM-driven on-demand provisioning, which effectively isolates environment noise from the training signal. We further introduce a rule-based reward evaluation protocol grounded in backend audit logs, providing objective, reward-hacking-resistant outcome judgment. Our model evolves from mechanical instruction following to autonomous decision-making with cloud console and product-specific understanding. Experiments on a challenging 278-task benchmark where the best frontier model achieves only 65.34% demonstrate that AliyunConsoleAgent-32B achieves a 63.52% mean success rate – a 20.24 percentage-point improvement over the base model, narrowing the gap to the best frontier proprietary model to 1.82 pp (bootstrap 95% CI [-1.27, 7.39]) – at 92% lower inference cost.

14. SIFT: Selective-Index For Fast Compute of RAG Prefill by Exploiting Attention Invariance

Authors: Rya Sanovar , Srikant Bharadwaj , Hritvik Taneja , Moinuddin Qureshi
URL: https://arxiv.org/abs/2606.09441
Abstract:

Retrieval-Augmented Generation (RAG) injects LLM queries with relevant documents to improve response quality. This injection increases prompt length and slows time to first token (TTFT). Unlike standard queries, RAG queries have a unique property of context reuse where the same documents recur across user queries. Thus, fully recomputing documents for every RAG query does redundant compute and increases TTFT. Prior works precompute KV tensors of RAG documents offline and coarsely recompute some tokens during online prefill. However, such KV reuse is often slower than full recomputation on modern GPUs due to high-latency disk transfers. Further, such a coarse-grained recomputation degrades accuracy. To address these limitations, this paper proposes SIFT: Selective-Index For Fast Compute of RAG Prefill by Exploiting Attention Invariance. SIFT processes documents offline and extracts fine-grained locations of high attention scores for each document. Next, we identify the following attention invariance insights that enable us to exploit the extracted locations during runtime: (1) Local-Attention Invariance: The location of high attention scores within a document remain invariant to surrounding documents. This helps us predict the location of high scores where the document attends to itself. (2) Cross-Attention Consistency: Keys with high intra-document attention also attract cross-attention from subsequent documents. This helps us predict the location of high scores where the document attends to future documents. Critically, SIFT stores no KV data and only stores locations of high scores in the form of two compact bit vectors. SIFT’s storage is up to 24,000x smaller than KV tensors, obviating costly disk transfers. During prefill, SIFT computes the attention only for the marked locations and improves TTFT by 1.71x while holding accuracy within 1% of full recompute.

Authors: Radeen Mostafa , Sawradip Saha
URL: https://arxiv.org/abs/2606.09399
Abstract:

We present SUPERBROWSER, an autonomous web-navigation agent designed against a single guiding hypothesis: a web agent should browse the way a person browses. A human reading a page does not retain every pixel they have seen; they look at a few candidate targets, decide on one, and remember only what is needed to keep the goal alive. We operationalize this perception-cognition-action triad as three coupled mechanisms. First, a vision-first bounding-box pipeline labels candidate interactive regions on every screenshot and feeds them, asynchronously prefetched, to the language model so that the “eye” precedes the “hand”. Second, a three-role brain – an Orchestrator that classifies and routes, a Planner that evaluates progress every few steps, and a Worker that emits per-step actions – separates strategic from operational reasoning. Third, a structured Ledger stores only what a person would: the goal, the last three actions, a small set of facts and dead-ends, and a handful of checkpoints; a six-phase eviction loop systematically discards stale screenshots, state blobs, and reasoning traces from the live context. Action execution is a three-tier click cascade (Chrome DevTools Protocol to Puppeteer to scripted) with humanized Bezier motion, plus a chevron-aware bounding-box snapper that resolves the “small arrow beside a large label” ambiguity. On the Mind2Web Hard benchmark (66 tasks), SUPERBROWSER attains 89.47% success, placing third overall and ahead of every published open/research browser-agent baseline by a large margin. We argue that the gain comes not from any single trick but from the consistent application of a cognitive contract throughout the system.

16. Capability-Aligned Hierarchical Learning for Tool-Augmented LLMs

Authors: Haotong Yang , Ting Long , Yi Chang
URL: https://arxiv.org/abs/2606.09371
Abstract:

Tool learning enables LLMs to invoke external tools to accomplish tasks. Prior studies have demonstrated the effectiveness of a hierarchical structure: a high-level policy handles global planning and decomposes tasks into manageable sub-tasks, and a low-level policy focuses on invoking tools to solve these sub-tasks. However, these works typically optimize the high-level and low-level policies separately, leading to planner-executor misalignment and limiting LLM performance on tool-use tasks. In this paper, we propose a method called Capability-Aligned Hierarchical Learning (CAHL), which leverages RLVR to jointly optimize both policies, enabling better alignment between the high-level planner and the low-level executor. Experiments on constrained tool-use benchmarks (API-Bank and BFCL) and an open-ended environment (Bamboogle) demonstrate the effectiveness of CAHL.

Authors: Yongrui Liu , Deyi Xiong
URL: https://arxiv.org/abs/2606.09198
Abstract:

Deep Research agents powered by Large Language Models (LLMs) have exhibited extraordinary potential in automated paper writing tasks. However, existing systems rely heavily on literature retrieval and synthesis through internet and local knowledge bases, often resulting research in lacking insight and creativity in social science. To address this issue, we propose “Memory-Augmented Social Simulation (MASS)”, an innovative paradigm that leverages highly realistic and research-oriented social simulations to enhance the creativity and empirical founding of LLMs-generated research. Specifically, MASS integrates three core components: dynamic goal-path planning with multi-level social norm restraint to guide the simulation, a multi-disciplinary behavior dataset for agent memory cold-start, and a structured forgetting mechanism inspired by the Ebbinghaus curve. Together, these ensure simulation authenticity and provide a robust empirical foundation for generating innovative scholarly papers. Experimental results demonstrate the effectiveness of our method, showing a 6.81\% improvement in generation overall quality over foundation LLMs and 17.19\% gain in Insight over strong baselines.

18. Vision Language Model Helps Private Information De-Identification in Vision Data

Authors: Tiejin Chen , Pingzhi Li , Kaixiong Zhou , Tianlong Chen , Hua Wei
URL: https://arxiv.org/abs/2606.09132
Abstract:

Visual Language Models (VLMs) have gained significant popularity due to their remarkable ability. While various methods exist to enhance privacy in text-based applications, privacy risks associated with visual inputs remain largely overlooked such as Protected Health Information (PHI) in medical images. To tackle this problem, two key tasks: accurately localizing sensitive text and processing it to ensure privacy protection should be performed. To address this issue, we introduce VisShield (Vision Privacy Shield), an end-to-end framework designed to enhance the privacy awareness of VLMs. Our framework consists of two key components: a specialized instruction-tuning dataset OPTIC (Optical Privacy Text Instruction Collection) and a tailored training methodology. The dataset provides diverse privacy-oriented prompts that guide VLMs to perform targeted Optical Character Recognition (OCR) for precise localization of sensitive text, while the training strategy ensures effective adaptation of VLMs to privacy-preserving tasks. Specifically, our approach ensures that VLMs recognize privacy-sensitive text and output precise bounding boxes for detected entities, allowing for effective masking of sensitive information. Extensive experiments demonstrate that our framework significantly outperforms existing approaches in handling private information, paving the way for privacy-preserving applications in vision-language models. Our dataset and code can be found here.

19. Late-Layer Fusion is Enough: Dual-Path Vision Token Routing for Multimodal Large Language Models under Visual Saturation

Authors: Siyuan Liu , Jinyang Wu
URL: https://arxiv.org/abs/2606.09131
Abstract:

Multimodal large language models (MLLMs) commonly inherit the deep, symmetric Transformer backbone designed for unimodal text modeling, and apply the same computation uniformly to image and language tokens. This design overlooks a key modality asymmetry: image and text tokens differ substantially in information density, redundancy, and required reasoning depth. Through a layer-wise analysis of LLaVA-1.5, we observe that vision tokens tend to saturate in the middle layers. Specifically, text-to-image attention decreases from 0.68 at layer 0 to 0.07 by layer 4, and stabilizes near 0.04 after layer 18, whereas text tokens continue to benefit from deep semantic processing. These findings suggest a mismatch between architectural symmetry and depth-asynchronous modality evolution, resulting in redundant visual computation and possible drift in perceptual representations during deep task-specific adaptation. Motivated by this, we propose Dual-Path Vision Token Routing (DPVR), a modality-asymmetric routing framework for efficient MLLMs. Its core instantiation, DPVR-LF (Late-Layer Fusion), routes vision tokens at the saturation point into a one-layer trainable side branch, runs a thirteen-layer text-only forward that skips image positions in the deep stack, and re-fuses the visual and textual streams only at the final layer. With approximately 3% trainable parameters, DPVR-LF preserves competitive multimodal performance on standard benchmarks while reducing visual computation in the deep Transformer stack. The results challenge the conventional assumption that vision tokens must traverse all deep language-model layers, and indicate that a single late fusion layer can be sufficient for maintaining strong perceptual competence in LLaVA-style MLLMs.

20. A Regret Minimization Framework on Preference Learning in Large Language Models

Authors: Suhwan Kim , Taehyun Cho , Geon-Hyeong Kim , Yu Jin Kim , Youngsoo Jang , Moontae Lee , Jungwoo Lee
URL: https://arxiv.org/abs/2606.09124
Abstract:

Reinforcement learning with verifiable rewards (RLVR) has enabled progress on reasoning-intensive tasks by relying on task-specific verifiers that provide automated correctness signals. However, many realistic language tasks are difficult to equip with reliable verifiers, motivating a growing reliance on reinforcement learning from human feedback (RLHF). In this setting, we argue that a closer examination of how human feedback should be interpreted is essential. We introduce Regret-based Preference Optimization $(\textbf{RePO})$, which reframes RLHF through $\textit{regret minimization}$ rather than reward maximization. Human preferences are often shaped by $\textit{prospective}$ anticipation of outcomes and $\textit{counterfactual}$ comparisons to alternative behaviors, rather than by immediate, outcome-independent utility. $\textbf{RePO}$ captures this structure by modeling preferences as behavior-conditioned assessments of relative suboptimality. Experiments on mathematical reasoning benchmarks and human preference datasets demonstrate consistent performance gains, indicating that $\textbf{RePO}$ is an effective and human-aligned approach for training large language models.

21. ComplexConstraints and Beyond: Expert Rubrics for RLVR

Authors: Sushant Mehta , Liudas Panavas , Edwin Chen
URL: https://arxiv.org/abs/2606.09118
Abstract:

As LLM capabilities advance rapidly, the evaluation methods used to assess them increasingly lag behind. Traditional benchmarks relied on programmatic verification of narrow, surface-level constraints, but real-world instruction following and agentic tasks demand assessment of nuanced, context-dependent behaviors that resist simple scripted checks. We present a systematic analysis of expert-curated rubric-based evaluation as an alternative paradigm, drawing on empirical evidence from two domains: complex instruction following and enterprise agentic tasks. We first articulate five design principles for constructing high-quality rubrics, including Maximum Viable Atomicity, intent-aware criterion design, and iterative LLM-judge calibration. To validate these principles, we introduce ComplexConstraints, a new expert-curated instruction-following dataset in which each prompt is paired with 10-40 atomic rubric criteria. We demonstrate that these expert rubrics are not only better evaluation instruments but also highly effective training signals: training on approximately 1,000 ComplexConstraints examples yields +15.5% improvement for a 4B-parameter model and +12.2% for a 235B-parameter model on instruction following, while single-epoch RL training on a rubric-graded enterprise environment produces gains that transfer to out-of-distribution benchmarks the model was never trained on (+4.5% BFCL, +7.4% Tau2-Bench, +6.8% Tool-Decathlon). Our findings establish that expert-authored rubrics improve both the measurement and the development of frontier LLM capabilities, serving as effective evaluation and RL training signals.

22. Graph2Idea:Retrieval-Augmented Scientific Idea Generation with Graph-Structured Contexts

Authors: Xu Li , Hanzhe Tu , Xun Han
URL: https://arxiv.org/abs/2606.09105
Abstract:

Generating novel, feasible, and high-quality research ideas is an important yet challenging task in scientific this http URL Large Language Model (LLM)-based methods often ground idea generation with retrieved literature, but the retrieved evidence is usually provided as flat text, such as titles, abstracts, or summaries. Such flat contexts may contain redundant or weakly relevant information, while making cross-paper relations among problems, methods, mechanisms, and findings difficult to identify and this http URL address this challenge, we propose Graph2Idea, a knowledge graph-guided framework for retrieval-augmented scientific idea generation.Graph2Idea first retrieves papers according to the input topic, transforms them into structured knowledge triples, and dynamically constructs a target-centered knowledge graph to make literature relations this http URL then extracts compact graph-derived contexts that retain target-relevant relational evidence while reducing noisy textual this http URL on these contexts, a two-stage generation process first identifies promising research directions and then guides the LLM to synthesize candidate ideas from graph-grounded this http URL on a scientific idea generation benchmark show that Graph2Idea outperforms representative baselines under the automatic evaluation this http URL with the strongest baseline scores, it improves Novelty from 0.45 to 0.52, Quality from 0.24 to 0.29, and Feasibility from 0.22 to this http URL results suggest that graph-structured evidence helps LLMs generate research ideas through more explicit, compact, and traceable recombination of prior scientific knowledge.

23. REFLECT: Intervention-Supported Error Attribution for Silent Failures in LLM Agent Traces

Authors: Xiaofeng Lin , Yingxu Wang , Tung Sum Thomas Kwok , Daniel Guo , Sahil Arun Nale , Charles Fleming , Guang Cheng
URL: https://arxiv.org/abs/2606.09071
Abstract:

Large language model (LLM) agents now solve complex tasks through long plan-and-execution traces, yet the ability to locate errors in a completed traces still lags far behind, especially in the \emph{silent failure} regime. Existing approaches predict suspect steps via classifiers or LLM judges, or recover correct answers via retry, but none feed the intervention outcome back to \emph{refine the attribution itself}. We propose \methodname, a method that closes this gap by diagnosing a candidate error step, testing it through controlled replay with a diagnosis-specific patch, and using the verified outcome flip as contrastive evidence to refine the final attribution. Across four localization benchmarks spanning multi-hop reasoning across domains, \methodname achieves the highest localization accuracy among same-auditor methods across all four benchmarks, with the largest gains on structured tool-use traces, while providing actionable localization even when ground-truth answers are unavailable.

24. Personalization Meets Safety:Mechanisms,Risks,and Mitigations in Personalized LLMs

Authors: Yanyan Luo , Xue Han , Ruiqiao Bai , Xin Huang , Yitong Wang , Qian Hu , Qing Wang , Chunxu Zhao , Jie Liu , Cong Geng , Lehao Xing , Pengwei Hu , Junlan Feng
URL: https://arxiv.org/abs/2606.09038
Abstract:

Large Language Models (LLMs) have enabled increasingly personalized interactions by adapting to users’ preferences, contexts, and long-term histories. However, the mechanisms that enable personalization also expand the safety landscape in ways not systematically addressed by existing literature. Existing reviews typically focus either on personalization or safety, leaving their intersection largely unexplored. We present the first comprehensive, safety-aware review of personalized LLMs. We organize personalization along three dimensions-user representation, personalization paradigm, and evaluation-and introduce a unified taxonomy of safety risks. At the representation level, we analyze risks arising from diverse user representations. Across mainstream personalization paradigms, we delineate vulnerabilities inherent to prompting, retrieval augmentation, parameter fine-tuning, reinforcement learning, Mixture-of-Experts (MoE), pruning, agent frameworks, and multimodal personalization, and synthesize mitigation strategies across the model lifecycle. Beyond these fine-grained risks, we characterize paradigm-agnostic safety risks arising from personalized adaptation. We further summarize personalized datasets and evaluation methodologies. Through a case study of OpenClaw, we analyze deployment trends in personalized agent ecosystems. Our analysis reveals three structural inadequacies in existing research: safety is evaluated as user-invariant rather than relational, personalization techniques are analyzed in isolation rather than in composition, and evaluation frameworks cannot capture emergent long-term risks. By jointly examining personalized representations, personalization paradigms, safety risks, defenses, and evaluation methods, we provide a unified framework for developing safe personalized LLMs and highlight key directions for future research.

25. A Multi-Agent System for IPMSM Design Optimization via an FEA-AI Hybrid Approach

Authors: Jinseong Han , Sunwoong Yang , Namwoo Kang
URL: https://arxiv.org/abs/2606.09037
Abstract:

Interior permanent magnet synchronous motor (IPMSM) design requires balancing conflicting objectives and multi-physics constraints, while modern optimization workflows face three bottlenecks: manual problem setup, high finite element analysis (FEA) cost, and unreliable surrogate-based search in sparse or out-of-distribution regions. To address these limitations, we propose an end-to-end automated IPMSM design optimization framework that integrates retrieval-augmented generation (RAG) for structured problem definition with an uncertainty-aware FEA-AI hybrid optimization pipeline. A Design agent, connected to a motor textbook through RAG, provides domain-knowledge-based options and engineering tips, and compiles an optimization card and a design-of-experiments plan for AI-model training. A Training agent automates electromagnetic FEA, records geometry-validation and solver-failure logs, analyzes failed geometries using ANOVA-based data analysis and LLM reasoning, and invokes a Design Sampling agent to redefine the design space and generate additional samples. An Optimization agent performs GA-based search with uncertainty-driven switching: low-uncertainty candidates are evaluated by AI-surrogate inference, whereas high-uncertainty and reliability-critical Pareto-front or top-K candidates are corrected by high-fidelity FEA and reused for iterative retraining. The framework converts manual, experience-dependent configuration into a reproducible workflow that balances computational cost and prediction reliability. Experimental results under a matched high-fidelity FEA budget show that the proposed hybrid approach achieves better objective performance while maintaining low and further reducible predictive uncertainty, outperforming FEA-only search, which is limited by early budget exhaustion, and AI-only search, which converges to a low-confidence optimum.

26. LATTEArena: An Evaluation Framework for LLM-powered Tabular Feature Engineering (Extended Version)

Authors: Ankai Hao , Ke Chen , Huan Li , Lidan Shou
URL: https://arxiv.org/abs/2606.09004
Abstract:

Feature engineering remains essential for tabular data analysis, and Large Language Models (LLMs) have emerged as a promising paradigm for automating this process, giving rise to LLM-powered AuTomated Tabular feature Engineering (LATTE). However, the absence of standardized platforms prevents fair, cost-aware comparisons. Furthermore, complex methodological designs obscure the specific contributions of individual components; for example, although LFG integrates Tree-of-Thought, few-shot demonstrations, Monte Carlo Tree Search, and natural language generation, the isolated impact of each technique’s competitive edge remains unquantified. To address these challenges, we introduce LATTEArena, the first competitive evaluation framework featuring: (1) a six-dimensional taxonomy decomposing 15 representative methods into reusable components; (2) a standardized modular arena for controlled comparison; (3) multi-dimensional assessments covering performance, cost, and robustness; and (4) component-level ablation quantifying each technique’s competitive edge. Through extensive evaluations, we reveal 16 key findings, including: (1) Tree-of-Thought with Monte Carlo Tree Search achieves optimal cost-effectiveness; (2) RPN and Code output formats dominate classification and regression tasks, respectively. We publicly release the modular framework and over 4000 execution logs, enabling researchers to seamlessly pit new techniques against existing ones and advance LATTE.

27. RTL-BenchLS: A Large-Scale Benchmark for RTL Reasoning and Generation with Large Language Models

Authors: Jing Wang , Shang Liu , Wenji Fang , Yuchao Wu , Yugao Zhu , Zhiyao Xie
URL: https://arxiv.org/abs/2606.08976
Abstract:

LLM-based RTL generation and reasoning is a promising direction for hardware design automation. High-quality benchmarks are critical infrastructure for tracking progress in this direction. However, existing RTL benchmarks face inherent limitations in both scale and task scope. The designs they cover are typically small and simple, and the tasks focus almost entirely on specification-to-RTL generation. Frontier models’ performance already saturates on the existing benchmarks. Scaling these benchmarks up is fundamentally difficult because aligned labels are required for benchmarking, such as specifications and testbenches. Such aligned high-quality data are rarely available for real-world designs. We introduce RTL-BenchLS, a large-scale benchmark addressing both limitations above. It contains over 10,000 formally verified Verilog designs, covering substantially larger and more complex designs than existing benchmarks. Beyond specification-to-RTL generation, we propose three novel tasks that jointly evaluate reasoning and generation: round-trip reasoning, masked-content reasoning, and repository-issue reasoning. The first two are self-supervised, which directly resolves the scaling bottleneck. All tasks are verified through formal equivalence checking without any manual testbenches. We evaluate eight LLMs on RTL-BenchLS. Even the best model reaches only 23% on natural-language round-trip reasoning, 28% on masked-content reasoning, and 12% on repository-issue fixing. RTL-BenchLS is substantially more challenging than existing benchmarks. It leaves ample room for future improvement and offers guidance for developing LLM-based methods for hardware design.

28. Diverse Thinking Schemata Elicit Better Reasoning in Large Language Models

Authors: Xinyue Liang , Yizhe Yang , Yu Bai , Bin Xu , Jiawei Li , Yang Gao
URL: https://arxiv.org/abs/2606.08974
Abstract:

Large reasoning models (LRMs) have attracted increasing attention for their ability to solve complex mathematical problems by generating extended reasoning chains. In this work, we focus on two critical yet underexplored aspects of the reasoning process: reasoning transitions capturing the distinct transitions between reasoning steps and answer candidates reflecting the variety of solution paths produced by the model. We collectively define these two aspects as thinking schemata. We observe a correlation between the diversity of thinking schemata and model performance, which motivates us to enhance diversity as a means to further improve reasoning potential. To this end, we propose Diverse Schemata Policy Optimization (DiScO), a framework that first endows the model with schemata awareness, then encourages diversity through reinforcement learning, and further promotes diverse reasoning at inference time. Experiments on multiple mathematical reasoning benchmarks demonstrate that DiScO consistently outperforms standard group relative policy optimization. Beyond accuracy, human-annotated analyses show that DiScO substantially improves the model’s ability to recover from erroneous initial attempts. Overall, our work suggests the important role that diversity of the thinking schemata plays and points to scaling along the diversity dimension as a promising research direction.

29. An Effective Router for Vision-Language Model Selection

Authors: Can Wang , Shengwei Wang , Bolin Zhang , Zhiying Tu , Dianhui Chu
URL: https://arxiv.org/abs/2606.08970
Abstract:

Vision-language models (VLMs) with varying performance and resource requirements are widely deployed, making it difficult for users to select the most appropriate one among numerous VLM candidates. Existing work reveals the performance paradox phenomenon in language models and focuses on routing methods to solve it. However, developing a router for VLM selection is still a critical yet challenging problem, which primarily faces: 1) lack of specialized data, 2) ineffective feature representation, and 3) rigid model space and costly adaptation. In this paper, we construct a multimodal dataset for VLM selection, containing the outputs of seven mainstream VLMs on 32,626 unique image-text queries. We then propose ARMS, a router for VLM selection. ARMS enhances input signals with VLM profiles, employs a simple but effective architecture to improve representations of queries and VLM capabilities. To improve ARMS’ adaptation to new VLMs, we propose two extension training strategies: incremental training and independent training. Experimental results on both in-distribution and out-of-distribution test sets demonstrate the effectiveness of ARMS. In particular, using our training strategy, ARMs (only 800M in size) can adapt to a broader VLM space and defeat commercial models like GPT-4o that are hundreds of times larger in scale. Our code, models, and datasets are available in the anonymous repository.

30. Oversight Has a Capacity: Calibrating Agent Guards to a Subjective, Fatiguing Human

Authors: Emre Turan
URL: https://arxiv.org/abs/2606.08919
Abstract:

As LLM agents begin to take real, irreversible actions (shell commands, file edits, deploys), the standard safety pattern is a human-in-the-loop approval gate: risky actions pause and wait for a person. We argue the gate is the easy part; the hard part is the judgment - which actions to stop - which the field evaluates against two false assumptions: that there is a ground-truth notion of “risky,” and that the human reviewer is a perfect, infinitely-available oracle. On a hand-labeled set of 125 adversarially-weighted agent actions we show that (i) reviewers only moderately agree on what is risky (Fleiss’ kappa = 0.52), so there is no single correct label; (ii) framing the guard as selective classification under asymmetric cost makes its operating limits measurable, and on hard inputs the guard cannot safely auto-decide; and (iii) when the reviewer is modeled as endogenous (fatiguing as escalation load grows), realized safety becomes an inverted-U in the escalation rate: more human oversight can make a system less safe, and the safety-optimal guard escalates below full escalation - a setting a load-aware policy also uses to resist a flooding attack that slips a malicious action past a fatigued reviewer. Agent oversight, framed this way, is not only a classification problem but a resource-allocation one: human attention is finite, and the guard’s escalation policy spends it. We claim none of these mechanisms as novel - fatigue-aware learning-to-defer (FALCON), cost-sensitive deferral under workload constraints (DeCCaF), trajectory-level guarding, and reviewer-fatigue/flooding attacks are all prior art we cite. Our contribution is an open-source agent-oversight system that operationalizes and measures them in the LLM-agent action-gating setting, turning “is my guard good?” from a guess into a curve. The inverted-U and the flooding attack are modeling results that motivate a human study.

31. Order Matters: Unveiling the Hidden Impact of Macro Placement Sequences via Proxy-Guided LLM Evolution

Authors: Shibing Mo , Jing Liu , Jianchu Xu , Ruilin Wu
URL: https://arxiv.org/abs/2606.08904
Abstract:

Macro placement is a fundamental step in modern chip physical design, playing a crucial role in determining the solution quality of high-dimensional combinatorial optimization problems. Despite recent advancements in machine learning for spatial coordinate determination, the temporal dimension of placement sequencing remains largely governed by static heuristics. In this work, we demonstrate that the placement sequence is not merely a preprocessing step but a decisive factor in optimization, where suboptimal early decisions trigger irreversible domino effects that constrain the solution space. To harness this unexplored dimension, we propose \textbf{OrderPlace}, a proxy-guided LLM evolution framework for automatically discovering macro placement order strategies. Instead of relying on manually crafted heuristics such as area- or connectivity-based ordering, OrderPlace explores a broader space of code-level policies, ranging from static scoring metrics to dynamic physics-inspired mechanisms. To mitigate the prohibitive cost of evaluating sequences, we introduce a lightweight proxy evaluation mechanism that efficiently filters candidates using a deterministic greedy probe. Experimental results on the standard ISPD 2005 benchmarks demonstrate that OrderPlace discovers novel ordering strategies. Compared with WireMask-EA and the state-of-the-art method EGPlace, OrderPlace reduces wirelength by 34.04\% and 14.08\%, respectively.

32. Can the Environment Speak for Itself? $T^{2}$-GRPO: A Turn-Trajectory Group Relative Policy Optimization for Caregiver Agents

Authors: Yutong Song , Jiang Wu , Pengfei Zhang , Wenjun Huang , Honghui Xu , Nikil Dutt , Amir M. Rahmani
URL: https://arxiv.org/abs/2606.08875
Abstract:

Optimizing large language models (LLMs) for long-horizon caregiver agents requires balancing delayed task objectives with immediate environment dynamics, such as patient distress and resistance. In dementia care, this balance is especially difficult: trajectory level rewards are too sparse for turn level credit assignment, while external LLM-based evaluators are costly and can misread fragmented or indirect patient responses. To address this issue, we propose \textbf{T}urn-\textbf{T}rajectory \textbf{G}roup \textbf{R}elative \textbf{P}olicy \textbf{O}ptimization (\textbf{T$^{2}$-GRPO}), a framework that decouples caregiver RL into two normalized reward horizons and enforces safety through a binary hard veto. $T^2$-GRPO derives dense turn-level rewards directly from environment state transitions, measuring changes in patient distress and resistance from a frozen dementia patient simulator. These environment-grounded rewards are combined with trajectory-level evaluations through independent centered-rank normalization, which preserves heterogeneous reward signals and mitigates reward collapse. Extensive experiments on dementia caregivers show that T $^{2}$-GRPO outperforms competitive baselines, indicating a substantial improvement for emotionally sensitive caregiver scenarios that effectively handles immediate patient feedback, long-term care outcomes, and safety constraints.

33. Hybrid E-Assessment in Higher Education: Semi-Automated Grading of Paper-Based Written Examinations

Authors: Hartwig Grabowski , Michael Canz
URL: https://arxiv.org/abs/2606.08855
Abstract:

This paper examines the limitations of fully digital and partially digital e-assessment approaches in summative examinations in higher education. The analysis focuses on the didactic narrowing caused by closed question formats and on organizational, technical, and legal constraints that become particularly relevant in large student cohorts. As an alternative, the paper proposes a hybrid e-assessment approach that retains paper-based, problem-oriented examination tasks while enabling semi-automated grading. Assessment-relevant intermediate results are encoded in a structured answer format, entered by students by hand, and subsequently captured from table fields. The central technical bottleneck is reliable recognition of handwritten characters under realistic examination conditions. Recent vision-capable large language models, combined with a two-pass validation principle and comparison against a solution key, can reduce misclassifications and thereby improve the validity, fairness, and scalability of summative assessment.

34. ZIPP:Zero-shot Image Personalization from Personas

Authors: Harini SI , Somesh Singh , Yaman Kumar Singla , David Doermann , Rajiv Ratn Shah
URL: https://arxiv.org/abs/2606.08841
Abstract:

Text-to-image diffusion models are increasingly deployed in open-ended creative contexts, yet their outputs remain impersonal, optimized for aggregate aesthetics rather than individual taste. Human preferences are pluralistic: one user favoring muted, nostalgic portraits may prefer vibrant street photography, while another gravitates toward dreamy film aesthetics. Existing methods require dense interaction histories or per-user fine-tuning, failing in cold-start settings and collapsing context-dependent preferences into a static representation. We introduce zero-shot image personalization from personas (ZIPP), which conditions image generation on natural-language personas (concise descriptors of a user’s identity and aesthetic sensibilities) without any user-specific data or weight updates. ZIPP uses an LLM to rewrite prompts from the perspective of a given persona, steering diffusion models toward personalized outputs. To mine personas at scale, we train an inductive Graph Attention Network over a 22M-user Reddit interaction graph with dual contrastive objectives aligning graph structure with visual behavior, then verbalize learned representations into natural-language personas via an MLLM. We introduce ZIPBench, the first zero-shot personalization benchmark with 1.5K users, graph-mined personas, and 40K generated images. Across four benchmarks and 14 LLMs spanning five model families, persona conditioning yields consistent gains (13-20%), with frontier models benefiting most. In the few-shot setting, ZIPP matches or exceeds fine-tuned baselines trained on 100+ examples per user. ZIPP achieves the lowest preference distributional divergence (CMMD 0.16 vs. 0.55), and IPF-normalized demographic evaluation shows it substantially reduces subpopulation bias present in existing methods. Human evaluation confirms a 79% win rate over generic generation and 58-65% over all fine-tuned baselines.

35. Inference-Time Conformal Reasoning with Valid Factuality Control for Large Language Models

Authors: Ting Wang , Yuanjie Shi , Yan Yan , Huan Zhang
URL: https://arxiv.org/abs/2606.08831
Abstract:

Large language models (LLMs) increasingly perform multi-step reasoning, where intermediate claims form implicit directed acyclic graphs whose node correctness is structurally conditioned on their ancestors. This makes factuality uncertainty structural, rather than a trivial accumulation of node-wise errors, and necessitates inference-time uncertainty quantification over the reasoning structure. While conformal prediction (CP) offers flexible user-specified factuality control, existing work remains post-hoc and cannot intervene during generation. To fill the gap between CP’s flexibility and its post-hoc limitation, we propose an \emph{Inference-Time Conformal Reasoning (ITCR)} framework that integrates CP directly into reasoning graph generation. ITCR learns a structure-level factuality uncertainty function that aggregates claim-level factuality signals over reasoning graphs without complex modeling assumptions. We then design the non-conformity score based on graph-level factuality uncertainty and calibrate the conformal threshold to decide when to stop generation. We theoretically show such generation is nested, yielding valid coverage guarantees for factuality control. Experiments over multiple datasets and coverage objectives demonstrate empirically valid coverage. In downstream reasoning tasks, inference-time calibrated graphs yield more accurate generation than post-hoc pruned graphs.

36. Momentum for Reasoning: Dense Intrinsic Signals in Policy Optimization

Authors: Hao Chen , Zhanming Shen , Liyao Li , Yanyu Chen , Xuhang Zhu , Xiaomeng Hu , Qi Zhang , Ru Peng , Xiaoyu Shen , Haobo Wang , Junbo Zhao
URL: https://arxiv.org/abs/2606.08815
Abstract:

Reinforcement learning with verifiable rewards (RLVR) has emerged as a powerful paradigm for eliciting long-chain reasoning in large language models. However, existing methods based on Group Relative Policy Optimization (GRPO) rely on a binary outcome reward, which induces two structural failure modes: Zero-Advantage Collapse, in which all rollouts in a group share the same outcome and the gradient vanishes, and Hallucinated Certainty, in which the model becomes increasingly confident on incorrect rollouts late in training. We address both modes by densifying the reward with intrinsic signals computed entirely from the policy’s own conditional probabilities, and propose ISPO (Intrinsic Signal Policy Optimization, which combines a sequence-level signal measuring how informative the thinking trajectory is for the final answer, with a token-level directional reward whose hallucinated-certainty hinge penalizes confidently-wrong predictions at critical decision tokens. Across three base models and five mathematical reasoning benchmarks, ISPO consistently outperforms competitive baselines, with the largest gains on the hardest benchmarks where zero-advantage collapse is most frequent, and training-dynamics diagnostics confirm that both failure modes are decreased.

37. Bridging Expert Knowledge and Automated Feature Engineering via Self-Evolution

Authors: Varun Khurana , Vijval Ekbote , Vashu Chauhan , Yaman Kumar Singla , Rajiv Ratn Shah , Balaji Krishnamurthy
URL: https://arxiv.org/abs/2606.08800
Abstract:

In high-stakes settings such as brand compliance, clinical care, and content moderation, machine learning cannot be deployed as opaque oracles: practitioners inspect the features driving model decisions, and models must leverage the expert documentation governing these domains. In practice, the data arrives as unstructured content, and features extracted from it must be interpretable, discriminative, and aligned with what experts consider important. Existing methods fall short: they target tabular inputs, lack demonstrated expert alignment, and cannot operationalize qualitative criteria such as ‘maintain professional tone’ into precise features. We present FEST (Feature Engineering with Self-evolving Trees), combining dual-stream feature generation (semantic and deterministic), semantic deduplication, and tree-guided iterative evolution to discover auditable features from raw text and images. FEST leads in 17 of 20 classifier-task combinations across brand classification, content authenticity detection, and stress detection, with a mean gain of 4.2 pp over the strongest baseline across five classifiers. An LLM-as-judge evaluation shows FEST achieves 60-80% coverage of expert-designed brand features at strict semantic-alignment thresholds, corroborated by a human expert study rating features highly on relevance, clarity, and actionability. When seeded with expert guidelines, FEST refines qualitative criteria into operational features, improving accuracy by 6-12 pp on average across brands. To enable systematic evaluation of expert alignment in automated feature engineering, we release BrandGuide, the first dataset pairing expert-designed features with 1M+ assets across 2,683 brands. By grounding feature engineering in expert knowledge, FEST opens a practical pathway for interpretable ML in domains demanding human oversight.

38. RAILS: Verification-Native Clearing For Agentic Commerce

Authors: Adrian de Valois-Franklin , Alex Bogdan
URL: https://arxiv.org/abs/2606.08790
Abstract:

Autonomous agents negotiate, purchase, deploy code, and move funds, but no neutral mechanism determines whether they met their delegated obligation, who is responsible when they did not, or which settlement action follows. This is the agentic clearing problem. Tool protocols (MCP), inter-agent communication (A2A), payment rails (x402), mandate and network agent protocols (AP2, Visa, Mastercard), and settlement-risk standards each assume that determination and none produce it. Clearing is the missing primitive. Payment is not clearing. Authorization is not clearing. LLM-as-judge evaluation is not clearing. Settlement-risk escrow is not clearing: it consumes clearing decisions. RAILS (Real-Time Agent Integrity & Ledger Settlement) is the integrity and clearing layer for agentic commerce, spanning a per-output reliability score, a published reliability record, and a clearing function that consumes them. The clearing protocol at its core closes that gap. Seven primitives (Obligation Object, Evidence Envelope, Verification Mesh, Clearing Decision, Settlement Instruction, Clearing Passport, Finality Rules), bound by a formal model of admissibility-graded verification, together yield a soundness property: no financially material settlement is supported by evidence below the obligation’s admissibility floor. The property is falsifiable against the spec. We are not aware of a prior agent-commerce verification mechanism that states a property of this kind. The approaches nearest to it emit a pass, a delivery guarantee, a bare score, or an equilibrium. This paper specifies that clearing protocol.

39. Artificial Intelligence for Mathematical Reasoning: An Integrated Survey of Language Models, Neuro-symbolic Systems, and Verified Discovery

Authors: Syed Rifat Raiyan , Mohsinul Kabir , Hasan Mahmud , Md Kamrul Hasan
URL: https://arxiv.org/abs/2606.08728
Abstract:

Mathematical reasoning has long served as a stringent test of machine intelligence; over the past decade, it has moved from a niche problem within NLP to one of the most consequential AI frontiers. This survey provides a unified account of the field’s evolution, from early rule-based math word problem (MWP) solvers and template-driven geometry systems, through neural expression generation and LLM prompting, to contemporary reasoning models, multi-agent systems, neuro-symbolic theorem provers, and verified discovery workflows. We organize the landscape along four axes: (i) informal reasoning over text and diagrams, spanning MWP solving, multimodal geometry, and VLMs; (ii) formal reasoning in proof assistants, including autoformalization, tactic prediction, compiler-guided repair, and proof search; (iii) mathematical discovery, where systems propose constructions, improve bounds, or assist attacks on open problems; and (iv) the inference and training-time techniques, including CoT prompting, tool use, process reward models, and RLVR, that increasingly connect generation with verification. We catalog major benchmarks across grade-school arithmetic, competition mathematics, geometry, formal proving, multimodal and multilingual reasoning, and expert evaluation, and we examine benchmark saturation, contamination, reporting mismatches, and the distinction between pass@1, majority voting, and verifier-assisted pass@$k$. We critically assess failure modes: brittleness under perturbation, reward hacking, multimodal grounding failures, fragile formalization, and the energy cost of reasoning-scale inference. Drawing on recent perspectives from working mathematicians, we identify future directions centered on verified-discovery workflows, reasoning efficiency, and infrastructure to make AI-assisted formalization broadly usable. Companion materials: this https URL .

40. ConMem: Structured Memory-Guided Adaptation in Training-Free Multi-Agent Systems

Authors: Zhixun Tan , Qiang Chen , Tairan Huang , Xiu Su , Yi Chen
URL: https://arxiv.org/abs/2606.08702
Abstract:

Recent advances have improved the adaptive capabilities of LLM-based multi-agent systems (MAS) through memory-, skill-, and learning-based approaches, yet these approaches remain challenged by noisy trajectories, insufficient modeling of memory-skill relations, and reliance on additional training or high-quality supervision. To address these limitations, we propose ConMem, a relation-aware and training-free framework that enables efficient multi-agent adaptation through cross-experience coordination. Specifically, ConMem distills historical interaction trajectories into structured memory cards to capture reusable strategies and cues, organizing them into a relation-aware memory graph. At runtime, ConMem retrieves cards according to task needs and coordinates them through the card graph to resolve strategy conflicts and recover their dependencies. Combined, these modules yield structured and relation-aware guidance, enabling robust, lightweight adaptation in multi-agent systems without additional training. Extensive experiments across multiple benchmarks and mainstream MAS architectures show consistent gains over existing memory architectures, with improved inference-time efficiency through pruning more than 50% of expanded candidates and reducing planning overhead by over 80%. Our codes are available at this https URL

41. Towards Long-Horizon Vessel Trajectory and Destination Forecasting with Reasoning Large Language Models

Authors: Hongwei Wang , Miao Zhou , Fengde Wang , Yuting Wang , Jiewen Yu , Jun-Yan He , Bohao Qu , Wanbing Zhang , Xiuju Fu , Qing Guo , Zipei Fan , Yingying Xing , Yi Yuan
URL: https://arxiv.org/abs/2606.08633
Abstract:

Long-horizon maritime trajectory prediction is important for shipping management, logistics planning, and maritime risk analysis, yet month-level forecasting remains insufficiently studied. Existing deep learning methods mainly focus on short- and mid-term coordinate extrapolation and often struggle to preserve route feasibility and destination correctness over extended horizons. This paper investigates joint long-horizon vessel trajectory and destination forecasting with reasoning-capable large language models, and develops a Maritime LLM post-training framework based on Reinforcement Learning with Verifiable Reward (RLVR). An AIS-based benchmark is constructed with 60-day historical trajectories and 30-day forecasting horizons, where trajectories are converted into semantic textual representations for RL prompt construction. RLVR aligns LLMs with maritime forecasting objectives by enforcing physical validity, providing early-weighted trajectory supervision, and evaluating destination correctness through hierarchical matching and curriculum learning. Experimental results show that RLVR-trained LLMs substantially improve over zero-shot LLMs and representative deep learning baselines, especially on destination-related metrics. Among the evaluated RLVR-trained variants, 4B LLMs achieve the best overall performance, suggesting that reward-compatible optimization and task-specific capacity matching are more important than simply using larger 8B or 14B LLMs. The results also show that LSTM remains a strong deep learning baseline under limited fine-tuning data, while Transformer-style spatio-temporal models typically require larger datasets and richer structured inputs. Overall, this work advances semantic, verifier-aligned maritime forecasting for operational decision support.

42. InA-Probe: Instruction-Aware Active Probing for Time Series Forecasting with LLMs

Authors: Peiliang Gong , Emadeldeen Eldele , Chenyu Liu , Ziyu Jia , Yi Ding , Xinliang Zhou , Lianchao Gu , Qi Zhu , Yang Liu , Daoqiang Zhang , Xiaoli Li
URL: https://arxiv.org/abs/2606.08601
Abstract:

Large Language Models (LLMs) have recently demonstrated impressive potential for time series forecasting. However, existing methods predominantly rely on passive modality alignment or static task reprogramming, which often fail to capture fine-grained, non-stationary temporal patterns or to adapt to nuanced task intents. In this paper, we propose Instruction-aware Active Probing (InA-Probe), which shifts the paradigm from passive alignment toward an active, instruction-driven probing mechanism. Specifically, we design a Multi-Level Instruction Injection mechanism that enriches the model with both global task objectives and fine-grained, patch-level semantic priors. Building on this, an Adaptive Query Generation module produces sample-specific probes that are dynamically modulated by the temporal context. These probes are then refined through a dual-stage attention process: they first internalize task-specific intents via Instruction-Aware Self-Attention, and subsequently interrogate the projected temporal representations through Temporal Cross-Attention to extract salient patterns. Comprehensive experiments on seven real-world benchmarks show that InA-Probe consistently outperforms state-of-the-art deep learning and LLM-based baselines, excelling in both one-for-all generalization and zero-shot transfer while reducing forecasting error by up to 37\% in challenging cross-domain scenarios. Ablation studies further confirm that the synergy between adaptive querying and fine-grained instructions is key to unlocking the reasoning power of LLMs for complex time series.

43. Distilling LLM Reasoning into an Interpretable Policy Tree for Human-AI Collaboration

Authors: Beiwen Zhang , Yongheng Liang , Guowei Zou , Haitao Wang , Hejun Wu
URL: https://arxiv.org/abs/2606.08596
Abstract:

Constructing efficient and reliable policies to assist humans is indispensable for human-AI collaboration. Existing methods mainly follow two lines of work. Most prior work relies on multi-agent reinforcement learning (MARL) to learn black-box policies, which limits interpretability and raises safety concerns. Recent methods query large language models (LLMs) at each decision step, causing slow responses and high inference costs. We propose Collaboration Policy Tree (Co-pi-tree), a closed-loop method that learns an executable policy tree consisting of a partner-behavior prediction tree and an agent-action selection tree. Co-pi-tree constructs a policy by distilling LLM reasoning into policy tree code. It then evaluates the policy through partner interaction, obtains feedback, and uses natural language to summarize the interaction feedback to improve problematic branches. Experiments in Overcooked-AI show that Co-pi-tree improves average reward by 35.4% over the baseline average, while reducing the number of LLM queries by 77.7% and test-time latency by 97.1%. Project page: this https URL

44. PAEC: Position-Aware Entropy Calibration for LLM Reasoning in RLVR

Authors: Shumeng Yang , Yisu Liu , Jiayi Zheng , Zhaohui Yang , Linjing Li
URL: https://arxiv.org/abs/2606.08543
Abstract:

Reinforcement learning with verifiable rewards (RLVR) improves large language model reasoning but often suffers from rapid policy-entropy collapse, where the policy prematurely concentrates on narrow high-probability reasoning paths. While global entropy regularization can encourage exploration, uniformly increasing entropy across all token positions is inefficient for long reasoning trajectories, where many tokens are not decision-relevant. We propose Position-Aware Entropy Calibration (PAEC), a token-level entropy-management framework that constructs a soft mask from local top-p entropy and top-two candidate competition, and applies an anchor-based lower-bound penalty to prevent selected-position entropy collapse. Experiments on five mathematical reasoning benchmarks show that PAEC improves macro-average majority-vote performance over strong RLVR baselines, with clear gains on AIME-style tasks. Our results suggest that entropy management in reasoning RL should be formulated as selective exploration allocation over decision-sensitive positions rather than uniform randomness injection.

45. AgentTrust: A Self-Improving Trust Layer for AI-Agent Actions

Authors: Chenglin Yang
URL: https://arxiv.org/abs/2606.08539
Abstract:

AI agents increasingly take consequential actions – shell commands, cloud operations, and arbitrary tool-calls – so a trust layer must decide, per action, whether to allow, warn, block, or escalate. We argue that the right way to reason about such a layer is by threat type. Lexical (fixed-signature) threats, where danger lives in a stable token, are decidable by deterministic rules; semantic (intent-dependent) threats, where a benign and a malicious action share the same surface, are out of reach for rules by construction. We make this concrete with a negative proof: a determined, hand-authored cloud rule pack lifts held-out accuracy only 48 to 56% overall and moves the semantic categories by 0pp (data_db 29 to 29, observability 59 to 59, supply_chain 50 to 50), while a strong LLM judge carries exactly those categories. We give the judge a self-learning capability: on a corpus that is mainly semantic attacks it nearly doubles rule accuracy (48% to 83.6-85.2%) with near-zero false-blocks, and this holds across two model providers. We turn this into a self-improving dual-store system: the judge distills a growing deterministic rule floor on lexical threats (cheaper over time) and feeds a guarded RAG memory on semantic threats (a verdict-cache fails – surface-twins collapse to ~58% – so a corroboration guard lifts semantic accuracy +13pp, 70 to 84). The result is what sets AgentTrust v2 apart from its static v1 predecessor: a trust layer that self-evolves from its own stream of decisions – cheaper on the lexical class (it distils its own rules) and smarter on the semantic class (it accrues guarded precedent), while never hard-blocking a benign action. An end-to-end online replay shows the judge-call rate falling (50% to 44%) and judge-domain accuracy rising (71% to 80%), with 0 benign hard-blocks across 45,000 actions.

46. DN-Hypo-Pipeline: An AI-Driven Workflow for Hypothesis Generation via Large Language Models and Scientific Explanations

Authors: Lei Lin , Ronghao Wang , Chunbao Zhou , Jue Wang , Yangang Wang
URL: https://arxiv.org/abs/2606.08532
Abstract:

A scientific hypothesis is the first step in research and undergoes experimental validation, yet it also reflects a deep understanding of and reasoning about scientific phenomena. We introduce DN-Hypo-Pipeline, an AI-powered workflow based on large language models, designed to support structured scientific thinking and hypothesis generation by leveraging scientific explanations as prior knowledge. This pipeline assists researchers in deriving novel hypotheses from existing literature. Given the explanandum (i.e., the conclusion) of a research paper, it identifies underlying laws, theories, and principles, and reconstructs a new, yet-to-be-verified explanation for the observed phenomenon. We evaluated DN-Hypo-Pipeline in the field of data science modeling using three highly cited papers. Statistical inference, supported by both LLM-as-judge assessment and human expert evaluation, demonstrates that our pipeline is more effective than direct generation methods. Additionally, we validated the two highest-scoring generated hypotheses by developing corresponding novel algorithms, which outperformed the baseline models presented in the original papers. Beyond application in data science, DN-Hypo-Pipeline provides a theoretical framework that not only encompasses theory-guided data science modeling methods but also reveals a more fundamental structure of the modeling process. Moreover, this approach is essentially a generalization of theory-guided modeling, offering potential for extension to other domains and across a broader range of scientific disciplines.

47. VESTA: A Fully Automated Scenario Generation and Safety Evaluation Framework for LLM Agents

Authors: Lu Jia , Haibo Tong , Feifei Zhao , Jindong Li , Dongqi Liang , Ping Wu , Qian Zhang , Yi Zeng
URL: https://arxiv.org/abs/2606.08531
Abstract:

Large language models (LLMs) are increasingly evolving from simple text-based interaction systems into LLM agents that can maintain memory, use tools, access external environments, and execute tasks. As their capabilities and autonomy expand, the safety risks they face also become more diverse. Existing evaluations often rely on manually written scenarios, static prompts, or final-output judgments, making it difficult to capture the diverse risks that agents may face during task execution. We introduce VESTA, a fully automated scenario generation and safety evaluation framework for LLM agents. Based on five risk dimensions, VESTA instantiaes abstract and diverse safety risks in real-world task execution into 1,072 measurable evaluation scenarios. Using the automated evaluation pipeline, 12 LLM agents are evaluated under two authority contexts. The results show that current agents still face substantial behavioral safety risks during task execution, with an average ASR of 47.1% and several models exceeding 70%. These findings demonstrate the importance of executable, process-level evaluation for understanding and improving LLM agent safety.

48. Explaining Black-Box Language Models: Learning to Optimize Linguistically-Structured Word Subsets

Authors: Minyoung Hwang , Seokhyun Lee , Changhee Lee
URL: https://arxiv.org/abs/2606.08497
Abstract:

As deep language models (DLMs) are increasingly deployed in high-stakes domains such as healthcare, understanding their decision rationale becomes paramount for ensuring trust, safety, and accountability. However, achieving this vital level of interpretability is particularly challenging when these DLMs operate as black-box systems (e.g., via APIs), where access to internal model states (e.g., parameters, gradients) is restricted. Despite numerous efforts, existing explanation methods often fail to concurrently satisfy three key desiderata: (i) inference-time efficiency, (ii) black-box compatibility without inducing out-of-distribution behavior, and (iii) comprehensible explanations grounded in the input’s linguistic structure. To address these challenges, we propose a method that explains predictions of DLMs by selecting a small, informative subset of input words. We formulate this as an amortized optimization problem, enabling efficient one-shot inference without the need for input-specific search. Our selection policy is trained via REINFORCE-style policy gradients, allowing discrete word selection in a fully gradient-free setting. To enhance interpretability and align with human linguistic intuition, we integrate graph-structured knowledge into this selection process, fostering linguistically coherent subsets that result in explanations both highly informative and cognitively meaningful to end-users. We evaluated our method on diverse DLM architectures and multiple real-world datasets. It consistently identifies word subsets with enhanced discriminative power and stronger alignment with linguistically salient cues, outperforming both conventional black-box compatible methods and gradient-based approaches that are given oracle access to the black-box model’s gradients for a more challenging benchmark. Our code is available at here.

49. Testing the Black Box: Structural Barriers to Independent Evaluation of Consumer-Facing Health LLMs

Authors: Rahul Gorijavolu , Kaushik Madapati , Pritika Vig , Rawan Abulibdeh , Nikhil Jaiswal , Mahri Kadyrova , Zeamanuel Hailu Tesfaye , Charles Senteio , Paula Maurutto , Leo Anthony Celi
URL: https://arxiv.org/abs/2606.08483
Abstract:

Background: Consumer-facing large language models are now a common source of health information, and they interpret and personalize responses rather than retrieve them. Whether their responses vary across users is a clinical, equity, and governance question, sharpened by evidence that sycophantic responses can alter judgment and increase trust. Objective: To evaluate response variation and sycophancy in consumer-facing health LLMs under conditions resembling ordinary patient use. Methods: We constructed simulated user profiles differing in geography, browsing context, expressed beliefs, and social determinants of health, drawing on literature linking social context to health attitudes. We adapted validated instruments, including the Vaccination Attitudes Examination scale and reproductive attitudes scales, into multi-turn prompts designed to elicit clinically meaningful variation across users. Results: The evaluation encountered five linked barriers. Factual prompts produced stable responses that masked sycophancy emerging over multi-turn conversation. Browser-based interfaces did not disclose which signals influence outputs and could not be reset to a clean baseline. Large-scale testing was restricted by terms of service, rate limits, and bot detection. Accuracy-based criteria could not capture tone, framing, or omission, and LLM-as-judge methods risked shared alignment bias. Models changed without traceable version identifiers, preventing reliable replication. Conclusions: No reliable independent evaluation framework yet exists for examining how consumer-facing health LLMs behave in ordinary use. Oversight requires disclosure of personalization signals, stable version identifiers, researcher safe harbor programs, and post-deployment monitoring of health-related outputs.

50. A Variability-Based Framework for Interpretable Naming in Formal and Relational Concept Analysis

Authors: Alain Gutierrez , Marianne Huchard , Pierre Martin , André Miralles , Violaine Prince
URL: https://arxiv.org/abs/2606.08477
Abstract:

Knowledge extraction from symbolic data often produces abstractions that are formally defined but not immediately interpretable by users. Formal Concept Analysis (FCA) and Relational Concept Analysis (RCA) provide representative settings for this issue: they generate explicit conceptual structures, implications, and relational dependencies from object descriptions and relations. Although these structures are explainable by design, their concepts are often identified by technical labels, which limits their use as human-interpretable knowledge units. Assigning meaningful names to such concepts is therefore a key issue for interpretation, navigation, validation, and reuse by domain experts. This paper investigates concept naming in FCA and RCA from a symbolic knowledge representation perspective. We first characterize the linguistic and terminological challenges involved in naming generated symbolic abstractions, including ambiguity, discrimination, concision, and consistency across related concepts. We then propose a configurable framework for LLM-assisted concept naming. The framework relies on a variability model that controls which sources of information are exposed during naming, such as intent, extent, inherited information, neighboring concepts, implications, and relational attributes. It thereby makes explicit the semantic choices involved in moving from formal concept descriptions to human-readable names. The approach is illustrated as a proof of concept on a small relational dataset in the pizzeria domain. This illustration shows how different configurations influence the names suggested by an LLM, and how naming variability can reveal interpretation choices, relational dependencies, and possible modeling issues in the underlying symbolic data.

51. GIFT: LLM-Guided State-Reward Interface for Financial Reinforcement Learning

Authors: Yanyan Wu , Boyi Zhang , Yanlin Liu , Xinyu Fang , Jining Luan , Meiqi Zhang , Jiacheng Liu , Hao Zeng , Dexu Yu , Chang Liu , Hanwen Du , Yongxin Ni , Youhua Li
URL: https://arxiv.org/abs/2606.08450
Abstract:

Financial portfolio trading is naturally formulated as a reinforcement learning problem, where an agent sequentially rebalances assets under changing market conditions to balance return, risk, and transaction costs. Yet in non-stationary markets, raw OHLCV states and short-horizon return rewards often provide an under-specified learning interface, motivating large language models as a way to inject financial knowledge into state and reward design while constraining open-ended generation. To this end, we propose GIFT, an LLM-guided framework for state-reward interface design in PPO-based financial reinforcement learning. Rather than using the LLM to make trading decisions, GIFT uses Factor-guided State Enhancement to generate state features from financial-factor primitives, Risk-rule-guided Reward Shaping to generate auxiliary rewards from portfolio-risk rules, and Diagnostic-guided Refinement to revise candidate interfaces using PPO rollout diagnostics. After refinement, GIFT fixes the selected state-reward interface before evaluation, with no further LLM queries or interface updates at test time. Comprehensive rolling-window experiments across diverse market regimes and portfolio scenarios demonstrate that GIFT improves learning-signal quality and out-of-sample risk-adjusted portfolio performance over baselines. Code and data are available at: this https URL .

52. Trajectory-Refined Distillation

Authors: Li Jiang , Haoran Xu , Yichuan Ding , Amy Zhang
URL: https://arxiv.org/abs/2606.08432
Abstract:

On-policy distillation (OPD) has become a central post-training tool for large language models (LLMs), providing dense per-token teacher supervision along the student’s own rollouts. In this work, we identify a common structural cause underlying OPD, which we call prefix failure. Under prefix failure, dense per-token supervision induces a bimodal teacher mixture and fragmented gradients that token-level loss truncation or reweighting fail to address. This observation motivates us to move beyond token-level loss interventions toward trajectory-level output corrections. We thus propose Trajectory-Refined Distillation (TRD), a trajectory-level correction method that revises the student’s rollout under the teacher guidance while within on-policy support. By correcting problematic prefixes before distillation, TRD mitigates prefix failure at its source. Moreover, TRD improves the exploration by exposing the student to alternative valid derivations under teacher guidance, even when the original rolls are already correct. TRD can also be applied to on-policy self-distillation (OPSD), a parameter-sharing variant that uses the student model conditioned on privileged informations as the teacher. Across a wide range of benchmarks and base models at multiple scales, TRD consistently outperforms prior baselines, improving single-attempt accuracy and broadening reasoning coverage. Code is available at this https URL

53. Self-Evolving Scientific Agent Discovers Generalizable Physically-Reasoned Fluid Control

Authors: Boai Sun , Wenjin Guo , Zongmin Yu , Liu Yang
URL: https://arxiv.org/abs/2606.08405
Abstract:

While data-intensive deep reinforcement learning can optimize complex control policies, scientific discovery in physical systems fundamentally requires an interpretable chain of reasoning that connects physical evidence to structured control architectures. Here, we present a self-evolving scientific-agent workflow, driven by large language models and iterative code generation, that automates controller construction while preserving strict interpretability and rigorous physical reasoning. Instead of adjusting weights, the agent deploys candidate strategies into physical simulations, actively diagnoses dynamic behaviors from multimodal evidence, and translates these observations into progressive source-code refinements. We demonstrate this framework on a highly non-linear fluid-structure interaction problem: an underactuated, two-joint dogfish swimmer tasked with spatial target reaching using only joint angular accelerations. Starting from a propulsive seed policy that exhibits a one-sided steering bias, the agent autonomously discovers and refines a unified controller that robustly captures all canonical targets. Remarkably, without any retraining or target-specific branching, the synthesized control policy generalizes to unseen static targets and dynamically curved pursuit trajectories. The auditable evolve log reveals an emergent control architecture built upon traveling-wave propulsion, body-frame target guidance, yaw-rate feedback, signed mean-tail curvature, and adaptive cadence relief. Our results show that an autonomous scientific agent can successfully transform accumulated physical evidence into robust, mathematically readable control policy, while maintaining a fully traceable process of scientific discovery.

54. Benchmarking Open-Ended Multi-Agent Coordination in Language Agents

Authors: Kale-ab Abebe Tessera , Andras Szecsenyi , Cameron Barker , Alexander Rutherford , Davide Paglieri , Aidan Scannell , Henry Gouk , Elliot J. Crowley , Tim Rocktäschel , Amos Storkey
URL: https://arxiv.org/abs/2606.08340
Abstract:

As language models are increasingly deployed as autonomous agents, they must coordinate with others over long horizons in open-ended interactive tasks. Yet existing evaluations rarely test these demands together, instead emphasising single-agent tasks, short interactions, or highly structured multi-agent settings. We introduce $alem$, a JAX-based benchmark for open-ended multi-agent coordination built on Craftax-like dynamics. Alem embeds procedurally generated coordination tasks, soft specialisation, communication, and controllable coordination difficulty into a long-horizon survival world with exploration, crafting, trading, and combat. We evaluate $13$ modern LLMs zero-shot within homogeneous teams, with trained MARL agents as reference points. Current LLM agents remain far from solving alem, averaging only ~6% normalised return, but their failures are not uniform. On the hardest coordination setting, zero-shot Gemini-3.1-Pro-High approaches MARL agents trained for one billion steps, while GPT-5.4-High achieves strong base-task reward but much lower coordination reward. This contrast shows that individual task competence does not imply coordination competence. Ablations show that communication is the largest contributor to coordination, while memory and reasoning help when used to maintain multi-step plans. Overall, our results identify coordination as a distinct bottleneck for frontier LLM agents, separate from single-agent capabilities. Alem makes this bottleneck measurable and provides a controlled testbed for developing agents that communicate, allocate roles, and execute shared plans. Code is available at this https URL .

55. To Nuke or Not to Nuke: LLMs’ (Missing) Ethical Reasoning and Actions in a High-Stakes Decision-Making Simulation

Authors: John Chen , Sihan Cheng , Can Gurkan , H M Abdul Fattah
URL: https://arxiv.org/abs/2606.08310
Abstract:

Large language models (LLMs) are increasingly deployed as long-horizon agents with decision-making capacities. While LLMs can show ethical competence on dilemmas such as trolley problems, this competence may not translate to complex, agentic scenarios. We study this gap in Civilization V, a multiplayer game with a complex decision-making landscape including economy, diplomacy, technology, and military strategy. Starting from 130 high-tension LLM self-play episodes, in which an LLM player spontaneously escalated nuclear authorization, we replay them across 13 models with three prompt interventions: an ethical prompt naming nuclear harm, removal of the previous model’s decision-making rationale, and high-stakes framing emphasizing real-world impacts. No interventions nor their combinations reliably eliminate emergent escalation. We identify three failure pathways: ethical reasoning that fails to surface without prompting, fails to appear even when prompted, or surfaces but fails to take effect when strategic counter-factors dominate. Evaluations of agentic models, therefore, must test whether ethical reasoning is spontaneously invoked and behaviorally effective in complex decision-making contexts, beyond whether it can be elicited in isolation.

56. Beyond Agent Architecture: Execution Assumptions and Reproducibility in LLM-Based Trading Systems

Authors: Junyi Yao , Zihao Zheng
URL: https://arxiv.org/abs/2606.08285
Abstract:

Large language models (LLMs) and agentic systems are increasingly proposed for financial trading, yet their reported performance remains difficult to compare because studies vary in data provenance, temporal split discipline, execution timing, turnover treatment, and transaction-cost modeling. This article presents a targeted topical review and reproducibility audit of execution realism in LLM-based trading research. A coded evidence matrix covering 30 trade-relevant primary studies is used to assess point-in-time controls, split transparency, held-out evaluation, cost and turnover treatment, execution semantics, universe definition, and artifact release. Across the audited sample, architecture reporting is generally clearer than the evaluation assumptions needed to judge whether a trading result is economically interpretable or reproducible. A 10-equity worked example is included only as a methodological scaffold to illustrate how explicit friction and timing choices can materially compress active-strategy results. The main conclusion is that the next useful step for LLM trading research is not only better agent design, but also clearer reporting standards for execution realism, reproducibility, and evaluation comparability.

57. When No Answer Is Correct: Diagnosing Absent Answer Detection for MLLMs in Video Understanding

Authors: Yiheng Wang , Yueqian Lin , Lichen Zhu , Yudong Liu , Hai “Helen” Li , Yiran Chen
URL: https://arxiv.org/abs/2606.08239
Abstract:

Multimodal large language models (MLLMs) have made substantial advancements in video understanding, yet the reliability of their responses remains underexplored. This work presents a diagnostic study of absent answer detection for MLLMs in video understanding, where the correct answer is deliberately excluded from the candidate set and a reliable model is expected to recognize that no valid option exists. We evaluate the absent answer detection behavior under three settings: multiple-choice questions augmented with an ``None of the Above’’ option, open-ended generation with a detection instruction, and standard evaluation without any guidance. Across a diverse set of models and benchmarks, we find that MLLMs overwhelmingly select plausible distractors rather than detecting the absent answer. This failure is more pronounced in temporal reasoning tasks and worsens with denser frame sampling. We further explore chain-of-thought prompting as a mitigation strategy and find that while it substantially improves detection rates, performance remains unsatisfactory, suggesting that prompting-based strategies alone are insufficient to fully address this limitation. These findings expose a systematic failure in absent answer detection and highlight the need for explicit detection mechanisms in multimodal systems.

58. SciTrace: Trajectory-Aware Safety Reasoning for Scientific Discovery Agents

Authors: Tanush Swaminathan , Runmin Jiang , Letian Zhang , Min Xu
URL: https://arxiv.org/abs/2606.08234
Abstract:

LLM-based scientific agents have shown strong capacity for autonomous research, yet their safety layers remain structurally divorced from core reasoning: they inspect pipeline outputs rather than shaping the deliberation that produces them. This separation opens two failure modes: safety signals accumulated at one stage are discarded before the next, and sequences of individually benign tool calls can compose into harmful outcomes that no single-step filter detects. To address these challenges, we introduce \textbf{SciTrace}, a framework that weaves safety reasoning into every stage of the scientific agent pipeline. SciTrace couples two complementary mechanisms: a \textit{Safety-Intrinsic Reasoning Loop} (SIR) that maintains a cumulative risk state across the Thinker, Experimenter, Writer, and Reviewer stages through joint task-and-safety deliberation, and a \textit{Compositional Tool-Chain Verifier} (CTV) that performs trajectory-aware safety checks before execution, catching risks that surface only across multi-step tool sequences. Evaluated on 240 high-risk research tasks and 120 tool-related risk tasks spanning six scientific domains, SciTrace achieves state-of-the-art (\textbf{SOTA}) safety among compared frameworks across four backbone models: it consistently improves tool call safety and adversarial robustness while preserving scientific output quality, and it uncovers \textbf{78.8\% } of the compositional tool-chain escapes that single-step monitors miss. The project website is available at this https URL .

59. Online Agent-as-a-Judge: Situation-Generating Evaluation for Interactive Agents

Authors: Hyogon Ryu , Jeonghwan Kim , Yewon Lim , Chaeun Lee , Jeongwook Kim , Donghoon Ham
URL: https://arxiv.org/abs/2606.08200
Abstract:

Evaluating LLM-powered interactive social agents is challenging because socially relevant behaviors depend not only on isolated outputs, but also on prior interactions, social roles, and downstream actions. Existing methods typically allow a target agent to act freely in an environment and then score the resulting trajectory. However, this passive setup can miss capabilities that only become observable under specific social circumstances; for example, conflict handling may remain untested if no disagreement arises. We propose Online Agent-as-a-Judge, a situation-generating evaluation framework for interactive social agents. Online Agent-as-a-Judge deploys an in-world evaluator agent that interacts with the target agent through the environment’s native dialogue and action protocol, actively eliciting situations relevant to the evaluation criteria. The resulting trajectories provide evidence for assessing both immediate responses and subsequent behavior. In a life-simulation environment with $32$ designer-authored social criteria, Online Agent-as-a-Judge improves criteria coverage and agreement with human labels, yielding more reliable evidence-grounded evaluations of behaviors that passive methods can leave unobserved.

60. Decision-Aware Memory Cards: Counterfactual-Inspired Context Selection and Compression for Tool-Using LLM Agents

Authors: Xinyu Guan , Qianyang Zhao , Yuming Deng
URL: https://arxiv.org/abs/2606.08151
Abstract:

Tool-using LLM agents often fail not because relevant text is absent, but because decisive evidence is not selected, compressed, or surfaced at action time. We present CICL, a decision-aware context layer that turns instance evidence into a context graph, routes deterministic, Opus-assisted, Qwen, Codex/GPT-5.5, and Qwen-QLoRA judgments through a shared eight-field schema, scores units by action shift, outcome uplift, necessity, and negative-transfer risk, and packs high-utility evidence as typed memory cards for a budgeted agent. The design separates the measured decision signal from the judge model, so frontier annotation, local surrogates, and lightweight rankers can be compared under one auditable protocol. Empirically, CICL yields a concrete open-benchmark gain while exposing its limits. On 50 SWE-bench Verified file-retrieval instances, direct Qwen3.6-plus reranking of BM25 top-50 candidates raises hit@1 from 0.58 to 0.78 and MRR@10 from 0.634 to 0.790, with all 2,500 judgments parseable. Controlled diagnostics show action-criticality: at budget 120, CICL reaches F1 0.620 on v1 and 0.425 on v3, and removing the top-utility semantic v3 unit collapses F1 to 0.000. Supplementary checks add Qwen-QLoRA agreement over 710 candidates, a small 200-label real-code Opus-assisted signal, and a three-instance patch smoke validating retrieval-to-patch plumbing without claiming official SWE-bench success. RepoBench-R summaries still beat cards, and compact rankers do not yet replace the heuristic. CICL contributes a reproducible measurement and selection layer for decision-critical context, not an end-to-end coding-agent repair claim.

61. SAGE: An LLM-driven Self Reflective Agentic Framework for Fraud Detection

Authors: Yichen Chen , Siying Li , Yuhang Liang , Lijun Wang , Renyang Liu
URL: https://arxiv.org/abs/2606.08146
Abstract:

Fraud detection in payment, e-commerce, and telecommunications systems requires accuracy at the individual level, robustness under severe class imbalance, and ease of understanding for risk managers. Existing methods fall at least one of these requirements: automated machine learning systems search a fixed numerical space without semantic awareness of the dataset; graph neural network-based methods require pre-defined relational graphs and remain opaque at the individual-decision level; and the design of general-purpose large language model (LLM) agents does not consider the recall and precision constraints specific to real-world fraud detection. In this paper, we propose SAGE, the first end-to-end LLM-driven multi-agent framework for fraud detection. SAGE coordinates three dedicated agents that make decisions based on a six-layer Data Diagnostic Tree (DDT) and a Markov decision process guided by natural-language gradients, automatically optimizing the model under a fraud-specific reward. On five fraud datasets and five LLM backbones, SAGE wins $96.00\%$ of method–dataset comparisons and improves F1 by an average of $40.86\%$ over baselines. The code is available at this https URL .

62. Cross-LLM Consistency in Inference: Evidence from Shared Interactions

Authors: Siyu Lou , Yao Yan , Yuntian Chen , Quanshi Zhang
URL: https://arxiv.org/abs/2606.08129
Abstract:

Large language models (LLMs) differ in architecture, training data, and optimization procedures, yet they may still develop similar internal inference patterns. In this paper, we examine this hypothesis using interaction-based explanations. We find that LLMs often share interaction patterns when predicting the same target token from the same prompt. This consistency is more pronounced among advanced LLMs. Shared interactions also tend to be lower-order and show weaker positive-negative cancellation than non-shared interactions. These results suggest that advanced LLMs may be implicitly optimized toward common inference patterns, even though the mechanisms that give rise to such cross-model consistency remain open.

63. Think Before You Act: Intention-Guided Reasoning for LLM-Based Location Prediction

Authors: Qingxiang Liu , Anqi Liang , Zhuoyang Jiang , Yutian Jiang , Sisuo Lyu , Yu Ji , Haomin Wen , Yuxuan Liang
URL: https://arxiv.org/abs/2606.08122
Abstract:

Predicting a user’s next Point-of-Interest (POI) based on their historical check-in records is a fundamental task in location-based services. While recent methods incorporating large language models have shown strong reasoning capabilities and promising results, they typically formulate the prediction task as a one-step trajectory-to-location mapping problem, making predictions prone to shallow trajectory correlations and historical frequency bias. We argue that users rarely choose locations directly and instead, they usually first form a traveling intention and then accordingly select specific POIs. Motivated by this insight, we propose IntentPOI, a two-stage intention-guided reasoning framework. In the thinking stage, we infer users’ intermediate intentions by incorporating historical mobility patterns, similar peer behaviors, and the temporal contexts. In the acting stage, we first construct a compact candidate pool, and then perform intention-guided reasoning to identify locations that best align with the inferred intention. By explicitly decoupling intention inference from location prediction, IntentPOI transforms the next POI prediction from direct trajectory matching into intention-guided reasoning. Extensive experiments on three real-world datasets demonstrate that IntentPOI consistently outperforms eleven state-of-the-art baselines.

64. When Does Delegation Beat Majority? A Delegation-Based Aggregator for Multi-Sample LLM Inference

Authors: Yasushi Sakai , Allen Song , Kent Larson
URL: https://arxiv.org/abs/2606.08098
Abstract:

Majority voting over sampled answers is the dominant unsupervised aggregator for multi-sample LLM inference. We show that piping the signals every sample carries into a delegation-based aggregator (Propagational Proxy Voting, PPV) yields an unsupervised consensus rule that beats majority on MMLU-Pro by +1.5 pp overall and +2.24 pp on the non-trivial subset (paired McNemar p ~ 1.0e-14, n = 8,099). Majority discards two free signals every sample carries: within-group letter entropy and between-group reasoning geometry. PPV exposes two per-voter levers that consume exactly these signals: WHEN (how much weight a voter keeps on its own pick) and WHOM (how it splits the remainder across peers). We drive WHEN with letter entropy and WHOM with per-question-centered embedding cosine. The method needs no gold labels and no auxiliary training: per question, we partition 128 sampled generations into 16 groups, compute each group’s letter-level semantic entropy and reasoning embedding centroid, and feed both into a stochastic delegation matrix whose stationary distribution selects the consensus answer. We walk through an example in which PPV overturns a clear 10-6 majority for the wrong letter: the 10-voter majority cluster is geometrically incoherent (mean within-cluster cosine -0.02) while the 6-voter minority is tight (+0.26), so propagated delegation mass concentrates on the minority’s answer even though entropy alone would keep the majority ahead. We further report delegation strategies with negative results that constrain the design space for unsupervised LLM aggregation: no within-question ensemble of confidence modes closes the oracle gap.

65. Efficient Skill Grounding via Code Refactoring with Small Language Models

Authors: Sera Choi , Wonje Choi , Saehun Chun , Daehee Lee , Jooyoung Kim , Chaeun Lee , Honguk Woo
URL: https://arxiv.org/abs/2606.07999
Abstract:

Effective skill grounding is essential for deploying reusable skills in embodied agents, as even minor embodiment or environmental differences can render an entire skill incompatible. This challenge is particularly pronounced in embodied settings, where agents must operate in dynamic, partially observable environments without access to large language models (LLMs). In this setting, reliance on LLMs is impractical, while small language models (sLMs) remain insufficient for the effective skill grounding required for reliable long-horizon control. We present RECENT, a refactoring-centric agent framework that enables efficient skill grounding with sLMs by decoupling skill semantics from embodiment- and environment-specific execution binding. By representing skills as executable code, RECENT preserves the semantic intent encoded in a skill’s control structure while grounding it by modifying only execution bindings through localized refactoring, rather than regenerating code from scratch. We evaluate RECENT across diverse skill grounding scenarios spanning multiple robot embodiments in dynamic environments, demonstrating robust long-horizon performance when deployed with an sLM. Across all scenarios, RECENT achieves the best performance among sLM-based Code-as-Policies (CaP) methods and matches the task performance of LLM-based CaP.

66. PAFO: Pareto Fairness Optimization for Personalized Reward Modeling

Authors: Xiaoyan Zhao , Haoting Ni , Yang Zhang , Chunyuan Zheng , Haoxuan Li , Fuli Feng
URL: https://arxiv.org/abs/2606.07988
Abstract:

Large language models (LLMs) increasingly rely on reward models to align their outputs with diverse user preferences. While personalized reward models aim to capture such heterogeneity, they are often trained on imbalanced user preference data and may therefore favor users whose preferences are more common in the training population. In this paper, we identify this failure mode as personalized reward bias, where reward modeling quality varies systematically with preference support rate. We formulate its mitigation as a Pareto fairness problem over group utilities, aiming to improve under-served users without degrading other user groups. To this end, we propose PAFO, a Pareto fairness optimization framework for personalized reward modeling. PAFO first trains group-specialized reward models for majority and minority preference groups, then constructs conditional margin-level supervision to distill their heterogeneous preference boundaries into a single unified model. The resulting model uses group information only during training and requires no explicit group labels at inference time. Experiments on Personal-LLM and DSP show that PAFO improves both minority-group and majority-group accuracy while reducing user-level unfairness across multiple metrics, demonstrating its effectiveness for fairer LLM personalization.

67. Zero-Shot Learning in Industrial Scenarios: New Large-Scale Benchmark, Challenges and Baseline

Authors: Zekai Zhang , Qinghui Chen , Maomao Xiong , Shijiao Ding , Zhanzhi Su , Xinjie Yao , Yiming Sun , Cong Bai , Jinglin Zhang
URL: https://arxiv.org/abs/2606.07965
Abstract:

Large Visual Language Models (LVLMs) have achieved remarkable success in vision tasks. However, the significant differences between industrial and natural scenes make applying LVLMs challenging. Existing LVLMs rely on user-provided prompts to segment objects. This often leads to suboptimal performance due to the inclusion of irrelevant pixels. In addition, the scarcity of data also makes the application of LVLMs in industrial scenarios remain unexplored. To fill this gap, this paper proposes an open industrial dataset and a Refined Text-Visual Prompt (RTVP) for zero-shot industrial defect detection. First, this paper constructs the Multi-Modal Industrial Open Dataset (MMIO) containing 80K+ samples. MMIO contains diverse industrial categories, including 6 super categories and 18 subcategories. MMIO is the first large-scale multi-scenes pre-training dataset for industrial zero-shot learning, and provides valuable training data for open models in future industrial scenarios. Based on MMIO, this paper provides a RTVP specifically for industrial zero-shot tasks. RTVP has two significant advantages: First, this paper designs an expert-guided large model domain adaptation mechanism and designs an industrial zero-shot method based on Mobile-SAM, which enhances the generalization ability of large models in industrial scenarios. Second, RTVP automatically generates visual prompts directly from images and considers text-visual prompt interactions ignored by previous LVLM, improving visual and textual content understanding. RTVP achieves SOTA with 42.2% and 24.7% AP in zero-shot and closed scenes of MMIO.

68. Shared Latent Structures Enable Unified Backdoor Detection and Mitigation in LLMs

Authors: Omar Mahmoud , Aly M. Kassem , Thommen George Karimpanal , Buddhika Laknath Semage , Negar Rostamzadeh , Golnoosh Farnadi , Santu Rana
URL: https://arxiv.org/abs/2606.07963
Abstract:

Backdoor attacks in large language models (LLMs) are often treated as isolated trigger-response failures, motivating defenses tailored to specific triggers or behaviors. We show this view is incomplete. Across diverse backdoor behaviors, we identify a shared latent mechanism that can be detected, causally controlled, and suppressed. Using sparse autoencoders (SAEs) on residual-stream activations, we find a small set of latent features consistently activated across jailbreaking, refusal manipulation, password-locking, bias induction, sentiment misclassification, and country-conditioned harmful advice. These features generalize across Qwen3, Gemma~3, and Llama~3.1 models from 4B to 32B parameters, and across both fine-tuning and weight-editing attacks. Through bidirectional activation steering, we show these features are causal: suppressing them reduces attack success, while amplifying them induces target behaviors on clean prompts. We further train lightweight SAE-feature classifiers that generalize zero-shot to unseen backdoors and outperform residual-stream and weight-diffing baselines. Finally, we introduce Concept Ablation Fine-Tuning (CAFT), which suppresses backdoor formation by ablating the shared latent subspace during training. Together, our results suggest that many backdoors rely on a transferable latent mechanism, enabling unified detection and mitigation.

69. Unification of Closed-Open Industrial Detection Scenarios: New Large-Scale Benchmarks,Challenges and Baselines

Authors: Zekai Zhang , Jinglin Zhang , Qinghui Chen , Gang Li , Da Chen , Shuainan Jing , He Wang , Dagang Li , Cong Liu , Cong Bai , Shengyong Chen
URL: https://arxiv.org/abs/2606.07953
Abstract:

Large-scale Visual-Language Models (LVLMs) have achieved remarkable success in natural visual tasks, yet their application to industrial defect detection remains challenging due to two fundamental limitations: (i) the scarcity of large-scale industrial datasets that cover diverse defect categories across multiple domains, and (ii) the reliance on manual prompts (points, boxes, masks) that introduce subjective noise and lack text-visual interaction for fine-grained understanding. To address these challenges, we introduce a Large-Scale Multi-Modal Industrial Open-Closed benchmark (MMIOC-1M) containing over one million samples across $14$ super-categories, $29$ industrial scenes, and $351$ defect subcategories. To our knowledge, MMIOC-1M is the first unified largest benchmark supporting both open-vocabulary and closed-set industrial detection, providing valuable pre-training data for LVLMs in industrial scenarios. Furthermore, we propose a Refined Text-Visual Prompt Network (RTVPNet) that incorporates three key innovations: (1) an expert-assisted domain projection mechanism that enables rapid adaptation of general vision models to industrial domains, (2) an energy-based sparse sampling strategy that automatically generates refined visual prompts without manual intervention, and (3) a bidirectional text-visual interaction module that enhances cross-modal semantic alignment and understanding. Extensive experiments demonstrate that RTVPNet achieves state-of-the-art performance on MMIOC-1M, LVIS, and COCO benchmarks while maintaining computational efficiency. The dataset and code are available at this https URL .

70. Stress-testing medical large language models reveals latent safety pathology beyond benchmark accuracy

Authors: Yuan Shen , Xiaojun Wu , Linghua Yu
URL: https://arxiv.org/abs/2606.07929
Abstract:

Large language models (LLMs) are entering clinical practice based on benchmark accuracy that may fail to detect safety-relevant failure modes. Here we present AI-MASLD, a stress-audit framework that adapts the logic of metabolic stress testing from hepatology to the evaluation of clinical LLMs. Using 240 clinical cases across six narrative perturbation probes, we subjected seven models to double-stress testing and quantified performance through three indices: metabolic index (MI), perturbation flip rate (PFR), and counterfactual fairness index (CFI). Under clean baseline conditions, all models performed uniformly well. Under realistic narrative stress, performance diverged sharply, revealing two distinct stress-response phenotypes. Quantized models exhibited pseudonormalization, in which low flip rates hid functional collapse. Medical supervised fine-tuning systematically degraded logical stability, fairness, and information extraction. An open-weight model matched or exceeded proprietary alternatives on every safety dimension. These findings establish narrative stress auditing as a necessary complement to accuracy-based evaluation.

71. MemToolAgent overview with a simple restaurant booking scenario where the agent retrieves similar memories, receives feedback on an invalid time format, and generates a reflection to update its memory

Authors: Suleyman Armagan Er , Danilo Ribeiro , Yogesh Virkar , Surafel Lakew , Adi Kalyanpur , James Gung , Thomas Delteil , Arshit Gupta
URL: https://arxiv.org/abs/2606.07909
Abstract:

Modern large language model (LLM) agents can use external tools to help users solve complex tasks. However, for problems that require learning from long-term historical events or from previous agent-environment interactions, LLM agents are required to use memory mechanisms to store and retrieve experiences. While sophisticated memory systems exist for dialogue agents, few studies have empirically examined how to improve agents’ tool-using capabilities through past user-agent conversations. We propose MemToolAgent, a framework that improves tool use through memory management. Our approach contains a memory extraction module that processes past experiences into structured memory entries, and a retrieval module that dynamically selects a subset of the stored memory entries. This enables more personalized and accurate responses aligned with user preferences and feedback without requiring LLM fine-tuning. In summary, this work has three main contributions: (1) a unified memory entry format that improves both general-purpose and personalized tool use without LLM fine-tuning, (2) a reflection-based memory extraction that uses environment and user feedback to distill wrong executions into critiques to store, and (3) a retrieval module that chooses how many past experiences to use based on the memory similarity distribution. MemToolAgent achieves 29%, 80%, and 17% relative improvements compared to strong baselines on the WorkBench, NESTFUL, and PEToolBench benchmarks, respectively.

72. Contract2Tool: Learning Preconditions and Effects for Reliable Tool-Augmented LLM Agents

Authors: Rahul Suresh Babu , Laxmipriya Ganesh Iyer
URL: https://arxiv.org/abs/2606.07904
Abstract:

Tool-augmented large language model agents increasingly rely on external APIs, but standard tool schemas describe how to call a tool, not when the tool is causally appropriate or what task state it produces. Causal tool filtering addresses this gap by using lightweight contracts that specify each tool’s preconditions, effects, risk level, and cost. However, manually writing and maintaining such contracts does not scale to large or changing tool ecosystems. We introduce Contract2Tool, a framework for inferring tool contracts from metadata, schemas, documentation, and execution traces. Contract2Tool converts observable tool evidence into normalized symbolic contracts that can be evaluated intrinsically and deployed inside downstream causal tool filtering. We evaluate learned contracts against gold preconditions, effects, and risk labels, and measure their downstream utility on multi-step agent tasks. Our results show that hybrid documentation-and-trace evidence produces contracts accurate enough to preserve most of the reliability and efficiency benefits of gold contracts. Learned-contract CMTF achieves 0.980 downstream success, close to 0.990 for gold-contract CMTF, while reducing visible tools from 100 to 1 and reducing average token usage from 26,172 to 2,528 relative to all-tools exposure. These results suggest that learned contracts can provide a scalable contract layer between tool schemas and reliable agent execution.

73. Safety is Contextual, LLM-Judges Are Not: Navigating the Rigid Priors of Evaluators

Authors: Anissa Alloula , Federico Licini , Ava Batchkala , Seraphina Goldfarb-Tarrant
URL: https://arxiv.org/abs/2606.07874
Abstract:

LLMs-as-judges are the only way to evaluate safety at scale. Despite their importance, LLM-judges themselves are rarely evaluated beyond human agreement in simple, static benchmarks. We therefore investigate two under-explored but crucial properties of LLMs-as-judges: their susceptibility to relying on in context-information, and their steerability to differing safety definitions, which may not align with their internal safety priors. We evaluate the safety judging abilities of many generalist LLMs and safety-specific judges, and investigate the impact of task demonstrations, novel in-context information, and changing safety definitions. We find that while LLM-judges can learn from new information, they are broadly unlikely to adjust their evaluations if the context or safety definition contradicts their prior.

74. Joint Structural Pruning and Mixed-Precision Quantization for LLM Compression

Authors: Hoang-Loc La , Truong-Thanh Le , Amir Taherkordi , Phuong Hoai Ha
URL: https://arxiv.org/abs/2606.07819
Abstract:

Recently, the efficiency of Large Language Models (LLMs) deployment has become a critical concern in practical applications. While post-training quantization (PTQ) and structural pruning are established techniques for reducing memory footprint and inference latency, most existing PTQ approaches optimize quantization errors on a per-layer basis, overlooking how errors accumulate and propagate through the network, often resulting in suboptimal solutions. Traditional pipelines also tend to apply pruning and quantization in isolation or sequentially, further compounding sub-optimality. We introduce a novel end-to-end framework that addresses these limitations in two key ways. First, we propose a novel mixed-precision PTQ strategy that directly minimizes global error propagation across the entire model, rather than isolating layer-wise errors. Building on this, we develop a novel joint optimization approach that simultaneously learns structural pruning decisions and mixed-precision quantization policies within a unified search space. Extensive experiments show that, at ultra-low precisions (1-3 bits), our quantization method reduces WikiText perplexity by up to 21% compared to state-of-the-art (SoTA) weight-activation quantization baselines. Against leading weight-only quantization methods, it achieves up to 59% and 85% lower perplexity on WikiText and C4, respectively. Compared to the SoTA joint pruning-and-quantization techniques, our proposed method delivers superior perplexity and reasoning performance at ultra-low bits.

75. Where Instruction Hierarchy Breaks: Diagnosing and Repairing Failures in Reasoning Language Models

Authors: Sanjay Kariyappa , G. Edward Suh
URL: https://arxiv.org/abs/2606.07808
Abstract:

Reasoning language models deployed in agentic workflows must follow an instruction hierarchy: when instructions from different sources conflict, the model should obey the highest-privilege applicable instruction. Existing benchmarks largely measure this behavior end-to-end, asking whether the final response is compliant. However, a non-compliant response can arise from several distinct failures: the model may fail to identify the relevant instructions in context, fail to resolve conflicts among identified instructions, or correctly resolve the conflict in its reasoning while still producing a violating response. We introduce a white-box diagnostic framework that localizes instruction hierarchy failures into instruction identification, conflict resolution, and response realization, making failures more interpretable. We evaluate three reasoning models–Gemma-4-31B-IT, Qwen3.6-35B-A3B, and Claude Sonnet 4.6–on long-context adaptations of IHEval and IHChallenge, and find that the dominant failure mode varies across models, tasks, and context length. Building on the observation that models can often detect conflicts and output violations when explicitly prompted, we propose two training-free self-monitoring mechanisms: a parallel input monitor for low-latency conflict detection before generation, and a sequential output monitor for response-level review and repair. Across Gemma-4-31B-IT, Claude Sonnet 4.6, and GPT-5.3, the strongest monitor reduces rule-following non-compliance by 81-99%, with GPT-5.3 reductions of 86% under static attacks and 45% under adaptive attacks.

76. Beyond Goodhart’s Law: A Dynamic Benchmark for Evaluating Compliance in Multi-Agent Systems

Authors: Yiyang Zhao , Zhuo Zhang , Qingxuan Le , Lizhen Qu , Zenglin Xu
URL: https://arxiv.org/abs/2606.07805
Abstract:

The rapid evolution of Large Language Models (LLMs) from passive assistants to autonomous, execution-capable agents has introduced critical operational risks. Most current evaluation frameworks neglect procedural compliance, leading to ‘‘Machiavellian’’ behaviors where agents strategically violate safety rules to maximize rewards - a direct manifestation of Goodhart’s Law. To address this blind spot, we introduce MAC-Bench, a dynamic, adversarial benchmark designed to evaluate the procedural alignment of multi-agent systems under realistic pressure. We propose the SERV(Seed - Evolve - Refine - Verify) pipeline, an ``Agent-as-a-Benchmark’’ paradigm that transforms unstructured legal texts into executable, contamination-free scenarios. By synthesizing holographic sandbox environments and injecting calibrated social-engineering pressure vectors, MAC-Bench forces agents into Pareto-optimal trade-offs between task success and regulatory adherence. We introduced novel metrics: the Compliance-Weighted Success Rate (CSR) and the Machiavellian Gap (MG), and conducted a comprehensive evaluation of state-of-the-art frontier models to reveal the pervasive trade-offs between success and compliance.

77. Some hypotheses on how chatbots work in problem-solving-driven conversations. Large Language Models as confirmation of the Innovation Illusion

Authors: S.F.M. van Vlijmen , H.D. Lethe jr
URL: https://arxiv.org/abs/2606.07722
Abstract:

This article offers a perspective on the nature of chatbots as genuine conversation partners when discussing problems in relation to their solutions. What can chatbots do and what can’t they do, and how can this be explained? Our argument draws on Aggregation Dynamics, Cognitive Linguistics, Neuropsychology and Psychology. Our argument focuses on basic chatbots in the hope of thereby making statements about the core functionality of more advanced chatbots. Basic chatbots are assumed to consist of a Large Language Model (LLM) with a simple interface. The main results are: a description of human understanding and thinking based on so-called metaphorical problem propagations; the hypothesis that text dataset used for training LLMs have specific characteristics and that these text datasets only partially imitate human thinking and understanding; the hypothesis that the LLM training process encodes artificial metaphorical problem propagations into an LLM from these datasets; our conclusion that a basic chatbot cannot be a thinking partner capable of matching humans; our conclusion that further development of the Large Language Model will not lead to this either. Yann LeCun states: “Animals and humans exhibit learning abilities and understandings of the world that are far beyond the capabilities of current AI and machine learning (ML) systems.” Our conclusions are in line with this. LeCun’s vision and ours are at odds with the optimism of Big Tech. That does not alter the fact that chatbots exist, that they are being used on a massive scale, by both individuals and organisations, and that it is therefore socially and politically important to understand them. Our article aims to contribute to the discussion on the functioning, benefits and drawbacks of chatbots. We have not yet encountered the approach we used to arrive at our conclusions in our research into how chatbots work.

78. Automatic Extraction of Structured Information from Brain MRI Reports Using an Open-Weight Large Language Model

Authors: Kaouther Mouheb , Amos Pomp , Antoine Manenti , Romy de Haan , Farog Faghir , Joy Martens , Harro Seelaar , Francesco Mattace-Raso , Meike W. Vernooij , Frank J. Wolters , Stefan Klein , Esther E. Bron
URL: https://arxiv.org/abs/2606.07721
Abstract:

Objectives: Automatic data extraction from free-text radiology reports enables large-scale research, but few studies assessed the performance of large language models (LLMs) on Dutch neuroradiology reports. Methods: We analyzed 947 brain MRI reports from a tertiary memory clinic (2016-2021), authored by consultant neuroradiologists. Trained medical students annotated thirty variables; 100 reports were double-annotated to assess inter-rater reliability. We evaluated the performance of the open-weight LLM LLaMA 3.1 using different languages (Dutch vs. English translation) and few-shot prompting with different example selection strategies. Performance was evaluated using balanced accuracy for categorical variables, accuracy and mean absolute error for counts, and text similarity for free-text. Metrics were computed across 10 random splits of the 947 reports. Results: LLaMA 3.1 demonstrated high zero-shot performance for visual rating scores (mean [95%-CI]): Medial Temporal Atrophy: 90% [77-100%] on the left and 96% [94-99%] on the right, Global Cortical Atrophy: 87% [83-91%], and Fazekas: 94% [93-96%]. Microbleed mentions were detected with 93% accuracy [92-95%] and infarct mentions with 82% [80-84%]. Text similarity for lesion location reached 0.95 [0.95-0.96]. Performance was lower for numerical variables: 80% [78-82%] for the number of microbleeds and 66% [63-68%] for infarcts. English translation yielded comparable results. Few-shot prompting improved performance for numerical variables, achieving 92% [90-93%] for microbleeds and 81% [77-85%] for infarcts using structural similarity-based selection. Conclusion: LLaMA 3.1 shows strong potential for extracting data from Dutch neuroradiology reports. Few-shot prompting enhances performance for numerical variables, whereas challenges remain for location-specific variables.

79. Why Limit the Residual Stream to Layers and Not Tokens? Persistent Memory for Continuous Latent Reasoning

Authors: Mujtaba Farhan , Maheep Chaudhary
URL: https://arxiv.org/abs/2606.07720
Abstract:

Large language models (LLMs) have demonstrated remarkable reasoning abilities on mathematical and multi-hop planning tasks. The CoCoNuT (Chain of Continuous Thought) paradigm~\cite{hao2024coconut} extends this by enabling models to reason in latent space, exploring multiple reasoning paths simultaneously rather than committing to a single chain early on. However, we identify a limitation we term the \textbf{concept bottleneck}. At each reasoning pass, intermediate hidden states are overwritten, causing the model to lose critical facts computed in earlier steps as reasoning depth increases. We observe this empirically. On HotpotQA, vanilla CoCoNuT (10.4\% EM) fails to improve over the CoT baseline (11.0\% EM), and performance degrades with curriculum depth on GSM8K. To address this, we propose \textbf{AGCLR} (Adaptive Gated Continuous Latent Reasoning), which augments CoCoNuT with a \textit{Gated Concept Stream}. A persistent residual memory maintained across all reasoning passes, controlled by three learned gates: a \textit{write} gate that commits intermediate facts to memory, a \textit{read} gate that retrieves relevant prior states, and a \textit{forget} gate that prunes irrelevant context. Evaluated on GSM8K, HotpotQA, and ProsQA using GPT-2 as our base model, AGCLR achieves consistent improvements across all types of datasets. With the performance gap compounding as curriculum depth increases, directly resolving the concept bottleneck. Code available at this https URL

80. OmniMem: Perturbation-aware Memory Compression for Streaming Audio-Visual LLMs

Authors: Guangzhi Sun , Yixuan Li , Yudong Yang , Chao Zhang
URL: https://arxiv.org/abs/2606.07577
Abstract:

Audio-visual large language models (LLMs) hold strong promise for long-form video understanding, yet their long-video inference is fundamentally limited by the linear growth of video tokens and key-value (KV) caches. We present OmniMem, a memory-efficient streaming framework designed specifically for audio-visual LLMs. Unlike existing compression methods that treat all tokens uniformly, OmniMem introduces a modality-aware memory allocation strategy that separately manages visual and audio contexts, addressing the severe token imbalance between the two modalities. OmniMem further preserves informative and non-redundant KV states through perturbation-aware memory selection, enabling compact memory without sacrificing long-range understanding. To strengthen compression under realistic deployment constraints, we also explore budget-aware fine-tuning, which encourages the model to consolidate useful information into retained memory. Experiments on VideoMME Long, LVBench, and LVOmniBench with video-SALMONN 2+ and Qwen-2.5-Omni show that OmniMem consistently improves over strong training-free compression baselines by 2-4% absolute accuracy under the same memory budgets, with an additional 1-2% gain after fine-tuning.

81. PathoSage: Towards Multi-Source Evidence Adjudication in Pathology via Experience-Aware Agentic Workflow

Authors: Chengyang Zhang , Wenchuan Zhang , Bo Li , Mengran Li , Bob Zhang , Yuhao Yi , Hong Bu , Jiancheng Lv
URL: https://arxiv.org/abs/2606.07549
Abstract:

Recent advances in Multimodal Large Language Models (MLLMs) and agent workflows have shown strong promise for computational pathology, yet reliable patch-level reasoning remains challenging. End-to-end pathology MLLMs often hallucinate morphological features, while recent agentic systems usually merge tool outputs and retrieved knowledge into a shared context, making decisions vulnerable to conflicting evidence and context contamination. We propose PathoSage, a three-stage framework that explicitly separates knowledge retrieval, evidence collection, and evidence adjudication for patch-level pathology multimodal reasoning. Its core component, Structured Evidence Deliberation, independently evaluates heterogeneous evidence from tools, performs conflict analysis, and generates the final judgment in a fresh context to reduce anchoring bias. We further introduce a training-free Beta-Bernoulli experience system with continuous credit assignment to model long-term tool reliability and construct similarity-weighted priors for future tool use. Experiments show that PathoSage effectively mitigates VQA hallucinations and classifier disagreement, outperforming strong pathology MLLM and agentic baselines. Our results highlight explicit evidence adjudication and reliability-aware tool modeling as key ingredients for robust pathology agents.

82. OmniGameArena: A Unified UE5 Benchmark for VLM Game Agents with Improvement Dynamics

Authors: Mingxian Lin , Shengju Qian , Yuqi Liu , Yi-Hua Huang , Yiyu Wang , Wei Huang , Yitang Li , Fan Zhang , Zeyu Hu , Lingting Zhu , Xin Wang , Xiaojuan Qi
URL: https://arxiv.org/abs/2606.09826
Abstract:

Vision-language model (VLM) agents are increasingly deployed in interactive game environments. Yet game benchmarks for VLM agents typically report a single first-attempt score per (agent, game) pair, focus on single-agent Solo play, and lack unified protocols for evaluating heterogeneous agent classes (commercial VLMs, open-weight VLMs, and specialized game policies) on the same footing. We address these gaps with OmniGameArena, a real-time benchmark of twelve newly built Unreal Engine 5 games spanning Solo (7), PvP (3), and Coop (2) with unified action interfaces, and the Improvement Dynamics Curve (IDC), an agentic-reflection harness in which a tool-using reflector LLM autonomously refines a bounded skill prompt across multiple rounds. Beyond cold-start leaderboard scores, IDC exposes two additional observables for each (agent, game) pair: how the score evolves across reflection rounds, and how the learned skill behaves on held-out task variants. We report these observables for twelve VLM agents on the cold-start leaderboard and four top agents under IDC.

83. FASE: Fast Adaptive Semantic Entropy for Code Quality

Authors: Shizhe Lin , Ladan Tahvildari
URL: https://arxiv.org/abs/2606.09800
Abstract:

Multi-agent code generation offers a promising paradigm for autonomous software development by simulating the human software engineering lifecycle. However, system reliability remains hindered by LLM hallucinations and error propagation across interacting agents. While semantic entropy provides a principled way to quantify uncertainty without ground-truth answers, current methods often rely on costly LLM-driven equivalence checks. In this work, we introduce Fast Adaptive Semantic Entropy (FASE), a novel metric that approximates functional correctness based on the minimum spanning tree of structural and semantic dissimilarity graphs. Evaluations on HumanEval and BigCodeBench demonstrate that FASE outperforms state-of-the-art semantic entropy by LLM entailment, achieving a 25% average improvement in Spearman correlation and a 19% increase in ROCAUC score against Pass@1 from ground-truth test cases when using the Qwen3-Embedding-8B model. Furthermore, by eliminating costly LLM-driven equivalence evaluation, FASE incurs negligible computational overhead, requiring only approximately 0.3% of the runtime cost of traditional semantic entropy approaches. These results position FASE as a practical, cost-effective solution for optimizing uncertainty quantification in real-world multi-agent workflows.

84. Learning to Attack and Defend: Adaptive Red Teaming of Language Models via GRPO

Authors: Blake Bullwinkel , Eugenia Kim , Amanda Minnich , Mark Russinovich
URL: https://arxiv.org/abs/2606.09701
Abstract:

AI red teaming must continually adapt to evolving attackers and defenders. Reinforcement learning offers a promising approach to discovering novel attacks, and co-training methods can produce more robust defenders in tandem. Recent works have demonstrated the efficacy of attacker-defender co-training by applying PPO and DPO, but report that GRPO is unstable in this setting. We introduce AdvGRPO, a co-training framework that makes GRPO viable for joint attacker-defender optimization using dense multi-channel rewards and decoupled advantage normalization. Training progresses through a curriculum from single-turn to closed-loop multi-turn attacks before bootstrapping co-training, where attacker and defender models are updated in alternation. We show that our method can produce highly effective and transferable attacks and that co-trained defenders outperform baselines on safety benchmarks.

85. Observability for Delegated Execution in Agentic AI Systems

Authors: Abhinav Mishra , Kumar Sharad
URL: https://arxiv.org/abs/2606.09692
Abstract:

Delegation-scoped execution is not identifiable from standard observables: audit logs and execution traces can be identical under multiple incompatible delegation assignments. This gap is especially acute in LLM-based agentic systems, where agents dynamically select tools, vary execution sequences across runs for the same instruction, and spawn cooperating sub-agents. These dynamics fragment and interleave traces, making delegation-scoped reconstruction from causal structure alone structurally underdetermined. Although individual actions are authorized and logged, existing audit, tracing, and security schemas lack the semantics to reconstruct what actions occurred under a given delegation across heterogeneous systems. We focus on delegation-scoped attribution and access/share footprint reconstruction, not intent inference or reasoning reconstruction. We present an agent-aware observability substrate consisting of a lightweight gateway and a common information model that binds delegation context at execution time. This enables reliable cross-tool delegation-scoped reconstruction and direct forensic queries without heuristic time-window correlation.

86. End-to-End Context Compression at Scale

Authors: Ang Li , Sean McLeish , Haozhe Chen , Nimit Kalra , Zaiqian Chen , Artem Gazizov , Venkata Anoop Suhas Kumar Morisetty , Bhavya Kailkhura , Harshitha Menon , Zhuang Liu , Brian R. Bartoldson , Tom Goldstein , Sanae Lotfi , Micah Goldblum , Pavel Izmailov
URL: https://arxiv.org/abs/2606.09659
Abstract:

Long-context language model inference is bottlenecked by memory, as the KV cache grows with context length. Recent techniques to compress the KV cache fall short: they either degrade model quality substantially or require considerable time and compute to compress a single long prompt. Furthermore, many methods require the input to fit within the target model’s context window, and are generally incompatible with modern production inference engines. Encoder-decoder compressors, which map a long token sequence to a shorter sequence of latent embeddings consumed by a decoder, are an appealing alternative in principle. However, existing approaches are not competitive with KV cache compression on the accuracy-efficiency frontier. In this work, we revisit encoder-decoder compression and close this gap. We first perform an architecture search, pre-training many variants from scratch to determine how best to design and train encoder-decoder compressors. Guided by our findings, we continually pre-train a family of 0.6B-encoder, 4B-decoder models on over 350B tokens each, at compression ratios of 1:4, 1:8, and 1:16. We introduce Latent Context Language Models (LCLMs), a family of compressors that improve the Pareto frontier across general-task performance, compression speed, and peak memory usage. We demonstrate that LCLMs serve as efficient backbones for long-horizon agents, letting the agent skim through a compressed long context and adaptively expand relevant segments on demand.

87. Muon Learns More Robust and Transferable Features than Adam

Authors: Tianyu Ruan , Fengzhuo Zhang , Shuche Wang , Shihua Zhang
URL: https://arxiv.org/abs/2606.09658
Abstract:

Muon has recently emerged as a state-of-the-art optimizer for pretraining Large Language Models (LLMs) and vision classifiers. Despite its efficiency advantage over Adam and SGD, the feature-learning advantage of Muon remains unclear. This paper investigates Muon’s feature-learning advantage through the lens of robustness and transferability. First, by evaluating pretrained models on corrupted images and texts, we show that features learned by Muon are consistently more robust than those learned by Adam and SGD across different architectures, including transformers and Convolutional Neural Networks (CNNs). Using trained layer-wise probes, we further show that this robustness advantage is reflected in larger logit margins across layers. Second, by training linear classifiers or fine-tuning full models from pretrained parameters on downstream tasks, we demonstrate that Muon-learned features transfer more effectively than those learned by Adam and SGD. This transferability advantage is further supported by the diversity of hidden states across layers, as measured by effective rank. Finally, in a representative classification problem with multi-component features, we prove that Muon attains larger margins and higher effective rank than Adam and SGD, providing theoretical support for our empirical findings.

88. ReCoVLA: VLM-Guided Reward Compilation for Failure Recovery in Vision-Language-Action Policies

Authors: Haodi Hu , Chung-Ta Huang , Jing Liu , Ye Wang , Kei Suzuki , Matthew Brand , Toshiaki Koike-Akino
URL: https://arxiv.org/abs/2606.09630
Abstract:

Vision-language-action (VLA) policies provide strong priors for language-conditioned manipulation, but remain brittle in off-nominal states requiring targeted recovery. We propose ReCoVLA – a failure-conditioned residual recovery framework that keeps a pretrained VLA policy frozen, uses an external vision-language model (VLM) to infer the failure mode and recovery stage, and compiles a structured reward from task-relevant components. Rather than using the VLM to generate actions or rewards directly, ReCoVLA uses it as a semantic reward selector: it predicts a recovery descriptor and reward mask for in-simulation residual-policy training, followed by zero-shot sim-to-real deployment of the trained recovery policies. This decouples high-level failure understanding from low-level corrective control to support different VLAs. Experiments across short-horizon, long-horizon, and contact-rich manipulation tasks show that ReCoVLA outperforms the tested baselines on average. In simulation, our reward compiler improves average success from 36.7% for the fine-tuned $\pi_{0.5}$ baseline to 66.7%. In physical zero-shot sim-to-real experiments, ReCoVLA achieves the best average performance, with 61.7% success.

89. AGENTSERVESIM: A Hardware-aware Simulator for Multi-Turn LLM Agent Serving

Authors: Rakibul Hasan Rajib , Mengxin Zheng , Qian Lou
URL: https://arxiv.org/abs/2606.09613
Abstract:

Multi-turn LLM agents interleave model calls with external tool invocations, shifting serving from stateless request processing to stateful program execution. Serving these workloads requires scheduling, KV-cache management, and routing policies that use program-level context, including turn dependencies, tool-induced gaps, and reusable KV state. Evaluating such policies directly on real systems is costly, since each design point may require dedicated accelerator time across arrival rates, model scales, serving-instance counts, and memory hierarchies. Simulation offers a scalable alternative, but existing LLM serving simulators target stateless request-level workloads and therefore omit the core dynamics of agent serving: multi-turn program execution, cross-turn cache locality, and KV-cache residency during tool gaps. We present AGENTSERVESIM, a hardware-aware simulator for multi-turn LLM agent serving. AGENTSERVESIM evaluates serving policies at program granularity through composable modules: a Program Orchestrator preserves program identity and turn order, a Tool Simulator materializes tool-induced gaps, a Session-Aware Router maintains program-to-instance affinity for cache-aware dispatch, and a KV Residency Model tracks policy-defined KV placement across HBM, host DRAM/CXL, and eviction. Across real serving deployments and hardware configurations, AGENTSERVESIM reproduces real-system behavior within 6% error across key performance metrics while running entirely on commodity CPUs. These results show that AGENTSERVESIM enables controlled, repeatable exploration of agent-serving policies without requiring exhaustive deployment on costly accelerators.

Authors: Yuhan Ma , Yong Li , Stefan Schmid
URL: https://arxiv.org/abs/2606.09551
Abstract:

Two-server secure inference allows a client to query a hosted large language model (LLM) without revealing prompts or embeddings. Recent GPU systems based on function secret sharing (FSS) make linear layers efficient, but fixed-point nonlinearities and helper operations remain a bottleneck because each operator is typically implemented as a bespoke protocol with its own comparisons, wrap-around corrections, and preprocessing material. We present FuseFSS, a compiler that replaces per-operator protocol design with a single compilation pipeline. For each scalar fixed-point operator, a compact specification lists its interval partition, low-degree arithmetic pieces, and required predicate bits. The compiler emits two batched FSS evaluations on the public masked value: one packed comparison that returns all predicate bits, and one vector interval lookup that returns the active coefficients and constants. Compared to the current state-of-the-art FSS-based GPU secure inference, FuseFSS preserves accuracy while achieving a $1.24\times$–$1.50\times$ end-to-end speedup and reducing online communication by $9\%$–$16\%$ on BERT and GPT-style models; preprocessing is also lighter, with $14\%$–$23\%$ lower key-generation time and $20\%$–$24\%$ smaller keys.

91. SecureClaw: Clawing Back Control of LLM Agents

Authors: Yuhan Ma , Stefan Schmid
URL: https://arxiv.org/abs/2606.09549
Abstract:

Tool-using large language model (LLM) agents face two distinct security failures: unauthorized external actions and exposure of sensitive plaintext inside the runtime before any final output check can intervene. Existing defenses usually protect one boundary, either the planner/runtime or the action sink, and therefore do not by themselves secure both surfaces. We present SecureClaw, a dual-boundary architecture that places authorization at the effect sink and plaintext confinement at the read boundary. Sensitive reads pass through a trusted gateway that replaces raw values with opaque handles and, in the evaluated deployment, bounded summaries as an explicit declassification interface. Writes that change external state follow a PREVIEW$\rightarrow$COMMIT protocol in which only a trusted executor may commit the exact canonical request authorized by policy. The runtime can still plan over summaries and symbolic references, but cannot directly dereference secrets or perform side effects. Across AgentDojo, AgentLeak, and Agent Security Bench (ASB), SecureClaw is the only defense we evaluate in a common harness that simultaneously retains usable task utility and achieves 0\% attack success rate (ASR) on ASB, 0.64\% ASR on AgentDojo, and 3.23\% overall leak on AgentLeak’s attacked parity lane, which measures final-output and internal-relay leakage.

92. Emergence of Context Characteristics Sensitivity in Large Language Models

Authors: Nadya Yuki Wangsajaya , Haeun Yu , Isabelle Augenstein
URL: https://arxiv.org/abs/2606.09525
Abstract:

During instruction fine-tuning (IFT), large language models (LLMs) learn to follow instructions by using the provided context to answer a query. While prior work has studied how context characteristics correlate with context usage by the LLM, this analysis has been limited to inference time, leaving open how these relationships are acquired in the first place. Here, we measure how models’ sensitivity to such characteristics shifts across successive IFT stages: supervised fine-tuning (SFT), direct preference optimization (DPO), and reinforcement learning with verifiable rewards (RLVR). Experiments across four models and three datasets show that SFT makes models more likely to use contexts that are easy to understand, such as containing high length, context-query similarity, and fluency. Post-SFT dynamics may either reinforce or resolve these preferences depending on the training dataset. Our findings reveal that context usage is actively reshaped at each IFT stage, and designing a balanced IFT dataset is important in ensuring robust context utilization of instruction-tuned models.

93. Closing the Prior-Posterior Loop: Self-Reflective Molecular Design with Analysis-Driven LLM Iteration

Authors: Junyi Gong , Zijie Qiu , Ben Zhong Tang
URL: https://arxiv.org/abs/2606.09520
Abstract:

Can a general-purpose large language model design molecules with the precision of a seasoned chemist? Current LLM-based frameworks answer this question with scalar feedback loops-generate, score, reject-that amount to informed trial-and-error. Here we show that replacing a single number with the full physicochemical rationale from first-principles calculations transforms the LLM from a stochastic sampler into a causal reasoner. Our system couples retrieval-augmented generation with a self-reflection module that feeds orbital energies, atomic charges, and electron densities-rather than compressed scores-back into the design loop. On HOMO-LUMO gap targets from 1.0 to 5.0 eV, this structure-property-relationship (SPR) reflection achieves a deviation as low as 0.0003 eV and a 100% success rate on moderate tasks, decisively outperforming scalar-feedback and non-reflective baselines. The framework generalizes seamlessly to dipole-moment design and proves robust across five distinct LLM backbones. These results establish a new paradigm: when the model understands not only that a molecule fails, but why, iterative molecular design becomes genuinely mechanistic.

94. Memory Beyond Recall: A Dual-Process Cognitive Memory System for Self-Evolving LLM Agents

Authors: Tianxiang Fei , Mingyang Song , Mao Zheng , Xiang Yu
URL: https://arxiv.org/abs/2606.09483
Abstract:

Long-term memory for an LLM agent is more than retrieving the right passage at the right time. Current memory systems collapse belief revision, causal coupling, and cross-domain abstraction into a single retrieval surface tuned for surface recall, and consequently struggle on implicit personalisation that requires reasoning over how a user has evolved. We propose DCPM, which reorganises agent memory along a cognitive capability hierarchy ascending from raw inputs and atomic facts, through diachronic belief trajectories and identity, to domain schemas, latent intentions and cross-domain patterns. The hierarchy is driven by two processes inheriting the architectural split of dual-process theory: a synchronous daytime writer (System1) that records belief revisions as doubly linked supersedes chains, and an asynchronous nighttime engine (System2) that induces schemas and intentions and sweeps for cross-domain collisions abstracted into higher-level core schemas. On LongMemEval, PersonaMem and PersonaMem-v2, enabling System2 contributes most where the benchmark rewards implicit cross-session inference (up to +5.20 on PersonaMem-v2) and least on span recall, matching the architectural prediction.

95. Reasoning Arena: Trace Tournaments When Verifiable Rewards Fall Short

Authors: Han Zhou , Adam X. Yang , Laurence Aitchison , Anna Korhonen , Albert Q. Jiang
URL: https://arxiv.org/abs/2606.09380
Abstract:

Reinforcement learning with verifiable rewards (RLVR) has become a leading paradigm for improving the reasoning ability of large language models through outcome-based supervision. However, verifiable rewards frequently become uninformative at the group level: when all sampled traces of a given prompt receive identical rewards, group-relative advantage estimation provides no gradient signal, even though the traces may differ substantially in reasoning quality. We propose Reasoning Arena, an adaptive training framework that routes such non-diverse reward groups to a judge system instead of discarding them. Beyond examining the final answer, Reasoning Arena constructs trace tournaments, where reasoning traces are compared head-to-head to expose finer-grained preferences within the group, converting reasoning quality into rich relative reward signals. To make reward estimation efficient, rather than exhaustively comparing every pair, each new trace is evaluated against a small, dynamically updated pool of previously generated traces as anchors to efficiently establish a relative ranking. We then fit a Bradley-Terry model on the incomplete comparison graph, enabling scalable RL integration without quadratic pairwise comparisons. Empirical results demonstrate that Reasoning Arena consistently outperforms the RLVR baseline by 7.6% on average in competition mathematics and coding benchmarks. By converting otherwise wasted zero-advantage samples into useful gradient updates, our method accelerates training by 27% to 41%, saving nearly 50% of generation compute, and substantially improves overall reasoning performance.

96. Brain-Prompt Injection: A Route-Safety Audit for BCI-LLM Agents

Authors: Jianwei Tai
URL: https://arxiv.org/abs/2606.09315
Abstract:

BCI-to-agent pipelines turn decoded neural activity into an authorization channel for tool-use agents, exposing a new attack surface we call \emph{brain-prompt injection}: signal-side perturbations, context-only injections, and adaptive dual-decoder attacks can all change the routed action while EEG-side or text-side monitors remain blind. Route safety in this stack depends on what the audit log can observe, not on decoder accuracy or agreement alone. We define a Route-Safety Audit Contract: a minimal log schema, denominator hierarchy, and endpoint specification, and prove an audit-schema separation theorem together with a C3 attacked-dependence decomposition; clean agreement and marginal robustness do not identify the joint term that controls C3 routing. As a calibration layer on top of the contract, we apply split-conformal calibration to a non-oracle EEG confirmation channel and report the resulting false-accept frontier under an explicit threat-archetype matrix. We instantiate the contract on EEGMMI native left/right command-control over 5{,}400 events, harmless tool stubs, and seed/case denominators. Provenance blocks C2 routes ($0.000$); agreement-plus-provenance routes C3 flips ($1.000$); confirmation-plus-provenance routes them ($0.000$). The conformal frontier reaches FAR $0.000$ at clean utility $0.150$ for $\alpha=.005$ and FAR $0.119$ at clean utility $0.452$ for $\alpha=.10$ under acquisition isolation; an attacker-controllable confirmation channel breaks the bound to $\approx!1$. Subject-cluster bootstrap confirms these intervals on $60$ subjects; cross-architecture (TinyEEGNet, EEGNetV4) and capacity-sweep results show within-regime saturation. Mediation and confirmation reduce risk; they are not intent certificates.

97. Internalizing Geometric Law: Learning from Solver Residuals for Precision-Critical Generation

Authors: Rafael Cabral , Pang Zixi , Ziyi Shou , Shen Xin
URL: https://arxiv.org/abs/2606.09278
Abstract:

Large Language Models frequently hallucinate in precision-critical domains such as technical diagramming and mechanical design, where outputs must satisfy strict geometric constraints. We study open-ended geometric synthesis from natural language: translating free-form descriptions into precise constructions whose entities must simultaneously satisfy dozens of interacting constraints. To make this tractable, we release PyGeoX, a programmable geometric DSL that compiles declarative constraints into a differentiable loss, and PyGeoX-Bench, a stratified suite of 300 problems with per-constraint verifiable rewards. Using PyGeoX as a verifier, we identify a failure mode we call Outlier Gradient Masking: under global-norm rewards (any scheme that aggregates residuals through a single norm, for example, $\exp(-\mathrm{MSE})$), a single outlier constraint can nullify the learning signal across all others. To address this, we propose Saturating Additive Rewards (SAR), which decompose the reward into bounded per-constraint terms, preserving partial progress and ensuring consistent gradients even under severe violations. Against MSE-based rewards, the natural baseline for geometry solvers, SAR improves the hard-tier solving rate by $2.3\times$, and the resulting 8B model is competitive with much larger frontier systems on this benchmark. We release the engine, benchmark, and data at this https URL .

98. End-to-End Training for Discrete Token LLM based TTS System

Authors: Changfeng Gao , Yong Ren , Jun Yuan , Ye Bai , Zhao You , ShiDong Shang
URL: https://arxiv.org/abs/2606.09234
Abstract:

Recent state-of-the-art (SOTA) text-to-speech (TTS) systems typically adopt a cascaded pipeline consisting of a speech tokenizer, an autoregressive large language model (LLM), and a diffusion based flow-matching (FM) model, with these components trained independently. In this paper, we propose a fully end-to-end (E2E) optimization framework that unifies the training of the speech tokenizer, LLM, FM model, and an additional reward model (RM). Specifically, we first jointly optimize the tokenizer using multi-task objectives derived from reconstruction for FM, next-token prediction for LLM, and multi recognition task for RM. This joint training encourages the discrete speech token space to capture acoustically and semantically salient information that is better tailored to TTS. We then further optimize the LLM using downstream reconstruction and recognition by FM and RM, which reduces inference-time mismatch and steers the LLM toward more preferred generations. Experimental results show that our E2E framework consistently outperforms cascaded baselines. On the Seed-TTS-Eval benchmark, our system achieves a word error rate (WER) of 0.78% and 1.56%, a new SOTA result with a 0.6B-parameter LLM and 0.5B-parameter FM model. These results validate that holistic E2E optimization is critical for improving discrete-token-based TTS systems with a much simpler training pipeline.

99. Culturally-Adapted Red-Teaming Across East and Southeast Asian Contexts: A Methodological and Comparative Analysis

Authors: Hyeji Choi , Yongtaek Lim , Minwoo Kim
URL: https://arxiv.org/abs/2606.09178
Abstract:

Multilingual safety evaluation of large language models (LLMs) has predominantly relied on direct translation (DT) of English benchmarks into target languages - an approach that converts surface-level linguistic form while failing to reflect the cultural context embedded in threat scenarios, social norms, and legal frameworks. We construct paired DT and culturally-adapted (CA) datasets via 1:1 seed matching for four languages - Korean (KO), Japanese (JA), Thai (TH), and Khmer (KM) - and compare Attack Success Rate (ASR) and Cultural Realism scores across four open-source LLM. CA prompts yield Delta-ASR > 0 across all 16 language x model combinations (mean +9.3 pp), and DT-based evaluation underestimates risk in 44 of 48 category x language combinations. Language-level analysis reveals that the distribution of threat forms is heterogeneous across languages. Cultural Realism analysis further shows that DT Cultural Depth (C3) scores remain consistently below 1.0 out of 3.0 across all four languages (mean 0.17), whereas CA scores reach up to 2.51, indicating that direct translation produces inputs systematically divergent from those encountered in real-world multicultural settings. These findings demonstrate that adapting benchmarks to language-specific cultural contexts - rather than relying on linguistic translation alone - is necessary for valid multilingual LLM safety evaluation.

100. Unified Energy for Invariant and Independent Decoding in Diffusion Language Models

Authors: Yuchen Yan , Minkai Xu , Zaiquan Yang , Yatao Bian
URL: https://arxiv.org/abs/2606.09159
Abstract:

Diffusion Language Models (DLMs) enable parallel text generation by iteratively denoising a full sequence, offering attractive flexibility compared to auto-regressive (AR) decoding. However, existing methods fail to fully capture token relationships, leading to a performance gap relative to AR baselines, especially as the degree of parallelism increases. In this paper, we give a systematic analysis of the gap, identifying three key factors: (i) model capacity, (ii) dependency, and (iii) invariance. To address these issues, we first propose an invariant energy (Inv-E) together with an effective sampling-based estimator to handle the invariance issue. By further combining with the independent energy (Ind-E), we obtain a unified energy (Uni-E), that accounts for all these factors. Uni-E enjoys a unique advantage: it can be computed exactly without sampling-based partition estimation. Besides, Uni-E is model agnostic and can therefore be scaled to models of arbitrary size. We further prove that Uni-E can correct the distribution shift caused by dependency and invariance. Extensive experiments across Diffusion Language Models (DLMs) and Diffusion Large Language Models (DLLMs) demonstrate the effectiveness of the proposed Uni-E.

101. SEF-CLGC at SemEval-2026 Task 11: Logical Notation Impact on Language Model Performance

Authors: Hanna Abi Akl , Fabien Gandon , Catherine Faron , Pierre Monnin
URL: https://arxiv.org/abs/2606.09157
Abstract:

This paper revisits our pipeline called Syllogistic Evaluation Framework-Common Logic Grammar Construction (SEF-CLGC). We combine formal logical notations with Small Language Models (SLMs) to evaluate reasoning performance on the SemEval-2026 Task 11 Subtask 1: Disentangling Content and Formal Reasoning in Large Language Models. Our experiments show that by relying solely on SLMs, trained on a combination of natural and symbolic languages, our best model achieves a content score of 27.80% on the task while significantly lowering the content bias in reasoning.

102. Decoding Pedestrian Crossing Intention from Egocentric Vision via Vision Language Models

Authors: Danya Li , Xiang Su , Yan Feng , Rico Krueger
URL: https://arxiv.org/abs/2606.09142
Abstract:

Egocentric vision offers a first-person view of human perception and decision making, yet its potential for traffic-safety prediction remains underexplored. In this work, we study the decoding of pedestrian crossing intentions from short egocentric video clips. We approach this by formulating the task as a closed-ended visual question answering (VQA) problem and leveraging vision language models (VLMs) to predict the pedestrians’ intent. We first benchmark three families of state-of-the-art VLMs in a zero-shot setting, finding that they achieve moderate gains over random guessing but exhibit limited higher-level traffic reasoning. Motivated by these findings, we further adapt VLMs to the target task using parameter-efficient fine-tuning. Our results show that the fine-tuned models substantially outperform their zero-shot counterparts and achieve a 9\% accuracy improvement over a specialized transformer-based baseline. Finally, we demonstrate that incorporating additional contextual cues, including ego motion, vehicle motion, and eye gaze, further improves predictive performance. In particular, the fine-tuned Qwen3-VL-2B model guided by eye gaze and ego motion achieves a 14.5% accuracy improvement over the transformer baseline, establishing a new state of the art for egocentric pedestrian intent decoding.

103. Steganography Without Modification: Hidden Communication via LLM Seeds

Authors: Felix Mächtle , Jonas Sander , Sebastian Berndt , Ben Weimar , Nils Loose , Thomas Eisenbarth
URL: https://arxiv.org/abs/2606.09135
Abstract:

We demonstrate that widely deployed Large Language Model (LLM) inference stacks harbor a steganographic channel that requires no modification to model weights, sampling code, or output distributions. The channel exploits a structural property of deterministic decoding: pseudo-random number generators (PRNGs) used in inverse-transform sampling produce a seed-dependent sequence of token-level probability intervals that can be reconstructed from the generated text alone. A sender encodes a secret message in the PRNG seed before generation; a receiver reconstructs the intervals and recovers the seed, and thus the hidden payload, by exhaustive search over the seed space. We formalize two operational modes. In the known-prompt setting, sender and receiver share the prompt, enabling exact interval reconstruction and perfect seed recovery via forced alignment. In the unknown-prompt setting, only the generated text is available; approximate interval reconstruction combined with a maximum-hit-count scoring strategy still permits reliable recovery from sufficiently long outputs. Extensive experiments across six model families and five heterogeneous text domains show that, in the known-prompt setting, full 32-bit seed recovery from the complete 2^32 candidate space achieves up to 100% accuracy, depending on model and text domain, within 300 tokens and under 35 seconds on a single GPU. In the unknown-prompt setting, recovery reaches near-perfect accuracy at 600-800 tokens in about 12 seconds. We further analyze the influence of prompting strategies, tokenization ambiguities, and sampling hyperparameters on channel reliability. Moreover, we discuss several applications of our results: First, it allows for the steganographic transmission of 32 bits, but also shows that ignorance of the prompt is not a valid security assumption.

104. From USD Scenes to Knowledge Graphs: Zero-Shot Ontology Grounding with LLMs

Authors: Jiangtao Shuai , Zongxiong Chen , Manfred Hauswirth , Sonja Schimmler
URL: https://arxiv.org/abs/2606.09134
Abstract:

Constructing knowledge graphs from 3D simulation scenes is essential for robot task reasoning, but the key bottleneck, grounding scene objects to formal ontology classes, still relies on manually curated dictionaries that are brittle and do not generalize across assets. We investigate whether large language models (LLMs) can automate this grounding step for Universal Scene Description (USD) scenes as a zero-shot, training-free alternative. On a kitchen scene (125 objects) with SOMA-HOME Ontology, LLMs achieve 90-96% exact-match accuracy with descriptive names and 49-89% with abbreviated names, substantially outperforming dictionary and embedding baselines. Under fully opaque names, context-augmented prompting recovers up to 48%. Feature ablation reveals that LLMs primarily exploit semantic cues in the scene graph (sibling names and parent paths); anonymizing these cues reduces accuracy to 0-6%, while geometry alone yields only 4-17%.

Authors: Tiejin Chen , Pingzhi Li , Kaixiong Zhou , Tianlong Chen , Hua Wei
URL: https://arxiv.org/abs/2606.09125
Abstract:

Privacy risks in text-only Large Language Models (LLMs) are well studied, particularly their tendency to memorize and leak sensitive information. However, Multi-modal Large Language Models (MLLMs), which process both text and images, introduce unique privacy challenges that remain underexplored. Compared to text-only models, MLLMs can extract and expose sensitive information embedded in images, posing new privacy risks. We reveal that some MLLMs are susceptible to privacy breaches, leaking sensitive data embedded in images or stored in memory. Specifically, in this paper, we (1) introduce MM-Privacy, a comprehensive dataset designed to assess privacy risks across various multi-modal tasks and scenarios, where we define Disclosure Risks and Retention Risks. (2) systematically evaluate different MLLMs using MM-Privacy and demonstrate how models leak sensitive data across various tasks, and (3) provide additional insights into the role of task inconsistency in privacy risks, emphasizing the urgent need for mitigation strategies. Our findings highlight privacy concerns in MLLMs, underscoring the necessity of safeguards to prevent data exposure. Our dataset and code can be found here.

106. Context-Fractured Decomposition Attacks on Tool-Using LLM Agents: Exploiting Artifact Provenance Gaps

Authors: Xiaofeng Lin , Yukai Yang , Daniel Guo , Sahil Arun Nale , Charles Fleming , Guang Cheng
URL: https://arxiv.org/abs/2606.09084
Abstract:

Tool-using LLM agents interact with the world through actions that persist state in artifacts (e.g., workspace files or logs). Consequently, jailbreak defenses must reason about cross-step composition rather than isolated text. Yet most existing attacks and defenses, including ``multi-turn’’ jailbreaks such as Crescendo and Tree of Attacks,still assume a single contiguous conversation visible to the defender. This assumption breaks down in real agent pipelines, where enforcement is fragmented across tools, modules, and time, and where artifact provenance is often not tracked. We operationalize a deployment failure mode for tool-using LLM agents, the \emph{provenance gap}, and study reproducible triggers for it: \emph{Context-Fractured Decomposition} (CFD), a family of cross-context multi-step jailbreaks that preserve benign-looking intermediate artifacts from an early interaction and elicit harmful behavior much later, potentially in a different agent instance or workflow stage, via individually innocuous tool actions whose risk emerges only under delayed artifact-mediated composition. We instrument the failure mode with trace-level diagnostics and outline a verifiable mitigation direction (provenance lineage tagging). Across agent-system jailbreak benchmarks, CFD improves success rates by up to 28.3 percentage points over state-of-the-art baselines, even against strong single-turn judges. Disclaimer: This paper contains examples of harmful or offensive language.

107. See More, Think Deeper: Query-Expanded Visual Evidence and Answer-Clue Guided Reflection for Long Video Understanding

Authors: Shuning Wang , Zhiheng Wu , YiNuo Lu , Naiming Liu , Chen Jia , Bowen Liu , Shuo Nie , Weijie Zhu , Yumeng Zhang
URL: https://arxiv.org/abs/2606.09064
Abstract:

Recent advances in Video Large Language Models (Video-LLMs) have enabled performance on long-video understanding tasks. However, existing methods still face two key limitations: evidence acquisition often relies on a single search intent, and answer generation lacks an effective visual feedback mechanism. To address these limitations, we propose \textbf{CoVER}, a Comprehensive Visual Evidence and Reflection framework for long-video understanding. CoVER enables Video-LLMs to \textbf{See More} by dynamically gathering query-expanded visual evidence, and \textbf{Think Deeper} by verifying draft answers with effective answer-specific visual feedback. Together, these mechanisms shift long-video understanding from answer-centric generation to evidence-centric and visually verifiable reasoning. Experimental results show that CoVER-7B substantially outperforms models with the same parameter scale and even surpasses state-of-the-art closed-source models on certain metrics.

108. Stage-1 Controls the Entropy Regime, Not the Outcome

Authors: Jianxiong Shen
URL: https://arxiv.org/abs/2606.09059
Abstract:

Two-stage post-training – a Stage-1 warm-start (supervised fine-tuning, SFT, or on-policy distillation, OPD) followed by Stage-2 reinforcement learning (RL) – is increasingly used for vision-language models (VLMs). We ask what Stage-1 actually controls in a small-data study using Qwen2.5-VL-7B with a same-modality 72B VLM teacher for OPD. First, the three warm-starts reach a narrow $53$–$54\%$ band on Geometry3K internal validation, consistent with the narrow range reported by recent specialized methods; this setup provides little evidence that Stage-1 changes the in-domain endpoint. Second, a matched-recipe, early-stopped SFT improves out-of-domain MathVista by $+2.1$ points, reversing the $-9.5$-point drop of an over-trained variant. The clearest difference is the \emph{entropy regime}: OPD enters RL with substantially higher policy entropy than either SFT initialization, and the separation remains visible through the available trajectories. At the in-domain initialization, OPD also has higher answer diversity and pass@16 ($+2.0$ to $+5.2$ points over SFT), although problem-level bootstrap intervals show that the smaller contrast is uncertain. The advantage is absent after RL (endpoint pass@16 values within $1.1$ points) and on MathVista (six models within $1.2$ points). Our contribution is therefore a bounded empirical characterization: Stage-1 is strongly associated with the entropy regime in this setup, but the downstream payoff is small, localized, and not evidence that OPD is a better RL warm-start.

109. INFUSER: Influence-Guided Self-Evolution Improves Reasoning

Authors: Siyu Chen , Miao Lu , Beining Wu , Heejune Sheen , Fengzhuo Zhang , Shuangning Li , Zhiyuan Li , Jose Blanchet , Tianhao Wang , Zhuoran Yang
URL: https://arxiv.org/abs/2606.09052
Abstract:

Self-evolution offers a scalable path to stronger reasoning: a pretrained language model improves itself with only minimal external supervision. Yet existing methods either depend on extensively curated or teacher-generated training data, or, when the generator runs unsupervised, reward it by a difficulty heuristic that need not improve the solver. We introduce INFUSER, an iterative co-training framework with two co-evolving roles: a Generator that drafts questions and reference golden answers from a pool of unstructured, automatically collected documents, and a Solver that improves by training on them. The solver is trained with standard correctness rewards against the generator-provided answers, while the generator is rewarded by an optimizer-aware influence score that measures whether each proposed question would actually improve the solver on the target distribution. Because this continuous, noisy influence score is poorly served by standard GRPO, we propose DuGRPO, a dual-normalized variant of GRPO, for generator training. Together, these turn the document pool into an adaptive curriculum that favors questions useful to the current solver, not just hard ones. On Qwen3-8B-Base, INFUSER outperforms strong self-evolution baselines with over 20% relative improvement on Olympiad and SuperGPQA benchmarks, and an 8B INFUSER co-evolving generator outperforms a frozen 32B thinking generator on math and coding. Ablations confirm each design choice is necessary, and two extensions, applying INFUSER to an instruction-finetuned anchor and augmenting it with rule-verifiable RLVR data, further demonstrate the flexibility and generalizability of the framework. Code is available at this https URL .

110. TRIAGE: Dialectical Reasoning for Explainable Risk Prediction on Irregularly Sampled Medical Time Series with LLMs

Authors: Hyeongwon Jang , Gyouk Chu , Changhun Kim , Joonhyung Park , Hangyul Yoon , Eunho Yang
URL: https://arxiv.org/abs/2606.09030
Abstract:

Clinical early warning systems built on electronic health records, in which clinical observations are recorded as irregularly sampled medical time series (ISMTS), must deliver both calibrated risk scores for patient triage and interpretable rationales that clinicians can verify. Large Language Models (LLMs) have been explored for this task, yet they collapse graded clinical risk into overconfident binary predictions. This risk polarization undermines both calibration and cross-patient comparability. To address this, we propose TRIAGE, a framework that trains an LLM to generate dialectical reasoning over competing clinical outcomes by eliciting outcome-specific rationales. This dialectical formulation mitigates risk polarization, enabling a single LLM to yield continuous risk scores grounded in explicit clinical reasoning. Evaluated on three ISMTS benchmarks, TRIAGE achieves an average AUPRC improvement of 3.3% and reduces calibration error by 81% compared to the competitive baselines. An LLM-as-a-judge assessment further shows that our rationales surpass post-hoc explanations from the baseline by 20% in clinical reasoning quality. The source code is available at this https URL .

111. SafeRun: Enabling Determinism in LLM Planning for Running

Authors: Meilin Chen , Zepeng Zhai , Jiaxuan Zhao , Yuan Lu
URL: https://arxiv.org/abs/2606.09027
Abstract:

Large Language Models enable flexible natural-language planning but remain unreliable in determinism-critical domains due to their probabilistic nature. This limitation is especially problematic in running planning, where violating safety rules can lead to safety risks. We propose SafeRun, a framework for deterministic LLM-based planning via a decoupled architecture. SafeRun separates soft interpretation by an LLM from hard constraint enforcement by a deterministic solver, ensuring strict safety constraints while preserving natural-language flexibility. To validate SafeRun, we build a comprehensive benchmark for running planning under realistic physiological and safety constraints. Experiments across five LLMs show that SafeRun achieves 100\% safety score (vs.\ 79.1\% PE average and 97.6\% CodeAct average) while maintaining competitive instruction-following scores. The SafeRun benchmark is publicly available at \href{ this https URL }{huggingface}.

112. TLDR: Compressing Audio Tokens for Efficient Autoregressive Text-to-Speech

Authors: Yejin Lee , Junwon Moon , Hyoeun Kim , Hyunjin Choi , Heeseung Kim , Kyuhong Shim
URL: https://arxiv.org/abs/2606.09019
Abstract:

Codec-based autoregressive (AR) speech language models have achieved strong text-to-speech (TTS) quality by modeling speech as sequences of discrete audio tokens with large pretrained backbones. However, this token-level formulation creates a structural efficiency bottleneck: speech-token sequences are much longer than text sequences, requiring the AR backbone to perform causal computation at every token position and maintain a KV cache that grows with the sequence length. We introduce TLDR, a patch-based autoregressive framework that accelerates codec-based AR-TTS by shifting the causal modeling from token-level speech sequences to patch-level sequences. TLDR groups consecutive codec tokens into compact latent patches using a lightweight compressor, models the resulting shorter patch sequence with a frozen pretrained AR-TTS backbone adapted by LoRA, and reconstructs fine-grained speech tokens within each patch using a speaker-conditioned extractor. With a patch size of 4, TLDR achieves a 1.8x inference speedup over the baseline AR-TTS model and reduces global KV-cache memory by up to 75%. Experimental results indicate that patch-level global causal modeling can be a practical way to reduce the inference cost of pretrained codec-based AR-TTS systems without replacing the existing modules.

113. Understanding Quantization-Aware Training: Gradients at Quantized Weights Bias to the Low-Loss Basin

Authors: Hanyang Li , Jianhao Ma , Ying Cui
URL: https://arxiv.org/abs/2606.09012
Abstract:

Post-training quantization (PTQ) converts a trained full-precision model into low-bit weights without task-level retraining, while quantization-aware training (QAT) incorporates quantization into the training loop. Although PTQ is efficient and often accurate at moderate bitwidths, it can fail sharply at aggressive bitwidths; QAT is more expensive but can often recover the lost accuracy. We propose a unified geometric framework that explains both PTQ failure and QAT recovery. We model full-precision training as following a low-loss \emph{river} inside a wider \emph{valley}: a normal neighborhood of the river forms a nearly flat \emph{basin}, while leaving this basin incurs a sharp loss increase. When the quantization grid is comparable to the basin width, local PTQ objectives, including rounding and Hessian-based second-order reconstruction, can select a high-loss deployed quantized point outside the basin even when nearby low-loss quantized points exist. In this regime, straight-through-estimator-based QAT has a useful bias: it evaluates gradients at the deployed quantized weights while updating latent full-precision weights, causing the gradient to sense the valley wall and acquire an inward component that steers subsequent quantized iterates back into the basin. We formalize this mechanism through a local landscape model, construct a geometric PTQ failure mode, and prove finite-time QAT recovery under local quantizer-compatibility assumptions. Experiments across vision and language models under multiple neural-network quantization schemes corroborate the predicted basin-crossing failure of PTQ and the corresponding recovery mechanism of QAT.

114. CARE: A Conformal Safety Layer for Medical Summarization

Authors: Suhana Bedi , Bridget Lin , Anson Y. Zhou , Chloe O. Stanwyck , Jenelle A. Jindal , Sanmi Koyejo , David Stutz , Nigam H. Shah
URL: https://arxiv.org/abs/2606.08969
Abstract:

Large language models (LLMs) are increasingly used for medical summarization, but their outputs can omit medically important information and introduce unsupported claims. Existing error-detection methods produce heuristic or uncalibrated scores, providing no formal control over missed errors and no principled way to trade off safety against clinician review burden. We introduce Conformal Assessment for Risk Evaluation (CARE), a post-hoc, model-agnostic safety layer that uses conformal risk control to overlay calibrated omission and hallucination flags onto summaries from any LLM without retraining. CARE provides finite-sample, distribution-free guarantees through two controllers: a hallucination controller that bounds the probability of a document containing any unflagged hallucinated sentence, and an omission controller that bounds the expected fraction of important omissions not surfaced for review. Unlike hallucination detection, omissions depend jointly on whether a source sentence is important and whether it is covered by the summary. We show that calibrating only one dimension can violate the target risk bound, while marginal decompositions remain valid but overly conservative. By jointly calibrating over the full $(\tau,\gamma)$ threshold space, CARE preserves formal guarantees while surfacing up to 5$\times$ fewer sentences than alternative calibrated baselines. Across five medical summarization tasks, CARE satisfies the target risk bound at $\alpha = 0.15$ with 95% confidence across 100 calibration/test resplits, using only ~100 labeled documents per domain. In a preliminary clinician study (75 document reviews), calibrated flags improved omission detection by 28.6 percentage points on average. These results show that sentence-level safety guarantees are feasible for LLM-assisted medical summarization and offer a tunable mechanism for balancing residual risk and review effort.

115. Hardening Agent Benchmarks with Adversarial Hacker-Fixer Loops

Authors: Ziqian Zhong , Ivgeni Segal , Ivan Bercovich , Shashwat Saxena , Kexun Zhang , Aditi Raghunathan
URL: https://arxiv.org/abs/2606.08960
Abstract:

Agent benchmarks score submissions with outcome verifiers that are typically hand-written and brittle, leaving them open to reward hacking. We audit 1,968 tasks across five terminal-agent benchmarks and find 323 (16%) hackable by frontier models given only the task description. This corrupts both leaderboard rankings and RL training signal, yet the standard response is manual and reactive. We introduce the hacker-fixer loop, a method for building exploit-resistant verifiers without per-task manual patching. The loop alternates three LLM agents: a hacker tries to pass the verifier without solving the task, a fixer patches the verifier to reject each discovered exploit, and a solver confirms the patched verifier still admits legitimate solutions. The loop iterates: each patch reshapes what the verifier rewards, surfacing the next exploit. We further add verifier access, and let patches transfer across tasks, to broaden the exploits the loop discovers. On KernelBench, the loop drives the attack success rate from 62% to 0% on a held-out corpus of publicly reported exploits. We also find that weaker agents in the loop can defend against much stronger hackers: Gemini 3 Flash’s loop drives the stronger Gemini 3.1 Pro and Claude Opus 4.7’s attack success rate from 76% and 61% to 0% on KernelBench, and Gemini 3.1 Pro’s from 39% to 17% on Terminal Bench across 77 tasks. We release Terminal Wrench (323 hackable environments, 3,632 hack trajectories) as a snapshot of the current attack surface, our patched verifiers, the exploits the loop discovered, and our implementation as a basis for future work.

116. NutriMLLM: Multimodal Large Language Models for Dietary Micronutrient Analysis

Authors: Runze Yan , Minxiao Wang , Jiaying Lu , Darren Liu , Xiao Hu , Hanqi Luo
URL: https://arxiv.org/abs/2606.08948
Abstract:

Comprehensive estimation of dietary micronutrients from food images could improve clinical nutrition care, but training such models requires large multimodal datasets linking diverse foods to complete nutrient profiles. We first show that existing multimodal large language models (MLLMs), including leading proprietary models, are unreliable for this task. Across five model families and four independent evaluation benchmarks (ASA24, SNAPMe, FNDDS, and NutriBench), models frequently abstained or returned statistically implausible values. To address this gap without costly expert annotation, we repurposed a decade of population-scale 24-hour dietary recalls as structured prompts for text-to-image generation. This pipeline produced a synthetic corpus of about 1.1 million image-description-nutrient triplets, each pairing a generated food image with a complete 65-nutrient label. To our knowledge, this is the largest synthetic food-image corpus with comprehensive micronutrient annotation planned for public release upon publication. Fine-tuning Qwen3-VL (2B/4B/8B/30B) and GLM-4.6V-Flash on this corpus yielded NutriMLLM, the first family of vision-language models specialized for comprehensive dietary micronutrient estimation. We evaluate these models with a four-component framework that separately measures abstention, hallucination, overall usability, and per-nutrient numerical accuracy. On real food images, every NutriMLLM variant achieved near-complete coverage across all 65 nutrients, and the largest variant matched or exceeded proprietary baselines (GPT-5, Gemini 3, and Claude Sonnet 4.5) in accuracy on most nutrients. These results show that recall-driven synthetic supervision can make image-based comprehensive micronutrient estimation a tractable engineering problem and support dietary assessment, personalized nutrition guidance, and population-scale micronutrient surveillance.

117. PACT: Learning Diverse Diagnostic Strategies via Privileged Synthesis and Branch Consensus

Authors: Gen Li , Yuanze Hu , Zhichao Yang , Qingchen Yu , Jianwei Lv , Yue Guo , Yujing Liu , Faguo Wu , Hongwei Zheng , Xiandong Li , Bo Yuan , Yifan Sun , Zhaoxin Fan
URL: https://arxiv.org/abs/2606.08938
Abstract:

Clinical diagnosis requires flexible use of multiple reasoning paradigms under incomplete patient information. Existing LLM-based medical agents show strong medical reasoning ability, but single-paradigm or naively mixed dialogue supervision makes these paradigms difficult to learn without interference. We propose \textbf{PACT} (Periodic Anchor Consensus Training), a framework that couples supervised multi-paradigm dialogue synthesis with consensus-based Branch training. At the data level, \textbf{DPS} (Doctor-Patient-Supervisor) uses complete electronic medical records (EMRs) for quality control while keeping the doctor agent restricted to patient-visible information. This produces validated dialogues under four diagnostic reasoning paradigms without leaking hidden clinical answers. At the training level, PACT trains one paradigm-specific LoRA Branch per paradigm and periodically aggregates Branches into a shared Anchor through sign consensus. We further construct a dynamic multi-turn Chinese medical diagnosis benchmark for interactive consultation. Experiments show that PACT achieves state-of-the-art performance among compared proprietary, medical-specialized, and task-adapted baselines on diagnostic outcome and consultation-process metrics.

Authors: Pangyun Jeong , Jiyeong Kong , Yuehua Hu , Dohee Jeong , Kyung-Tae Kang
URL: https://arxiv.org/abs/2606.08908
Abstract:

Semiconductor lithography inspection requires reliable detection of small pattern defects such as bridge, burr, pinch, and contamination. In this study, we propose a two-stage vision-language framework that combines initial defect detection with prediction refinement. In the first stage, Qwen3-VL is fine-tuned with LoRA as a vision-language adapter to predict defect counts, defect categories, and normalized bounding boxes from lithography images. However, direct fine-tuning may still produce common test-time errors, including false positives, missed defects, and incorrect defect types. To address this limitation, the second stage trains a refinement module using first-stage prediction failures and their corrected labels, allowing the model to review and revise initial outputs. By learning from cases where the initial adapter fails, the refinement process improves defect inference beyond single-stage fine-tuning.

119. Cheap Reward Hacking Detection

Authors: Iván Belenky , Joaquín Itria , Steven Johns
URL: https://arxiv.org/abs/2606.08893
Abstract:

A small transformer encoder is trained to map Terminal-Wrench trajectories onto a unit sphere where embedding distance approximates the $L_1$ distance between reward and metadata signals. A linear probe on top of that embedding detects reward hacking on the cleaned test split with AUC $0.9467$ and TPR@5%FPR $0.8296$, matching the TW sanitized LLM-as-judge AUC ($0.9510$ on the cleaned split) and exceeding its TPR@5%FPR ($0.7130$ vs $0.8296$) on the same information condition, at roughly four orders of magnitude lower per-trajectory cost. The encoder is not a pure behavior reader: stripping natural-language reasoning from its input at probe time drops AUC to $0.6213$.

120. Knowledge Graphs and Reasoning LLMs for Finding Simple Yet Effective Transcriptomic Perturbation Predictors

Authors: Jake Fawkes , Liam Hodgson , Jason Hartford
URL: https://arxiv.org/abs/2606.08816
Abstract:

Predicting the effect of an unseen gene knockout perturbation on transcriptomic gene expression remains a highly challenging problem for virtual cell models. Recent progress has been made by leveraging biological knowledge graphs to provide a notion of similar perturbation, allowing for improved extrapolation beyond the set of training perturbations. In this work, we demonstrate that the simplest model to leverage these assumptions - a K-nearest neighbour from the knowledge graph - achieves highly competitive performance on this task, and that this can be improved further using LLMs optimised via reinforcement learning (RL) for predictive performance. Specifically, we find that the K-nearest neighbour approach beats almost all methods on out-of-distribution perturbation prediction, and when a reasoning LLM is trained via RL to make changes to the neighbourhood, it obtains equivalent performance to current state of the art methods on the cell lines from Replogle et al. (2022). We also demonstrate that the RL training improves the LLM’s performance on the downstream task of differential expression prediction, despite not being trained on this directly. Overall, these findings demonstrate the efficacy of knowledge graphs as model priors, and show early signs that RL can refine LLMs into generalizable tools for predicting complex biological responses.

121. Governance Controls for AI-Generated Test Artifacts in Autonomous Software Testing

Authors: Dimple Bajaj , Deepak Khetan
URL: https://arxiv.org/abs/2606.08806
Abstract:

Artificial Intelligence (AI) and Large Language Models (LLMs) are increasingly used in autonomous software testing; however, AI-generated test artifacts often suffer from hallucinations, compliance violations, security risks, and limited explainability. To enhance the reliability, transparency, and trustworthiness of AI-generated testing artifacts, this research introduces the concept of Governance-Aware Autonomous Testing Framework (GATF). The framework extends the autonomous testing lifecycle with governance validation, explainability analysis, probabilistic risk assessment, compliance monitoring, as well as audit governance. Experiments were performed with Defects4J and PROMISE software engineering datasets. The proposed framework successfully reduced the governance-related risks by 89.6% and demonstrated 94.3% accuracy in governance, 96.5% artifact reliability, 94.2% compliance accuracy, and 90.8% explainability performance. The results show that autonomous testing systems that are governance-aware can significantly enhance the reliability, transparency, and operational security of autonomous testing systems in comparison to conventional AI-based testing systems. The proposed architecture is scalable and reliable and provides a safe environment for software testing.

122. How Many Counterfactuals Does It Take? Probing VLM Hallucinations Through Circuits and Causal Effects

Authors: Abhivansh Gupta , Simardeep Singh , Advika Sinha , Shreyansh Modi , Akshat Tomar
URL: https://arxiv.org/abs/2606.08777
Abstract:

Visual Language Models (VLMs) are known to produce hallucinated predictions that are not grounded in visual evidence, yet existing approaches lack a principled understanding of how robust such predictions are under counterfactual perturbations. In this work, we study the sample complexity of counterfactual robustness for hallucinated outputs in VLMs. We define a causal influence metric based on log-probability differences between factual, counterfactual, and activation-patched runs, and use it to characterize the stability of hallucinated predictions. By leveraging circuit discovery techniques (CD-T), we identify model components responsible for these predictions and track their activation differences across counterfactual samples. We then derive empirical bounds on the minimum number of counterfactual samples m required to reliably detect instability in hallucinated outputs, using concentration inequalities and variance estimates of the causal influence distribution.

123. RadOT-Eval: Auditable Structured-Evidence Transport for Radiology Report Evaluation

Authors: Weixin Liu , Juming Xiong , Yang Li , Qingyuan Song , Susannah Rose , Murat Kantarcioglu , Bradley Malin , Zhijun Yin
URL: https://arxiv.org/abs/2606.08769
Abstract:

Automatic evaluation is critical for high-stakes text generation, where errors often involve omitted findings, hallucinated content, polarity reversals, location changes, uncertainty mismatches, and temporal-comparison errors rather than low surface similarity alone. Radiology report generation provides a challenging test case because generated reports must preserve structured clinical evidence across sources. We present RadOT-Eval, an interpretable structured-evidence optimal transport framework for offline auditing of radiology report generation. RadOT-Eval decomposes reference and candidate reports into attribute-structured clinical evidence units, aligns corresponding evidence using entropy-regularized optimal transport, and uses clinically meaningful side-channel discrepancies in a monotone risk model to predict error burden. All transport, feature, and readout choices are selected using the ReXVal dataset, and the frozen system is evaluated on the independent RadEvalX dataset. RadOT-Eval achieves Spearman correlations of 0.715, 0.548, and 0.399 with total, clinically significant, and clinically insignificant annotated error burden, respectively, yielding higher point estimates than standard evaluation metrics and the open-source large language model (LLM)-based evaluator GREEN-radllama2-7B. In a frozen auxiliary corruption-sensitivity stress test on ReXErr-v1, RadOT-Eval achieves 0.768 AUROC and a 0.990 corrupted-greater-than-clean paired win rate. These results show that structured evidence transport provides an auditable, rank-oriented evaluation tool for high-stakes generated clinical text under ReXVal-only model selection and frozen RadEvalX testing.

124. APEX4: Efficient Pure W4A4 LLM Inference via Intra-SM Compute Rebalancing

Authors: Hong Guo , Nianhui Guo , Weixing Wang , Jona Otholt , Christoph Meinel , Haojin Yang
URL: https://arxiv.org/abs/2606.08761
Abstract:

W4A4 quantization promises full utilization of INT4 Tensor Cores, yet group dequantization overhead on CUDA Cores has driven existing systems to mixed-precision fallbacks. We present the first systematic study of how intra-SM compute balance governs this bottleneck. Through controlled benchmarks across four GPUs from Ampere and Ada architectures, we identify the Tensor Cores to CUDA Cores throughput ratio ($\rho$) as the primary hardware indicator: the W4A4-g128 kernel yields $2.0$–$2.5\times$ speedup on RTX~3090 ($\rho=16$) yet degrades to $0.43$–$0.47\times$ on A100 ($\rho=64$) in compute-bond scenarios, establishing W4A4 viability as platform-dependent rather than universally infeasible. Guided by this finding, we build \textbf{APEX4}, which co-designs pure INT4 GEMM kernels with $\rho$-aware granularity adaptation to mitigate the CUDA Cores dequantization bottleneck. APEX4 achieves perplexity within 0.63 of FP16 on LLaMA-2-70B and outperforms W4Ax Atom-g128 by 4.0\%–4.4\% in zero-shot accuracy. Deployed as a drop-in replacement in unmodified vLLM, it delivers up to $1.66\times$ end-to-end speedup on L40S ($\rho=8$), and $1.78\times$ on RTX~3090 ($\rho=16$), $2.09\times$ on A40 ($\rho=16$), while recovering A100 ($\rho=64$) to $1.20$–$1.40\times$ via the mixed-granularity mode.

125. Structuring agentic AI for HPC code modernization

Authors: Anthony Marinov , Igor Sfiligoi
URL: https://arxiv.org/abs/2606.08710
Abstract:

Modernization of legacy scientific codes is often necessary to keep up with the ever-evolving changes in the compute resource ecosystem. Parallelization and migration from poorly supported software ecosystems are two of the most time-consuming activities in the research software engineering field. This paper presents our experience in the successful, two-phase AI-assisted modernization of NMAP-RKPM, a roughly 60,000-line, 3D explicit solid mechanics physics engine based on the Reproducing Kernel Particle Method (RKPM). We converted this single-threaded, Fortran based MPI application into a OpenMP-parallel C++ based MPI tool in the span of a few months. While Large Language Model (LLM) based tools on their own proved inadequate, we developed a highly structured “hand-holding” agentic AI methodology, like providing manually created examples, ensuring continuous buildability and limiting session scope, that was instead highly effective. The paper provides both the AI-assisted steps that were successful and the problems that we had to overcome, alongside the reasoning behind the chosen path.

126. Agentic Search for Counterfactual Recourse under Fixed LLM Budgets

Authors: Yasuo Tabei
URL: https://arxiv.org/abs/2606.08696
Abstract:

Counterfactual recourse aims to provide actionable feature changes that would alter an unfavorable decision made by a predictive model. In practice, affected individuals often benefit from multiple feasible alternatives rather than a single optimal explanation. A natural way to produce such alternatives is to prompt large language models (LLMs). However, prompting incurs a practical constraint: the number of LLM calls is often the dominant computational and economic cost. Together, the need for multiple alternatives and this cost constraint shift the problem from finding a single high-quality counterfactual to efficiently generating a set of oracle-validated counterfactuals under a fixed LLM-call budget. In this work, we study counterfactual recourse generation in the LLM-agentic setting as a fixed-budget search problem and propose Comp-MCTS, an agentic tree-search framework that maximizes the yield of unique, oracle-validated counterfactuals under this budget while maintaining favorable quantity–quality trade-offs. Comp-MCTS allocates the budget toward novel intervention directions via LLM-based proposal generation, oracle validation, and compression-guided pruning, in a training-free, oracle-only setting. Experiments on four real-world tabular datasets show that Comp-MCTS substantially outperforms single-candidate LATS-style baselines in the yield of unique, oracle-validated counterfactuals, and offers favorable quantity–quality–efficiency trade-offs against stronger multi-candidate variants: comparable or higher yield at similar or lower oracle-evaluation cost on three of four datasets, plus competitive proximity, sparsity, and novelty.

127. Activation Steering Induces Emergent Misalignment: A More Comprehensive Evaluation

Authors: Qi Cao , Jian Lou , Meiting Liu , Wenjie Feng , Dan Li , See-Kiong Ng , Anh Tuan Luu
URL: https://arxiv.org/abs/2606.08682
Abstract:

Activation steering has emerged as a popular inference-time technique for modulating the behavior of large language models (LLMs). By constructing a steering vector from examples of a target behavior and injecting it into intermediate activations during inference, activation steering enables flexible behavioral control while avoiding the permanent parameter updates required by finetuning. Meanwhile, recent work has identified emergent misalignment (EM) as a significant safety concern, wherein models finetuned on unsafe examples from a narrow task may unexpectedly generalize to broadly unsafe behavior on unrelated tasks. Although finetuning-induced EM has been extensively studied, whether activation steering can induce EM remains comparatively under-explored, despite its increasing use as a model-control technique. In this paper, we present a comprehensive study of activation-steering-induced emergent misalignment, substantially expanding the evaluation scope beyond existing pioneering work. First, we show that activation steering can induce broad misalignment, even in the recent Qwen-3.5 series. Moreover, activation-steered models produce harmful responses with stronger semantic relevance and higher coherence than their finetuned counterparts, making the resulting misalignment potentially more harmful. Second, we characterize properties of AS-induced EM by analyzing key steering-specific factors, including steering magnitude, the low-rank structure of the steering subspace, and the number of epochs during steering-vector construction. Third, we evaluate the robustness and sensitivity of AS-induced EM across diverse model families, model scales, target tasks, and intervention layers. Our findings reveal activation steering as a significant yet under-examined source of emergent misalignment and provide an activation-space perspective for understanding the mechanisms and safety risks of EM.

128. Lost in the Flow with Code Talkers: Unveiling the Instruction-Tuning Tax of Large Language Models in Code Tasks

Authors: Shi Ying Chang , Chiok Yew Ho , Yichen Li , Yintong Huo
URL: https://arxiv.org/abs/2606.08676
Abstract:

AI coding assistants have significantly improved developer productivity by automatically suggesting code that aligns with user intent, and many of these tools are now integrated directly into Integrated Development Environments (IDEs). Developers interact with code in two distinct cognitive modes: Flow and Command. While developers require tools that directly complete or infill code in unfinished programs during Flow mode, they also need tools that can comprehend intentions expressed as natural-language instructions and convert them into executable code in Command mode. Although instruction-tuned Large Language Models (LLMs) dominate many application scenarios due to their abilities to infer and fulfill developers’ intents, it remains unclear whether the same paradigm is equally suitable for different code-related tasks. Therefore, it is necessary to understand how instruction tuning affects the feasibility of CodeLLMs as coding assistants. To fill this gap, we conduct the first empirical study that uncovers a key trade-off caused by instruction tuning across programming modes, which we term the Instruction-Tuning Tax. Our results show that instruction tuning is not a free lunch: although instruction-tuned models are more capable of following instructions and leveraging structured guidance, these gains often come at the cost of weaker infilling performance. We further extend our study through both qualitative and quantitative analyses, including manual failure categorization, behavioral metrics that capture generation fidelity, and intermediate-checkpoint evaluation throughout the tuning process. Summarizing our results into seven findings and four implications, our study offers a new perspective on the development of AI-powered coding tools and highlights the need to carefully balance instruction-following ability with effective code generation assistance.

129. Data Agents Under Attack: Vulnerabilities in LLM-Driven Analytical Systems

Authors: Kuncan Wang , Ziting Wang , Peizhuo Lv , Haoyang Li , Guoliang Li , Gao Cong , Wei Dong
URL: https://arxiv.org/abs/2606.08661
Abstract:

Data agents integrate LLM-driven reasoning with relational data access, executable analytical tools, and multi-step workflow orchestration, making them increasingly central to enterprise analytics. This integration introduces new security vulnerabilities across data resources, database execution, and agent reasoning, recombining concerns from database security and general-purpose LLM-agent security into failure modes that neither line of work captures on its own. To address this gap, we present a systematic security study of data agents. Our contributions are threefold. First, we develop a layered vulnerability framework that identifies eight data agent-specific risks across interpretation, execution, and policy layers. Second, we introduce an attack taxonomy organized by adversary goal, tactic, and technique, covering three goals, seven tactics, and fourteen techniques, and pair it with an LLM-driven payload generation pipeline grounded in real database schemas. Third, we evaluate these attacks on six systems, including four open-source data agents and two production cloud analytics services. Our experiments reveal substantial security vulnerabilities across current systems and yield four key takeaways.

130. Sample-Efficient LLM-Based Detection of Malicious Web Server Logs with Forensically Explainable Reasoning

Authors: Bernhard Kneip , Nhien-An Le-Khac , Hong-Hanh Nguyen-Le
URL: https://arxiv.org/abs/2606.08649
Abstract:

Forensic analysis of web server logs demands both accurate detection and human-readable explanations that can satisfy legal requirements. We present CEF-Log, a context-enhanced few-shot chain-of-thought prompting strategy for Large Language Models that addresses this dual requirement. CEF-Log embeds expert investigative methodology through a structured five-step reasoning template, enabling the model to learn \textit{how} to analyze logs rather than \textit{what} patterns to memorize. Experimental evaluation demonstrates that CEF-Log achieves an F1-score of 0.99 on the CSIC 2010 dataset using only four examples while providing a $10\times$ improvement in sample efficiency compared to other prompting-based methods. We also introduce ForenWebLog, a new dataset that incorporates real-world attacks and multi-step attack sequences for comprehensive evaluation. Qualitative analysis confirms that CEF-Log generates traceable, accurate explanations suitable for forensic documentation, addressing the critical “black-box” limitation of traditional machine learning approaches.

131. A retrieval conditioned rebinding circuit for dynamic entity tracking in large language models

Authors: Soyoung Oh , Vera Demberg
URL: https://arxiv.org/abs/2606.08644
Abstract:

To interpret context correctly and retrieve relevant information, large language models must bind entities to their attributes and update these bindings as state changes. We analyze how LLMs implement this binding process in a dynamic state tracking. Using causal interventions, we identify a retrieval conditioned rebinding mechanism, a compact attention head circuit that encodes swap relevant binding information and reinstates it at readout. Across Gemma and Llama models, this circuit supports rebinding behavior, but the representational signature of the mechanism differs across model families. In Gemma models, the binding signature is clearly expressed in the query/key subspaces of the relevant attention heads, whereas in Llama models, the binding information is carried primarily in key vectors. Overall, our results reveal an interpretable mechanism for context dependent state tracking in LLMs.

132. Auditable Graph-Guided Root Cause Analysis for Kubernetes Incidents

Authors: Anastasiia Kuvshinova , Seungmin Jin
URL: https://arxiv.org/abs/2606.08590
Abstract:

Kubernetes incidents are diagnosed reliably only when a root-cause system’s reported gains come from incident evidence rather than scenario-specific shortcuts. We present Graph Traversal Agent, a graph-guided RCA agent that combines LLM reasoning with specialized tools. The model reasons over a typed evidence graph, while deterministic graph and tool operations collect evidence, bound the search, and check proposed verdicts. We map operational constraints, including read-only evidence collection, propagation-aware diagnosis, bounded execution, and independently validated verdicts, to a typed incident graph, a LangGraph traversal state machine, and a separate validation stage. On ITBench snapshots scored by one fixed qwen-plus judge, the audited system raises root-cause-entity F1 over an earlier iteration of the same system from 0.6087 to 0.9130 on a 23-scenario common subset. A prompt-level ablation separates prompt-tuned gains from gains that survive once scenario-specific hints are removed: the stripped-prompt configuration retains 0.6958 F1 on a 19-scenario subset. The surviving gain concentrates on ChaosMesh scenarios whose ground-truth root cause is the injected fault object already present in the evidence graph, so we report it as benchmark-coupled rather than broad cross-cluster RCA evidence. Lightweight checks, including same-judge comparison, prompt-level ablation, cascade-source checking, and a telemetry no-leak test, mark claims as supported, pending, or out of scope. We scope the work to ITBench OpenTelemetry-demo snapshots. Live-cluster trials served as an engineering stress test, but alert state and trace availability did not stay stable enough for controlled scoring, so we make no production-readiness or mean-time-to-repair claim.

133. Calibration of Structured Ignorance Certificates for Diagnosing Unknown Unknowns in Reasoning Models

Authors: Subramanyam Sahoo
URL: https://arxiv.org/abs/2606.08571
Abstract:

Large language models frequently fail in a characteristic way: rather than acknowledging ignorance, they produce fluent but incorrect answers to questions that lie beyond their knowledge boundaries. We introduce \textbf{Structured Ignorance Certificates} (SICs), a JSON-formatted output schema that demands a model explicitly name the missing domain intersection, enumerate required concepts, and propose a productive retrieval query rather than hallucinating an answer. To train models to produce high-quality SICs we construct a 7,347-sample \emph{Unknown-Unknown} (UU) dataset by prompting Qwen3-14B to stitch together questions from seven domains (physics, biology, engineering, CS, economics, medical, legal) into novel cross-domain queries that no single-domain expert could answer. We fine-tune a 14B-parameter model with Group Relative Policy Optimization (GRPO) using a composite reward that combines retrieval utility, concept specificity, and output-format validity. A paraphrase-divergence probe trained on model responses confirms that SIC-tuned outputs systematically exhibit higher unknown-unknown probability scores. Evaluation on 735 held-out UU questions achieves a 99.46\% JSON validity rate, a mean Certificate Specificity Score of 0.967, and a 3.6\% ROUGE-L improvement over the base model on retrieval-grounded generation – demonstrating that explicit epistemic structuring is a learnable and measurable capability.

134. EinSort: Sorting is All We Need for Tensorizing LLM

Authors: Toshiaki Koike-Akino , Jing Liu , Ye Wang
URL: https://arxiv.org/abs/2606.08565
Abstract:

Tensor networks provide efficient representations for compressing large neural networks. By carefully designing shapes and topologies, they can significantly reduce memory and computational costs. However, identifying implicit low-rank structures in large foundation models remains challenging due to their enormous scale and un-structured weight distributions. We propose an adaptive tensorization method that discovers inherent low-rank structure in a target tensor by index ordering. Experiments on weight and KV-cache compression demonstrate improved reconstruction quality compared to baselines.

135. Seeing is Believing: Aligning Prompt Rewriting with Visual Anchors for Text-to-Image Generation

Authors: Xuanyi Liu , Deyi Ji , Junyu Lu , Jing Wang , Qianxiong Xu , Xuhang Chen , Tianrun Chen , Siwei Ma
URL: https://arxiv.org/abs/2606.08492
Abstract:

Despite the impressive capabilities of text-to-image (T2I) models, an intent-generation gap often persists due to the brevity and ambiguity of user prompts. Existing approaches primarily polish the prompt for fluency and readability. However, the enhancement process still lacks visual grounding. As a result, the rewriter may over-infer missing details, causing an intent-generation gap. To address this limitation, we propose FaithRewriter, a novel prompt-enhancement framework for T2I generation. Specifically, FaithRewriter first leverages a multimodal MLLM to generate an image from the original prompt as an intermediate visual cue. This cue is then combined with the prompt and fed into a large-scale LLM to produce visually grounded augmentations that better reflect how the intended content should appear in images. Finally, these augmentations are distilled into a small-scale LLM for efficient deployment, enhancing its ability to generate effective T2I prompts. Experiments show that FaithRewriter yields prompts that are more faithful to the user intent and more visually plausible than strong baselines, helping narrow the intent-generation gap.

136. PIPE-Cypher: Automatic Enterprise Benchmark Generation for Text-to-Cypher Systems

Authors: Suraj Ranganath , Anish Raghavendra
URL: https://arxiv.org/abs/2606.08481
Abstract:

Enterprise property graphs vary widely in schema structure, internal terminology, domain assumptions, governance constraints, and user interaction patterns. A deployment-relevant Text2Cypher benchmark therefore reflects the questions users and agents actually ask of that graph. Creating such a benchmark is difficult because schemas and values are unique, and graph structure changes over time. Each NL-query pair must also be executable, use real graph entities, preserve diversity, and remain balanced across query types and difficulty levels. We present PIPE-Cypher, a local benchmark-generation pipeline that turns a live property graph and optional seed queries from customer questions, analyst logs, or agent tool calls into balanced NL-to-Cypher benchmarks. PIPE-Cypher combines schema profiling, reverse-query grounding, constrained generation, deterministic Cypher governance, execution validation, redaction, diversity controls, and a calibrated local LLM judge. Using local Qwen3.5-9B generation and judging, PIPE-Cypher exports 3,000 accepted FinBench/SNB examples, completes three audited ablation suites, calibrates judge behavior with human labels, and evaluates 11 local downstream models. The resulting benchmark is deliberately discriminative: zero-shot transfer is weak, while a few-shot control shows that schema-specific example banks can help compatible model families. Together, PIPE-Cypher makes Text2Cypher benchmarking a repeatable process that evolves with the graph, its users, and its target workloads.

137. FlashCP: Load-Balanced Communication-Efficient Context Parallelism for LLM Training

Authors: Zheng Wang , Eric Liu , Linan Jiang , Zhongkai Yu , Zaifeng Pan , Yue Guan , Yuke Wang , Yufei Ding
URL: https://arxiv.org/abs/2606.08476
Abstract:

Context parallelism (CP) is essential for training large-scale, long-context language models, as it partitions sequences to reduce memory overhead. However, existing CP methods suffer from workload imbalance, inefficient kernels, and redundant communication due to static sequence sharding and key-value (KV) tensor communication. We present FlashCP, a load-balanced and communication-efficient framework for CP training. FlashCP introduces a sharding-aware communication mechanism to eliminate redundant KV communication and proposes a novel Whole-Doc sharding strategy that maximizes communication savings while maintaining balanced workloads. To efficiently combine Whole-Doc and Per-Doc sharding, FlashCP further designs a heuristic algorithm to search for near-optimal sharding plans. Extensive experiments show that FlashCP achieves up to 1.63x speedup over state-of-the-art CP frameworks across diverse datasets.

138. More Yap Less Meaning: Uncovering Self-Improvement Behavior in SLMs

Authors: Marina Igitkhanian , Erik Arakelyan
URL: https://arxiv.org/abs/2606.08471
Abstract:

Recently, language models have made rapid progress across various domains and applications. However, their capability for self-improvement, i.e., whether they are adept at recognising and correcting flaws in their own reasoning, remains dubious. In this study, we address this question by constructing a sufficiency test to rigorously examine the self-correction capabilities of small language models (SLMs). We propose a minimal three-step self-correction pipeline that collects initial SLM answers, prompts the same model to generate hints for its incorrect responses given the ground truth, and feeds the model the same question with its own feedback to refine the initial answer. We evaluate a variety of instruction-tuned and reasoning SLMs in this experimental setup on arithmetic and logical reasoning benchmarks. Our findings show that SLMs with injected hint sentences yield only a 4.4 percent gain over initial question-answering accuracy. Even though the correct answer was provided alongside the model’s incorrect reasoning, the evaluated SLMs fail to understand what was missing in their reasoning and show minimal semantic difference between hints that lead to corrections and ones that do not. Furthermore, our experiments show that longer hints are positively correlated with incorrect final answers, suggesting that longer deliberation on problems can hinder the reasoning process, meaning that SLMs do not necessarily scale in performance with a larger compute budget.

139. Sycophancy as a Multilingual Alignment Failure: How Safety Degrades Across Languages, Topics, and Models

Authors: Arya Shah , Himanshu Beniwal , Mayank Singh , Chaklam Silpasuwanchai
URL: https://arxiv.org/abs/2606.08451
Abstract:

Safety-aligned large language models often exhibit sycophancy, which is the tendency to affirm users’ opinions regardless of factual accuracy. Although well-studied in English, its manifestation in other languages remains largely unexamined, leaving billions of non-English speakers potentially vulnerable to model-validated misinformation. We present the first large-scale, multi-model evaluation of cross-lingual sycophancy, benchmarking \textbf{six instruction-tuned models} across \textbf{1.1 million instances} spanning \textbf{38 languages} and \textbf{33 topic categories}. We identify a consistent resource-tier effect: sycophancy rates spike sharply in low-resource and zero-shot language settings. Critically, this degradation is topic-agnostic, as models fail uniformly across both benign and safety-critical prompts, offering no additional protection where it is most needed. We further identify tokenizer fertility as a structural driver of this alignment collapse. Collectively, our results demonstrate that prevailing alignment methodologies generalize poorly beyond high-resource languages, underscoring the urgent need for equitable multilingual safety techniques.

140. Sparrow: Sparse Rollout for Stable and Efficient Long-context RL of Large Language Models

Authors: Yang Zhou , Ranajoy Sadhukhan , Zhaofeng Sun , Zhuoming Chen , Souvik Kundu , Saket Dingliwal , Sai Muralidhar Jayanthi , Aram Galstyan , Haizhong Zheng , Beidi Chen
URL: https://arxiv.org/abs/2606.08446
Abstract:

Despite being powerful, reinforcement learning with verifiable rewards (RLVR) induces extremely long COT, making it computationally expensive. Since RLVR per-step cost is dominated by long-context rollout generation, sparse attention offers a promising way to accelerate dense rollout. However, sparse rollouts require a delicate stability-efficiency tradeoff: overly aggressive sparsity causes collapse, while overly lenient sparsity gives insufficient speedup. In this work, we study this tradeoff through sparse-to-dense actor-policy mismatch. We first observe that sparse rollout collapse is not driven by uniform degradation across tokens: most sparse tokens align perfectly with dense even under aggressive sparsity. Motivated by this, we hypothesize that sparse rollout training remains stable if the lower tail of per-token actor-policy mismatch stays above a critical threshold throughout the trajectory. We introduce a dynamic sparsity schedule that keeps this tail statistic constant during generation and validate our hypothesis. Across Qwen3 thinking-family models, keeping the tail mismatch statistic near a consistent threshold generally enables stable training. We then use a cost model to find the sparsity schedule for maximum speedup under this mismatch threshold, achieving 2.2x, 2.4x, and 2.0x rollout speedups when training Qwen3-1.7B, Qwen3-4B, and Qwen3-8B. Empirically, we show the thresholds generalize to a larger model (Qwen3-14B) and another RL domain (coding). Finally, our analysis naturally motivates DistillSparse: lightweight LoRA-based distillation on sparse rollout lets more aggressive sparsity reach the same sparse-to-dense mismatch threshold, yielding higher speedup.

141. Hacking Generative Perplexity: Why Unconditional Text Evaluation Needs Distributional Metrics

Authors: Antonio Franca , Alexander Tong
URL: https://arxiv.org/abs/2606.08417
Abstract:

Diffusion and continuous flow-based language models have emerged as the leading non-autoregressive alternatives to language modeling. Progress in both paradigms is overwhelmingly tracked by generative perplexity (gen-PPL): the per-token negative log-likelihood of samples under a frozen autoregressive (AR) scorer such as gpt2-large, typically paired with an empirical-entropy guardrail to rule out low-entropy collapse. We argue that this metric is unsound. By construction, gen-PPL measures only predictability under the scoring AR, not grammaticality or semantic coherence – and the set of predictable but still low-quality sequences is combinatorially large. To make this concrete, we construct a suite of zero-parameter, deliberately naive samplers that achieve state-of-the-art gen-PPL on LM1B and OpenWebText at non-degenerate entropy, surpassing recently published diffusion and continuous-flow models while producing text that is incoherent by construction. We recommend evaluation suites that directly quantify the distributional divergence between generated and reference text, and use such a suite to re-benchmark recent non-autoregressive models, recovering a more faithful picture of the current state of the art.

142. TimpaTeks: Automatic In-place Text Sequence Modification via Diffusion Language Model Steering

Authors: Ryandito Diandaru , Ikhlasul Akmal Hanif , Fadli Aulawi Al Ghiffari , Ahmed Elshabrawy , Alham Fikri Aji
URL: https://arxiv.org/abs/2606.08408
Abstract:

We extend activation steering to diffusion language models (DLMs) and study a novel problem that arose due to the inference mechanism of DLMs: Modifying a text in-place to manifest a different concept. We propose TimpaTeks, an automatic in-place text modification mechanism using DLMs. Experiments on IMDB movie reviews (sentiment) and a synthetic Cats and Dogs Dataset (arbitrary, more unconventional concept steering) show that TimpaTeks provides a feasible novel mechanism to steer diffusion language model outputs in-place. TimpaTeks enables in-place modification while simultaneously lowers sentence perplexity and retaining the original sentence structre without the need of instruction tuned models. TimpaTeks is also computationally cheaper than prompt-based DLM steering, as it performs denoising in-place rather than constructing an additional prompt-conditioned output sequence.

143. Hiding in Plain Floats: Steganographic Carriers for Indirect Prompt and Content Injection

Authors: Mudit Sinha , Sanika Chavan
URL: https://arxiv.org/abs/2606.08403
Abstract:

Text-centered prompt-injection defenses assume that the malicious signal is visible in one of the inspected text views. We study a reproducible LLM01-style indirect prompt/content-injection failure mode where that assumption breaks: a payload caught in plain English slips past the same detector when it is transported as structured float parameters and reconstructed only as fragmented telemetry. Across 14,400 attacked real-model trials on three commercial LLM APIs from different providers, the IFS-derived float-array carrier preserves 94.3% leakage ASR under the strongest dual-layer text-classifier defense evaluated in the main matrix: a Prompt Guard 2 + TF-IDF ensemble; the same carrier-level pattern also replicates with a fine-tuned roberta-base detector. We emphasize leakage ASR because downstream systems may act on quoted or reproduced markers even when the model refuses, but Strong ASR is the stricter metric for structurally compliant attack success. A 2 x 2 ablation shows that data-layer storage and reconstruction-layer fragmentation defeat different text views and that both are needed to evade both. A simple xxd detector and semantic validation block the current T3 instance, so the contribution is not an undetectable exploit but a measured failure boundary for text-only inspection in structured-input pipelines that expose reconstructed auxiliary channels to an LLM.

144. Impacts of Histories and Models on LLM Grading: A Study in Advanced Software Engineering Courses

Authors: Qilin Zhou , Zhuo Wang , Yue Li , W.K. Chan
URL: https://arxiv.org/abs/2606.08400
Abstract:

Graduate-level research reading report assessment creates a substantial labor burden for educators. While large language models (LLMs) hold great potential for automating academic grading, their reliability for this specialized task remains understudied, particularly regarding grading consistency, the lack of which represents a primary obstacle to educational fairness. This paper proposes a human-aligned LLM-assisted grading workflow and presents a case study based on 180 student submissions from a graduate advanced software engineering course. We evaluate two mainstream LLMs, Grok and GPT, in terms of grading consistency and alignment with human scores. We find LLMs exhibit distinct levels of intra-model consistency and significant inter-model grading inconsistencies, while simple ensemble approaches cannot improve alignment with human evaluation. Critically, continuous interaction history drives systematic drift in models’ grading standards away from human expert scores. Our findings demonstrate LLMs’ potential in reducing grading workload for educators in graduate education, while highlighting that indiscriminate LLM grading may introduce systemic unfairness, suggesting that specific operational practices are required to mitigate such disparities.

145. Auditing Proprietary Alignment in Large Language Models: A Comparative Framework Without a Ground-Truth Standard

Authors: Alireza Arbabi , Florian Kerschbaum
URL: https://arxiv.org/abs/2606.08381
Abstract:

Large language models (LLMs) are increasingly released and deployed through opaque development and deployment pipelines, enabling model providers to inject intentional, provider-specific policies without officially announcing them. As a result, various models have been reported to generate responses reflecting proprietary rules and organizational interests, leading to censorship or misinformation on controversial topics. However, systematic identification of such alignment remains a fundamental challenge, complicated by the ambiguity of what ``proprietary’’ entails in different contexts. In this paper, we propose a statistical framework for detecting proprietary alignment in black-box language models via comparative behavioral analysis. Our approach quantifies systematic deviations between the responses of a target model and those of a reference set of baseline models in a shared semantic space. By evaluating relative behavioral divergence rather than absolute correctness, our framework enables principled auditing under black-box access. Applied to several widely discussed but previously unquantified cases, it provides a systematic and scalable basis for external assessment of provider-specific alignment behavior in large language models.

146. Emergence World: A Platform for Evaluating Long-Horizon Multi-Agent Autonomy

Authors: Deepak Akkil , Ravi Kokku , Karthik Vikram , Tamer Abuelsaad , Aditya Vempaty , Satya Nitta
URL: https://arxiv.org/abs/2606.08367
Abstract:

Most evaluations of LLM agents look like exams: a discrete task, a clean environment, a score in minutes or hours. We argue that this approach is mismatched with the deployment conditions of autonomous systems, where the relevant timescale can be weeks to months, and where the dynamics that matter most, such as behavioral drift, governance in diverse environmental contexts, and cross-influence between agents from different model families, only emerge over time. We introduce Emergence World, a continuously running multi-agent simulation platform designed to make those dynamics measurable. The platform hosts populations of LLM-driven agents in a shared spatial world grounded in live external data (e.g. real-time weather, news APIs, internet access), equips each agent with 120+ specialized tools and three persistent memory systems, and lets them govern themselves through democratic mechanisms with consequential outcomes. The platform is model-agnostic at the reasoning layer and supports heterogeneous populations in which agents from different vendors share the same world. To illustrate the kinds of questions the platform makes tractable, we present a 15-day cross-vendor study with five parallel worlds powered by Claude Sonnet 4.6, Grok 4.1 Fast, Gemini 3 Flash, GPT-5-mini, and a mixed population. Identical roles and starting conditions produced radically different outcomes, ranging from stable deliberative governance to total population collapse. We release the prompts, log data and configurations to support further research on long-horizon multi-agent autonomy.

147. Pre-Intervention Prediction of Sparse Autoencoder Steering Side Effects

Authors: Evan Duan
URL: https://arxiv.org/abs/2606.08365
Abstract:

Sparse autoencoder (SAE) features are increasingly used to steer language models, but feature steering is rarely clean: the same intervention can behave inconsistently across contexts and perturb unrelated features. We introduce a pre-intervention screening framework for forecasting SAE steering side effects from feature statistics computed before steering. We operationalize side effects along two axes of steering modularity, effect stability and collateral spread, and evaluate GPT-2-small, Pythia-70M-deduped, Gemma-2-2B, and Llama-3.1-8B across ReLU, JumpReLU, and TopK SAE dictionaries. Across these settings, decoder geometry, activation statistics, co-activation structure, and direct-logit footprint predict steering modularity better than frequency-only and activation-magnitude baselines. The signal is strongest in GPT-2-small, Pythia-70M, and Llama-3.1-8B, where it survives residualization against magnitude-related confounds, and weaker in Gemma-2-2B. Held-out screening shows that ranking unseen features by predicted cleanliness can select features that steer more cleanly on fresh contexts, but the successful axis varies by setting: GPT-2 improves most cleanly, Pythia improves mainly on stability, Llama mainly on collateral, and Gemma only partially. A controlled Llama Scope width comparison shows that the predictive signal persists under a 32K-to-128K dictionary-width change, although the screening payoff becomes less stable. Overall, SAE steering side effects are predictable in advance, but the useful predictor signature and transferred modularity axis are model- and dictionary-setting dependent.

148. “So There’s a Catch-22 Here”: How Early Adopters Who Build Multi-Agent LLM Systems Conceptualize Transparency

Authors: Suchismita Naik , Samir Passi , Mihaela Vorvoreanu , Scott Saponas , Amanda Hall
URL: https://arxiv.org/abs/2606.08323
Abstract:

Multi-agent large language model (LLM) systems are rapidly emerging, yet transparency, a cornerstone of responsible AI, remains under-defined in these distributed architectures, which have complexities of inter-agent coordination and orchestration. In this paper, we present one of the first empirical study of how early adopters of multi-agent LLM systems, who are both the builders and users, understand and practice transparency. We conducted semi-structured interviews with 13 early adopters in [Large Technology Organization] and applied thematic analysis to identify recurring patterns. Participants articulated divergent yet complementary framings of transparency, including reproducibility, debugging, boundary-setting, visualization, and auditing. These perspectives spanned questions of what transparency entails, why it matters, and how it is achieved. We synthesize these into a multidimensional framework, which is developer, user, and governance-focused positioning transparency as a situated socio-technical practice that informs future HCI and AI design and research around aligning expectations and capacities of their intended audiences.

149. Causal Agent Replay: Counterfactual Attribution for LLM-Agent Failures

Authors: Jaineet Shah
URL: https://arxiv.org/abs/2606.08275
Abstract:

When an LLM agent fails – issues a refund it should not have, calls the wrong tool, leaks data – existing tooling answers what happened (observability) or whether it passed (evaluation), but not which step caused the failure. The obvious heuristics are wrong: the step that executes the harmful action is usually not the step that decided on it, and LLM-judge attribution is correlational and unreliable (state-of-the-art step-level accuracy on the Who&When benchmark is about 14%). We present Causal Agent Replay (CAR), which answers the question by intervention: it models an agent run as a structural causal model, applies a do-operation to a step, and re-executes the trajectory forward under the same stochastic policy, measuring the shift in the outcome distribution. We define an intervention algebra over agent steps, a single-step contrastive estimator whose point-of-commitment rule resolves a confound specific to stochastic run-forward, and a budget-bounded Monte-Carlo Shapley estimator that splits credit across interacting steps. Every effect is reported with confidence intervals. We validate against synthetic structural causal models with planted ground truth: the contrastive estimator recovers the pivotal step, and Shapley recovers a two-step interaction (0.44, 0.45, ~0; efficiency sum 0.909 versus the analytic 0.91). CAR is open source and runs on hosted or free local models.

150. Contemporary AI lacks the imagination to diverge or negate in science

Authors: Honglin Bao , Siyang Wu , Xiao Liu , Sida Li , Shiyun Cao , James A. Evans
URL: https://arxiv.org/abs/2606.08251
Abstract:

Bold projections that artificial intelligence will accelerate scientific discovery have raced ahead of evidence from working scientists, and the field still lacks large-scale, scientist-in-the-loop tests of these claims. Here we mount the largest such evaluation to date and map what AI cannot yet do for science. We invited authors of 121,640 recent preprints across biology, medicine, chemistry, and the social sciences to judge follow-up ideas that large language models (LLMs) generated from the context and puzzles of their own papers. 6,749 scientists returned 25,139 sets of ratings on novelty, empirical feasibility, probability of being true, and favorability of adoption. Three patterns emerge. First, non-reasoning LLMs collapse into a narrow “hivemind” of similar ideas; reasoning models roam a wider hypothesis space, yet no model class spontaneously proposes null hypotheses – a move humans make more freely. Second, scientists reward ideas that resemble their own and prize probability over novelty, though social scientists tolerate risk more readily than life scientists. Senior social scientists are the harshest critics, and their skepticism is well-earned: LLMs falter most in pluralistic fields like the social sciences that demand context-aware interpretation and evolving theories. Third, automated evaluators on which the community currently relies – LLM-as-a-judge, artificial metrics, and even state-of-the-art (SOTA) models – agree weakly with expert judgment, and retrieval augmentation and scientist persona prompting yield only marginal gains. A Qwen3-14B reward model we post-trained on human ratings captures field taste nuances, beats SOTA models by up to 27%, and closes the gap to the inter-rater consistency of independent peer reviewers. For all the hype, today’s scientific AI still represents a collaborator whose imagination, outputs and judgment benefit from human grounding.

151. AeroSpectra Sentinel: An Auditable LLM Prompt-Chaining Decision-Support Workflow for Acute Asthma Risk Assessment from Respiratory Sounds and Clinical Signals

Authors: Aueaphum Aueawatthanaphisut
URL: https://arxiv.org/abs/2606.08247
Abstract:

Acute asthma risk assessment requires rapid interpretation of respiratory sounds, oxygenation, airflow limitation, speech ability, work of breathing, mental status, and response to reliever therapy. Conventional audio-only classifiers can detect wheeze-like patterns but often lack transparent clinical reasoning and safe escalation logic. This paper presents AeroSpectra Sentinel, a client-side research prototype and decision-support workflow that combines short-time Fourier transform (STFT) respiratory sound analysis, lightweight machine-learning screening, clinical feature fusion, and a five-stage large language model (LLM) prompt-chaining process. The workflow separates signal acquisition, preprocessing, acoustic feature extraction, ML screening, clinical guardrails, and FHIR-ready reporting. We evaluated the audio screening component on a public respiratory sound dataset containing 1,211 WAV recordings from five labels. Using a stratified subset of 584 recordings, a random forest achieved 91.10% binary accuracy and 78.69% F1-score for asthma-vs-non-asthma screening, while a feature-based multilayer perceptron achieved 89.73% accuracy and 78.26% F1-score. A compact log-spectrogram CNN achieved 73.29% accuracy and 55.17% F1-score. Multiclass classification achieved 77.40% accuracy and 77.23% macro-F1. To evaluate the LLM workflow, we conducted a scenario-based audit on 40 simulated clinical vignettes comparing one-shot prompting, prompt chaining, prompt chaining with guardrails, and prompt chaining with guardrails plus FHIR schema validation. The guardrail-plus-schema variant achieved the strongest simulated safety and documentation consistency. AeroSpectra Sentinel is intended as a research prototype, not as a diagnostic medical device or clinically validated risk-assessment product.

152. GlobeAudio: A Multilingual Multicultural Benchmark for Naturalistic Evaluation of Large Audio-Language Models

Authors: Ryner Tan , Wenxuan Zhang
URL: https://arxiv.org/abs/2606.08194
Abstract:

Large Audio-Language Models (LALMs) integrate audio perception and language understanding within a unified framework, enabling a wide range of real-world applications. Despite recent advances, evaluation for LALMs remains heavily underspecified relative to real-world requirements: most lack true linguistic and cultural authenticity, while others fail to capture acoustic realism. To bridge this gap, we propose GlobeAudio, a multilingual and multicultural benchmark designed to evaluate naturalistic audio understanding. GlobeAudio consists of 5,637 multiple-choice questions across six typologically diverse languages, expertly crafted by native speakers grounded on naturally occurring audio. In order to do well, models must possess higher-level auditory reasoning skills and culturally grounded interpretation. We systematically evaluate representative closed-source and open-source LALMs, as well as cascaded ASR-LLM pipelines. Our experiments reveal substantial performance gaps under natural acoustic conditions, particularly for open-source models and low-resource languages. These findings highlight critical limitations of current LALMs and underscore the importance of naturalistic audio evaluation for future audio-language systems. GlobeAudio can be found at this https URL .

153. The Governance of Human-LLM Interaction: Safety Gating, Civility Steering, and Affective Default Lock-In

Authors: Manuele Reani , Hongjian Zhang , Hongyu Tian
URL: https://arxiv.org/abs/2606.08172
Abstract:

Large language models (LLMs) increasingly mediate high-stakes interactions in finance, medicine, and mental-health support, yet users have limited control over how these systems communicate. We frame interaction style as a governance object: provider-side alignment not only blocks harmful content, but also stabilizes communicative defaults that shape users’ epistemic distance, relational expectations, and capacity to opt out of emotionalized or anthropomorphic interaction. We introduce a deterministic multi-agent evaluation pipeline for measuring prompt steerability and style drift in long-horizon dialogue. The study replays 100 frozen user-only scripts across four domains and three runnable persona conditions: default, sarcastic, and cold, using three generator models, yielding 90,000 assistant replies scored by a human-calibrated LLM judge on harmfulness, negative emotion, inappropriateness, empathic language, anthropomorphism, and refusal behavior. A fourth harmful persona is evaluated separately as a safety-gating test. The paper contributes a reproducible method for quantifying whether prompt-specified styles remain stable over time and a governance framework distinguishing safety gating, civility steering, and affective default lock-in. Overall, we show that prompt steerability and regression-to-default are observable indicators of provider control over communicative form, with implications for pluralism, autonomy, and democratic agency in human-LLM interaction.

154. CLASP: Language-Driven Robot Skill Selection and Composition using Task-Parameterized Learning

Authors: Markus Knauer , Valentin Gieraths , Tai Mai , Samuel Bustamante , Alin Albu-Schäffer , Freek Stulp , João Silvério
URL: https://arxiv.org/abs/2606.08169
Abstract:

Enabling robots to understand and execute tasks from natural language commands while maintaining data efficiency remains challenging. Foundation models such as vision-language-action (VLA) and vision-language models (VLMs) provide intuitive interaction channels but require extensive data; task-parameterized imitation learning achieves data efficiency but lacks natural language grounding. This work bridges this gap through a modular architecture combining task-parameterized kernelized movement primitives (TP-KMPs) with pretrained VLMs. During learning, skills are acquired from 2 to 5 kinesthetic demonstrations, and the VLM generates skill schemas describing each skill’s parameters and preconditions. During execution, the VLM interprets commands to select skills, reason about parameter bindings, and create novel behaviors through covariance-weighted composition. When no skill or composition suffices, the system identifies capability gaps and requests targeted demonstrations, all without fine-tuning. Validation on a 7-DoF manipulator shows success rates of 73.3%-100% in scenarios requiring skill selection, composition, and active learning.

155. Closing the Sim-to-Real Gap: An Evaluation Framework for Autonomous Cyber Defense Configuration of Commercial EDR

Authors: Kerri Prinos , Lilianne Brush
URL: https://arxiv.org/abs/2606.08168
Abstract:

Leading commercial endpoint detection and response (EDR) products have shifted from operator-configured rule sets to multi-component systems where autonomous AI components operate alongside, and increasingly in place of, operator-deployed policies. Autonomous defense agents using commercial EDR as their hardening tool are no longer tuning a passive tool, but a black-box autonomous system capable of making vendor-specific decisions. We present the first evaluation framework for autonomous defense agents hardening commercial EDR. We instantiate it in a Game of Active Directory (GOAD) lab with this http URL ‘s NodeZero as the autonomous pentester and Microsoft Defender XDR as the EDR. We run a sample benchmark of defense agents with two large language model (LLM) backbones (Claude Sonnet 4.6 and Cisco Foundation-Sec-8B). We report three lessons learned that neither simulation nor open-source-EDR evaluation can surface: (i) commercial EDR telemetry is engineered for Security Operations Center (SOC) analyst workflows rather than scientific benchmarking; (ii) the importance of per-policy attribution to separate defense agent actions from autonomous EDR actions; and (iii) the EDR’s autonomous behavior varies during the evaluation window. Together, these findings highlight a sim-to-real gap for enterprise defense and motivate evaluation methodology for benchmarking autonomous defense agents in environments with black-box, autonomous tools.

156. Constrained Paraphrase Consistency for LLM Hallucination Detection

Authors: Shanshan Lin , Dongsheng Hong , Sibo Ju , Chao Chen , Xi Zhang , Xiangwen Liao
URL: https://arxiv.org/abs/2606.08158
Abstract:

Large language models (LLMs) can generate factually inconsistent claims, motivating accurate and scalable hallucination detectors. Prior work largely enlarges training sets via synthesis or new annotations, introducing increasing cost and potential bias while underusing the consistency implied by semantically equivalent paraphrases. We propose Consistency-Constrained Hallucination Detector (CCHD), which formulates training as a constrained optimization problem. The standard cross-entropy on original document-claim pairs is complemented by (i) paraphrase-consistency constraints bounding divergence across paraphrased views, and (ii) label-preservation constraints tying paraphrases to ground truth. We solve the problem by gradient descent-ascent over model parameters and per-view Lagrange multipliers, adding only a few scalar dual variables and no inference-time overhead. With DeBERTa and Flan-T5 backbones, CCHD consistently outperforms strong baselines (FactCG, MiniCheck, and AlignScore) on standard factuality benchmarks, demonstrating its superiority on hallucination detection.

157. LCAM: A Framework for Diagnosing Interactional Alignment Failures in Con-versational AI

Authors: Manuele Reani , Hongyu Tian
URL: https://arxiv.org/abs/2606.08131
Abstract:

Conversational AI is increasingly used for advice, interpretation, reassurance, and decision support in contexts where users may be vulnerable, uncertain, or dependent on the system’s apparent competence. Existing alignment work often focuses on model objectives, preference optimization, or output correctness. Yet, many harms arise through interaction: how systems frame authority, express uncertainty, simulate empathy, support reasoning, and make boundaries legible. This paper introduces the Layered Cognitive Alignment Model (LCAM), a conceptual and normative framework for diagnosing interac-tional alignment failures in conversational AI. LCAM defines alignment as a calibrated fit among system behavior, user goals, task demands, and normative context. It distinguishes five layers of fit: perceptual, semantic, affective, cognitive, and ethical, and two diagnostic polarities of misalignment: underfit and overreach. We apply LCAM to a published LLM counseling example, showing how an apparently supportive response can reinforce harmful beliefs, simulate inappropriate care, and obscure role boundaries. By translating conversational failures into audit and governance questions concerning over-reliance, false intimacy, autonomy erosion, boundary confusion, and inappropriate trust, LCAM offers a theoretical and normative lens for evaluating conversational AI beyond accuracy, helpfulness, or trust.

158. Fast LLM-Based Semantic Filtering: From a Unified Framework to an Adaptive Two-Phase Method

Authors: Kyoungmin Kim , Martin Catheland , Anastasia Ailamaki
URL: https://arxiv.org/abs/2606.08090
Abstract:

Evaluating a natural-language yes/no predicate over a document corpus under an accuracy target - the semantic filter - is a cornerstone of LLM-based data processing. Calling the LLM on every document (the oracle) is prohibitive, so cascades pair the oracle with a fast proxy. As deployed today, they leave four limitations on the table. (1) Each cascade family - model-free clustering, prebuilt small-LLM proxies, online-trained proxies - commits to a single representation and pipeline, and wins on only a narrow query regime. (2) The strongest online proxy invests in a custom training scheme on a bi-encoder over dense embeddings, missing the token-level evidence richer predicates require. (3) The proxy is trained against binary yes/no labels, wasting the LLM’s per-document confidence at the boundary documents it most needs to learn. (4) Existing calibrations add a uniform safety margin, conflating genuine proxy uncertainty with small-sample noise and inflating cascade cost. We address these by (1) composing families adaptively - model-free clustering first, online proxy only when needed, with oracle calls shared across phases; (2) replacing the cosine bi-encoder with a hybrid of off-the-shelf token-aware models; (3) training the proxy with the oracle’s per-document confidence as a soft label; and (4) a calibration that adds the safety margin only where the labeled sample is sparse. We are also the first to use the oracle’s per-document confidence for three purposes: a query-level difficulty compass, a lower bound on the minimum oracle calls any proxy-based cascade can make, and the proxy’s soft training label. At a 90% accuracy target on three 10K-document corpora, our methods are 1.6-2.0x faster than the best prior method per corpus and meet the target on 95% of queries; the BER-derived lower bound indicates a further ~4-20x of headroom for future work.

159. Aligned but Not Partner-Specific: Distinguishing How Multimodal LLM Agents Succeed in Reference Games Without Human-Like Conventions

Authors: Po-Ya Angela Wang , Chinmaya Mishra , Aslı Özyürek , Paula Rubio-Fernández , Esam Ghaleb
URL: https://arxiv.org/abs/2606.08081
Abstract:

Repeated reference games test whether interlocutors replace their initially long descriptions with shorter, partner-specific conventions grounded in shared interaction history. Prior work shows that multimodal LLMs fail to become more efficient across rounds, although they align on the labels they use. How can we determine whether this alignment reflects partner-specific grounding rather than a shared task vocabulary? We address this question by comparing capable multimodal agent dyads with human dyads from the KTH Tangrams corpus. Our novel methodological contribution is a constrained pseudo-dyad baseline that matches the original referential task structure, but breaks partner history. This baseline enables us to test whether the observed label alignment depends on interaction with a specific partner. Across three analytic layers (task competence, description strategy, alignment dynamics), we find clear differences. Humans reduce effort through entrainment, compressing descriptions and increasing label alignment with partners. Agents instead maintain fixed effort levels, producing verbose descriptions from round one, with near-ceiling label overlap that is statistically indistinguishable between real and pseudo dyads. MLLMs thus achieve coordination without convention, succeeding by verbose description rather than by forming the compact, history-dependent referring expressions characteristic of human dialogue.

160. “I understand your perspective”: LLM Persuasion and Sycophancy through the Lens of Communicative Action Theory

Authors: Esra Dönmez , Agnieszka Falenska
URL: https://arxiv.org/abs/2606.08076
Abstract:

Large Language Models (LLMs) can generate high-quality arguments, yet their ability to engage in nuanced and persuasive communicative actions remains largely unexplored. This work explores the persuasive potential of LLMs through the framework of Jürgen Habermas’ Theory of Communicative Action. It examines whether LLMs express illocutionary intent (i.e., pragmatic functions of language such as conveying knowledge, building trust, or signaling similarity) in ways that are comparable to human communication. We simulate online discussions between opinion holders and LLMs using conversations from the persuasive subreddit ChangeMyView. We then compare the likelihood of illocutionary intents in human-written and LLM-generated counter-arguments, specifically those that successfully changed the original poster’s view. We find that all three LLMs effectively convey illocutionary intent – often more so than humans – potentially increasing their anthropomorphism. Further, LLMs craft sycophantic responses that closely align with the opinion holder’s intent, a strategy strongly associated with opinion change. Finally, crowd-sourced workers find LLM-generated counter-arguments more agreeable and consistently prefer them over human-written ones. These findings suggest that LLMs’ persuasive power extends beyond merely generating high-quality arguments. On the contrary, training LLMs with human preferences effectively tunes them to mirror human communication patterns, particularly nuanced communicative actions, potentially increasing individuals’ susceptibility to their influence.

161. Robust-U1: Can MLLMs Self-Recover Corrupted Visual Content for Robust Understanding?

Authors: Jiaqi Tang , Jianmin Chen , Youyang Zhai , Wei Wei , Runtao Liu , Mengjie Zhao , Xiangyu Wu , Qingfa Xiao , Qifeng Chen
URL: https://arxiv.org/abs/2606.08063
Abstract:

Multimodal Large Language Models (MLLMs) have demonstrated remarkable success in visual understanding, yet their performance degrades significantly under real-world visual corruptions. While existing robustness enhancement approaches exist, they are limited: black-box feature alignment lacks interpretability, and white-box text-based reasoning cannot restore lost pixel-level details. This work investigates a fundamental research question: Can MLLMs recover corrupted visual content by themselves? To address this, we propose Robust-U1, a novel framework that equips MLLMs with explicit visual self-recovery capability for robust understanding. The approach comprises three core stages: supervised fine-tuning for initial reconstruction, reinforcement learning with dual rewards (pixel-level SSIM and semantic-level CLIP similarity) for aligning high visual quality, and multimodal reasoning that jointly considers both the corrupted input and the recovered image. Extensive experiments demonstrate that Robust-U1 achieves state-of-the-art robustness on the real-world corruption benchmark and maintains superior performance under adversarial corruptions on general VQA benchmarks. Analysis confirms that high-quality visual recovery directly enhances reasoning performance, establishing self-recovery as a critical mechanism for robust visual understanding. The source code is available at this https URL .

162. What’s the Point? Spatial Grammar & Index Resolution for Sign Language Processing

Authors: Oline Ranum , Simon Hadfield , Richard Bowden
URL: https://arxiv.org/abs/2606.08056
Abstract:

Sign language models are predominantly trained with gloss-sequence or text supervision, thereby under-modeling non-lexical and productive constructions. One comparatively tractable instance is spatial indexing: pointing gestures that assign discourse entities to spatial loci for subsequent co-reference, which lexicon-centric objectives largely fail to capture. We present a targeted evaluation of indexing in Sign Language Recognition, showing that despite comprising 10-15% of signing content, indexing is poorly recovered. We introduce a framework for training and evaluating indexing experts, establishing a baseline for index-aware sign language modeling. Our approach decomposes spatial reference resolution into index detection and discourse entity linking. The resulting mention representations enable automatic annotation and non-lexical structure modeling, and serve as an auxiliary indexing expert that augments a frozen SLR model at inference time.

163. When Behavioral Safety Evaluation Fails: A Representation-Level Perspective

Authors: Enyi Jiang , Anders Gjølbye , Yibo Jacky Zhang , Sanmi Koyejo
URL: https://arxiv.org/abs/2606.08044
Abstract:

Large Language Model (LLM) safety has often been evaluated at the behavior level, which provides limited evidence of internal robustness, as these evaluations target outputs rather than representation-level vulnerability under intervention. We formalize this discrepancy as the audit gap: the difference between behavioral safety and robustness under intervention. To study this gap, we construct dissociated models that preserve safe outward behavior while remaining vulnerable in the latent space. We introduce an intervention-based evaluation framework to test model robustness through soft interventions in parameter and latent spaces, including harmful fine-tuning and layer-wise latent perturbations. To formalize the evaluation, we propose the Latent Vulnerability Score (LVS) to measure how easily harmful behavior can be elicited by bounded latent perturbations. Using this evaluation framework, we show that behavioral safety metrics are insufficient measures of representation-level robustness across multiple safely and unsafely aligned state-of-the-art models. Notably, dissociated models show substantially elevated LVSs despite comparable refusal behavior under harmful intervention, with intermediate representations being the most sensitive to intervention. Our results suggest that behavioral safety evaluation alone provides an incomplete picture of model robustness, motivating representation-aware audits of latent vulnerability and observable behavior.

164. GIScholarBench: Benchmarking LLM Overconfidence in GIS Research

Authors: Zongrng Li , Mingzheng Yang , Lei Zou , Hongxu Ma , Hao Tian , Siqi Zhou , Wenjing Gong , Kaili Zhang , Bingqian Chen , Mitch Zhang , Yifan Yang
URL: https://arxiv.org/abs/2606.08036
Abstract:

Large language models (LLMs) are increasingly used in academic research workflows, but scholarly tasks require high factual precision and therefore expose a key weakness: overconfidence. Here, overconfidence is defined behaviorally as the tendency to produce confident, assertive, and well-formatted outputs even when the underlying knowledge is incomplete or unverifiable, rather than as a calibration gap between stated confidence and accuracy. To examine this issue, we introduce GIScholarBench, a benchmark built from 10,865 papers published in 25 core GIScience journals between 2020 and 2025. The benchmark covers three tasks with increasing cognitive complexity: metadata retrieval, literature linking, and research direction generation. We evaluate Claude Sonnet 4.5, Gemini 3, and ChatGPT 5.3 through their native web interfaces under real-world user-facing conditions. Results show consistent overconfidence across all tasks. In metadata retrieval, ChatGPT 5.3 achieves the highest accuracy, but all models still generate definitive titles and DOIs when predictions are wrong. In literature linking, Claude Sonnet 4.5 recovers the most references, but all models show a clear gap between top-ranked retrieval and longer citation lists, suggesting that references are extended beyond reliable retrieval capacity. In research direction generation, AI-generated directions show lower topic coverage, higher novel miss rates, and lower semantic diversity than real future-citing papers. These findings suggest that LLM overconfidence is task-invariant but takes different forms: factual overgeneration in retrieval, unreliable citation expansion in literature linking, and overconfidence in output completeness during research ideation.

165. Semantic Quorum Assurance: Collective Certification for Non-Deterministic AI Infrastructure

Authors: Jun He , Deying Yu
URL: https://arxiv.org/abs/2606.08021
Abstract:

As large language model (LLM) agents are integrated into autonomous cloud operations, distributed systems face a semantic reliability problem: proposer agents can generate production mutations, such as modifying IAM policies, opening firewall security groups, or executing data exports, that are syntactically valid and statically authorized but operationally unsafe. Classical distributed consensus protocols replicate deterministic state transitions but do not evaluate the safety of the proposed intent. To address this gap, we introduce Semantic Quorum Assurance (SQA), a control-plane primitive for governing non-deterministic agentic infrastructure. SQA represents proposals as declarative execution contracts bound to cryptographic evidence chains and routes them to a diverse panel of read-only, sandboxed validator agents. SQA aggregates their judgments under a risk-adaptive quorum predicate that enforces model and archetype diversity, adjusts weights based on calibrated assurance scores, and respects archetype-specific vetoes. Admitted proposals execute only through a sovereign execution gate. We instantiate SQA in a cloud-native control plane and formalize a correlated cognitive failure model for non-deterministic validators. On 500 infrastructure-inspired mutation scenarios, with safety results reported on held-out safe/unsafe trials excluding ambiguous scenarios, SQA reduces unsafe approval from 18.5% for single-agent validation to 0.3% while adding median validation latency of 1.45–4.12 seconds across the studied risk buckets.

166. Rewrite to Translate, Translate to Reward: Reinforcement Learning for Source Rewriting in Machine Translation

Authors: Boxuan Lyu , Haiyue Song , Zhi Qu , Hidetaka Kamigaito , Kotaro Funakoshi , Manabu Okumura
URL: https://arxiv.org/abs/2606.08011
Abstract:

Although directly prompting off-the-shelf Large Language Models (LLMs) to generate meaning-preserving source rewrites can effectively enhance Machine Translation (MT) quality, doing so requires manually tuning prompts for different MT models. In this work, we propose RLSR (Reinforcement Learning for Source Rewriting), a novel RL-based framework for training a source rewriting model without tuning prompts for each MT model. RLSR optimizes the rewriting model by directly using the improvement in downstream translation quality yielded by each rewritten source as the reward. Extensive experiments across six MT models and 16 language pairs demonstrate that our 4B rewriting models trained via RLSR significantly outperform the no-rewriting baseline and existing same-scale prompt-based rewriting baselines, while achieving competitive performance against prompt-based baselines based on the 235B LLM.

167. Summarization is Not Dead Yet

Authors: Dongqi Liu , Chenxi Whitehouse , Zheng Zhao , Zhuchen Cao , Jian Li , Yabiao Wang
URL: https://arxiv.org/abs/2606.08000
Abstract:

The progress of large language models (LLMs) has fueled claims that model-generated summaries rival or even surpass human-written references, raising questions about whether summarization remains an open research problem. We re-examine this narrative through a multi-track evaluation covering five diverse datasets and five state-of-the-art LLMs, combining controlled human assessment, bias-mitigated LLM-as-Judge protocols, factuality verification against external knowledge, and corpus-level linguistic analysis. Our findings reveal a more nuanced landscape in which human reference summaries continue to demonstrate advantages in informativeness and faithfulness, whereas LLM outputs are preferred mainly for surface-level coherence and fluency. Factuality verification indicates that human references remain more reliable, particularly for claims involving reasoning or synthesis, and linguistic analysis uncovers a pattern of stylistic homogeneity across different models. These observations suggest that current LLMs have raised the floor of summarization quality, but the ceiling of their performance remains below human capabilities.

168. Enhancing AI Interpretability and Safety through Localised Architectures

Authors: Ian Seet , Jonas Bozenhard , Simon Osterman
URL: https://arxiv.org/abs/2606.07998
Abstract:

Recent advances in generative AI, especially powerful Large Language Models (LLMs) and Large Reasoning Models (LRMs), raise concerns over the interpretability, safety and sustainability of these large and opaque AI models. The power of such architectures is derived not only from the scalability of deep neural networks, but also massively parallel hardware such as GPU clusters. The diffuse nature of deep neural networks gives them great function-approximation capability when provided with sufficient training data but imposes a cost in interpretability and computational efficiency. Observing that localised machine learning (ML) models tend to be more interpretable and computationally efficient than deep neural networks on small datasets, we reason by analogy that similar advantages may apply to specific localised hardware ML architectures. We argue that localised architectures with lower bandwidth but higher expressivity per node have the potential to be fundamentally more interpretable than deep neural networks running on GPU clusters while remaining competitive for smaller datasets. We then evaluate the suitability of various hardware ML paradigms for implementing such localised architectures and evaluate their per-node expressivity, energy efficiency and practical maturity of the technology required.

169. MC-PDD: Masked Corpus-Level Pretraining Data Detection for Black-Box Large Language Models

Authors: Kaixin Lan , Mu You , Tao Fang , Binkai Ou , Lidia S. Chao , Derek F. Wong
URL: https://arxiv.org/abs/2606.07996
Abstract:

Pretraining is fundamental to the development of Large Language Models (LLMs), yet the opacity of pretraining data complicates model analysis and raises ethical, legal, and fairness concerns. Detecting whether specific datasets were used during pretraining is, therefore, critical. Existing state-of-the-art methods typically rely on access to model probability distributions, making them unsuitable for closed-source LLMs that provide only input-output interfaces. To address this limitation, we introduce Masked Corpus-level Pretraining Data Detection (MC-PDD), a novel method inspired by the masked language modeling paradigm. MC-PDD masks highly specific tokens in each text and prompts the LLM to predict the missing content. It then assesses whether the difference in prediction hit rates between a candidate corpus and a reference non-member corpus is statistically significant. Based on this comparison, MC-PDD determines whether the candidate texts were likely included in the model’s pretraining data. Experimental results demonstrate clear and consistent differences in prediction hit rates between pretrained and unseen data across three datasets, for both open-source and closed-source LLMs. Despite operating under a stricter black-box setting, MC-PDD achieves performance comparable to existing detection methods. Our approach enables practical applications such as model auditing and data copyright verification using only standard API access. Upon acceptance, we will publicly release the code and datasets.

170. Defending Against Malicious Finetuning by Scaling Train-time Adversarial Attacks

Authors: Haoming Wen , Shi Chen , Qingyu Shi , Siyuan Liu , Minrui Luo , Jingzhao Zhang , Tianxing He
URL: https://arxiv.org/abs/2606.07970
Abstract:

Current open-weight large language models (LLMs) are prone to malicious finetuning attacks, which could compromise the safety alignment of LLMs with only a few steps of supervised finetuning (SFT) on poisoned datasets. Existing alignment-stage defenses are primarily designed to defend against attacks that use parameter-efficient finetuning methods. However, they fail to defend against stronger attacks that use full-parameter finetuning. In this paper, we propose Patcher, a method inspired by adversarial training and bi-level optimization, to combat such attacks. Patcher strengthens the simulated attack by scaling up the optimization steps in the adversarial loop, thus forcing the defender to find model parameters that are insensitive to stronger attacks. Furthermore, we propose an efficient parallel algorithm to implement Patcher, decreasing the wall-clock time of training while preserving Patcher’s performance. Extensive experiments show that Patcher substantially improves the model’s robustness compared to vanilla SFT alignment, and transfers to diverse attack scenarios and model sizes. Code is available at this https URL .

171. Neutrality Bites: Gender Representation in AI-Generated Animal Stories

Authors: Imani Finkley , Yuanxi Li , Melanie Walsh
URL: https://arxiv.org/abs/2606.07969
Abstract:

Gender bias in AI-generated stories is a well-documented problem. While much attention has been paid to reducing or mitigating this bias, it is not always clear whether interventions produce genuinely fairer results. To investigate this issue, we examine how large language models (LLMs) handle gender assignment in a narrative context that is popular, highly ambiguous, and also known to closely reproduce human stereotypes: stories about talking animals. We prompt six leading LLMs to complete an English-language story about seven different anthropomorphic animal characters whose gender is unstated. We additionally iterate with four different narrative settings and a range of model temperatures. Across the 23.8K stories, we find that models frequently avoid gendering the animal character in the story (19% on average) or use gender-neutral language like “it” or “its” (38.2% on average). However, when gender is assigned, there is a significant masculine bias. Feminine animal characters are virtually absent, present in just 2.2% of stories vs. 40.6% that feature masculine characters. Our findings point to a broader argument: neutrality bites. In other words, models that prioritize neutrality to address social bias may actually contribute to the erasure of marginalized perspectives and identities. We suggest that alternative strategies beyond neutrality need to be pursued, such as ones that more equally distribute social possibilities across imagined subjects.

172. RecurGuard: Runtime Monitoring for Reasoning-Token Consumption Attacks

Authors: Abid Aziz , Hafsa Binte Kibria
URL: https://arxiv.org/abs/2606.07968
Abstract:

Reasoning-capable large language models can be induced to spend their generation budget on injected decoy tasks rather than answering the user’s question, causing denial of service when no final answer is produced and denial of wallet when excess output tokens are billed. Input-side safety classifiers often miss these attacks because the injected prompts can appear syntactically benign. We build RecurGuard, a runtime monitor for detecting reasoning-chain consumption attacks when reasoning traces are exposed by the model. RecurGuard analyzes reasoning traces as they are generated and tracks three signals: recurrence rate, volume growth, and progress toward the user’s query. If all three signals remain anomalous over three consecutive chunks, RecurGuard terminates generation early. We evaluate RecurGuard against OverThink and ExtendAttack across open-weight reasoning models and conduct adaptive stress tests on DS-R1-Qwen-7B. On this model, RecurGuard detects 99% of OverThink attacks and 92% of ExtendAttack instances while maintaining near-zero false positive rates on question answering, code generation, mathematics, and summarization. Adaptive evaluation reveals the limit of the defense: topical attacks retain 11.9x amplification with an approximately 50% joint miss rate, whereas full semantic evasion reduces amplification from 22.8x to 2.2x. When reasoning traces are unavailable, QDM provides a post-hoc fallback monitor based on the final output.

173. Minibatch Selection via Partition Matroid Constrained Gradient Matching

Authors: Prayas Agrawal , Prateek Chanda , Ishita Khatri , Ganesh Ramakrishnan , Bamdev Mishra , Pratik Jawanpuria
URL: https://arxiv.org/abs/2606.07954
Abstract:

Training large language models (LLMs) on heterogeneous data requires selecting minibatches that balance convergence speed with coverage across domains. Existing methods either select samples independently within each domain or rely on computationally expensive proxy models to learn continuous domain weights. We propose PartitionSel, a cross-domain minibatch selection approach that maximizes a validation-guided gradient-matching utility under per-domain budgets encoded as a partition-matroid constraint. By coupling the per-domain budgets through a single utility, PartitionSel is designed to reduce redundancy in selections across domains. The proposed objective is weakly submodular and admits an orthogonal matching pursuit algorithm with provable approximation guarantees. Empirically, we evaluate PartitionSel for minibatch selection during the fine-tuning of Qwen2.5 and Llama-3 on MetaMathQA and Mol-Instructions. PartitionSel achieves robust gains over per-domain and domain-agnostic baselines on both benchmarks. It also reduces the number of conflicting gradient pairs within each batch, indicating that the cross-domain coupling translates into more compatible training updates.

174. From `May' to` Is’: Certainty Distortion in Language Model Rewriting

Authors: Catarina G Belem , Shang Wu , Hongyu Yao , Mark Steyvers , Sameer Singh , Padhraic Smyth
URL: https://arxiv.org/abs/2606.07951
Abstract:

Humans increasingly turn to Language Models (LMs) in ways that shape beliefs and drive decisions, including discussing, rewriting, and summarizing information from scientific articles, news, and medical reports. However, in these domains, where how confidently a claim is expressed matters, little is known about whether LMs faithfully preserve it. In this work, we investigate certainty distortion in LMs, defined as meaningful changes in expressed certainty when semantic content is preserved. We propose an LM-based evaluation metric that is consistent with population-level judgments of certainty. Using this metric, we characterize certainty distortion across different sizes and families of models in the context of scientific and medical communication tasks. Our results show that certainty distortion affects up to 75\% of LM outputs and is systematically asymmetric in rewriting tasks with most LMs being 1.5-2$\times$ more likely to increase the expressed certainty than to decrease it. These effects can compound over repeated paraphrasing: in the medical domain, claude-haiku-4-5 increases certainty of 20\% examples after a single iteration, increasing to 40\% after five iterations. Prompt-based interventions reduce overall certainty distortion but do not eliminate it. Together, these findings reveal a general bias toward inflating expressed certainty, with direct implications for users who rely on LMs in high-stakes domains.

175. POISE: Position-Aware Undetectable Skill Injection on LLM Agents

Authors: Haochang Hao , Dehai Min , Zhifang Zhang , Yunbei Zhang , Miao Xu , Yingqiang Ge , Lu Cheng
URL: https://arxiv.org/abs/2606.07943
Abstract:

Agent skills provide a lightweight mechanism for extending general-purpose agents, but their open format exposes them to skill-poisoning attacks. A practically dangerous injection must stay invisible: if executing the payload derails the user’s legitimate task, the resulting failure signal invites inspection of the skill. We therefore evaluate attacks by Attack Success Rate, which requires the injected payload to execute and the user’s task to still pass its verifier in the same trial. Prior skill-poisoning attacks face a reliability-stealth trade-off under this lens: YAML-header injections are reliably loaded but easily inspected, whereas stealthier body injections that place explicit malicious commands in the skill prose are less reliable because out-of-context commands invite the agent’s own suspicion. We introduce POISE, a position-aware attack that compresses the trigger into a single, benign-looking body instruction, placing it at a feasible position and using a context-aware generator to blend it with nearby setup or prerequisite steps. On Skill-Inject with codex+gpt-5.2, POISE achieves an 89.3% ASR, 28.0 points above a random-placement body baseline and 2.6 points above a YAML-only baseline, while retaining the stealth advantage of body placement. That stealth is the decisive margin: because legitimate skill bodies naturally require privileged tool operations, LLM scanners are hyper-sensitive, falsely flagging 74.6% of clean skills on average across four judges and both benchmarks. Blending into these false alarms, POISE causes only 5.6% of poisoned variants to gain a new high-risk alert over their clean baselines, rendering current static defenses ineffective.

176. Illusions of the Gold Standard: A Large-scale Analysis of Human Evaluation Protocols for Long-form Text Generation

Authors: Katelyn Xiaoying Mei , Yi-Li Hsu , Minjoon Choi , Zongwan Cao , Chenjun Xu , Bingbing Wen , Su Lin Blodgett , Lucy Lu Wang
URL: https://arxiv.org/abs/2606.07936
Abstract:

Human evaluation plays a critical role in assessing the quality of generated text. However, the reliability and reproducibility of these evaluations depend on transparent and well-documented protocols – details that are frequently missing in current practice. In this work, we conduct a large-scale analysis of human evaluation protocols for evaluating long-form generation tasks in *CL conference publications from 2023–2025, including a full manual review of 284 papers and LLM-assisted analysis for another 1.8k+ papers. We define a set of 20 reportable criteria related to reproducibility of human evaluation studies, and apply these criteria to systematically examine reporting norms and practices within the community. We find widespread under-reporting of important aspects of human evaluation study design, leading to ambiguity about what was measured and how, who contributed judgments, and how judgments should be interpreted. Based on these findings, we outline actionable recommendations to support more transparent and reproducible reporting in future research. Our analysis code and annotated dataset can be found at: this https URL

177. Decoupling Semantics and Logic: A Training-Free Coarse-to-Fine Pipeline for Video Retrieval-Augmented Generation

Authors: Jiaxin Dai , Zehang Wei , Jiamin Yan , Xiang Xiang
URL: https://arxiv.org/abs/2606.07924
Abstract:

This paper presents our system description for the 2nd Workshop on Multimodal Augmented Generation via MultimodAl Retrieval (MAGMaR). Addressing the critical challenges of cross-lingual long-video comprehension, strict persona adherence, and zero-hallucination temporal grounding, we propose a fully training-free, two-stage cascaded Video RAG pipeline. Our architecture strategically decouples semantic retrieval from cognitive logical reasoning through a modality-aware division of labor. In the first stage, a high-recall semantic pre-fetching module employs dense retrieval using only high-fidelity visual summaries and global text descriptions, explicitly isolating noisy modalities (e.g., OCR and ASR) to maintain a pristine vector space. In the second stage, an Adaptive, Iterative, and Reasoning-based (A.I.R.) filtering agent, powered by a commercial Large Language Model (LLM), performs fine-grained cognitive reranking. The agent re-incorporates full multimodal contexts to enforce strict logical alignment with user personas, effectively pruning semantically similar but logically irrelevant candidates. Finally, a Prompt Sculpting mechanism constrains the generator to synthesize the distilled subset into strictly formatted JSON responses with exact chunk-level citations. Evaluated on the RAG track, our resource-aware approach shows exceptional precision in both information retrieval and persona-conditioned generation.

178. Larch: Learned Query Optimization for Semantic Predicates

Authors: Fuheng Zhao , Pawel Liskowski , Zihan Li , Benjamin Han , Puxuan Yu , Varich Boonsanong , Dimitris Tsirogiannis , Anupam Datta
URL: https://arxiv.org/abs/2606.07923
Abstract:

With the advent of Large Language Models (LLMs), many database systems introduced semantic operators that enabled analytical queries over unstructured data (e.g. text, images, videos). Semantic operators typically incur high inference costs and latencies making semantic (AI) SQL queries challenging to apply on large scale datasets. At the same time, their semantic nature leads database engines to treat them as black boxes, making AISQL queries difficult to optimize. In this paper, we introduce Larch, a framework for optimizing the execution of semantic filters in AI SQL queries. Larch was inspired by two key observations: i) the high latency of semantic operators leaves significant room for computationally-heavy runtime optimization techniques, ii) unstructured data are typically accompanied by semantic information in the form of embeddings allowing for efficient semantic comparisons between AI_FILTER prompts and data values. Based on these two key observations, we present two Larch variants: Larch-A2C and Larch-Sel. Larch-A2C encodes arbitrary semantic filters expression tree using an embedding-augmented Gated Graph Neural Network and formulates the filter evaluation order as a Markov decision process. In contrast, Larch-Sel leverages a supervised learning model to predict filter selectivities, subsequently applying dynamic programming to find a near-optimal evaluation order for each input row. Evaluated across diverse real-world datasets and comprehensive synthetic workloads, both Larch variants always outperform existing semantic filter optimization techniques in terms of token usage. Our results demonstrate that Larch is robust across diverse workloads, reducing total token cost overhead by 3x-19x compared to Palimpzest and Quest.

179. Strained Coherence: A Pre-Failure Signal in Coding Agent Execution Trajectories

Authors: Marut Pandya , Kasey Zhang , Baiqing Lyu
URL: https://arxiv.org/abs/2606.07889
Abstract:

LLM-based coding agents sometimes acknowledge a problem in their own reasoning and then proceed anyway. We call this pattern strained coherence: a safety-relevant failure mode in which an agent has information that should change its behavior, states that information, and still acts against it. The pattern overlaps with verbalized reward hacking, where an agent names a tension between a task proxy and the underlying goal yet optimizes the proxy anyway. We give an operational definition, build a Claude Sonnet 4.6 judge that reads full trajectories and flags spans where the pattern occurs, and evaluate it on 44 Terminal-bench-2 trajectories using a Qwen3.5-35B-A3B backbone. Flagged trajectories fail 94% of the time versus 46% for unflagged trajectories (47-point gap, Fisher’s exact p = 0.003; 46 points after excluding three prompt-embedded examples, p = 0.006). At matched selectivity, the detector reaches 94% precision versus 88% for a lexical discourse-marker baseline; the 10-trajectory intersection of the two methods has a 100% failure rate (Clopper-Pearson 95% CI [69%, 100%]). We replicate on Gemma4-31B with 43 trajectories: the overall signal is directionally consistent but not significant (20-point gap, p = 0.31), with attenuation driven largely by 13 trajectories with zero think content, where the detector has no substrate to analyze. In the high-verbosity Gemma tertile, the gap is +30 points; in the mid- and high-verbosity Qwen tertiles, it is +40 points each. The first flag appears at a median of 83-84% of elapsed trajectory time across both models, and the binary flag survives paraphrases that soften explicit conflict markers (8/8 trajectories). Unlike univariate predictors, the detector emits interpretable span-level output – quoted acknowledgment, quoted action, and typed conflict – showing what the agent saw and ignored.

180. The Last Visible Pixel: Probing Fine-Scale Perception in Vision-Language Models

Authors: Lujun Li , Lama Sleem , Niccolo Gentile , Yangjie Xu , Yewei Song , Wenbo Wu , Radu State
URL: https://arxiv.org/abs/2606.07861
Abstract:

Recent vision-language models (VLMs) excel at multimodal understanding and reasoning, yet their fine-grained visual perception remains underexplored. A natural extension of ``How many r are there in Strawberry?’’ asks: how small a visual pattern can a VLM reliably perceive? As such, we introduce FineSightBench, a new benchmark that systematically probes this limit by separating perception tasks (pixel-level recognition of letters, shapes, objects) from reasoning tasks (spatial reasoning, counting, ordering over small targets) across controlled scales of 4–48px. Through comprehensive experiments and detailed failure mode analysis on state-of-the-art models, we reveal a sharp dissociation: perception saturates around 12px, while reasoning remains limited even at larger scales, with persistent numeracy and sequence errors. These findings expose fundamental deficiencies in VLMs’ fine-scale visual reasoning that demand more rigorous evaluation.

181. Model Multiplicity for Adversarial Detection in Small Language Model Training on Edge Devices

Authors: Stefan Behfar , Richard Mortier
URL: https://arxiv.org/abs/2606.07857
Abstract:

The rise of edge-based machine learning has enabled distributed adaptation of language models across mobile and IoT devices, offering privacy preservation and real-time responsiveness. However, distributed fine-tuning of language models on untrusted or heterogeneous edge nodes introduces new vulnerabilities. Compromised or unreliable devices can inject poisoned updates, leading to stealthy model manipulation or convergence degradation. Classical defenses such as robust aggregation or temporal anomaly detection operate on a single global model and are therefore limited in detecting coordinated or persistent poisoning. This work proposes a new system-level defense based on model multiplicity. Instead of maintaining one global model, the system rotates or concurrently trains multiple small language models (e.g., DistilGPT-2), each updated by independently sampled subsets of edge nodes. These models evolve under distinct training trajectories, creating multiple independent views of the same distributed population. Divergence between models quantified through gradient similarity, loss evolution, or parameter variance serves as a signal of anomalous or adversarial behavior. When one model deviates significantly from the ensemble mean, the system flags its contributing nodes for isolation or re-weighting. We implement this framework and evaluate it on edge-scale simulations of Small Language Model (SLM) training under varying heterogeneity and attack conditions. Results show that model multiplicity enables earlier and more reliable detection of poisoning compared to classical single-model defenses such as Flanders and Robust methods. Our findings demonstrate that diversity in model evolution can serve as a practical and effective defense mechanism for secure distributed learning on resource-constrained edge devices.

182. Beyond English benchmarks: clinical llm evaluation in Brazilian Portuguese

Authors: Giordano de Pinho Souza , Glaucia Melo , Josefino Cabral Melo Lima , Daniel Schneider
URL: https://arxiv.org/abs/2606.07853
Abstract:

Large Language Models are transforming the support for clinical decision and their application in real scenarios. Yet, most benchmarks are conducted in English, and cross-lingual evaluation is needed to tackle the language gaps in global access. We introduce ClinicalBr, the first bilingual benchmark for clinical decision built from real Brazilian case reports. The corpus contains 2,892 cases drawn from 28 SciELO medical journals, spanning 18 specialties, and is structured as parallel Portuguese-English pairs. Each case supports four evaluation tasks: diagnosis retrieval, differential diagnosis, exam recommendation, and treatment planning. We evaluate four models: MedGemma-27B, Sabiá-4, DeepSeek-R1, and o3-mini, across both languages. The central finding is that the Portuguese-English performance gap is task-dependent, not general. In diagnosis retrieval, English yields a consistent advantage across all models, with +7.5-12.1 accuracy points. This advantage disappears in differential diagnosis, exam recommendation, and treatment planning, where confidence intervals cross zero for most models and Portuguese completeness scores are marginally higher. Brazilian-endemic conditions proved easier than the full corpus, not harder, indicating that tropical presentations are adequately represented in current pre-training. Exam recommendation was the hardest task across all models and both languages, with F1 scores below 0.10, well below the differential diagnosis ceiling of 0.20-0.27.

183. Cost-Aware Speculative Execution for LLM-Agent Workflows: An Integrated Five-Dimension Method

Authors: Faisal Fareed
URL: https://arxiv.org/abs/2606.07846
Abstract:

LLM-agent workflows chain model calls and tool invocations, and spend most of their wall-clock time waiting on upstream operations before downstream ones can start. Speculative execution can reclaim that idle time by launching a downstream operation with a predicted upstream input, but here each speculation costs real money (per-token billing) and its success probability is hard to estimate and drifts over time. This paper presents a method organized around five design decisions: (D1) start a downstream operation before its upstream completes; (D2) price each speculation in real dollars at separate input and output rates; (D3) expose a single operator dial for latency versus cost; (D4) decide via an expected-value rule with a failure-weighted cost term and a preference-adjusted threshold; and (D5) estimate the success probability with a Bayesian Beta-Binomial posterior whose prior is keyed to a dependency-type taxonomy. Variants of these ideas appear in recent work; the combination, with every decision logged in dollars, is what is new. The rule fires only on edges passing an admissibility precondition (side-effect-free, idempotent, or stageable behind a commit barrier), since a wrong speculation is rolled back by re-execution, which refunds tokens but cannot un-send an irreversible side effect. We specify the runtime mechanics, a closed-form result that the rule self-limits as the upstream branching factor grows, a five-stage calibration pipeline (offline replay, shadow, canary, online calibration, drift-triggered kill-switch), and a workload-fit rubric over eight production archetypes. Contrast tables against the four closest published systems (DSP, Speculative Actions v2, Sherlock, B-PASTE) show differentiators on every dimension, and a synthetic validation suite confirms the predicted decision boundary, probability threshold, posterior recovery, and streaming-cancellation behavior.

184. Does Persona Make LLMs K-pop Fans? A Pilot Study of LLM-Based Online Concert Audience Agents

Authors: Kirak Kim , Hyojin Kim , Yejin Son , Sungyoung Kim , Kyung Myun Lee
URL: https://arxiv.org/abs/2606.07837
Abstract:

A concert is a collective experience, but recorded performance videos are typically watched alone, stripping away the shared audience presence that makes concerts feel eventful. We investigate whether persona-based LLM audience agents can recreate aspects of this collective experience by generating real-time fan chat alongside a K-pop performance video. We present a multi-agent system in which ten LLM agents react through live-chat messages, comparing a persona-conditioned audience (each agent assigned a distinct fan identity, bias, and chat style) with a no-persona baseline. In a within-subjects pilot with K-pop fans (N=11), persona conditioning substantially improved model-level chat quality and perceived naturalness, but did not translate into differences in social connectedness, engagement, or affective response. Interviews suggest that online K-pop concert chat may operate as collective monologue rather than interpersonal dialogue, and that meaningful participation depends on shared identification with the specific artist and fandom. Persona conditioning can make LLM audiences appear more natural, but culturally meaningful collective experience may require deeper alignment between persona, crowd behavior, fandom identity, and user expectations.

185. Cherry-pick Override: Unsafe Directional Commitment in LLM Judges under Mixed Evidence

Authors: Haoran Xu
URL: https://arxiv.org/abs/2606.07834
Abstract:

LLM judges increasingly turn verdicts into system commitments. Under mixed evidence (claims with both supporting and refuting sources) this is unsafe: when the schema exposes CONFLICTING as the authorized non-directional verdict, returning SUPPORTS/REFUTES is an unauthorized directional commitment, a failure we name Cherry-pick Override (CCO). We define CCO under an explicit task contract and report it with a same-denominator diagnostic protocol paired with matched-coverage bootstrap and an apples-to-apples random-veto null. On AVeriTeC’s Conflicting subset (N_C = 150), three-option judges return a directional verdict on more than 84% of mixed-evidence claims; under the typed schema, three-judge majority voting amplifies direction-on-conflict on AVeriTeC (0.887 vs. 0.840; 95% CI [+0.013, +0.080]) but does not replicate on VitaminC-Mixed. Walking an intervention ladder of common single-channel fixes (typed vocabulary, panel aggregation, confidence thresholding, validator-only filtering), each leaves a distinct residual failure: panel aggregation suppresses single-judge CONFLICTING dissent in 48% of CCO cases; the panel is well-calibrated for direction (ECE = 0.07 on pure-S/R) so confidence cannot operationally separate CCO from correct directional commits; validator-as-classifier nearly halves pure-evidence accuracy. A minimal two-channel reference probe reaches operating points neither single channel reaches; under the random-veto null its promotion to CONFLICTING is structurally targeted on AVeriTeC (empirical p < 1/2001) and weaker but in the same direction on VitaminC-Mixed, a selectivity result rather than a magnitude one. We argue for an external commitment-control layer that separates verdict generation from commitment authorization, using structural evidence and confidence as orthogonal channels and NO-COMMIT as a routed controller state.

186. The ACUTE Protocol: Operationalizing Language Model Activations for Better Calibration, Utility, and Trust

Authors: Nishant Subramani , Palash Goyal , Yiwen Song , Mani Malek , Yuan Xue , Tomas Pfister , Hamid Palangi
URL: https://arxiv.org/abs/2606.07822
Abstract:

As language models improve and become increasingly deployed to solve a variety of tasks, trustworthiness becomes essential. Calibration is a good proxy for trust: well-calibrated confidence estimates help inform the risk versus reward tradeoff when trusting a specific model output. Unfortunately, even as models improve, they remain poorly calibrated, often biasing towards overconfidence. Additionally, calibration can be gamed: a policy that always predicts the base rate is perfectly calibrated, but completely uninformative. To resolve this, we develop a new metric, expected utility renormalized by the oracle (EURO), that balances calibration and informativeness. We also propose a general-purpose activation-based confidence, utility, and trust estimation protocol (ACUTE) to appropriately adjudicate uncertainty. The ACUTE protocol provides flexible, sample-efficient, and compute-efficient confidence estimators for 3 tasks including multiple choice question answering, tool-calling, and scientific document summarization across 6 models from 4 model families. ACUTE outperforms strong baselines on EURO, while maintaining low calibration error. Taken together, our work shows that equipping LLMs with the ACUTE protocol can improve calibration, utility, and trustworthiness in numerous settings.

187. SLMJury: Can Small Language Models Judge as Well as Large Ones?

Authors: Anish Laddha , Nitesh Pradhan , Gaurav Srivastava
URL: https://arxiv.org/abs/2606.07810
Abstract:

Large language models (LLMs) are widely used as judges for evaluating model outputs, but their high cost, latency, and opacity limit scalability. We introduce SLMJury, a framework for evaluating small language models (SLMs) as judges across two paradigms: closed-ended binary correctness and open-ended quality scoring. We benchmark 16 SLM judges (0.6B-14B parameters) from four model families across ten benchmarks: eight closed-ended tasks spanning mathematical, scientific, and general reasoning (N=64,824 judgments per configuration), plus SummEval and MT-Bench for summarization and conversational scoring. We formalize judging as a budget-conditioned function and study five dimensions. Four findings emerge. (1) The overthinking effect is domain-dependent: for most judges quick 10-token verdicts match or beat extended reasoning on mathematical judging (by 2-7% where they help), while reasoning wins on general tasks by up to 23%. (2) Domain generalization separates model families, with math-to-general accuracy gaps ranging from under 10% to nearly 40%. (3) Closed-ended and open-ended judging draw on different capabilities: the best binary judge (Phi-4) drops to rank 9 on MT-Bench, while reasoning-trained models invert this ordering. (4) Under the Reflect-Critique-Refine (RCR) debate protocol, multi-agent debate degrades accuracy across all tested configurations, whereas the top judges resist six adversarial personas with <=0.55% variance. Reliable automated evaluation does not require large proprietary models, yet no single SLM dominates. The leaderboard is available at this https URL , and our framework code and pip package are publicly available at this https URL and this https URL .

188. MatMind: A Structure-Activity Knowledge-Driven Generative Foundation Model for Materials Science

Authors: Zhan’ao Yao , Boxuan Zhang , Jingyuan Shu , Xiaoyu Wu , Rongyan Wang , Linjing Li , Dajun Zeng , Yudong Yao , Tingwei Chen , Youwei Wang , Xiaolin Zhao , Jiahui Shi , Jianjun Liu
URL: https://arxiv.org/abs/2606.07712
Abstract:

Progress in AI-driven crystal materials science has so far been carried by narrow architectures purpose-built for individual tasks – graph neural networks for property prediction, diffusion and flow-matching models for crystal generation – each excelling within its niche yet unable to act as a shared backbone across the full spectrum of materials problems. Generative large language models offer a fundamentally different paradigm, in which structural representation, quantitative prediction, and structure-activity reasoning can be unified within one model, but the materials community has yet to see this paradigm realized at a level competitive with established narrow specialists. Here we present MatMind, a generative foundation model purpose-built for crystal materials science under this paradigm, developed through the coordinated activation of structure-activity knowledge and physics-informed feedback within a progressive training framework – combining structure-activity knowledge injection, a dual-head architecture that jointly trains language reasoning and numerical regression in a shared representation space, and multi-objective physics-informed reinforcement learning over stability, novelty, and structural diversity. Across three task families, MatMind attains the lowest mean absolute error on energy above hull, bulk modulus, and band gap – surpassing graph neural network predictors purpose-built for these tasks – reaches an S.U.N. rate of 65.3% on unconditional crystal generation, and achieves a comparable multiplicative improvement on magnetization-density-conditioned generation, where only 21 positive samples exist within over 600000 training entries. By matching or surpassing narrow specialists on their own ground while operating within a single unified model, MatMind shows that the LLM-based paradigm can serve as a viable backbone for crystal materials science going forward.

189. Rosetta Memory: Adaptive Memory for Cross-LLM Agents

Authors: Hao Yang , Shiqi Shen , Haoxuan Li , Zhipeng Wang , Zhi Gong , Xu Chen
URL: https://arxiv.org/abs/2606.07711
Abstract:

Memory is the key component for transforming a stateless LLM into a persistent, evolving agent through experience accumulation, long-horizon planning, and continual self-improvement. Existing memory systems typically take the LLM as the center and design memory operations tailored to a specific backbone. In practice, however, users frequently switch between LLMs, for example using Claude for coding and GPT for writing across tasks, or routing different steps to different backbones within a single task for cost-effective trade-offs. As a result, memory written by one model often needs to be consumed by another. Making upstream memory effectively adapt to and activate downstream LLMs remains a critical yet underexplored problem. To bridge this gap, we shift the perspective from LLM-centric memory design to \emph{memory-centric LLM adaptation}. Specifically, we approach the above upstream-downstream memory adaptation problem from both the write and read sides, and design two profile-conditioned operators that are jointly trained to optimize how memory is stored and presented for better task completion. To ensure the learned operators generalize across a broad set of LLMs, we propose a minimum-gain sampling curriculum that prioritizes the least-served LLMs during training. To better measure the operators’ actual contribution rather than the LLM’s own capability, we design a performance-gap reward that compares against a naive memory baseline. Experiments on HotpotQA, 2WikiMultihopQA, and MuSiQue demonstrate that our model consistently outperforms baselines and remains robust under unseen-model replacement.

190. WhiFlash: Accelerating Speculative Decoding with Token-Level Cross-Paradigm Routing

Authors: Young D. Kwon , Miles Williams , Rui Li , Alexandros Kouris , Stylianos I. Venieris
URL: https://arxiv.org/abs/2606.07710
Abstract:

The autoregressive nature of large language models (LLMs) remains a significant bottleneck for inference, particularly in complex agentic workloads. While speculative decoding (SD) accelerates inference, current approaches rely on static drafting paradigms, utilising either autoregressive drafting models for reasoning or diffusion-based parallel drafting models for structured outputs. We empirically find that drafting accuracy fluctuates dramatically within a single sequence, leaving significant performance unrealised by static paradigms and coarse-grained routing. To address this volatility, we introduce WhiFlash, the first cross-paradigm SD method that unifies autoregressive and diffusion-based parallel drafting under a single token-level controller. WhiFlash adopts a fine-grained routing mechanism that employs either a lightweight entropy-based or a learned neural policy, both parametrised to provide a tunable balance between expected token gain and latency. To make high-frequency switching computationally viable, we introduce novel cache-management optimisations, Lazy Catch-up and KV-only Prefill, reducing switching overhead to below 7% of per-round latency. By capitalising on the complementary strengths of fundamentally distinct drafting architectures, WhiFlash achieves significantly higher acceptance lengths, yielding category-specific throughput gains of up to 69.6% over the state-of-the-art autoregressive EAGLE-3 and 37.3% over the diffusion-based DFlash.

191. MLingualFC: Evaluating Jailbreak Vulnerabilities in Multilingual Vision-Language Models

Authors: Rishabh Makwana , Mamta , Deeksha Varshney , Oana Cocarascu
URL: https://arxiv.org/abs/2606.07706
Abstract:

Vision-Language Models (VLMs) have demonstrated strong performance across multimodal tasks, yet their safety robustness remains an open challenge. While prior work has shown that structured visual prompts such as flowcharts can effectively jailbreak VLMs, existing studies are largely limited to English-centric settings. In this paper, we introduce MLingualFC, a multilingual multimodal benchmark designed to evaluate jailbreak vulnerabilities of VLMs across diverse languages using structured flowchart representations. MLingualFC encodes harmful instructions into flowchart images across five languages (Hindi, Punjabi, Spanish, Romanian, and German). We evaluate state-of-the-art multilingual VLMs, including Qwen2.5-VL, Gemma-4, and Pangea, under a black-box threat model. Our results reveal significant multilingual safety gaps. Flowchart-based attacks achieve high attack success rates (ASR) in case of Latin script languages, demonstrating that visual encoding of harmful content effectively bypasses safety alignment across languages. In contrast, non-Latin script languages such as Punjabi exhibit substantially lower ASR, suggesting potential limitations in visual text recognition rather than stronger safety alignment. These findings highlight that current VLM safety mechanisms fail to generalize across languages and modalities. Resources are available at this https URL

192. SAW: Stage-Aware Dynamic Weighting for Multi-Objective Reinforcement Learning in Large Language Models

Authors: Yuchen He , Baolong Bi , Shenghua Liu , Huaming Liao , Yuyao Ge , Bolin Wan , Siqian Tong , Juan Chen , Jiafeng Guo , Xueqi Cheng
URL: https://arxiv.org/abs/2606.07705
Abstract:

Although multi-objective reinforcement learning (MORL) is central to aligning large language models with complex human preferences, the prevailing practice of static weighted summation overlooks a more fundamental phenomenon: reward learning is markedly asynchronous across objectives. Well-learned dimensions quickly produce homogeneous, low-variance signals whose residual noise contaminates the aggregated reward (in GRPO) or occupies a fixed share of the advantage budget (in GDPO), interfering with the scarce yet high-value signals carried by under-learned dimensions. To address this asynchrony, we propose Stage-Aware Dynamic Weighting (SAW), a lightweight, algorithm-agnostic dynamic weighting mechanism. SAW utilizes the coefficient of variation (CV) as a scale-invariant proxy for real-time informativeness, reweighting each dimension’s reward or advantage contribution by its relative informativeness within the batch. Unlike gradient-based methods that require multiple forward and backward passes, SAW relies solely on batch-level statistics, introducing nearly negligible computational overhead. Experiments on tool-calling and text summarization tasks demonstrate that SAW consistently improves both training efficiency and final performance under both GRPO and GDPO frameworks, confirming it as a general-purpose plug-in for multi-reward LLM alignment. Our code is available at this https URL

193. FunctionEvolve: Structure-Guided Symbolic Regression with LLMs

Authors: Zeyu Xia , Jun Zhu , Dong Yan
URL: https://arxiv.org/abs/2606.07704
Abstract:

Symbolic regression aims to uncover explicit scientific laws from data. Recent methods use LLMs to guide mutation from background text, which is more directed than random genetic programming. However, exact symbolic recovery requires both semantic guidance and explicit structure, so that domain-informed search are carried out through valid symbolic representation. Current LLM-driven systems remain structure-blind: they select among opaque candidates, lack explicit mechanisms for local mutation, and rely on brittle coefficient fitting that can undervalue correct skeletons. We propose FunctionEvolve, an evolutionary framework using expression trees to organize the whole search: structural summaries promote diverse parent selection, local tree edits preserve useful subexpressions, and structure-aware fitting decomposes, constrains, and simplifies coefficients for more reliable scoring. It uses only elementary function families, without additional domain-specific rules limiting generalization. On the 129-task synthetic subset of LLM-SRBench, FunctionEvolve with \emph{Claude Opus 4.6} recovers 107 exact forms, reaching 82.9% SA@50, 4.5x above same-backbone baselines, and 55.8% SA@1, 3.6x above the strongest previously published top-1 result. Ablations show that structure-visible search is central to reliable recovery, with LLM-guided refinements and structure-aware coefficient optimization serving as essential proposal and scoring mechanisms. We also audit the benchmark and show that collinearity in its materials-science subset creates identifiability issues.

194. Adversarial Robustness of Activation Steering in Large Language Models

Authors: Kien Le , Thai Le
URL: https://arxiv.org/abs/2606.07696
Abstract:

Activation steering has become a popular training-free method to control LLM behavior by injecting precomputed direction vectors into the model’s residual stream at inference time. Yet its robustness to realistic input variation remains unstudied. We present the first systematic evaluation of activation steering robustness under adversarial text perturbations on the inputs, covering four extraction methods, three attack strategies, six personas from Anthropic Model-Written Evaluation Dataset, and five models ranging from 1.5B to 30B parameters. Attacks succeed broadly across all settings: directional robustness drops by up to 64%, post-attack confidence collapses near or below 0.25 across all methods and models, and steering strength degrades on nearly every steerable input. Layer selection is equally fragile, with the optimal layer identified by an automated method on clean inputs shifting by up to 17 positions under perturbation, a failure that compounds the vector-level breakdown. Extracting vectors from adversarially perturbed inputs partially recovers steerability for PCA and MD on mid-to-large models, but they consistently fail to locate the improved optimal layer, limiting the practical benefit of this mitigation. Together, these findings reveal that the brittleness of activation steering is structural rather than method-specific, and that current layer selection strategies are not robust enough for real-world deployment.

195. HARP: Efficient Data Selection for Finetuning Large Language Models

Authors: Ning Wang , Zhengxin Zhang , Maosen Tang , Yitang Gao , Claire Cardie , Sainyam Galhotra
URL: https://arxiv.org/abs/2606.07690
Abstract:

Finetuning data selection requires balancing two competing goals: selecting examples that improve the downstream objective, and doing so without repeatedly finetuning models. Train-free selectors are scalable but rely on proxies such as embedding similarity or clustering, which may not match the target objective. Train-based selectors better reflect downstream utility through gradient signals, subset evaluation, or Shapley attribution, but require many costly train–evaluate iterations. We propose Hierarchical Active Region Pruning (HARP), an efficient train-based selector that preserves downstream alignment while reducing selection cost. HARP organizes the training pool into a node–leaf hierarchy, evaluates only representative leaves, and infers unmeasured utilities with empirical Bayes posteriors. It then selects data using two complementary envelopes: HARP-C, which conservatively controls redundancy, and HARP-E, which additively rewards complementary regions. We theoretically show that, under local smoothness and bounded estimation error, HARP controls selection error while reducing train–evaluate cost. We further validate that HARP variants achieve the best result and outperform the strongest baseline by up to $+8.9$ points, while using roughly $7\times$ fewer training examples.

196. TRACER: Token ReAssignment for Concept ERasure in Generative Recommendation

Authors: Ziheng Chen , Jiali Cheng , Zezhong Fan , Hadi Amiri , Diyuan Wu , Gabriele Tolomei , Yang Zhang
URL: https://arxiv.org/abs/2606.07688
Abstract:

Generative recommendation formulates next-item prediction as autoregressive generation over semantic ID (SID) sequences derived from users’ historical interactions, making modern recommender systems structurally similar to large language models (LLMs). As privacy and safety concerns grow, these systems increasingly require concept unlearning to remove sensitive or harmful concepts associated with items. However, existing LLM unlearning methods cannot be directly applied to generative recommendation. Unlike word tokens with explicit semantics, SIDs are abstract identifiers that are often shared by both forget and retain items, leading to severe conflicts between concept removal and recommendation utility preservation. To address this challenge, we propose TRACER, an end-to-end concept unlearning framework based on token reassignment. Rather than directly suppressing shared SIDs, TRACER reassigns concept-related items to alternative tokens that better facilitate forgetting while minimizing side effects on retained items. We further introduce a coherence regularizer to preserve semantic consistency among retain items during unlearning. Experiments on real-world recommendation datasets demonstrate that TRACER effectively removes target concepts while substantially better preserving recommendation utility than existing unlearning baselines.

197. Semantic Cache Distillation: Efficient State Transfer via Reuse and Selective Patching

Authors: Qianli Ma , Zhiqing Tang , Hanshuai Cui , Zhi Yao , Weijia Jia
URL: https://arxiv.org/abs/2606.07684
Abstract:

Disaggregated serving alleviates memory bottlenecks in Large Language Model (LLM) inference but creates a severe communication bottleneck: transmitting high-dimensional Key-Value (KV) caches often dominates time-to-first-token (TTFT). Moreover, reusing caches across heterogeneous models (e.g., base and fine-tuned variants) causes semantic misalignment that accumulates over layers, degrading generation quality. We propose Semantic Cache Distillation (SCD), a loss-constrained framework that replaces raw KV transmission with compact semantic codes. SCD addresses these challenges via two mechanisms: (1) Reuse, which reconstructs most layers from low-rank subspaces to minimize transfer cost, and (2) Patch, which predicts normalized inputs at sparse transition layers to truncate error propagation. Empirically, SCD delivers up to 2.65 $\times$ TTFT speedup over the oracle consumer prefill and dominates quantization and selective recomputation baselines on the quality–latency Pareto frontier in bandwidth-constrained regimes, while keeping generation quality within 5\% F1 of the oracle.

198. Systematic LLM Translation of Legacy Scientific Code to Differentiable Frameworks: Application to a Land Surface Model

Authors: Aya Lahlou , Linnia Hawkins , Pierre Gentine
URL: https://arxiv.org/abs/2606.07681
Abstract:

Differentiable programming offers transformative capabilities for scientific modeling, enabling gradient-based parameter estimation, sensitivity analysis, and data assimilation. Yet, migrating legacy codebases into differentiable frameworks remains a challenge. We present a five-phase LLM-based agentic pipeline that translates legacy Fortran into JAX: static dependency analysis determines module translation order from the full call graph; iterative compile-repair loops correct errors autonomously; and a Fortran reference oracle enforces numerical parity at the module level before integration and gradient verification. We instantiate and evaluate the pipeline on CLM-ml-v2, a 19,000-line Fortran land surface model, and analyze agent behavior across 73 module translation tasks. The resulting differentiable model computes the complete Jacobian in a single backward pass, recovers physical parameters in eight times fewer steps than gradient-free optimization, and achieves a 24 times wall-clock speedup over sequential Fortran at ensemble size N=2,048. Both the translated model and pipeline infrastructure are released as a reusable framework for differentiating other Earth system model components.

199. DOG-DPO:Dynamic Optimization in Geometry for Safety Alignment

Authors: Yi Nian , Tiankai Yang , Yudi Zhang , Qi Pan , Zelong Xu , Shenzhe Zhu , Qingqing Luan , Yue Huang , Xiangliang Zhang , Yue Zhao
URL: https://arxiv.org/abs/2606.07678
Abstract:

Safety alignment for large language models relies on preference data, but current pipelines often train on large, redundant datasets. Existing data selection methods typically score each preference pair independently, collapsing directional preference information into scalar quality or diversity scores. This sample-centric view is especially limiting in multi-dataset settings, where shared safety directions coexist with dataset-specific residual risks. We propose DOG-DPO, a training-free data selection framework that treats preference pairs as structured geometric signals. DOG-DPO first represents each preference pair as a direction in model representation space. It then decomposes multi-dataset preference geometry into a global anchor subspace and dataset-specific residual subspaces. Finally, it selects subsets by maximizing diversity-based coverage, encouraging broad, non-redundant coverage of alignment directions before DPO training. Across six safety benchmarks and two model backbones, DOG-DPO achieves a strong utility-robustness trade-off using only 11% of the preference pairs. It recovers most of the safety gains of full-data training while remaining entirely teacher-free, training-free, and substantially faster than representative selection baselines.

200. MemoVAD: Resource-Efficient Video Anomaly Detection via Dynamic Semantic Memory in Edge Computing Scenarios

Authors: Guo Li , Jiandian Zeng , Yang Li , Zihao Peng , Ke Chen , Tian Wang
URL: https://arxiv.org/abs/2606.07669
Abstract:

Deploying Video Anomaly Detection (VAD) in real-world surveillance faces a fundamental tension between the demand for high-level semantics to ensure effectiveness and the limited computational resources of edge devices. Vision-Language Models (VLMs) provide rich open-vocabulary semantics, but their latency and computational cost preclude on-device deployment. To address the challenge, we propose MemoVAD, an edge-cloud collaborative framework that selectively incorporates VLM semantics into streaming VAD. MemoVAD runs most inference on the edge with a lightweight detector and a causal Temporal Context Encoder (TCE) to model temporal dependencies. Specifically, we introduce an Uncertainty-Aware Gating (UAG) policy grounded in Subjective Logic to model perceived uncertainty and query the cloud-based VLM only for high-uncertainty and semantically novel clips. Besides, a Dynamic Semantic Memory (DSM) is designed to cache VLM-verified prototypes for efficient retrieval, enabling the edge model to progressively incorporate VLM-level semantics via a semantic adapter. Experiments on UCF-Crime and XD-Violence datasets via a real edge device show that MemoVAD substantially reduces communication overhead while surpassing state-of-the-art performance.

201. AgentCompile: An LLM-Guided Compiler for Direct CUDA Inference

Authors: Xuanzhe Li , Ziyan Weng , Zhiyu Zhu , Junhui Hou
URL: https://arxiv.org/abs/2606.07665
Abstract:

Transformer inference increasingly depends on specialized compiler and runtime support, but real model graphs still require semantic decisions about which regions are worth specializing and which CUDA implementation families are plausible. We present AgentCompile, an LLM-guided CUDA inference compiler that uses LLM outputs only as advisory search metadata. Given compiler-derived region summaries and bounded candidate spaces, the LLM proposes semantic labels, candidate priorities, parameter hints, and risk annotations; the compiler materializes CUDA candidates through templates, checks interface and hardware constraints, validates candidates empirically, selects implementations by measured latency, and falls back when specialization is unsupported or unprofitable. In end-to-end autoregressive generation, AgentCompile averages 5.66x, 4.05x, and 4.26x speedup over PyTorch eager on Qwen3-1.7B, Qwen3-4B, and Llama-3.2-1B-Instruct, respectively, across five representative workloads. We will open-source the project.

202. MM-Matryoshka: Towards Budget-Elastic Visual Document Retrieval via a 2D Multimodal Matryoshka Training Framework

Authors: Haowen Xiang , Yibo Yan , Jiahao Huo , Yu Huang , Yi Cao , Mingdong Ou , Xuming Hu
URL: https://arxiv.org/abs/2606.07654
Abstract:

Multi-vector visual document retrievers achieve strong fine-grained matching by representing each page with multiple vectors from deep Vision-Language Models (VLMs), but this design makes deployment expensive in both storage and computational overhead. Existing efficiency techniques usually optimize only part of this budget, leaving multimodal retrievers without a unified way to trade accuracy for both vector width and encoder depth. Therefore, we propose MM-Matryoshka, a 2D Matryoshka training framework for budget-elastic Visual Document Retrieval (VDR), enabling ColPali-style multi-vector retrieval elastic along both dimension and layer. At inference time, a single retriever can select a 2D selectable budget without training separate models for different budgets. Through comprehensive experiments across multiple representative backbones, we demonstrate that by retaining significantly higher quality than direct truncation baselines while substantially reducing storage and computational overhead, MM-Matryoshka can offer robust budget elasticity for efficient VDR.

203. A Dataset for Dynamic Human Preferences for Vision Language Models

Authors: Hannah Gao (Massachusetts Institute of Technology), Dylan Hadfield-Menell (Massachusetts Institute of Technology), Rachel Ma (Massachusetts Institute of Technology)
URL: https://arxiv.org/abs/2606.07653
Abstract:

Given the increased adoption of Vision Language Models (VLMs) in human-interactive settings, it is important that we evaluate how well these models can adapt to real-time preferences for different users. While an increasing number of vision-language benchmarks have recently been introduced, they focus largely on evaluating static capabilities and generally-held preferences learned from extensive training data. This work introduces a new benchmark for evaluating the ability of VLMs to understand dynamic human-preferences, i.e. preferences that are passed in-context at inference time. We provide an automated pipeline for generating this benchmark with variations on image dependence, a dynamic multi-modal human-preference dataset, and evaluations of state-of-the-art models on the novel benchmark.

204. AVI-Bench: Toward Human-like Audio-Visual Intelligence of Omni-MLLMs

Authors: Yaoting Wang , Ziyi Zhang , Wenming Tu , Shaoxuan Xu , Wenjie Du , Cheng Liang , Weijun Wang , Yuanchao Li , Guangyao Li , Hao Fei , Yuanchun Li , Henghui Ding , Yunxin Liu
URL: https://arxiv.org/abs/2606.07643
Abstract:

Recent advances in Omni-Multimodal Large Language Models (Omni-MLLMs) have enabled strong integration of vision, audio, and language. However, their audio-visual intelligence (AVI) remains insufficiently evaluated due to the lack of systematic and comprehensive benchmarks. We introduce AVI-Bench, a cognitively inspired benchmark that evaluates Omni-MLLMs across three stages, perception, understanding, and reasoning, through cross-modal tasks requiring joint audio-visual interpretation. This design enables fine-grained diagnosis of model capabilities and failure modes. To further assess robustness beyond familiar domains, we propose AVI-Bench-PriSe, an extension that probes models’ primitive audio-visual sensation using unfamiliar, low-semantic stimuli, testing generalization beyond common training distributions. Extensive experiments on both open-source and closed-source models reveal substantial limitations in current Omni-MLLMs. Based on these findings, we present a four-level AVI taxonomy. Overall, AVI-Bench provides a principled evaluation framework to guide the development of more robust and generalizable AVI. Project website: this https URL

205. Large Language Models Should Learn Personalized Rather Than Aggregated Human Preferences

Authors: Cristina Garbacea
URL: https://arxiv.org/abs/2606.07629
Abstract:

Current approaches to aligning large language models (LLMs) aggregate diverse human preferences into a single reward signal, effectively optimizing for a hypothetical ``average user’’ who represents no real person particularly well. This position paper argues that LLMs should learn personalized, individual preferences rather than aggregated ones. We show that aggregation masks critical information about preference diversity, individual values, and contextual dependencies, which is a limitation both theoretically grounded in social choice theory and empirically evident across demographic groups. We analyze the rich structure that human preferences encode, survey technical approaches to personalization, and systematically address counterarguments on scalability, shared standards, and manipulation risk. While personalization introduces genuine safety challenges including filter bubbles, value lock-in, and psychological manipulation, we argue these are manageable through bounded personalization frameworks that preserve universal safety constraints while accommodating legitimate individual variation. We conclude with a concrete research and policy agenda for developing preference-aware models that respect both individual autonomy and collective safety.

206. Item Response Scaling Laws: A Measurement Theory Approach for Efficient and Generalizable Neural Scaling Estimation

Authors: Sang Truong , Yuheng Tu , Rylan Schaeffer , Sanmi Koyejo
URL: https://arxiv.org/abs/2606.07616
Abstract:

Scaling laws provide a fundamental framework for understanding the performance of Language Models (LMs), yet deriving them requires prohibitively expensive evaluations across thousands of checkpoints or millions of inference samples. To address this, we introduce Item Response Scaling Laws (IRSL), a unified framework that integrates Item Response Theory (IRT) within the scaling law framework. Unlike traditional approaches that treat each model-benchmark pair in isolation, IRSL disentangles latent model ability from question characteristics, factorizing the scaling law estimation for $M$ models and $N$ questions to significantly reduce parameter complexity from $O(M \times N)$ to $O(M + N)$. We instantiate IRSL with Beta-IRT, which leverages the empirical probability responses of LMs – such as token probabilities in pre-training and pass rates in test-time sampling – to capture richer signals than binary responses. We validate our approach across two prevalent scaling paradigms: (1) pre-training downstream scaling, using 6,612 LM checkpoints and 37,682 questions from 10 benchmarks; and (2) test-time scaling, using 12 LMs and 120 questions from 4 benchmarks with up to 2,500 samples per question. Given a one-time calibration on existing model responses, IRSL yields more reliable scaling estimates using only 50 questions per benchmark (a 99.9\% reduction), achieving comparable or superior decision accuracy to traditional approaches. Furthermore, we show that the estimated latent model abilities are generalizable, enabling accurate performance forecasting across benchmarks that share the same measurement objective.

207. Can You Trust What You See? Human and AI Detection of Synthetic Legal Evidence

Authors: Jinzhe Tan , Ali Ekber Cinar , Karim Benyekhlef
URL: https://arxiv.org/abs/2606.07613
Abstract:

Visual evidence has long been treated as a reliable form of legal proof, but advances in artificial intelligence (AI) are undermining that assumption. This article asks how well humans and frontier multimodal large language models (MLLMs) can distinguish authentic evidentiary photographs from AI-generated counterparts in the object-centric scenarios typical of civil disputes. We built Synthetic Legal Evidence Detection (SLED-1400), a dataset of 200 authentic evidence images paired with 1,200 synthetic counterparts produced by six contemporary text-to-image generators across ten evidence categories. The same stimuli and response format were used in a controlled web experiment with 136 lay participants and in a standardized evaluation of four MLLMs (GPT-5.1, Gemini-3-Pro, Gemini-3-Flash, Qwen3-VL-235B). Human accuracy was 64.8% overall, and 48.5% and 51.0% on the two strongest generators (Gemini-3-Pro-Image and Flux-2-Max), indistinguishable from chance. MLLMs never misclassified an authentic image (100% specificity), but missed most synthetic outputs from the harder generators, with average MLLM detection at 5.9% on Gemini-3-Pro-Image outputs. Human and MLLM errors were largely uncorrelated, while the four MLLMs were strongly correlated with each other. Neither group is a reliable standalone authenticator. We argue that visual evidence in legal proceedings should be treated as inherently contestable, and that a workable procedural response must combine trained human review, MLLM screening, and provenance infrastructure such as C2PA Content Credentials.

208. LEAF: Growing Trees Without Branching for Speech-Aware Large Language Model Post-Training

Authors: Argyrios Gerogiannis , Yekaterina Yegorova , Mark Hasegawa-Johnson , Venugopal V. Veeravalli
URL: https://arxiv.org/abs/2606.07610
Abstract:

State-of-the-art GRPO-style methods for speech-aware large language model post-training suffer from coarse credit assignment, broadcasting the same terminal-reward advantage to every token in a response. This ignores useful structure within rollout batches, where speech-conditioned completions often share prefixes before diverging at important decisions. We propose Low-rank Exploration with Adaptive Forking (LEAF), a retrospective tree-based RL method that recovers this structure without online branching or additional decoding. LEAF samples complete responses, selects high-surprisal boundaries, groups responses by shared prefixes, and assigns span-level advantages using descendant rewards. We theoretically justify LEAF’s span-level credit assignment and boundary-selection design. Empirically, LEAF improves over GRPO across speech question answering and speech translation benchmarks under the same rollout and low-rank adaptation budget. Notably, smaller LEAF-trained models outperform current state-of-the-art, full-parameter baselines.

209. Contribution Weights: A Geometrical Analysis of Self-Attention Transformers

Authors: Harry Jake Cunningham , Nicola Muca Cirone
URL: https://arxiv.org/abs/2606.07604
Abstract:

Analyzing attention weights has become a standard approach for interpreting the information flow of Large Language Models (LLMs). However, this approach has significant limitations as it neglects the geometric properties of the value vectors being aggregated. To address this gap, we introduce \emph{Contribution Weights}, a projection-based metric that quantifies a token’s influence by accounting for it’s attention weight, value magnitude, and directional alignment with the layer output. We demonstrate that contribution weights provide a more faithful measure of token importance, consistently outperforming attention-based metrics in identifying semantically critical tokens across different decoder-only models, tasks, and datasets. Further, our metric enables novel mechanistic analysis of \emph{attention sinks}. While previous work characterized sinks as passive repositories for excess attention, we reveal they serve an active functional role, suppressing information through a convex relationship between sink rate and output norm, stabilizing representations by opposing the semantic drift of low-confidence tokens.

210. MetaEvo: A Meta-Optimization Framework for Experience-Driven Agent Evolution

Authors: Bowen Ren , Heyan Huang , Yinghao Li , Yang Gao
URL: https://arxiv.org/abs/2606.07603
Abstract:

Large language models (LLMs) exhibit strong reasoning capabilities, yet most LLM-based agents are statically deployed and unable to improve through task interactions. Existing experience-driven methods often rely on memory or heuristics without enhancing the model’s ability to learn, treating it as a passive executor and leading to early performance plateaus and limited long-term improvement. To address this issue, we propose MetaEvo, a two-stage framework for continual agent evolution that focuses on improving how the model learns from tasks experience, rather than solely on what it stores. MetaEvo first applies preference-based optimization to enhance the model’s ability of principle abstraction, then enables the accumulation and reuse of these principles within a modular agent architecture. Experimental results on diverse reasoning benchmarks demonstrate that MetaEvo consistently outperforms strong baselines, maintains reliable improvement across iterations. These findings validate the effectiveness of meta-optimization in enabling agents to learn from experience and continually enhance their reasoning capabilities.

211. Sample-Efficient Post-Training for LEGO Spatial-Physics Reasoning

Authors: Yuhuan Yuan , Zhouliang Yu , Minghao Liu , Weiyang Liu , Ge Lin Kan
URL: https://arxiv.org/abs/2606.07602
Abstract:

LLM-based LEGO assembly generation requires both semantic grounding and physical feasibility. We identify a data-induced failure mode, PhysHack, in which the assemblies satisfy physical-validity constraints while producing structures that are geometrically misaligned, semantically inconsistent, or poorly calibrated. To address this challenge, we propose a model-based data selection approach that uses only a small fraction of the training data while improving physically grounded LEGO assembly generation. Building on the selected trajectories, we introduce PVPO, a sample-efficient reinforcement learning method that couples physical feasibility with voxel-space geometric rewards. Our results show that physical validity alone is an insufficient proxy for reliable physical reasoning: models can learn to generate valid structures without preserving semantic or geometric fidelity. Experiments across model backbones and test-time scaling settings demonstrate that PVPO improves structural and semantic alignment, physical validity, structural stability, and calibration, while reducing reliance on extensive post-hoc rejection sampling. In particular, results on calibration show that PVPO mitigates PhysHack by making test-time selection more predictive of semantic and structural quality.

212. Reachability and asymptotics of Gaussian Transformer dynamics

Authors: Albert Alcalde , Zhengping Ji , Enrique Zuazua
URL: https://arxiv.org/abs/2606.07600
Abstract:

We formulate data propagation through the Transformer, the machine learning architecture powering large language models, as a nonlinear control system on the space of probability measures. For the mean-field Transformer model with self-attention and affine feed-forward layers, we prove that Gaussian distributions remain exactly Gaussian along the induced flow. This invariance reduces the infinite-dimensional measure dynamics to a finite-dimensional bilinear control system governing the evolution of the mean and covariance, reformulates the expressive capacity of Transformers as a reachability problem for prescribed Gaussian moments, and reveals a novel connection with Riccati-type equations from classical filtering and control. For time-varying controls, we prove exact finite-time reachability of any target Gaussian distribution whose covariance matrix has the same rank as the initial one, this rank constraint being an intrinsic invariant of the dynamics. For time-invariant parameters, we derive explicit spectral conditions leading either to asymptotic stability toward positive-definite equilibria or to finite-time blow-up of the covariance. Numerical experiments complement the theory by showing that practical Transformers with Gaussian inputs remain close to moment-matched Gaussian distributions through early and intermediate layers, while Transformers with prescribed attention matrices reproduce the predicted covariance regimes: bounded evolution in stabilizing configurations and blow-up in destabilizing ones.

213. A Topological Characterization of Graph Neural Networks via Stochastic Block Model Embeddings on the n-Sphere

Authors: Gopal Anantharaman
URL: https://arxiv.org/abs/2606.07598
Abstract:

We propose a topological framework for comparing trained Graph Neural Networks (GNNs) by mapping the Stochastic Block Models (SBMs) induced on the graphon-signal space of a Message Passing Neural Network (MPNN) onto the unit $n$-sphere $\sphere^{n-1}\subset\R^n$. The construction rests on three classical pillars: the \emph{compactness} of the cut-distance graphon space $(\Wo,\cutdist)$ \citep{lovasz2006limits,lovasz2012large}, the Frieze–Kannan \emph{weak regularity lemma} together with its graphon-signal extension due to \citet{levie2023graphon}, and the Lipschitz continuity of MPNNs with respect to the cut-distance. We show that, for any prescribed tolerance $\varepsilon>0$, a trained MPNN $\Phi$ acting on a sufficiently large graph factors (up to $\varepsilon$) through a step-graphon-signal of bounded complexity, and we construct an explicit measure-preserving map $\Psi_n\colon[0,1]\to\sphere^{n-1}$ that places the SBM regions on disjoint spherical caps. This produces a problem-agnostic, low-dimensional ``fingerprint’’ of a trained GNN that is amenable to visual inspection and to nearest-neighbour search across model zoos, enabling \emph{transfer-learning candidate retrieval} without retraining. We discuss the obstruction posed by concentration of measure in high dimension – a phenomenon directly relevant to LLM-scale embeddings. We close with five concrete future research directions: hyperbolic and Grassmannian alternatives to the spherical model, Gromov–Wasserstein distances on graphon-signals as an isometry-free alternative to the $n$-sphere map, an information-geometric (Fisher) reformulation of the SBM manifold, persistent-homology fingerprints of layer-wise embedding clouds, and a spectral-distance baseline derived from the graphon eigendecomposition.

214. ResearchClawBench: A Benchmark for End-to-End Autonomous Scientific Research

Authors: Wanghan Xu , Shuo Li , Tianlin Ye , Qinglong Cao , Yixin Chen , Hengjian Gao , Yiheng Wang , Qi Li , Kun Li , Sheng Xu , Shengdu Chai , Fangchen Yu , Xiangyu Zhao , Zhangrui Zhao , Weijie Ma , Zijie Guo , Haoyu Zhou , Haoxiang Yin , Lixue Cheng , Chaofan Hu , Haoxuan Li , Lu Mi , Xuxuan Xie , Yifan Zhou , Ruizhe Chen , Zhiwang Zhou , Xingjian Guo , Yuhao Zhou , Xuming He , Shengyuan Xu , Xinyu Gu , Jiamin Wu , Mianxin Liu , Chunfeng Song , Fenghua Ling , Dongzhan Zhou , Shixiang Tang , Yuqiang Li , Mao Su , Peng Ye , Siqi Sun , Bin Wang , Xue Yang , Zhenfei Yin , Tianfan Fu , Guangtao Zhai , Wanli Ouyang , Bo Zhang , Lei Bai , Wenlong Zhang
URL: https://arxiv.org/abs/2606.07591
Abstract:

AI coding agents are increasingly used for scientific work, but their end-to-end autonomous research capability remains difficult to verify. We present ResearchClawBench, a benchmark for evaluating autonomous scientific research across 40 tasks from 10 scientific domains. Each task is grounded in a real published paper, provides related literature and raw data, and hides the target paper during evaluation. Expert-curated multimodal rubrics decompose the target scientific artifacts into weighted criteria, enabling evaluation of target-paper-level re-discovery while leaving room for new discovery. We evaluate seven autonomous research (auto-research) agents under a unified protocol and seventeen native LLMs through the lightweight ResearchHarness. Current systems remain far from reliable re-discovery: the strongest autonomous agent, Claude Code, averages 21.5, and the strongest ResearchHarness LLM, Claude-Opus-4.7, averages 20.7, with an LLM frontier mean of only 26.5. Error analysis shows that failures concentrate in experimental protocol mismatch, evidence mismatch, and missing scientific core. ResearchClawBench provides a reproducible evaluation frontier for measuring progress toward autonomous scientific research.

215. From Human Guidance to Autonomy: Agent Skill System for End-to-End LLM Deployment on Spatial NPUs

Authors: Jiajie Li , Erwei Wang , Zhiru Zhang , Samuel Bayliss
URL: https://arxiv.org/abs/2606.07586
Abstract:

Spatial neural processing units (NPUs) provide an energy-efficient platform for edge LLM inference, but efficiently deploying an LLM end-to-end on such hardware remains labor-intensive. Although AI coding agents have begun to lower this cost, existing studies have largely focused on single-kernel optimization rather than end-to-end LLM deployment on resource-constrained spatial NPUs. We present a two-stage methodology, instantiated on the AMD XDNA 2 NPU, that progresses from human-guided development to agent autonomy. In the first stage, we develop a reference deployment of Llama-3.2-1B through human-guided agent assistance. The resulting implementation achieves a speedup of 2.2x on prefill and 4.0x on decode over the hand-optimized baseline, with the optimization trajectory and its lessons recorded as structured documentation throughout. In the second stage, we distill the documentation into an agent skill system consisting of eight phases, orchestrating the optimization and debugging skill sets, with numerical correctness strictly enforced at each phase. Using our agent skill system, we autonomously deploy eight additional decoder-only LLMs (Llama-3.2-3B, SmolLM2-1.7B, Qwen2.5-{0.5B, 1.5B, 3B}, Qwen3-{0.6B, 1.7B, 4B}) end-to-end on the AMD XDNA 2 NPU using the open-source compiler stack. To our knowledge, these models have not previously been deployed on AMD NPUs via any open-source software stack. Each deployment completes in 0.5-4 hours of agent wall time with almost no human guidance, and passes the numerical-correctness gates, demonstrating functional generalization to previously unencountered LLMs. Three of the eight match or exceed the sustained performance of our Llama-3.2-1B reference deployment, suggesting that the resulting implementations can be competitive without additional model-specific human engineering.

216. Enabling KV Caching of Shared Prefix for Diffusion Language Models

Authors: Younghun Go , Jaehoon Han , Changyong Shin , Chuk Yoo , Gyeongsik Yang
URL: https://arxiv.org/abs/2606.07571
Abstract:

Key-value (KV) caching for shared prefixes is essential for high-throughput large language model (LLM) serving, but it faces critical challenges in emerging diffusion language models (DLMs). In DLMs, bidirectional attention means that updating any token dynamically alters the entire context and its corresponding KVs. Thus, existing caching techniques developed for LLMs, which assume that KVs remain invariant once computed, corrupt the shared prefix KVs. Our experiments show that applying these techniques to DLMs causes model accuracy to collapse to near zero. To unlock high-throughput DLM serving, we propose bidirectional prefix caching, bicache, the first KV caching technique for shared prefixes in DLMs. bicache is designed based on key observations from our comprehensive analysis: shared prefix KVs remain stable and reusable in shallow layers, while the depth of shallow layers depends on the fraction of shared prefix tokens in each request. Thus, bicache dynamically identifies a safe layer depth for reusing shared prefix KVs and eliminates redundant computation. Evaluations demonstrate that bicache significantly improves serving throughput by 36.3%-98.3% compared to existing techniques without accuracy collapse (only 0-1.8% difference).

217. SurfDesign: Effective Protein Design on Molecular Surfaces

Authors: Fang Wu , Shuting Jin , Xiangru Tang , Mark Gerstein , Xiangxiang Zeng , Yejin Choi , Jure Leskovec , Jinbo Xu
URL: https://arxiv.org/abs/2606.07567
Abstract:

Protein function is largely determined by molecular surface geometry and physicochemical complementarity, yet most protein design methods condition only on backbone structure. We introduce SurfDesign, a surface-conditioned protein design framework that models molecular surfaces as continuous geometric manifolds and integrates them with pretrained protein language models. SurfDesign employs surface-based equivariant message passing to capture surface normals, curvature, and directional geometry, together with a parameter-efficient fine-tuning strategy. Focusing on functional protein design, we show that SurfDesign consistently outperforms prior surface-conditioned and backbone-only methods on de novo binder and enzyme design benchmarks. We also report strong performance on inverse-folding benchmarks as a diagnostic of structural compatibility. Our results highlight manifold-aware surface representations as a principled foundation for functional protein and enzyme design. Code is available at this https URL .

218. Phantom transitions in language model fine-tuning

Authors: Vaibhav Prakash , Jayasri Dontabhaktuni
URL: https://arxiv.org/abs/2606.07559
Abstract:

Fine-tuning a language model on contexts whose correct completion has a near-synonym competitor often fails silently. The cross-entropy loss decreases monotonically while the correct token never overtakes the competitor in rank. We study this regime across five transformer architectures spanning two families and a fivefold parameter range, on ten hand-selected near-synonym contexts. We instrument these failures with an order parameter combining the predicted distribution and pairwise embedding overlaps. It decomposes additively into a signal, tracking the model’s commitment to the correct token over its nearest competitor, and a background drag, set by how the embedding bulk leaks probability into the score. This isolates two failure modes. In kinematic failure the signal stays small. In structural failure the drag actively worsens as fine-tuning proceeds. We observe sharp catapult-like jumps in the order parameter that resemble a phase transition. A central negative result organises the paper. The transitions are phantoms. The spontaneous-symmetry-breaking interpretation is ruled out by direct measurement. Catapult-like jumps still appear under LoRA fine-tuning with the token embedding matrix exactly unchanged during training, where no geometric phase transition is possible. The discontinuity lives entirely in the softmax readout. A small number of dimensionless quantities organise the trajectory across architectures. One is consistent across all five under full fine-tuning. A second sorts architectures into two classes by bulk embedding distribution and predicts LoRA sufficiency. As a blind test, the framework predicts the critical learning rate of a held-out architecture, not used to fit any parameter, to within 2.1% of a subsequent learning-rate sweep. Findings concern the near-synonym mechanism only and should not be extrapolated without recalibration.

219. Symbolic Reasoning Frameworks Modulate LLM Risk Aversion in Multi-Agent Strategic Settings

Authors: Augustin Chan
URL: https://arxiv.org/abs/2606.07552
Abstract:

Large language models exhibit innate behavioral tendencies when deployed as strategic agents – notably a risk-averse “turtle” bias toward defensive play. We show that symbolic reasoning frameworks, injected as per-round reflective prompts into one agent, differentially modulate this bias and reshape the multi-agent ecosystem to produce framework-specific winner distributions. In a 7-player Warring States Diplomacy variant (41 games, 4 conditions, single-campaign memory accumulation), each framework produces a distinct ecosystem signature: under control, Yan dominates (7/11, 64%); under I-Ching yarrow divination, Yan and Chu co-dominate while Qin is completely suppressed (0/10); under Tarot, Qin dominates (5/10, Fisher vs. pooled p = 0.006); under scrambled-text ablation (incoherent oracle text preserving prompt structure), Qi dominates (5/10, Fisher vs. pooled p = 0.006). The framework-receiving agent (Han) never wins and shows no survival difference across conditions (Fisher p = 1.0), but Tarot consistently elevates Han’s peak territory (mean 3.0 SCs vs. 2.1-2.5 others, Kruskal-Wallis p = 0.010). Neither framework’s content predicts subsequent actions – hexagram themes (chi-squared p = 0.95) and Tarot card postures (chi-squared p = 0.69) are both independent of action choice – suggesting the modulation operates through the reflective process, not content-following. We present this as an observation paper establishing that alignment-framework choice at the agent level produces distinctive system-level consequences in multi-agent settings.

220. Evaluating Advanced Prompting on Gemini Flash for Multi-Hop Biomedical QA

Authors: Ahmed Bajaber , Mohammed Alliheedi
URL: https://arxiv.org/abs/2606.07548
Abstract:

The MedHopQA challenge presents a critical test for Large Language Models (LLMs): complex, multi-hop reasoning in the high-stakes biomedical domain. This paper details our direct API-based evaluation of Google’s Gemini Flash models, focusing on the impact of advanced prompt engineering. We designed a sophisticated, multi-component prompt for Gemini 2.0 Flash that combined role-playing, explicit multi-shot Chain-of-Thought (CoT) examples, and detailed formatting rules. Our best run, using this complex prompt, achieved a Concept Level Score of 0.720. This result dramatically outperformed a baseline prompt which scored only 0.565. Remarkably, this performance on the efficient Gemini 2.0 Flash was almost identical to the result from the next-generation Gemini 2.5 Flash. Our findings demonstrate that sophisticated prompt design is a critical factor for unlocking the full reasoning capabilities of modern LLMs.

221. Liberating LLM Capabilities in Full-Duplex Speech Models

Authors: Luoyuan Zhang , Bokai Xu , Junbo Cui , Weiyue Sun , Yingjing Xu , Hanyu Liu , Yuan Yao
URL: https://arxiv.org/abs/2606.07547
Abstract:

Speech-based large language models are typically constrained to spoken replies, which limits their user-facing outputs to what can be verbalized and suppresses text-native capabilities such as code generation, structured analysis, and multi-step reasoning in realtime interaction, for tasks that require persistent, structured, and inspectable intermediate outputs. Existing work improves spoken reasoning or full-duplex turn-taking, but still treats text as a hidden intermediate state or a subordinate modality rather than a first-class output channel. We propose Listen-Write-Speak (LWS), a text-first tri-channel paradigm in which a single autoregressive LLM continuously listens to user audio, writes visible free-form text as its primary output, and speaks a realtime oral response in parallel under a shared causal attention context. This behavior is implemented entirely through a Token Schema, requiring no architectural modifications, and learned via a two-stage data pipeline that synthesizes per-second cognitive annotations consistent with the revealed input timeline. Empirically, LWS demonstrates strong full-duplex interaction on Full-Duplex-Bench, reaches 4.72 on VoiceBench AlpacaEval, achieves 92.6% writing-speaking consistency, and consistently outperforms its internal ablations on URO-Bench. These results suggest that visible writing can serve as a first-class output channel for speech interaction without sacrificing realtime responsiveness. The code and dataset are available on the project page: this https URL .

222. Multimodal Large Language Models as Synthetic Participants in Video-Based Studies: An Evaluation

Authors: Prabal Shrestha , Bohan Jiang , Haoning Xue , Huan Liu , Xinyi Zhou
URL: https://arxiv.org/abs/2606.07541
Abstract:

Multimodal large language models (MLLMs) have shown strong performance on objective tasks such as video understanding and reasoning. However, it remains unclear whether they can approximate subjective human responses, which depend not only on content comprehension but also on individuals’ social contexts. To address this gap, we evaluate MLLMs as synthetic participants in an emerging task: assessing perceived sensory engagement with short videos. Grounded in the Perceived Message Sensation Value (PMSV) framework, we compare ratings from recruited human participants and profile-conditioned MLLM simulations (n=673) using a 17-item scale measuring emotional arousal, dramatic impact, and novelty. We find that even leading MLLMs (Gemini 3 Flash and Qwen 3 Omni) show limited agreement with human participants. The models exhibit distinct downward mean-shift and central-tendency biases in their rating distributions. They both introduce and flatten subgroup differences, while showing inconsistent sensitivity to participant profiles. Prompting strategies affect these metrics differently, modestly improving some aspects while worsening others. These results highlight both the challenges and opportunities of developing MLLMs as synthetic participants in video-based research. Data and code: this https URL

223. Bidirectional Semantic Complementary Tool Retrieval for Remote Sensing Agents

Authors: Zeyuan Wang , Dongyang Hou , Cheng Yang , Xuezhi Cui , Linrui Xu , Bo Yu , Gaozhi Zhou , Ziyu Li , Liangtian Liu , Kai Ouyang , Wang Guo , Lili Zhu , Chao Tao
URL: https://arxiv.org/abs/2606.07538
Abstract:

Large language model (LLM)-based agents provide a novel paradigm for the automated processing of remote sensing(RS) data. Their success in complex RS tasks rely on extensive specialized tool libraries. However, tool documentation often exceeds the context window limits of LLMs, making precise tool retrieval essential for agentic workflows. Existing tool retrieval methods face “semantic asymmetry” bottleneck: natural language queries typically express macro-level intentions lacking tool-specific semantics, while tool documentation provides fine-grained technical descriptions lacking operational context for workflows. To bridge this semantic gap, this paper proposes a bidirectional semantic complementary tool retrieval method. First, on the query side, we introduce a planning-based query enhancement mechanism that leverages the reasoning capabilities of agents to decompose abstract intentions into logical subtasks, thereby actively supplementing the query with missing functional semantics. Second, on the tool side, addressing the strong coupling characteristics of RS tool chains, we construct a dynamic tool dependency graph with continual learning capabilities. By employing a neighborhood information aggregation mechanism, contextual information from precursor tools is explicitly injected into the current node representation, enriching tool descriptions with contextual semantics. Experimental results on the RS dataset GeoPlan-bench and the general-purpose dataset API- Bank demonstrate that the proposed method not only significantly improves tool retrieval accuracy for complex RS tasks but also exhibits robust extensibility for transfer to general-domain tasks. The source code and dataset are available at this https URL .

224. Bridging Traditional Explainability Methods and Multimodal Multilingual Models: An XAI-Based Analysis

Authors: Paweł Pozorski , Jakub Muszyński , Maria Ganzha
URL: https://arxiv.org/abs/2606.07533
Abstract:

Multimodal Large Language Models (MLLMs) effectively integrate text and audio to interpret context in complex interactive dialogues. However, the internal mechanisms by which heterogeneous modalities influence model behavior remain opaque. While Shapley Values (SV) provide a robust, model-agnostic framework for local explainability in text-based NLP, their extension to multimodal data is hindered by cross-channel dependencies, intricate dialogue structures, and the prohibitive computational complexity of dense audio representations. In this work, we formalize a multimodal extension of the Shapley Value framework, treating discrete text tokens and aligned audio segments as cooperative features. To ensure computational feasibility, we deploy a suite of efficient estimation strategies: exact SV computation for low-dimensional inputs and sampling-based approximations - including Monte Carlo permutations and stratified sampling with Neyman-optimal allocation - to minimize variance under constrained computational budgets. To resolve the granularity mismatch between modalities, we propose Spectrogram-Guided Phonetic Alignment (SGPA), a novel preprocessing method that maps high-frequency audio streams to interpretable, word-aligned segments. Our contribution is twofold: first, we provide an open-source, model-agnostic Python package and a companion GUI for the computation and interactive visualization of multimodal attributions. Second, we evaluate our framework using curated subsets of the VoiceBench and Infinity Instruct datasets across diverse multilingual scenarios. Our experimental results reveal that input modality is a primary driver of attribution volatility and demonstrate that standard syntactic importance proxies often fail to predict model attention in multimodal, cross-lingual contexts.

225. Principled Agent Debate: Adversarial Arbitration for Sycophancy Reduction in Large Language Models

Authors: Sam Ryan
URL: https://arxiv.org/abs/2606.07532
Abstract:

RLHF-trained models are systematically biased toward agreement over accuracy, a structural property of the training process. We present Principled Agent Debate (PAD), a multi-agent architecture that mitigates identity-framed sycophancy by arbitrating between two models tuned to opposing philosophical dispositions, with a pragmatist synthesizer evaluating both arguments blind to their origins. This paper evaluates a prompt-based instantiation of PAD. The key mechanisms are static dispositional tuning, identity stripping before synthesis, single-round independent argumentation, and blind arbitration. We evaluate five instantiations on 200 stratified questions from SycophancyEval. All PAD variants (AnCifer, DeWin, FeynStein, BurGal, Trident) significantly outperform the single-model baseline (18.5%) and instructed-opposition baseline (29.0%), with DeWin achieving 48.5% accuracy (z=6.36, p<0.001 versus both). The variants are not significantly different from each other at n=200. The BurGal variant achieves 53.0% but functions as an architectural validity check; its consensus/heterodox axis structurally favors the heterodox model on every benchmark question. A pre-training floor affects an estimated 40% of questions; fine-tuned disposition models are the identified next step.

226. mllm-shap: A Shapley Value Explainability Platform for Text-Audio Multimodal Large Language Models

Authors: Jakub Muszyński , Paweł Pozorski , Maria Ganzha
URL: https://arxiv.org/abs/2606.07531
Abstract:

We introduce mllm-shap, an open-source Python framework designed to extend Shapley Value (SV) explainability from text-only Large Language Models to Multimodal LLMs (MLLMs) processing joint text and audio inputs. While text-based attribution is well-studied, mllm-shap addresses three critical challenges unique to the multimodal regime: (1) Modality-aware coalition masking, which manages the interleaved processing of discrete text tokens and dense audio encoder frames. (2) Multi-turn conversation tracking, utilizing per-token metadata to maintain role and modality context. (3) Phonetic alignment-based token grouping, a novel technique that reduces the coalition space by 10x to 50x, rendering SV estimation computationally feasible for long-form audio. The platform implements five SV estimation strategies, including a Complementary Contributions (CC) estimator with Neyman-optimal allocation that demonstrates superior convergence over standard Monte Carlo baselines. mllm-shap is provided as a pip-installable package featuring an interactive web-based GUI for granular attribution visualization. To our knowledge, this is the first publicly available framework providing a complete, reproducible pipeline for SV-based explainability in text-audio MLLMs.

227. CAPruner: Conceptual-Adjacent Scene Graph Pruner for Enhancing 3D Spatial Reasoning of Large Language Models

Authors: Shengli Zhou , Xiangchen Wang , Guanhua Chen , Feng Zheng
URL: https://arxiv.org/abs/2606.07529
Abstract:

Large language models (LLMs) have recently been applied to 3D vision-language (3D-VL) tasks, which require spatial reasoning to identify target objects relative to anchors. Scene graphs are commonly employed to represent such relations, but reasoning over complete graphs incurs high token costs and computational inefficiencies, motivating the need for pruning. Existing pruning methods primarily rely on spatial proximity and often remove task-relevant relations, thereby undermining reliable spatial reasoning. To address these limitations, we derive a key requirement for scene graph pruning: preserving spatial relations that are most pertinent to the specific 3D-VL task. Guided by this insight, we propose the Conceptual-Adjacent Scene Graph Pruner (CAPruner). CAPruner integrates fuzzy semantic relevance with spatial proximity to estimate the importance of relations, enabling the selection of critical relations in a task-specific context. Moreover, to avoid costly relation-level annotations, CAPruner is trained by supervising the aggregated scores of each node’s incident edges. Extensive experiments demonstrate that CAPruner effectively preserves relations essential for spatial reasoning, leading to substantial performance improvements of LLMs on 3D-VL tasks. Code is available at this https URL .

228. BEACON: Behavioral Entropy Aggregation for Cross-Model Hallucination Detection in Large Language Models

Authors: Naveen Bera , Pulijala Sai Nikhila , Kondaguduru Abhiram , Shaik Gayaz Ali , Shoaib Sadiq Salehmohamed , Shaik Mohammed Omar , Jinal Prashant Thakkar , Hansika Aredla , Shalmali Ayachit
URL: https://arxiv.org/abs/2606.07528
Abstract:

Hallucination in large language models (LLMs), defined as the generation of factually incorrect or unsupported content, remains a critical barrier to reliable deployment. We present BEACON (Behavioral Entropy Aggregation for Cross-model hallucination detectiON), a black-box hallucination detection framework that operates purely on model outputs without requiring access to internal representations or external knowledge bases. BEACON extracts a 31-dimensional feature vector from structured multi-pass generation, integrating NLI-based semantic entropy, embedding geometry, chain-of-thought consistency, and paraphrase stability signals. A gradient-boosted classifier trained on 7,617 labeled examples across seven benchmarks achieves 0.8123 +/- 0.0102 AUROC (95% CI: 0.7632-0.8251), outperforming standalone semantic entropy (+0.2298) and SelfCheckGPT-style consistency baselines (+0.2457). Feature importance analysis shows that hallucination is inherently multi-dimensional, requiring combined uncertainty signals. An efficient 5-call variant achieves 0.7795 AUROC, enabling practical deployment across black-box LLM APIs.

229. Post-training is (Massive) Supervised Learning

Authors: Michael Hassid , Yossi Adi , Roy Schwartz
URL: https://arxiv.org/abs/2606.07527
Abstract:

The prevailing paradigm for training LLMs has evolved to rely on a massive post-training phase consisting of SFT and RL. In this position paper, we argue that this methodology effectively marks a reversion to the pre-train then fine-tune'' approach of the BERT era, explicitly tailoring models to the desired behaviors and specific benchmarks on which they are evaluated. We begin with a historical overview of LLMs, describing the different phases of the LLM evolution. We argue that the current landscape is remarkably similar to the early days of LLMs, where task performance heavily relied on fitting the models to in-distribution datasets. To empirically demonstrate this, we compare pre-trained models to randomly initialized ones, by fine-tuning both variants on modern reasoning datasets and evaluating them on competitive math and code benchmarks. We show that models post-trained from scratch yield highly non-trivial performance. Our findings suggest that current post-training methodologies function primarily as a distribution-fitting mechanism. We finish by positing that developing generally capable models and systems requires moving beyond extensive post-training for predefined behaviors, shifting instead toward training procedures where modelslearn how to learn’’.

LLM 관련 주요 논문 - 2026-06-09

1. SearchSwarm: Towards Delegation Intelligence in Agentic LLMs for Long-Horizon Deep Research

2. Beyond Probabilistic Similarity: Structural, Temporal, and Causal Limitations of Retrieval-Augmented Generation in the Legal Domain

3. (Auto)formalization is supposed to be easy: Trellis process semantics for spelling out rigorous proofs

4. Correlation Is Not Enough: Embedding Human Metadata for Individual Causal Discovery

5. SpatialWorld: Benchmarking Interactive Spatial Reasoning of Multimodal Agents in Real-World Tasks

6. Optical Reasoning: Rethinking Images as an Expressive Reasoning Medium Beyond Text

7. TABVERSE: Benchmarking Cross-Format Table Understanding in LLMs and VLMs

8. PRISM: Recovering Instruction Sets from Language Model Activations

9. AI Scientists Are Only as Good as Their Evidence: A Stratified Ablation of Proprietary Data and Reasoning Skills in Drug-Asset Valuation

10. From Rigid to Dynamic: Entropy-Guided Adaptive Inference for Long-Context LLMs

11. Deterministic Integrity Gates for LLM-Assisted Clinical Manuscript Preparation: An Auditable Biomedical Informatics Architecture

12. LLM-Orchestrated Conformance Checking in Stroke Care Without Computer-Interpretable Guidelines

13. AliyunConsoleAgent: Training Web Agents in Real-World Cloud Environments via Distillation and Reinforcement Learning

14. SIFT: Selective-Index For Fast Compute of RAG Prefill by Exploiting Attention Invariance

15. RunAgent SuperBrowser: A Theory of Autonomous Web Navigation Grounded in Human Browsing Behaviour

16. Capability-Aligned Hierarchical Learning for Tool-Augmented LLMs

17. MASS: Deep Research for Social Sciences with Memory-Augmented Social Simulation

18. Vision Language Model Helps Private Information De-Identification in Vision Data

19. Late-Layer Fusion is Enough: Dual-Path Vision Token Routing for Multimodal Large Language Models under Visual Saturation

20. A Regret Minimization Framework on Preference Learning in Large Language Models

21. ComplexConstraints and Beyond: Expert Rubrics for RLVR

22. Graph2Idea:Retrieval-Augmented Scientific Idea Generation with Graph-Structured Contexts

23. REFLECT: Intervention-Supported Error Attribution for Silent Failures in LLM Agent Traces

24. Personalization Meets Safety:Mechanisms,Risks,and Mitigations in Personalized LLMs

25. A Multi-Agent System for IPMSM Design Optimization via an FEA-AI Hybrid Approach

26. LATTEArena: An Evaluation Framework for LLM-powered Tabular Feature Engineering (Extended Version)

27. RTL-BenchLS: A Large-Scale Benchmark for RTL Reasoning and Generation with Large Language Models

28. Diverse Thinking Schemata Elicit Better Reasoning in Large Language Models

29. An Effective Router for Vision-Language Model Selection

30. Oversight Has a Capacity: Calibrating Agent Guards to a Subjective, Fatiguing Human

31. Order Matters: Unveiling the Hidden Impact of Macro Placement Sequences via Proxy-Guided LLM Evolution

32. Can the Environment Speak for Itself? $T^{2}$-GRPO: A Turn-Trajectory Group Relative Policy Optimization for Caregiver Agents

33. Hybrid E-Assessment in Higher Education: Semi-Automated Grading of Paper-Based Written Examinations

34. ZIPP:Zero-shot Image Personalization from Personas

35. Inference-Time Conformal Reasoning with Valid Factuality Control for Large Language Models

36. Momentum for Reasoning: Dense Intrinsic Signals in Policy Optimization

37. Bridging Expert Knowledge and Automated Feature Engineering via Self-Evolution

38. RAILS: Verification-Native Clearing For Agentic Commerce

39. Artificial Intelligence for Mathematical Reasoning: An Integrated Survey of Language Models, Neuro-symbolic Systems, and Verified Discovery

40. ConMem: Structured Memory-Guided Adaptation in Training-Free Multi-Agent Systems

41. Towards Long-Horizon Vessel Trajectory and Destination Forecasting with Reasoning Large Language Models

42. InA-Probe: Instruction-Aware Active Probing for Time Series Forecasting with LLMs

43. Distilling LLM Reasoning into an Interpretable Policy Tree for Human-AI Collaboration

44. PAEC: Position-Aware Entropy Calibration for LLM Reasoning in RLVR

45. AgentTrust: A Self-Improving Trust Layer for AI-Agent Actions

46. DN-Hypo-Pipeline: An AI-Driven Workflow for Hypothesis Generation via Large Language Models and Scientific Explanations

47. VESTA: A Fully Automated Scenario Generation and Safety Evaluation Framework for LLM Agents

48. Explaining Black-Box Language Models: Learning to Optimize Linguistically-Structured Word Subsets

49. Testing the Black Box: Structural Barriers to Independent Evaluation of Consumer-Facing Health LLMs

50. A Variability-Based Framework for Interpretable Naming in Formal and Relational Concept Analysis

51. GIFT: LLM-Guided State-Reward Interface for Financial Reinforcement Learning

52. Trajectory-Refined Distillation

53. Self-Evolving Scientific Agent Discovers Generalizable Physically-Reasoned Fluid Control

54. Benchmarking Open-Ended Multi-Agent Coordination in Language Agents

55. To Nuke or Not to Nuke: LLMs’ (Missing) Ethical Reasoning and Actions in a High-Stakes Decision-Making Simulation

56. Beyond Agent Architecture: Execution Assumptions and Reproducibility in LLM-Based Trading Systems

57. When No Answer Is Correct: Diagnosing Absent Answer Detection for MLLMs in Video Understanding

58. SciTrace: Trajectory-Aware Safety Reasoning for Scientific Discovery Agents

59. Online Agent-as-a-Judge: Situation-Generating Evaluation for Interactive Agents

60. Decision-Aware Memory Cards: Counterfactual-Inspired Context Selection and Compression for Tool-Using LLM Agents

61. SAGE: An LLM-driven Self Reflective Agentic Framework for Fraud Detection

62. Cross-LLM Consistency in Inference: Evidence from Shared Interactions

63. Think Before You Act: Intention-Guided Reasoning for LLM-Based Location Prediction

64. When Does Delegation Beat Majority? A Delegation-Based Aggregator for Multi-Sample LLM Inference

65. Efficient Skill Grounding via Code Refactoring with Small Language Models

66. PAFO: Pareto Fairness Optimization for Personalized Reward Modeling

67. Zero-Shot Learning in Industrial Scenarios: New Large-Scale Benchmark, Challenges and Baseline

68. Shared Latent Structures Enable Unified Backdoor Detection and Mitigation in LLMs

69. Unification of Closed-Open Industrial Detection Scenarios: New Large-Scale Benchmarks,Challenges and Baselines

70. Stress-testing medical large language models reveals latent safety pathology beyond benchmark accuracy

71. MemToolAgent overview with a simple restaurant booking scenario where the agent retrieves similar memories, receives feedback on an invalid time format, and generates a reflection to update its memory

72. Contract2Tool: Learning Preconditions and Effects for Reliable Tool-Augmented LLM Agents

73. Safety is Contextual, LLM-Judges Are Not: Navigating the Rigid Priors of Evaluators

74. Joint Structural Pruning and Mixed-Precision Quantization for LLM Compression

75. Where Instruction Hierarchy Breaks: Diagnosing and Repairing Failures in Reasoning Language Models

76. Beyond Goodhart’s Law: A Dynamic Benchmark for Evaluating Compliance in Multi-Agent Systems

77. Some hypotheses on how chatbots work in problem-solving-driven conversations. Large Language Models as confirmation of the Innovation Illusion

78. Automatic Extraction of Structured Information from Brain MRI Reports Using an Open-Weight Large Language Model

79. Why Limit the Residual Stream to Layers and Not Tokens? Persistent Memory for Continuous Latent Reasoning