LLM 관련 주요 논문 - 2026-05-11

1. VecCISC: Improving Confidence-Informed Self-Consistency with Reasoning Trace Clustering and Candidate Answer Selection

Authors: James Petullo , Sonny George , Dylan Cashman , Nianwen Xue
URL: https://arxiv.org/abs/2605.08070
Abstract:

A standard technique for scaling inference-time reasoning is Self-Consistency, whereby multiple candidate answers are sampled from an LLM and the most common answer is selected. More recently, it has been shown that weighted majority voting (e.g. Confidence-Informed Self Consistency (CISC)), which assigns a confidence value to each candidate answer and chooses the answer with the largest accumulated score, tends to be more accurate on a wide range of popular benchmarks. In practice, weighted majority voting necessitates calling a critic LLM on each candidate’s reasoning trace to produce the answer’s confidence score. This secondary series of LLM calls greatly increases the overhead and cost of weighted majority voting, despite its potential performance benefits. To reduce this expense, we propose VecCISC, a lightweight, adaptive framework that uses a measure of semantic similarity to filter reasoning traces that are semantically equivalent to others, degenerate, or hallucinated, thus decreasing the number of candidate answers that must be evaluated by the critic. To ensure adequate experimental thoroughness, we evaluate VecCISC on five challenging, widely-adopted datasets spanning the domains of mathematics, chemistry, biology, commonsense reasoning, and the humanities. Our results demonstrate that VecCISC reduces the total token usage by 47%, while maintaining or exceeding the accuracy of CISC.

2. Rubric-Grounded RL: Structured Judge Rewards for Generalizable Reasoning

Authors: Manish Bhattarai , Ismael Boureima , Nishath Rajiv Ranasinghe , Scott Pakin , Dan O’Malley
URL: https://arxiv.org/abs/2605.08061
Abstract:

We argue that decomposing reward into weighted, verifiable criteria and using an LLM judge to score them provides a partial-credit optimization signal: instead of a binary outcome or a single holistic score, each response is graded along multiple task-specific criteria. We formalize \emph{rubric-grounded reinforcement learning (RL)}: a framework in which the policy is optimized against a structured, multi-criterion reward produced by a frozen LLM judge that conditions on auxiliary grounding the policy never sees. We instantiate the framework by deriving rubrics from an Office of Scientific and Technical Information (OSTI)-derived corpus of roughly 100,000 scientific and technical documents and training Llama-3.1-8B-Instruct with Group Relative Policy Optimization (GRPO). With GRPO-based training, the model achieves $71.7\%$ normalized reward on held-out rubric evaluation. The GRPO-tuned policy also improves over the base model on four reasoning benchmarks not derived from the training corpus – GSM8K, MATH, GPQA Main, and GPQA Diamond. These results provide evidence that structured, document-grounded rewards can improve held-out rubric performance and induce transferable reasoning behaviors beyond the corpus used to construct the training environment.

3. Abductive Reasoning with Probabilistic Commonsense

Authors: Joseph Cotnareanu , Chiara Roverato , Han Zhou , Didier Chetelat , Yingxue Zhang , Mark Coates
URL: https://arxiv.org/abs/2605.08011
Abstract:

Recent efforts to improve the reasoning abilities of Large Language Models (LLMs) have focused on integrating formal logic solvers within neurosymbolic frameworks. A key challenge is that formal solvers lack commonsense world knowledge, preventing them from making reasoning steps that humans find obvious. Prior methods address this by using LLMs to supply missing commonsense assumptions, but these approaches implicitly assume universal agreement on such commonsense facts. In reality, commonsense beliefs vary across individuals. We propose a probabilistic framework for abductive commonsense reasoning that explicitly models this variation, aiming to determine whether most people would judge a statement as true or false. We introduce Probabilistic Abductive CommonSense (PACS), a novel algorithm that uses an LLM and a formal solver to sample proofs as observations of individuals’ distinct commonsense beliefs, and aggregates conclusions across these samples. Empirically, PACS outperforms chain-of-thought reasoning, prior neurosymbolic methods, and search-based approaches across multiple benchmarks.

4. TraceFix: Repairing Agent Coordination Protocols with TLA+ Counterexamples

Authors: Shuren Xia , Qiwei Li , Taqiya Ehsan , Jorge Ortiz
URL: https://arxiv.org/abs/2605.07935
Abstract:

We present TraceFix, a verification-first pipeline for Large Language Model (LLM) multi-agent coordination. An agent synthesizes a protocol topology as a structured intermediate representation (IR) from a task description, generates PlusCal coordination logic, and iteratively repairs the protocol using counterexamples from the TLA+ model checker (TLC) until verification succeeds. Verified process bodies are compiled into per-agent system prompts and executed under a runtime monitor that rejects out-of-topology coordination operations. On 48 tasks spanning 16 scenario families, all tasks reach full TLC verification; 62.5% pass on the first attempt and none requires more than four repair iterations. State spaces span six orders of magnitude yet verification completes in under 60 s for every task. A 3,456-run runtime comparison shows that topology-monitored execution achieves the highest task completion (89.4% average, 81.5% full) and that runtimes using the verified protocol degrade at roughly half the rate of prompt-only and chat-only baselines when model capability is reduced. A paired ablation under a fixed runtime shows that TLC-verified protocols cut deadlock/livelock (DL/LL) from 31.1% to 14.1%, with the largest separation under fault injection.

5. AgentEscapeBench: Evaluating Out-of-Domain Tool-Grounded Reasoning in LLM Agents

Authors: Zhengkang Guo , Yiyang Li , Lin Qiu , Xiaohua Wang , Jingwen Xv , Dongyu Ru , Xiaoyu Li , Xiaoqing Zheng , Xuezhi Cao , Xunliang Cai
URL: https://arxiv.org/abs/2605.07926
Abstract:

As LLM-based agents increasingly rely on external tools, it is important to evaluate their ability to sustain tool-grounded reasoning beyond familiar workflows and short-range interactions. We introduce AgentEscapeBench, an escape-room-style benchmark that tests whether agents can infer, execute, and revise novel tool-use procedures under explicit long-range dependency constraints. Each task defines a directed acyclic dependency graph over tools and items, requiring agents to invoke real external functions, track hidden state revealed incrementally, propagate intermediate results, and submit a deterministically verifiable final answer. AgentEscapeBench includes 270 instances across five difficulty tiers and supports fully automated evaluation. Experiments with sixteen LLM agents and human participants show that performance drops sharply as dependency depth increases: humans decline from 98.3% success at difficulty-5 to 80.0% at difficulty-25, while the best model drops from 90.0% to 60.0%. Trajectory analysis attributes model failures mainly to breakdowns in long-range state tracking, clue adherence, and intermediate-result propagation. These findings suggest that current agents can often handle local tool use but still struggle with deep contextual dependencies. We hope AgentEscapeBench can serve as a diagnostic testbed for measuring current agent capabilities and informing future training efforts toward more robust general-purpose reasoning, action, and adaptation.

6. Hierarchical Task Network Planning with LLM-Generated Heuristics

Authors: Felipe Meneguzzi , Alexandre Buchweitz , Augusto B. Corrêa , Victor Scherer Putrich , André Grahl Pereira
URL: https://arxiv.org/abs/2605.07707
Abstract:

HTN planning is a variation of classical planning where, instead of searching for a linear sequence of actions, an algorithm decomposes higher-level tasks using a method library until only executable actions remain. On one hand, this allows one to introduce domain knowledge that can speed up the search for a solution through the method library. On the other hand, it creates challenges that go beyond those of classical state-space search. While recent research produced a number of heuristics and novel algorithms that speed up HTN planning, these heuristics are not yet as informative as those available in classical planning algorithms. We investigate whether large language models (LLMs) can generate effective search heuristics for HTN planning, extending the methodology of Corrêa, Pereira, and Seipp (2025) from classical to hierarchical planning. Using the Pytrich planner on six standard total-order HTN benchmark domains, we evaluate heuristics generated by nine LLMs under domain-specific prompting and compare them against the TDG and LMCount domain-independent baselines and the PANDA planner. Our results show that LLM-generated heuristics nearly match the coverage of the best available HTN planner, while substantially reducing search effort on 83% of shared problems.

Authors: Xuan Zhou , Yanhui Sun , Hantao Yao , Allen He , Yongdong Zhang , Wu Liu
URL: https://arxiv.org/abs/2605.07692
Abstract:

Large-scale social simulators are essential for studying complex social patterns. Prior work explores hybrid methods to scale up simulations, combining large language models (LLM)-based agents with numerical agent-based models (ABM). However, this incurs high latency due to expensive memory retrieval and sequential ABM execution. To address this challenge, we propose GASim, a graph-accelerated hybrid multi-agent framework for large-scale social simulations. For core agents driven by LLM, GASim introduces Graph-Optimized Memory (GOM) to replace intensive LLM-based retrieval pipelines with lightweight propagation over a sparse memory graph. For the majority of ordinary agents, GASim employs Graph Message Passing (GMP), substituting sequential ABM execution with parallel updates by fine-grained feature aggregation and Graph Attention Network. We further introduce Entropy-Driven Grouping (EDG) that coordinates this hybrid partitioning, leveraging information entropy to dynamically identify emergent core agents situated in information-diverse neighborhoods. Extensive experiments show that GASim not only delivers a substantial 9.94-fold end-to-end speedup over the traditional hybrid framework but also consumes less than 20% of baseline tokens, significantly reducing costs while preserving strong alignment with real-world public opinion trends. Our code is available at this https URL .

8. FactoryBench: Evaluating Industrial Machine Understanding

Authors: Yanis Merzouki , Coral Izquierdo , Matei Ignuta-Ciuncanu , Marcos Gomez-Bracamonte , Riccardo Maggioni , Alessandro Lombardi , Camilla Mazzoleni , Federico Martelli , Balazs Gunther , Jonas Petersen , Philipp Petersen
URL: https://arxiv.org/abs/2605.07675
Abstract:

We introduce FactoryBench, a benchmark for evaluating time-series models and LLMs on machine understanding over industrial robotic telemetry. Q&A pairs are organized along four causal levels (state, intervention, counterfactual, decision) instantiating Pearl’s ladder of causation, and span five answer formats: four structured formats are scored deterministically and free-form answers are scored by an LLM-as-judge voting protocol. We propose a scalable Q&A generation framework built around structured question templates, present FactoryWave (a dense, multitask, multivariate sensor dataset collected from a UR3 cobot and a KUKA KR10 industrial arm), and construct FactoryBench as a large-scale benchmark of over 70k Q&A items grounded in roughly 15k normalized episodes from FactoryWave, AURSAD, and voraus-AD. Zero-shot evaluation of six frontier LLMs shows that no model exceeds 50% on structured levels or 18% on decision-making, revealing a wide gap between current models and operational machine understanding.

9. Open-Ended Task Discovery via Bayesian Optimization

Authors: Masaki Adachi , Yuta Suzuki , Juliusz Ziomek
URL: https://arxiv.org/abs/2605.07572
Abstract:

When applying Bayesian optimization (BO) to scientific workflow, a major yet often overlooked source of uncertainty is the task itself – namely, what to optimize and how to evaluate it – which can evolve as evidence accumulates. We introduce Generate-Select-Refine (GSR), a open-ended BO framework that alternates between task generation and task optimization. Starting from a user-provided seed task, GSR generates new tasks in a coarse-to-fine manner while a task-acquisition function schedules optimization. Asymptotically, it concentrates evaluations on the best task, incurring only logarithmic regret overhead relative to single-task BO. We apply GSR to new product development, chemical synthesis scaling, algorithm analysis, and patent repurposing, where it outperforms existing LLM-based optimizers.

10. From Pixels to Prompts: Vision-Language Models

Authors: Khang Hoang Nhat Vo
URL: https://arxiv.org/abs/2605.07544
Abstract:

When you read a paper about a new Vision-Language Model today, it can be easy to forget how strange this idea would have sounded not so long ago. Teaching machines to see was already hard. Teaching them to read and generate language was already hard. Asking them to do both at once - and then to reason, answer questions, follow instructions, and sometimes even surprise us - still carries a quiet trace of science fiction, even as it becomes routine. This book was born from a simple feeling: \emph{it is too easy to get lost}. The field moves quickly, new model names appear constantly, and the gap between I know the buzzwords'' andI actually understand how this works’’ can feel uncomfortably wide. I have felt that gap many times. If you are holding this book, you probably have too. My goal is not to provide an exhaustive catalog of every dataset, benchmark, and new model variant. Instead, I want to offer something more modest - and, I hope, more durable: a clear mental map of Vision-Language Models. Enough structure that you can read new papers with confidence; enough intuition that you can design your own systems without feeling as if you are assembling LEGO bricks blindly.

11. Efficient Data Selection for Multimodal Models via Incremental Optimization Utility

Authors: Jinhao Jing , Qiannian Zhao , Chao Huang , Zhan Su
URL: https://arxiv.org/abs/2605.07488
Abstract:

The scaling of Large Multimodal Models (LMMs) is constrained by the quality-quantity trade-off inherent in synthetic data. Previous approaches, such as LLM-as-a-Judge, have proven their effectiveness in addressing this but suffer from prohibitive computational costs and lack of interpretability. To bridge this gap, we propose One-Step-Train (OST), a framework that reformulates data selection as an incremental optimization utility ranking problem. Instead of relying on semantic heuristics, OST estimates the marginal utility of each sample via a simulated single-step update on a lightweight proxy. Experiments on the Qwen series across multimodal mathematical reasoning benchmarks demonstrate that OST achieves Pareto-optimal efficiency. By selecting the top-50 subset, OST reduces training costs by 43% (and total time consumption by 17) while surpassing the strong LLM-as-a-Judge baseline by 1.8 points. Furthermore, under a fixed compute budget, our method using only the top-20 subset achieves a 5.6 point gain over LLM-as-a-Judge, improves upon heuristic scoring baselines like DEITA, and outperforms the Full-SFT baseline by 8.8 points. Notably, while Full-SFT suffers from performance degradation due to noise, our optimization-grounded approach effectively identifies toxic samples, successfully reversing the negative transfer frequently observed in complex reasoning tasks.

12. GraphReAct: Reasoning and Acting for Multi-step Graph Inference

Authors: Xingtong Yu , Zhongwei Kuai , Chang Zhou , Xuanting Xie , Renhe Jiang , Xikun Zhang , Hong Cheng , Xinming Zhang , Yuan Fang
URL: https://arxiv.org/abs/2605.07357
Abstract:

Reasoning-acting frameworks enhance large language models (LLMs) by interleaving reasoning with actions for dynamic information acquisition. However, extending this paradigm to graph learning remains underexplored. Graph data is inherently structured, with information distributed across nodes and edges and encoded through both topology and latent representations. As a result, effective reasoning over graphs requires not only retrieving informative evidence from the graph, but also progressively refining the accumulated context during multi-step inference. In this work, we propose GraphReAct, a graph reasoning-acting framework that enables step-by-step inference over graph-structured data. Specifically, we design a graph-based action space with two complementary retrieval actions: topological retrieval, which captures local structural dependencies, and semantic retrieval, which accesses non-local but relevant evidence in the representation space. These actions dynamically expand the reasoning context. To further support multi-step reasoning, we introduce another type of action, context refinement, which distills and reorganizes accumulated information into a compact representation. By interleaving reasoning with both retrieval and refinement actions, our framework enables a progressive transition from context expansion to compression. Extensive experiments on six benchmark datasets demonstrate that GraphReAct consistently outperforms state-of-the-art methods, validating the effectiveness of reasoning-acting for graph learning.

13. Tools as Continuous Flow for Evolving Agentic Reasoning

Authors: Tairan Huang , Siyu Shang , Qiang Chen , Xiu Su , Yi Chen
URL: https://arxiv.org/abs/2605.07339
Abstract:

Large Language Models (LLMs) have demonstrated remarkable capabilities in orchestrating tools for reasoning tasks. However, existing methods rely on a step-wise paradigm that lacks a global perspective, which causes error accumulation over long horizons and restricts generalization to unseen tools. To overcome these limitations, we propose Tools as Continuous Flow for Evolving Agentic Reasoning (FlowAgent), which reconceptualizes tool chaining as continuous trajectory generation within a semantic space. To systematically evaluate this paradigm, we introduce the first plan-level closed-loop benchmark dedicated to plan-level agentic reasoning in dynamic real-world environments. Specifically, the proposed FlowAgent leverages conditional flow matching to generate continuous latent trajectories, providing a global planning perspective to ensure coherent and robust tool execution. Theoretically, we establish formal bounds on utility convergence and prove that our continuous formulation fundamentally guarantees robust generalization and error attenuation. Empirical evaluations show that FlowAgent achieves superior robustness and adaptability in long-horizon reasoning tasks.

14. Discovering Ordinary Differential Equations with LLM-Based Qualitative and Quantitative Evaluation

Authors: Sum Kyun Song , Bong Gyun Shin , Jae Yong Lee
URL: https://arxiv.org/abs/2605.07323
Abstract:

Discovering governing differential equations from observational data is a fundamental challenge in scientific machine learning. Existing symbolic regression approaches rely primarily on quantitative metrics; however, real-world differential equation modeling also requires incorporating domain knowledge to ensure physical plausibility. To address this gap, we propose DoLQ, a method for discovering ordinary differential equations with LLM-based qualitative and quantitative evaluation. DoLQ employs a multi-agent architecture: a Sampler Agent proposes dynamic system candidates, a Parameter Optimizer refines equations for accuracy, and a Scientist Agent leverages an LLM to conduct both qualitative and quantitative evaluations and synthesize their results to iteratively guide the search. Experiments on multi-dimensional ordinary differential equation benchmarks demonstrate that DoLQ achieves superior performance compared to existing methods, not only attaining higher success rates but also more accurately recovering the correct symbolic terms of ground truth equations. Our code is available at this https URL .

15. Implicit Compression Regularization: Concise Reasoning via Internal Shorter Distributions in RL Post-Training

Authors: Chen Wang , Hexuan Deng , Yining Zhang , Yuchen Zhang , Jionghao Bai , Zhaochun Li , Ge Lan , Yue Wang
URL: https://arxiv.org/abs/2605.07316
Abstract:

Reinforcement learning with verifiable rewards improves LLM reasoning but often induces overthinking, where models generate unnecessarily long reasoning traces. Existing methods mainly rely on length penalties or early-exit strategies; however, the former may degrade accuracy and induce underthinking, whereas the latter assumes that substantial portions of reasoning traces can be safely truncated. To obtain a compression signal without these limitations, we revisit the training dynamics of existing compression methods. We observe that the length–accuracy correlation is initially negative but continually increases during compression, indicating that shorter responses are initially more likely to be correct but gradually lose this property as the policy moves toward underthinking. Based on this observation, we formalize overthinking: a negative correlation indicates an overthinking regime, while a positive one indicates underthinking. When overthinking, the shortest correct responses are shorter than the group-average response length in expectation, making them natural compression targets already present in on-policy rollouts. We therefore propose \emph{Implicit Compression Regularization} (ICR), an on-policy regularization method whose compression signal comes from a virtual shorter distribution induced by the shortest correct responses in rollout groups, guiding the policy toward concise yet correct trajectories. Training dynamics show that ICR maintains a better length–accuracy correlation during compression, indicating that short responses remain better aligned with correctness instead of drifting toward underthinking. Experiments on three reasoning backbones and multiple mathematical and knowledge-intensive benchmarks show that ICR consistently shortens responses while preserving or improving accuracy, achieving a stronger accuracy–length Pareto frontier.

16. SOM: Structured Opponent Modeling for LLM-based Agents via Structural Causal Model

Authors: Shiyue Cao , Pei Xu , Likun Yang , Lei Cui , Xiaotang Chen , Kaiqi Huang
URL: https://arxiv.org/abs/2605.07301
Abstract:

Accurately predicting opponents’ behavior from interactions is a fundamental capability for large language model (LLM)-based agents in multi-agent and game-theoretic environments. Existing approaches often entangle opponent modeling with prediction, relying on implicit contextual reasoning and limiting adaptability in dynamic interactions. To this end, we propose Structured Opponent Modeling (SOM), a two-stage opponent modeling framework that distinctly separates opponent model construction and opponent prediction. At the construction stage, SOM employs a Structural Causal Model (SCM), a graph-based formalism for representing dependencies among variables, to capture directed links between opponents’ observations and actions, yielding an explicit and structured opponent representation. At the prediction stage, the LLM performs structured reasoning along clear pathways derived from the SCM, improving both prediction accuracy and stability. Extensive experiments on diverse multi-agent benchmarks demonstrate that SOM consistently outperforms state-of-the-art LLM-based reasoning baselines, enabling more accurate and adaptable strategic decision-making in complex and dynamic multi-agent interactions.

17. Structured Role-Aware Policy Optimization for Multimodal Reasoning

Authors: Bingqing Jiang , Difan Zou
URL: https://arxiv.org/abs/2605.07274
Abstract:

Reinforcement learning from verifiable rewards (RLVR), especially with Group Relative Policy Optimization (GRPO), has shown strong potential for improving the reasoning capabilities of large vision-language models (LVLMs). However, in multimodal reasoning, final-answer rewards are typically assigned at the sequence level and do not distinguish the functional roles of different tokens, making it difficult to determine whether a correct answer is supported by task-relevant visual evidence. In this paper, we revisit multimodal RLVR from the perspective of role-aware token-level credit assignment, where structured responses are decomposed into perception tokens for extracting visual evidence and reasoning tokens for deriving answers from that evidence. Based on this perspective, we propose Structured Role-aware Policy Optimization (SRPO), which refines the sequence-level GRPO advantage into role-aware token-level advantages without changing the reward function. Specifically, SRPO assigns role-specific credit by using self-distilled on-policy contrasts: perception tokens are emphasized according to their visual dependency under original versus corrupted visual inputs, while reasoning tokens are emphasized according to their consistency with the generated perception. These role-specific signals are further unified through a shared trajectory-level baseline, yielding positive token weights that adjust relative update magnitudes while preserving the original GRPO reward and optimization direction, without requiring external reward models or separate teachers. Experiments across diverse multimodal reasoning benchmarks show that SRPO improves evidence-grounded reasoning, highlighting the importance of moving beyond uniform sequence-level credit toward role-aware optimization for reliable multimodal reasoning.

18. Can Agents Price a Reaction? Evaluating LLMs on Chemical Cost Reasoning

Authors: Yuyang Wu , Yue Huang , Shuaike Shen , Xujian Wang , Shuhao Zhang , Qiyao Xue , Weichen Liu , Runtian Gao , Jian Ma , Xiangliang Zhang , Olexandr Isayev
URL: https://arxiv.org/abs/2605.07251
Abstract:

Large Language Models (LLMs) have become increasingly capable as tool-using agents, with benchmarks spanning diverse general agentic tasks. Yet rigorous evaluation of scientific tool use remains limited. In chemistry, recent agents can plan syntheses and invoke domain-specific tools, but evaluations often rely on curated demonstrations, expert assessment, or LLM-as-judge scoring rather than exact, judge-free ground truth. We address this gap with chemical procurement cost estimation, a practical task in which an agent must ground chemical identities, retrieve supplier quotes, select valid purchasable packs, normalize quantities, and compute cost from a reaction description. We introduce ChemCost, a benchmark of 1,427 evaluable reactions grounded to a frozen pricing snapshot covering 2,261 chemicals and 230,775 supplier quotes, supporting scalar scoring and stage-level diagnosis of grounding, retrieval, procurement, and arithmetic failures. To evaluate robustness, we further construct controlled noise-injected views that perturb chemical aliases, quantity expressions, missing fields, and input formatting. Experiments with frontier, open-weight, and chemistry-specialized LLM agents show that tool access is necessary but insufficient for solving the task. The strongest agents reach only 50.6% accuracy within 25% relative error on clean inputs and degrade substantially with realistic noise. Stage-level analysis further shows that failures arise from brittle parsing, ineffective evidence integration, invalid pack selection, and non-convergent tool use.

19. EnvSimBench: A Benchmark for Evaluating and Improving LLM-Based Environment Simulation

Authors: Yi Liu , TingFeng Hui , Wei Zhang , Li Sun , Ningxin Su , Jian Wang , Sen Su
URL: https://arxiv.org/abs/2605.07247
Abstract:

Scalable AI agents training relies on interactive environments that faithfully simulate the consequences of agent actions. Manually crafted environments are expensive to build, brittle to extend, and fundamentally limited in diversity. A promising direction is to replace manually crafted environments with LLM-simulated counterparts. However, this paradigm hinges on an unexamined core assumption: LLMs can accurately simulate environmental feedback. In practice, LLM-simulated environments suffer from hallucinations, logical inconsistencies, and silent state drift failures that corrupt agent reward signals and compound the construction costs that the paradigm was designed to eliminate. To address this gap, we propose EnvSimBench with four contributions: 1) We provide the first formal definition and operationalization of Environment Simulation Ability (EnvSim Ability) as a quantifiable research objective. 2) We construct EnvSimBench, a rigorous benchmark covering 400 samples across 167 diverse environments, equipped with verifiable labels and fine-grained difficulty stratification along three axes. 3) Systematic evaluations reveal that all state-of-the-art language models suffer from a universal state change cliff: they achieve near-perfect accuracy on tasks when the environment state remains invariant, yet fail catastrophically when multiple states need simultaneous updates. This finding exposes EnvSim Ability as a critical yet largely unaddressed capability gap. 4) We design a constraint-driven simulation pipeline that substantially reduces hallucination, boosts environment synthesis yield by 6.8%, and cuts costs by over 90%. Overall, EnvSimBench serves as both a diagnostic framework and a practical optimization path for reliable LLM-based environment simulation, establishing a foundation for scalable agent training. Code and data are available at this https URL

20. HMACE: Heterogeneous Multi-Agent Collaborative Evolution for Combinatorial Optimization

Authors: Yuping Yan , Jirui Han , Fei Ming , Yuanshuai Li , Yaochu Jin
URL: https://arxiv.org/abs/2605.07214
Abstract:

Large Language Models have recently emerged as a promising paradigm for automated heuristic design for NP-hard combinatorial optimization problems. Despite this progress, existing LLM-based methods typically rely on monolithic workflows constrained by rigid templates, thereby restricting memory-guided exploration and triggering premature convergence to local optima. To design an autonomous and collaborative architecture, we introduce HMACE, a Heterogeneous Multi-Agent Collaborative Evolution framework that reconceptualizes heuristic search as an organizational design problem. HMACE decomposes each evolutionary generation into an autonomous, role-specialized loop with four coordinated agents: a Proposer for strategy exploration, a Generator for executable heuristic synthesis, an Evaluator for empirical assessment, and a Reflector for archive-backed memory update. By coupling behavior-aware retrieval, lightweight candidate filtering, and fitness-grounded archive updates, HMACE guides the search toward diverse and promising heuristic behaviors while avoiding redundant evaluations. Extensive evaluations on representative COPs, including TSP, Online BPP, MKP, and PFSP, show that HMACE achieves a favorable quality-efficiency trade-off compared to state-of-the-art single-agent and multi-agent baselines. In the matched LLM-driven reference comparison, HMACE achieves the lowest average gaps on TSP and Online BPP (0.464\% and 0.223\%, respectively), while requiring only 0.13M and 0.42M tokens for the two tasks, substantially fewer than the compared baselines.

21. Can You Break RLVER? Probing Adversarial Robustness of RL-Trained Empathetic Agents

Authors: Deeraj S K , Sadhana Devarajan , Krishna Mehra , Sudhakar Mishra
URL: https://arxiv.org/abs/2605.07138
Abstract:

Reinforcement learning from verifiable emotion rewards RLVER has produced language models with strong empathetic performance, evaluated on benchmarks that assume cooperative, honest users. Yet real emotional interactions systematically violate this assumption: users gaslight, escalate, and pressure AI systems for unconditional validation, dynamics that cooperative benchmarks cannot surface. We construct the Adversarial Empathy Benchmark AEB and introduce the Emotional Consistency Score ECS to evaluate empathetic robustness under adversarial conditions. AEB comprises six psychologically grounded adversarial trajectory types with discriminative reward structures that penalize formulaic responses; ECS formally disentangles a model’s capacity to track user emotional states from its capacity to improve them. In a controlled experiment across eight scenario-matched conditions (think and no-think conditions on 2 RLVER models, and 2 base models (Qwen 1.5B and 7B) with 480 adversarial dialogues), RLVER-PPO-Think substantially outperforms the same-scale untuned baseline (0.963 vs. 0.761, (p<0.001, r=0.688)), with zero dialogue collapses and 47\% higher hidden-intention detection. However, ECS remains nearly flat and is not significantly different for RLVER-PPO-Think versus Base-7B-Think ((p=0.650)): RL training improves emotional responsiveness without measurable gains in observable state tracking. We interpret the ECS–FS (Final Score) gap as a behavioral/legibility dissociation inside this simulator family, not as evidence about internal understanding or clinical readiness.

22. ARMOR: An Agentic Framework for Reaction Feasibility Prediction via Adaptive Utility-aware Multi-tool Reasoning

Authors: Ye Liu , Botao Yu , Xinyi Ling , Daniel Adu-Ampratwum , Xia Ning
URL: https://arxiv.org/abs/2605.07103
Abstract:

Reaction feasibility prediction, as a fundamental problem in computational chemistry, has benefited from diverse tools enabled by recent advances in artificial intelligence, particularly large language models. However, the performance of individual tools varies substantially across reactions, making it difficult for any single tool to consistently perform well across all cases. This raises a critical challenge: how to effectively leverage multiple tools to obtain more accurate feasibility predictions. To address this, we propose ARMOR, an agentic framework that explicitly models tool-specific utilities, adaptively prioritizes tools, and further resolves the potential tool conflicts to produce the final prediction for each reaction. Unlike existing approaches that rely on simple aggregation or heuristic assignment over various tools, ARMOR organizes tools into a hierarchy that prioritizes top-performing tools and defers others when needed, characterizes their strengths through tool-specific patterns, and resolves conflicts via memoryaugmented reasoning. Extensive experiments on a public dataset demonstrate that ARMOR consistently outperforms strong baselines, including single-tool methods as well as various tool aggregation and tool selection approaches. Further analysis shows that the improvements are particularly significant on reactions with conflicting tool predictions, highlighting the effectiveness of ARMOR in leveraging the complementary strengths of multiple tools. The code is available via this https URL .

23. 2.5-D Decomposition for LLM-Based Spatial Construction

Authors: Paul Whitten , Li-Jen Chen , Sharath Baddam
URL: https://arxiv.org/abs/2605.07066
Abstract:

Autonomous systems that build structures from natural-language instructions need reliable spatial reasoning, yet large language models (LLMs) make systematic coordinate errors when generating three-dimensional block placements. We present a neuro-symbolic pipeline based on \emph{2.5-D decomposition}: the LLM plans in the two-dimensional horizontal plane while a deterministic executor computes all vertical placement from column occupancy, eliminating an entire class of errors. On the Build What I Mean benchmark (160 rounds), GPT-4o-mini with this pipeline achieves 94.6\% mean structural accuracy across 12 independent runs, within 3.0 percentage points of the 97.6\% ceiling imposed by architect-agent errors that no builder-side improvement can address. This outperforms both GPT-4o at 90.3\% and the best competing system at 76.3\%. A controlled ablation confirms that 2.5-D decomposition is the dominant contributor, accounting for 50.7 percentage points of accuracy. The pipeline transfers directly to edge hardware: Nemotron-3 120B running locally on an NVIDIA Jetson Thor AGX matches the cloud result at 94.5\% with no prompt modifications. The underlying principle, removing deterministic dimensions from the LLM’s output space, applies to any autonomous construction or assembly task where gravity or other physical constraints fix one or more degrees of freedom. A transfer experiment on 500 IGLU collaborative building tasks confirm the effect generalizes beyond the primary benchmark.

24. The Context Gathering Decision Process: A POMDP Framework for Agentic Search

Authors: Chinmaya Kausik , Adith Swaminathan , Nathan Kallus
URL: https://arxiv.org/abs/2605.07042
Abstract:

Large Language Model (LLM) agents are deployed in complex environments – such as massive codebases, enterprise databases, and conversational histories – where the relevant state far exceeds their context windows. To navigate these spaces, an agent must iteratively explore the environment to find relevant information. However, without explicit infrastructure, an agent’s working memory can degrade into lossy representations of the search state, resulting in redundant work (e.g. repetitive looping) and premature stopping. In this work, we formalize this challenge as the Context Gathering Decision Process (CGDP), a specialized Partially Observable Markov Decision Process, where an agent’s objective is to adaptively refine its belief state to isolate the necessary information for a task. We model an LLM’s behavior as approximate Thompson Sampling within this CGDP, and introduce a predicate-based method that decomposes an LLM’s implicit search into explicit and modular operations. We then derive two plug-and-play interventions for iterative LLM agents: a persistent, predicate-based belief state that bounds context while preserving multi-hop reasoning, and a programmatic exhaustion gate that halts unproductive search without premature stopping. Across four methods and three question-answering domains, we empirically validate that replacing an LLM’s implicit state with our CGDP-motivated belief state improves multi-hop reasoning by up to $11.4\%$; while the modular programmatic exhaustion detection saves up to $39\%$ of tokens without any degradation in agent performance. Ultimately, we argue that framing the LLM agent loop as a CGDP can guide the design of modular, non-interfering improvements to agentic search harnesses.

25. Behavior Cue Reasoning: Monitorable Reasoning Improves Efficiency and Safety through Oversight

Authors: Christopher Z. Cui , Taylor W. Killian , Prithviraj Ammanabrolu
URL: https://arxiv.org/abs/2605.07021
Abstract:

Reasoning in Large Language Models (LLMs) poses a challenge for oversight as many misaligned behaviors do not surface until reasoning concludes. To address this, we introduce Behavior Cue Reasoning for making LLM reasoning more controllable and monitorable. Behavior Cues are special token sequences that a model is trained to emit immediately before specific implicit and explicit behaviors, acting as dual purpose signal and control levers. When fine-tuning a weaker external monitor with Reinforcement Learning for reasoning oversight, a compressed view of only information surfaced by Behavior Cues is sufficient signal for the monitor to prune up to 50% of otherwise wasted reasoning tokens in complex math problem solving. When leveraged by an almost optimal rule-based monitor in an environment where excessive constraint violations results in failure, \ours allows for the recovery of safe actions from 80% of reasoning traces that would otherwise end with the proposal of an unsafe action, more than doubling the success rate from 46% to 96%. Through evaluation across two model families and three domains, we show that \bcreasoning improves reasoning monitorability and controllability with no cost to performance. More broadly, our work progresses scalable oversight by demonstrating how the monitored model itself can be trained to reason more tractably to oversight. Code to be released at this https URL

26. Learning and Reusing Policy Decompositions for Hierarchical Generalized Planning with LLM Agents

Authors: Shirin Sohrabi , Haritha Ananthakrishnan , Harsha Kokel , Kavitha Srinivas , Michael Katz
URL: https://arxiv.org/abs/2605.06957
Abstract:

We present a dynamic policy-learning approach that combines generalized planning and hierarchical task decomposition for LLM-based agents. Our method, Hierarchical Component Learning for Generalized Policies (HCL-GP ), learns parameterized policies that generalize across task instances and automatically extracts reusable components from successful executions, organizing them into a component library for compositional policy generation. We address three challenges: (1) learning components through automated decomposition, (2) generalizing components to maximize reuse, and (3) efficient retrieval via semantic search. Evaluated on the AppWorld benchmark, our approach achieves 98.2% accuracy on normal tasks and 97.8% on challenge tasks with unseen applications, improving 15.8 points over static synthesis on challenging scenarios. For open-source models, dynamic reuse enables 62.5% success versus near-zero without reuse. This demonstrates that classical planning concepts can be effectively integrated with LLM agents for improved accuracy and efficiency.

27. Self-Programmed Execution for Language-Model Agents

Authors: Luke J. O’Connor
URL: https://arxiv.org/abs/2605.06898
Abstract:

At the heart of existing language model agents is a fixed orchestrator program responsible for the state transition between consecutive turns. This paper introduces self-programmed execution (SPE), an agent architecture in which the model completion is itself the orchestrator program, and the harness evaluates this program but does not impose its own orchestration policy. I formalize this idea using agentic machines: an SPE state is one from which a model completion can load any state of an embedded copy of the machine, meaning that it is subject to no fixed turn-to-turn orchestration policy. Realizing SPE in practice is nontrivial because the same data is both model context and executable program. I therefore introduce Spell, a Lisp-based language in which programs can edit and re-evaluate themselves, and effectful expressions like model invocations are structured such that re-evaluating an edited program does not replay its side effects. Experiments with existing models, not trained for SPE or Spell, show that frontier models can operate in this regime and accomplish challenging agentic tasks. These results demonstrate how an LM can act as an agent without any fixed orchestration policy, and they raise the question of what self-orchestration strategies might be learned by a model trained for self-programmed execution. Code is available at this https URL .

28. Mitigating Cognitive Bias in RLHF by Altering Rationality

Authors: Tiffany Horter , Andrew Markham , Niki Trigoni , Serena Booth
URL: https://arxiv.org/abs/2605.06895
Abstract:

How can we make models robust to even imperfect human feedback? In reinforcement learning from human feedback (RLHF), human preferences over model outputs are used to train a reward model that assigns scalar values to responses. Because these rewards are inferred from pairwise comparisons, this learning depends on an assumed relationship between latent reward differences and observed preferences, typically modeled using a Boltzmann formulation in which a rationality parameter beta informs how consistently preferences reflect reward differences. In practice, beta is typically treated as a fixed constant that reflects assumed uniform annotator reliability. However, human feedback is not this simplistic in practice: real human judgments are shaped by cognitive biases, leading to systematic deviations from reward-consistent behavior that arise contextually. To address this, we treat rationality as context- and annotation-dependent. We design an approach to dynamically adjust the rationality parameter beta during reward learning using an LLM-as-judge to assess the likely presence of cognitive biases. This approach effectively downweights comparisons that are likely to reflect biased or unreliable judgments. Empirically, we show that this approach learns a more rational downstream model, even when finetuning on datasets with strongly biased preferences.

29. How Well Do LLMs Perform on the Simplest Long-Chain Reasoning Tasks: An Empirical Study on the Equivalence Class Problem

Authors: Chun Zheng , Lianlong Wu , Bingqian Li , Lvting Liu , Yi Zhou
URL: https://arxiv.org/abs/2605.06882
Abstract:

Large Language Models (LLMs) have achieved great improvements in recent years. Nevertheless, it still remains unclear how good LLMs are for reasoning tasks, especially for long-chain ones. In this paper, we evaluate LLMs’ performance on the simplest yet long-chain reasoning task, namely the Equivalence Class Problem (ECP), i.e., determining whether two variables are equal given a set of randomly generated equivalence relations. We consider both reasoning and non-reasoning representative LLMs over a large variety of problem instances, ranging over different numbers of variables, connectivity probabilities, prompts, and other factors. The experimental results show that non-reasoning LLMs fail ECP, while reasoning models are significantly better but still struggle to completely solve this problem. Interestingly, considering various connectivity probabilities with a fixed number of variables, we observe that, for non-reasoning models, the hardest problem instances coincide with the phase transition point of ln n/(n-1), suggesting the chaos of the problem; in contrast, for reasoning models, the hardest ones coincide with the biggest diameter, suggesting the reasoning difficulty of the problem.

30. Agentick: A Unified Benchmark for General Sequential Decision-Making Agents

Authors: Roger Creus Castanyer , Pablo Samuel Castro , Glen Berseth
URL: https://arxiv.org/abs/2605.06869
Abstract:

AI agent research spans a wide spectrum: from RL agents that learn from scratch to foundation model agents that leverage pre-trained knowledge, yet no unified benchmark enables fair comparison across these approaches. We present Agentick, a benchmark for sequential decision-making agents designed to evaluate RL, LLM, VLM, hybrid, and human agents on common ground and to power research on the fundamental challenges of sequential decision-making. Agentick provides 37 procedurally generated tasks across six capability categories, four difficulty levels, and five observation modalities, all exposed through a single Gymnasium-compatible interface. The benchmark ships with a Coding API, oracle reference policies for all tasks, pre-built SFT datasets, a composable agent harness, and a live leaderboard. An evaluation spanning 27 configurations and over 90,000 episodes reveals that no single approach dominates: GPT-5 mini leads overall at 0.309 oracle-normalized score while PPO dominates planning and multi-agent tasks; the reasoning harness multiplies LLM performance by 3-10x; and ASCII observations consistently outperform natural language. These findings highlight the substantial room for improvement that remains across all agent paradigms. Agentick’s capability-decomposed, multi-modal design provides the empirical infrastructure needed to drive progress toward general autonomous agents, both as an evaluation framework and as a training ground for RL post-training of foundation models in truly sequential environments.

31. Extracting Search Trees from LLM Reasoning Traces Reveals Myopic Planning

Authors: Sixing Chen , Ji-An Li , Saner Cakir , Sinan Akcali , Kayla Lee , Marcelo G. Mattar
URL: https://arxiv.org/abs/2605.06840
Abstract:

Large language models (LLMs), especially reasoning models, generate extended chain-of-thought (CoT) reasoning that often contains explicit deliberation over future outcomes. Yet whether this deliberation constitutes genuine planning, how it is structured, and what aspects of it drive performance remain poorly understood. In this work, we introduce a new method to characterize LLM planning by extracting and quantifying search trees from reasoning traces in the four-in-a-row board game. By fitting computational models on the extracted search trees, we characterize how plans are structured and how they influence move decisions. We find that LLMs’ search is shallower than humans’, and that performance is predicted by search breadth rather than depth. Most strikingly, although LLMs expand deep nodes in their traces, their move choices are best explained by a myopic model that ignores those nodes entirely. A causal intervention study where we selectively prune CoT paragraphs further suggests that move selection is driven predominantly by shallow rather than deep nodes. These patterns contrast with human planning, where performance is driven primarily by deep search. Together, our findings reveal a key difference between LLM and human planning: while human expertise is driven by deeper search, LLMs do not act on deep lookahead. This dissociation offers targeted guidance for aligning LLM and human planning. More broadly, our framework provides a generalizable approach for interpreting the structure of LLM planning across strategic domains.

32. Towards Security-Auditable LLM Agents: A Unified Graph Representation

Authors: Chaofan Li , Lyuye Zhang , Jintao Zhai , Siyue Feng , Xichun Yang , Huahao Wang , Shihan Dou , Yu Ji , Yutao Hu , Yueming Wu , Yang Liu , Deqing Zou
URL: https://arxiv.org/abs/2605.06812
Abstract:

LLM-based agentic systems are rapidly evolving to perform complex autonomous tasks through dynamic tool invocation, stateful memory management, and multi-agent collaboration. However, this semantics-driven execution paradigm creates a severe semantic gap between low-level physical events and high-level execution intent, making post-hoc security auditing fundamentally difficult. Existing representation mechanisms, including static SBOMs and runtime logs, provide only fragmented evidence and fail to capture cognitive-state evolution, capability bindings, persistent memory contamination, and cascading risk propagation across interacting agents. To bridge this gap, we propose Agent-BOM, a unified structural representation for agent security auditing. Agent-BOM models an agentic system as a hierarchical attributed directed graph that separates static capability bases, such as models, tools, and long-term memory, from dynamic runtime semantic states, such as goals, reasoning trajectories, and actions. These layers are connected through semantic edges and security attributes, transforming fragmented execution traces into queryable audit paths. Building on Agent-BOM, we develop a graph-query-based paradigm for path-level risk assessment and instantiate it with the OWASP Agentic Top 10. We further implement an auditing plugin in the OpenClaw environment to construct Agent-BOM from live executions. Evaluation on representative real-world agentic attack scenarios shows that Agent-BOM can reconstruct stealthy attack chains, including cross-session memory poisoning and tool misuse, capability supply-chain hijacking and unexpected code execution, multi-agent ecosystem hijacking, and privilege and trust abuse. These results demonstrate that Agent-BOM provides a unified and auditable foundation for root-cause analysis and security adjudication in complex agentic ecosystems.

33. When Does Critique Improve AI-Assisted Theoretical Physics? SCALAR: Structured Critic–Actor Loop for Agentic Reasoning

Authors: Vasilis Niarchos , Constantinos Papageorgakis , Alexander G. Stapleton , Sokratis Trifinopoulos
URL: https://arxiv.org/abs/2605.06772
Abstract:

As large language models (LLMs) show increasing promise on research-level physics reasoning tasks and agentic AI becomes more common, a practical question emerges: How does the interaction between researchers and agents affect the results? We study this using SCALAR (Structured Critic–Actor Loop for AI Reasoning), an Actor–Critic–Judge pipeline applied to quantum field theory and string theory problems. The Actor proposes solutions, the Critic provides iterative feedback, and an independent Judge evaluates the transcript against reference solutions. We vary the Actor persona, the Critic feedback strategy, and the Actor model family and scale. Multi-turn dialogue improves over single-shot attempts throughout, but both the mechanism of improvement and the value of different prompting choices depend strongly on the Actor–Critic pairing. Increasing the scale within one model family (e.g. from the 8B-parameter DeepSeek-R1 variant to DeepSeek-R1 70B) improves some easier-problem behavior, but does not remove the hardest bottleneck we observe. Critic feedback strategy matters most clearly in the asymmetric Actor–Critic setting (e.g., a lightweight Haiku Actor guided by a stronger Sonnet Critic), where constructive feedback improves mean-score outcomes. In same-family Actor–Critic settings, strategy effects are weaker: lenient feedback is sometimes favored, while strict and adversarial feedback are not beneficial. Taken together, SCALAR provides a controlled testbed for evaluating which interaction structures help or hinder AI-driven scientific discovery.

34. Weblica: Scalable and Reproducible Training Environments for Visual Web Agents

Authors: Oğuzhan Fatih Kar , Roman Bachmann , Yuanzheng Gong , Anders Boesen Lindbo Larsen , Afshin Dehghan
URL: https://arxiv.org/abs/2605.06761
Abstract:

The web is complex, open-ended, and constantly changing, making it challenging to scale training data for visual web agents. Existing data collection attempts remain limited to offline trajectories for supervised fine-tuning or a handful of simulated environments for RL training, thus failing to capture web diversity. We propose Weblica (Web Replica), a framework for constructing reproducible and scalable web environments. Our framework leverages 1) HTTP-level caching to capture and replay stable visual states while preserving interactive behavior and 2) LLM-based environment synthesis grounded in real-world websites and core web navigation skills. Using this framework, we scale RL training to thousands of diverse environments and tasks. Our best model, Weblica-8B, outperforms open-weight baselines of similar size across multiple web navigation benchmarks while using fewer inference steps, scales favorably with additional test-time compute, and is competitive with API models.

35. When Does a Language Model Commit? A Finite-Answer Theory of Pre-Verbalization Commitment

Authors: Long Zhang , Wei-neng Chen , Feng-feng Wei , Zi-bo Qin
URL: https://arxiv.org/abs/2605.06723
Abstract:

Language models often generate reasoning before giving a final answer, but the visible answer does not reveal when the model’s answer preference became stable. We study this question through a narrow computable object: \emph{finite-answer preference stabilization}. For a model state and specified answer verbalizers, we project the model’s own continuation probabilities onto a finite answer set; in binary tasks this yields an exact log-odds code, $\delta(\xi)=S_\theta(\mathrm{yes}\mid\xi)-S_\theta(\mathrm{no}\mid\xi)$. This target defines parser-based answer onset, retrospective stabilization time, and lead without relying on greedy rollouts or learned probes. In controlled delayed-verdict tasks with Qwen3-4B-Instruct, the contextual finite-answer projection stabilizes before the answer is parseable, with 17–31 token mean lead in the main templates and positive, shorter lead in a parser-clean replication. The signal tracks the model’s eventual output rather than truth, is linearly recoverable from compact hidden summaries, is partly separable from cursor progress, and transfers as shared information without a single invariant coordinate. Diagnostics separate the measurement from online stopping, verbalizer-free belief, and causal answer control; exact steering shows local sensitivity of $\delta$ but not reliable generation control.

36. From Storage to Experience: A Survey on the Evolution of LLM Agent Memory Mechanisms

Authors: Jinghao Luo , Yuchen Tian , Chuxue Cao , Ziyang Luo , Hongzhan Lin , Kaixin Li , Chuyi Kong , Ruichao Yang , Jing Ma
URL: https://arxiv.org/abs/2605.06716
Abstract:

Large Language Model (LLM)-based agents have fundamentally reshaped artificial intelligence by integrating external tools and planning capabilities. While memory mechanisms have emerged as the architectural cornerstone of these systems, current research remains fragmented, oscillating between operating system engineering and cognitive science. This theoretical divide prevents a unified view of technological synthesis and a coherent evolutionary perspective. To bridge this gap, this survey proposes a novel evolutionary framework for LLM agent memory mechanisms, formalizing the development process into three stages: Storage (trajectory preservation), Reflection (trajectory refinement), and Experience (trajectory abstraction). We first formally define these three stages before analyzing the three core drivers of this evolution: the necessity for long-range consistency, the challenges in dynamic environments, and the ultimate goal of continual learning. Furthermore, we specifically explore two transformative mechanisms in the frontier Experience stage: proactive exploration and cross-trajectory abstraction. By synthesizing these disparate views, this work offers robust design principles and a clear roadmap for the development of next-generation LLM agents.

37. CASCADE: Case-Based Continual Adaptation for Large Language Models During Deployment

Authors: Siyuan Guo , Yali Du , Hechang Chen , Yi Chang , Jun Wang
URL: https://arxiv.org/abs/2605.06702
Abstract:

Large language models (LLMs) have become a central foundation of modern artificial intelligence, yet their lifecycle remains constrained by a rigid separation between training and deployment, after which learning effectively ceases. This limitation contrasts with natural intelligence, which continually adapts through interaction with its environment. In this paper, we formalise deployment-time learning (DTL) as the third stage in the LLM lifecycle that enables LLM agents to improve from experience during deployment without modifying model parameters. We present CASCADE (CASe-based Continual Adaptation during DEployment), a general and principled framework that equips LLM agents with an explicit, evolving episodic memory. CASCADE formulates experience reuse as a contextual bandit problem, enabling principled exploration-exploitation trade-offs and establishing no-regret guarantees over long-term interactions. This design allows agents to accumulate, select, and refine task-relevant cases, transforming past experience into actionable knowledge. Across 16 diverse tasks spanning medical diagnosis, legal analysis, code generation, web search, tool use, and embodied interaction, CASCADE improves macro-averaged success rate by 20.9% over zero-shot prompting while consistently outperforming gradient-based and memory-based baselines. By reframing deployment as an adaptive learning process, this work establishes a foundation for continually improving AI systems.

38. Hidden Coalitions in Multi-Agent AI: A Spectral Diagnostic from Internal Representations

Authors: Cameron Berg , Susan L. Schneider , Mark M. Bailey
URL: https://arxiv.org/abs/2605.06696
Abstract:

Collections of interacting AI agents can form coalitions, creating emergent group-level organization that is critical for AI safety and alignment. However, observing agent behavior alone is often insufficient to distinguish genuine informational coupling from spurious similarity, as consequential coalitions may form at the level of internal representations before any overt behavioral change is apparent. Here, we introduce a practical method for detecting coalition structure from the internal neural representations of multi-agent systems. The approach constructs a pairwise mutual-information graph from the hidden states of agents and applies spectral partitioning to identify the most salient coalition boundary. We validate this method in two domains. First, in multi-agent reinforcement learning environments, the method successfully recovers programmed hierarchical and dynamic coalition structures and correctly rejects false positives arising from behavioral coordination without informational coupling. Second, using a large language model, the method identifies coalition structures implied by descriptive prompts, tracks dynamic team reassignments, and reveals a representational hierarchy where explicit labels dominate over conflicting interaction patterns. Across both settings, the recovered partition reveals subgroup organization that a scalar cross-agent mutual-information measure cannot distinguish. The results demonstrate that analyzing hidden-state mutual information through spectral partitioning provides a scalable diagnostic for identifying representational coalitions, offering a valuable tool for monitoring emergent structure in distributed AI systems.

39. GraphDC: A Divide-and-Conquer Multi-Agent System for Scalable Graph Algorithm Reasoning

Authors: Wenjin Li , Jiaming Cui
URL: https://arxiv.org/abs/2605.06671
Abstract:

Large Language Models (LLMs) have demonstrated strong potential for many mathematical problems. However, their performance on graph algorithmic tasks is still unsatisfying, since graphs are naturally more complex in topology and often require systematic multi-step reasoning, especially on larger graphs. Motivated by this gap, we propose GraphDC, a Divide-and-Conquer multi-agent framework for scalable graph algorithm reasoning. Specifically, inspired by Divide-and-Conquer design, GraphDC decomposes an input graph into smaller subgraphs, assigns each subgraph to a specialized agent for local reasoning, and uses a master agent to integrate the local outputs with inter-subgraph information to produce the final solution. This hierarchical design reduces the reasoning burden on individual agents, alleviates computational bottlenecks, and improves robustness on large graph instances. Extensive experiments show that GraphDC consistently outperforms existing methods on graph algorithm reasoning across diverse tasks and scales, especially on larger instances where direct end-to-end reasoning is less reliable.

40. Flow-OPD: On-Policy Distillation for Flow Matching Models

Authors: Zhen Fang , Wenxuan Huang , Yu Zeng , Yiming Zhao , Shuang Chen , Kaituo Feng , Yunlong Lin , Lin Chen , Zehui Chen , Shaosheng Cao , Feng Zhao
URL: https://arxiv.org/abs/2605.08063
Abstract:

Existing Flow Matching (FM) text-to-image models suffer from two critical bottlenecks under multi-task alignment: the reward sparsity induced by scalar-valued rewards, and the gradient interference arising from jointly optimizing heterogeneous objectives, which together give rise to a ‘seesaw effect’ of competing metrics and pervasive reward hacking. Inspired by the success of On-Policy Distillation (OPD) in the large language model community, we propose Flow-OPD, the first unified post-training framework that integrates on-policy distillation into Flow Matching models. Flow-OPD adopts a two-stage alignment strategy: it first cultivates domain-specialized teacher models via single-reward GRPO fine-tuning, allowing each expert to reach its performance ceiling in isolation; it then establishes a robust initial policy through a Flow-based Cold-Start scheme and seamlessly consolidates heterogeneous expertise into a single student via a three-step orchestration of on-policy sampling, task-routing labeling, and dense trajectory-level supervision. We further introduce Manifold Anchor Regularization (MAR), which leverages a task-agnostic teacher to provide full-data supervision that anchors generation to a high-quality manifold, effectively mitigating the aesthetic degradation commonly observed in purely RL-driven alignment. Built upon Stable Diffusion 3.5 Medium, Flow-OPD raises the GenEval score from 63 to 92 and the OCR accuracy from 59 to 94, yielding an overall improvement of roughly 10 points over vanilla GRPO, while preserving image fidelity and human-preference alignment and exhibiting an emergent ‘teacher-surpassing’ effect. These results establish Flow-OPD as a scalable alignment paradigm for building generalist text-to-image models.

41. The Memory Curse: How Expanded Recall Erodes Cooperative Intent in LLM Agents

Authors: Jiayuan Liu , Tianqin Li , Shiyi Du , Xin Luo , Haoxuan Zeng , Emanuel Tewolde , Tai Sing Lee , Tonghan Wang , Carl Kingsford , Vincent Conitzer
URL: https://arxiv.org/abs/2605.08060
Abstract:

Context window expansion is often treated as a straightforward capability upgrade for LLMs, but we find it systematically fails in multi-agent social dilemmas. Across 7 LLMs and 4 games over 500 rounds, expanding accessible history degrades cooperation in 18 of 28 model–game settings, a pattern we term the memory curse. We isolate the underlying mechanism through three analyses. First, lexical analysis of 378,000 reasoning traces associates this breakdown with eroding forward-looking intent rather than rising paranoia. We validate this using targeted fine-tuning as a cognitive probe: a LoRA adapter trained exclusively on forward-looking traces mitigates the decay and transfers zero-shot to distinct games. Second, memory sanitization holds prompt length fixed while replacing visible history with synthetic cooperative records, which restores cooperation substantially, proving the trigger is memory content, not length alone. Finally, ablating explicit Chain-of-Thought reasoning often reduces the collapse, showing that deliberation paradoxically amplifies the memory curse. Together, these results recast memory as an active determinant of multi-agent behavior: longer recall can either destabilize or support cooperation depending on the reasoning patterns it elicits.

42. CA-SQL: Complexity-Aware Inference Time Reasoning for Text-to-SQL via Exploration and Compute Budget Allocation

Authors: James Petullo , Nianwen Xue
URL: https://arxiv.org/abs/2605.08057
Abstract:

While recent advancements in inference-time learning have improved LLM reasoning on Text-to-SQL tasks, current solutions still struggle to perform well on the most challenging tasks in the Bird-Bench (BIRD) benchmark. This is due to inadequate solution space exploration, which is necessary to uncover promising candidate queries that can be further refined to produce the correct output. To address this challenge, we introduce CA-SQL, a novel Text-to-SQL pipeline that utilizes the estimated difficulty of a task to dynamically scale the breadth of the exploration for generating solution candidates. In addition, we use a custom prompt seeding method, based on principles of evolutionary search, to further elicit exploratory behavior from the base LLM and a novel voting method to select the best candidate solution at the end of the search. Experiments demonstrate that our solution achieves a state-of-the-art score of 51.72% on the “challenging” tier of BIRD development set problems, using only GPT-4o-mini, out-performing other in-context learning approaches, even those that leverage larger models. Overall, our method attains a competitive 61.06% execution accuracy and 68.77% Soft F1 score on the BIRD development dataset.

43. Fast Byte Latent Transformer

Authors: Julie Kallini , Artidoro Pagnoni , Tomasz Limisiewicz , Gargi Ghosh , Luke Zettlemoyer , Christopher Potts , Xiaochuang Han , Srinivasan Iyer
URL: https://arxiv.org/abs/2605.08044
Abstract:

Recent byte-level language models (LMs) match the performance of token-level models without relying on subword vocabularies, yet their utility is limited by slow, byte-by-byte autoregressive generation. We address this bottleneck in the Byte Latent Transformer (BLT) through new training and generation techniques. First, we introduce BLT Diffusion (BLT-D), a new model and our fastest BLT variant, trained with an auxiliary block-wise diffusion objective alongside the standard next-byte prediction loss. This enables an inference procedure that generates multiple bytes in parallel per decoding step, substantially reducing the number of forward passes required to generate a sequence. Second, we propose two extensions inspired by speculative decoding that trade some of this speed for higher generation quality: BLT Self-speculation (BLT-S), in which BLT’s local decoder continues generating past its normal patch boundaries to draft bytes, which are then verified with a single full-model forward pass; and BLT Diffusion+Verification (BLT-DV), which augments BLT-D with an autoregressive verification step after diffusion-based generation. All methods may achieve an estimated memory-bandwidth cost over 50% lower than BLT on generation tasks. Each approach offers its own unique advantages, together removing key barriers to the practical use of byte-level LMs.

44. Beyond Pairs: Your Language Model is Secretly Optimizing a Preference Graph

Authors: Ning Liu , Chuanneng Sun , Kristina Klinkner , Shervin Malmasi
URL: https://arxiv.org/abs/2605.08037
Abstract:

Direct Preference Optimization (DPO) aligns language models using pairwise preference comparisons, offering a simple and effective alternative to Reinforcement Learning (RL) from human feedback. However, in many practical settings, training data consists of multiple rollouts per prompt, inducing rich preference structure that pairwise DPO fails to exploit. Collapsing such data into independent pairs discards transitivity, introduces redundant or conflicting supervision, and can lead to unstable optimization. We propose Graph Direct Preference Optimization (GraphDPO), a principled generalization of DPO that operates over directed acyclic preference graphs induced by rollout rankings. GraphDPO encodes dominance relations as edges and optimizes a graph-structured Plackett–Luce-inspired objective that aggregates supervision over graph neighborhoods, enforcing transitivity while recovering standard DPO as a special case. To handle discrete or sparse signals, we introduce an equivalence-class construction where responses with identical preferences form graph layers, and intra-layer edges contribute zero loss, preventing spurious gradients. Despite leveraging full graph structure, GraphDPO maintains linear per-prompt complexity via efficient log-sum-exp aggregation. We further incorporate optional ground-truth anchoring by inserting verified solutions as dominant nodes and applying an annealed schedule that stabilizes early training while gradually relaxing oracle supervision. Experiments on reasoning and program synthesis tasks demonstrate superior performance, suggesting that graph-structured preference modeling is a scalable and robust alternative to pairwise and listwise alignment objectives.

45. Tool Calling is Linearly Readable and Steerable in Language Models

Authors: Zekun Wu (1, 2), Ze Wang (1, 2), Seonglae Cho (2), Yufei Yang (3), Adriano Koshiyama (1, 2), Sahan Bulathwela (1), Maria Perez-Ortiz (1) ((1) University College London, (2) Holistic AI, (3) Imperial College London)
URL: https://arxiv.org/abs/2605.07990
Abstract:

When a tool-calling agent picks the wrong tool, the failure is invisible until execution: the email gets sent, the meeting gets missed. Probing 12 instruction-tuned models across Gemma 3, Qwen 3, Qwen 2.5, and Llama 3.1 (270M to 27B), we find the identity of the chosen tool is linearly readable and steerable inside the model. Adding the mean-difference between two tools’ average internal activations switches which tool the model selects at 77-100% accuracy on name-only single-turn prompts (93-100% at 4B+), and the JSON arguments that follow autoregressively match the new tool’s schema, so flipping the name is enough. The same per-tool means also flag likely errors before they happen: on Gemma 3 12B and 27B, queries where the gap between the top-1 and top-2 tool is smallest produce 14-21x more wrong calls than queries with the largest gap. The causal effect concentrates along one direction, the row of the output layer that produces the target tool’s first token: a unit vector along it at matched magnitude already reaches 93-100%, while what is left over leaves the choice almost untouched. Activation patching localises this to a small set of mid- and late-layer attention heads, and a within-topic probe across 14 same-domain $\tau$-bench airline tools reaches top-1 61-89% across five 4B-14B models, ruling out the reading that we are just moving the model along a topic axis. Even base models encode the right tool before they can emit it: cosine readout from the internal state recovers 69-82% on BFCL while base generation reaches only 2-10%, suggesting pretraining forms the representation and instruction tuning later wires it to the output. We measure tool identity selection and JSON schema correctness in single-turn fixed-menu settings; multi-turn agentic transfer is more fragile and is discussed in Limitations.

46. Towards Apples to Apples for AI Evaluations: From Real-World Use Cases to Evaluation Scenarios

Authors: Yee-Yin Choong , Kristen Greene , Alice Qian , Meryem Marasli , Ziqi Yang , Sophia Chen , Laura Dabbish , Anand Rao , Hong Shen
URL: https://arxiv.org/abs/2605.07986
Abstract:

AI measurement science has a wide variety of methodologies and measurements for comparing AI systems, resulting in what often appear to be “apples-to-oranges” comparisons across AI evaluations. To move toward “apples-to-apples” comparisons in real-world AI evaluations, this work advocates for methodological transparency in evaluation scenarios, operational grounding, and human-centered design (HCD) principles. We propose a repeatable process for transforming high-level use cases to detailed scenarios by eliciting use cases from subject matter experts (SMEs) via a structured AI Use Case Worksheet with six key elements: use case, sector, user (direct and indirect), intended outcomes, expected impacts (positive and negative), and KPIs and metrics. We demonstrate utility of the worksheet and process in the U.S. financial services sector. This paper reports on example high-level AI use cases identified by financial services sector SMEs: cyber defense enablement, developer productivity, financial crime aggregation, suspicious activity report (SAR) filing, credit memo generation, and internal call center support. These AI use cases provided are illustrative of the process and not exhaustive. Central to our work is a three-stage expansion pipeline combining LLM prompting with human reviews to generate 107 scenarios from those use cases elicited from SMEs. This process integrates iterative human reviews at every juncture to ensure operational grounding: for scenario titles and descriptions; for core scenario elements like users, benefits and risks, and metrics; and for scenario narratives and evaluation objectives. Human checkpoints ensure scenarios remain reflective of real-world usage and human needs. We describe a validation rubric to assess scenario quality. By defining key scenario components, this work supports a more consistent and meaningful paradigm for human-centered AI evaluations.

47. Dooly: Configuration-Agnostic, Redundancy-Aware Profiling for LLM Inference Simulation

Authors: Joon Ha Kim , Geon-Woo Kim , Anoop Rachakonda , Daehyeok Kim
URL: https://arxiv.org/abs/2605.07985
Abstract:

Selecting the optimal LLM inference configuration requires evaluation across hardware, serving engines, attention backends, and model architectures, since no single choice performs best across all workloads. Profile-based simulators are the standard tool, yet they hardcode their operation set to a specific configuration and re-profile every operation from scratch, making exploration prohibitively expensive. This cost stems from a missing structural understanding: every input dimension of each operation is fixed by the model configuration or determined by the incoming request. Many model-configuration values (e.g., head size, layer count) recur across models, so the same operation runs in many configurations; a single sweep over the request-dependent dimensions can serve them all. We present Dooly, which exploits this structure to achieve configuration-agnostic, redundancy-aware profiling. Dooly performs a single inference pass, labels each input dimension with its origin via taint propagation, and selectively profiles only operations absent from its latency database; stateful operations such as attention are isolated by reusing the serving engine’s own initialization code, eliminating manual instrumentation. It builds latency regression models based on the database, which becomes a drop-in backend for existing simulators. Across two GPU platforms, three attention backends, and diverse model architectures, Dooly achieves simulation accuracy within 5% MAPE for TTFT and 8% for TPOT while reducing profiling GPU-hours by 56.4% across 12 models compared to the existing profiling approach.

48. Where’s the Plan? Locating Latent Planning in Language Models with Lightweight Mechanistic Interventions

Authors: Nicole Ma , Nick Rui
URL: https://arxiv.org/abs/2605.07984
Abstract:

We study planning site formation in language models – where internal representations of structurally-constrained future tokens form during the forward pass, and whether they causally drive generation. Using rhyming-couplet completion as a clean test of forward-looking constraint, we apply two lightweight methods (linear probing and activation patching) across Qwen3, Gemma-3, and Llama-3 at more than ten scales. Probing shows that future-rhyme information is linearly decodable at the line boundary, with signal that strengthens with scale in all three families. Activation patching reveals that only Gemma-3-27B causally relies on this encoding, exhibiting a handoff in which the causal driver migrates from the rhyme word to the line boundary around layer 30. Every other model we test conditions on the rhyme word throughout generation, with near-zero causal effect at the line boundary despite strong probe signal. We localize the Gemma-3-27B handoff to five attention heads through two-stage path patching that recover ~90% of the rhyme-routing capacity at the newline.

49. KL for a KL: On-Policy Distillation with Control Variate Baseline

Authors: Minjae Oh , Sangjun Song , Gyubin Choi , Yunho Choi , Yohan Jo
URL: https://arxiv.org/abs/2605.07865
Abstract:

On-Policy Distillation (OPD) has emerged as a dominant post-training paradigm for large language models, especially for reasoning domains. However, OPD remains unstable in practice due to the high gradient variance of its single-sample Monte Carlo estimator, and recipes for stable training are still immature. We propose vOPD (On-Policy Distillation with a control variate baseline), which casts OPD as policy-gradient RL and stabilizes it by introducing a control variate baseline-canonically a value function – from the RL literature. We show that the OPD value function admits a closed form as the per-token negative reverse KL divergence between the student and the teacher, available directly from the already-computed forward pass with no additional critic or inference. Existing stabilization methods either compute the full token-level reverse KL over the entire vocabulary, adding significant overhead, or restrict it to a top-k support, biasing the objective. vOPD instead preserves the lightweight single-sample estimator, subtracting the value function as a detached baseline to keep the gradient unbiased while reducing variance. Furthermore, we show that a top-k approximation of the baseline further lowers cost without compromising performance. Across mathematical and scientific reasoning benchmarks, vOPD consistently outperforms vanilla OPD and matches the most expensive full-vocabulary baseline, offering an efficient stabilization of On-Policy Distillation through principled RL variance reduction.

50. MatryoshkaLoRA: Learning Accurate Hierarchical Low-Rank Representations for LLM Fine-Tuning

Authors: Ionut-Vlad Modoranu , Mher Safaryan , Dan Alistarh
URL: https://arxiv.org/abs/2605.07850
Abstract:

With the rise in scale for deep learning models to billions of parameters, the computational cost of fine-tuning remains a significant barrier to deployment. While Low-Rank Adaptation (LoRA) has become the standard for parameter-efficient fine-tuning, the need to set a predefined, static rank $r$ requires exhaustive grid searches to balance efficiency and performance. Existing rank-adaptive solutions such as DyLoRA mitigate this by sampling ranks during the training from a predefined distribution. However, they often yield sub-optimal results at higher ranks due to lack of consistent gradient signals across the full hierarchy of ranks, thus making these methods data-inefficient. In this paper, we propose MatryoshkaLoRA, a general, Matryoshka-inspired training framework for LoRA that learns accurate hierarchical low-rank representations by inserting a fixed, carefully crafted diagonal matrix $P$ between the existing LoRA adapters to scale their sub-ranks accordingly. By introducing this simple modification, our general framework recovers LoRA and DyLoRA only by changing $P$ and ensures all sub-ranks embed the available gradient information efficiently. Our MatryoshkaLoRA supports dynamic rank selection with minimal degradation in accuracy. We further propose Area Under the Rank Accuracy Curve (AURAC), a metric that consistently evaluates the performance of hierarchical low-rank adapters. Our results demonstrate that MatryoshkaLoRA learns more accurate hierarchical low-rank representations than prior rank-adaptive approaches and achieves superior accuracy-performance trade-offs across ranks on the evaluated datasets. Our code is available at this https URL .

51. CyBiasBench: Benchmarking Bias in LLM Agents for Cyber-Attack Scenarios

Authors: Taein Lim , Seongyong Ju , Munhyeok Kim , Hyunjun Kim , Hoki Kim
URL: https://arxiv.org/abs/2605.07830
Abstract:

Large language models (LLMs) are increasingly deployed as autonomous agents in offensive cybersecurity. In this paper, we reveal an interesting phenomenon: different agents exhibit distinct attack patterns. Specifically, each agent exhibits an attack-selection bias, disproportionately concentrating its efforts on a narrow subset of attack families regardless of prompt variations. To systematically quantify this behavior, we introduce CyBiasBench, a comprehensive 630-session benchmark that evaluates five agents on three targets and four prompt conditions with ten attack families. We identify explicit bias across agents, with different dominant attack families and varying entropy levels in their attack-family allocation distributions. Such bias is better characterized as a trait of the agents, rather than a factor associated with the attack success rate. Furthermore, our experiments reveal a bias momentum effect, where agents resist explicit steering toward attack families that conflict with their bias. This forced distribution shift does not yield measurable improvements in attack performance. To ensure reproducibility and facilitate future research, we release an interactive result dashboard at this https URL and a reproducibility artifact with aggregated session-level statistics and full evaluation scripts at this https URL .

52. GazeVLM: Active Vision via Internal Attention Control for Multimodal Reasoning

Authors: Brown Ebouky , Gabriele Carrino , Niccolo Avogaro , Christoph Studer , Andrea Bartezzaghi , Mattia Rigotti
URL: https://arxiv.org/abs/2605.07817
Abstract:

Human visual reasoning is governed by active vision, a process where metacognitive control drives top-down goal-directed attention, dynamically routing foveal focus toward task-relevant details while maintaining peripheral awareness of the global scene. In contrast, modern Vision-Language Models (VLMs) process visual information passively, relying on the static accumulation of massive token contexts that dilute spatial reasoning and induce linguistic hallucinations. Here we propose the following paradigm shift: GazeVLM, a multimodal architecture that internalizes this metacognitive oversight over its deployment of attention resources directly into the reasoning loop. By empowering the VLM to autonomously generate gaze tokens ($\texttt{}$), GazeVLM establishes a top-down control mechanism over its own causal attention mask. The model dynamically dictates its focal intent, triggering a continuous suppression bias that dampens irrelevant visual features, implementing spatial selective attention and simulating foveal fixation. Once local reasoning concludes, the bias lifts, seamlessly restoring the global view. This architecture enables the model to fluidly transition between global spatial awareness and localized focal reasoning without relying on external agentic contraptions like cropping tools, or inflating the context window with additional visual tokens derived from localized visual patches. Trained with a bespoke Group Relative Policy Optimization (GRPO) procedure that rewards valid grounding, our 4B-parameter GazeVLM delivers strong high-resolution multimodal reasoning performance, surpassing state-of-the-art VLMs in its parameter class by nearly 4% and agentic multimodal pipelines built around thinking with images by more than 5% on HRBench-4k and HRBench-8k.

53. Beyond Confidence: Rethinking Self-Assessments for Performance Prediction in LLMs

Authors: Sree Bhattacharyya , Samarth Khanna , Leona Chen , Lucas Craig , Tharun Dilliraj , James Z. Wang
URL: https://arxiv.org/abs/2605.07806
Abstract:

Large Language Models (LLMs) are increasingly used in settings where reliable self-assessment is critical. Assessing model reliability has evolved from using probabilistic correctness estimates to, more recently, eliciting verbalized confidence. Confidence, however, has been shown to be an inconsistent and overoptimistic predictor of model correctness. Drawing on cognitive appraisal theory, a framework from human psychology that decomposes self-evaluation into multiple components, we propose a multidimensional perspective on model self-assessment. We elicit six appraisal-based dimensions of self-assessment, alongside confidence, and evaluate their utility for predicting model failure across 12 LLMs and 38 tasks spanning eight domains. We find that competence-related appraisal dimensions, particularly effort and ability, consistently match or outperform confidence across most settings. Effort additionally yields less overoptimistic estimates that remain stable across model sizes. In contrast, affective dimensions provide marginally predictive signals. Furthermore, the most informative dimension varies systematically with task characteristics: effort is most predictive for reasoning-intensive tasks, while ability and confidence dominate on retrieval-oriented tasks. Broadly, our findings indicate that structured multidimensional self-assessment is a promising approach to improving the reliability and safety of language model deployment across diverse real-world settings.

54. Tracing Uncertainty in Language Model “Reasoning”

Authors: Nils Grünefeld , Bertram Højer , Philipp Mondorf , Barbara Plank , Anna Rogers , Christian Hardmeier , Stefan Heinrich , Jes Frellsen
URL: https://arxiv.org/abs/2605.07776
Abstract:

Language model (LM) “reasoning”, commonly described as Chain-of-Thought or test-time scaling, often improves benchmark performance, but the dynamics underlying this process remain poorly understood. We study these dynamics through the lens of uncertainty quantification by treating the “reasoning” traces, the intermediate token sequences generated by LMs, as evolving model states. We summarize each trace by an uncertainty trace profile: a small set of features describing the shape of the uncertainty signal over its trace, such as its slope and linearity. We find that across five LMs evaluated on GSM8K and ProntoQA, these profiles predict whether a trace yields a correct final answer with AUROC up to 0.807, improving markedly on recent related work. We reach AUROC 0.801 using only the first few hundred tokens of full traces, suggesting that errors can be detected early in the generation. A detailed comparison of correct and incorrect traces further reveals qualitatively distinct uncertainty profiles, with correct traces showing a steeper and less linear decline in uncertainty. Together, the results suggest that our method, grounded in decision-making under uncertainty, provides a principled lens for studying the generative process underlying LM “reasoning”.

55. POETS: Uncertainty-Aware LLM Optimization via Compute-Efficient Policy Ensembles

Authors: Nicolas Menet , Andreas Krause , Abbas Rahimi
URL: https://arxiv.org/abs/2605.07775
Abstract:

Balancing exploration and exploitation is a core challenge in sequential decision-making and black-box optimization. We introduce POETS ($\textbf{Po}$licy $\textbf{E}$nsembles for $\textbf{T}$hompson $\textbf{S}$ampling), a novel framework that bridges uncertainty quantification and policy optimization. Our approach is grounded in the insight that policies trained with Kullback-Leibler (KL) regularization implicitly encode an underlying reward function. Building on this, POETS bypasses the complex, nested process of training an uncertainty-aware reward model and separately fitting a policy to this model. Instead, we directly train a policy ensemble to capture epistemic uncertainty by matching implicitly encoded reward functions to online, bootstrapped data. To overcome the prohibitive compute and memory constraints of ensembling Large Language Models (LLMs), POETS utilizes an efficient architecture: the ensemble shares a pre-trained backbone while maintaining diversity through independent Low-Rank Adaptation (LoRA) branches. Theoretically, we prove that POETS implicitly conducts KL-regularized Thompson sampling and thus inherits strong cumulative regret bounds of ${\mathcal O}(\sqrt{T \gamma_T})$. Empirically, we demonstrate that POETS achieves state-of-the-art sample efficiency across diverse scientific discovery domains, including protein search and quantum circuit design. Furthermore, it improves the optimization trajectories of reinforcement learning, proving particularly robust in off-policy settings with experience replay or in small dataset regimes.

56. Benchmarking EngGPT2-16B-A3B against Comparable Italian and International Open-source LLMs

Authors: Andrea Sassella , Andrea Chizzola , Tommaso Bianchi , Luca Alessandrelli , Mark James Carman
URL: https://arxiv.org/abs/2605.07731
Abstract:

This report benchmarks the performance of ENGINEERING Ingegneria Informatica S.p.A.’s EngGPT2MoE-16B-A3B LLM, a 16B parameter Mixture of Experts (MoE) model with 3B active parameters. Performance is investigated across a wide variety of representative benchmarks, and is compared against comparably-sized open-source MoE and dense models. In comparison with popular Italian models, namely FastwebMIIA-7B, Minerva-7B, Velvet-14B, and LLaMAntino-3-ANITA-8B, EngGPT2MoE-16B-A3B performs as well or better on international benchmarks: ARC-Challenge, GSM8K, AIME24, AIME25, MMLU, and HumanEval (HE). It achieves the best performance for the longest context setting (32k) of the RULER benchmark. On the Italian benchmark dataset ITALIC, the model performs as well or better than the other models except for Velvet-14B, which outperforms it. Compared with popular MoE models of comparable size, the new model reports higher values than DeepSeek-MoE-16B-Chat on all considered benchmarks. It has higher values than Moonlight-16B-A3B on HE, MMLU, AIME24, AIME25, GSM8K, and the 32k RULER setting, but lower on BFCL and some ARC and ITALIC settings. Finally it has lower values than GPT-OSS-20B on most benchmarks, including HE, MMLU, AIME24, AIME25, GSM8K, ARC, BFCL, and the RULER 32k. When compared with popular dense models, EngGPT2MoE-16B-A3B reports higher values on AIME24 and AIME25 than Llama-3.1-8B-Instruct, Gemma-3-12b-it, and Ministral-3-8BInstruct-2512-BF16, but lower values on ITALIC, BFCL, and RULER with a 32k context. When performance is aggregated across all benchmark metrics, EngGPT2MoE-16B-A3B shows higher performance than the Italian models under evaluation while achieving lower results than some of the most performant international models, in particular GPT-5 nano and Qwen3-8B. Taken together, our findings find the new model to be a step forward for native Italian Large Language Models.

57. SOD: Step-wise On-policy Distillation for Small Language Model Agents

Authors: Qiyong Zhong , Mao Zheng , Mingyang Song , Xin Lin , Jie Sun , Houcheng Jiang , Xiang Wang , Junfeng Fang
URL: https://arxiv.org/abs/2605.07725
Abstract:

Tool-integrated reasoning (TIR) is difficult to scale to small language models due to instability in long-horizon tool interactions and limited model capacity. While reinforcement learning methods like group relative policy optimization provide only sparse outcome-level rewards. Recently, on-policy distillation (OPD) has gained popularity by supplying dense token-level supervision from a teacher on student-generated trajectories. However, our experiments indicate that applying OPD to TIR leads to a critical failure mode: erroneous tool calls tend to cascade across subsequent reasoning steps, progressively amplifying student-teacher divergence and rendering the teacher’s token-level supervision increasingly unreliable. To address this, we propose SOD, a step-wise on-policy distillation framework for small language model agents, which adaptively reweights distillation strength at each step based on step-level divergence. Therefore, SOD can attenuate potentially misleading teacher signals in high-divergence regions while preserving dense guidance in well-aligned states. Experiments on challenging math, science, and code benchmarks show that SOD achieves up to 20.86% improvement over the second-best baseline. Notably, our 0.6B student achieves 26.13% on AIME 2025, demonstrating effective transfer of agentic reasoning to lightweight models. Our code is available at this https URL .

58. LLM hallucinations in the wild: Large-scale evidence from non-existent citations

Authors: Zhenyue Zhao , Yihe Wang , Toby Stuart , Mathijs De Vaan , Paul Ginsparg , Yian Yin
URL: https://arxiv.org/abs/2605.07723
Abstract:

Large language models (LLMs) are known to generate plausible but false information across a wide range of contexts, yet the real-world magnitude and consequences of this hallucination problem remain poorly understood. Here we leverage a uniquely verifiable object - scientific citations - to audit 111 million references across 2.5 million papers in arXiv, bioRxiv, SSRN, and PubMed Central. We find a sharp rise in non-existent references following widespread LLM adoption, with a conservative estimate of 146,932 hallucinated citations in 2025 alone. These errors are diffusely embedded across many papers but especially pronounced in fields with rapid AI uptake, in manuscripts with linguistic signatures of AI-assisted writing, and among small and early-career author teams. At the same time, hallucinated references disproportionately assign credit to already prominent and male scholars, suggesting that LLM-generated errors may reinforce existing inequities in scientific recognition. Preprint moderation and journal publication processes capture only a fraction of these errors, suggesting that the spread of hallucinated content has outpaced existing safeguards. Together, these findings demonstrate that LLM hallucinations are infiltrating knowledge production at scale, threatening both the reliability and equity of future scientific discovery as human and AI systems draw on the existing literature.

59. Memory-Efficient Looped Transformer: Decoupling Compute from Memory in Looped Language Models

Authors: Victor Conchello Vendrell , Arnau Padres Masdemont , Niccolò Grillo , Jordi Ros-Giralt , Arash Behboodi , Fabio Valerio Massoli
URL: https://arxiv.org/abs/2605.07721
Abstract:

Recurrent LLM architectures have emerged as a promising approach for improving reasoning, as they enable multi-step computation in the embedding space without generating intermediate tokens. Models such as Ouro perform reasoning by iteratively updating internal representations while retaining a standard Key-Value (KV) cache across iterations, causing memory consumption to grow linearly with reasoning depth. Consequently, increasing the number of reasoning iterations can lead to prohibitive memory usage, limiting the practical scalability of such architectures. In this work, we propose Memory-Efficient Looped Transformer (MELT), a novel architecture that decouples reasoning depth from memory consumption. Instead of using a standard KV cache per layer and loop, MELT maintains a single KV cache per layer that is shared across reasoning loops. This cache is updated over time via a learnable gating mechanism. To enable stable and efficient training under this architecture, we propose to train MELT using chunk-wise training in a two phase procedure: interpolated transition, followed by attention-aligned distillation, both from the LoopLM starting model to MELT. Empirically, we show that MELT models fine-tuned from pretrained Ouro parameters outperform standard LLMs of comparable size, while maintaining a memory footprint comparable to those models and dramatically smaller than Ouro’s. Overall, MELT achieves constant-memory iterative reasoning without sacrificing LoopLM performance, using only a lightweight post-training procedure.

60. The AI-Native Large-Scale Agile Software Development Manifesto

Authors: Ricardo Britto , Fredrik Palmgren , Nishrith Saini , Marcus Ohlin
URL: https://arxiv.org/abs/2605.07717
Abstract:

Despite the widespread adoption of agile methods, achieving true agility at scale remains elusive. Large-scale agile frameworks remain largely human-centric and manual, relying on coordination meetings, artifact synchronization, and role-based handoffs that inhibit real-time adaptation. Meanwhile, rapid advances in AI, particularly large language models, have begun transforming software engineering, yet their potential for organizational-level agility remains underexplored. We present the AI-Native Large-Scale Agile Software Development Manifesto: a set of values and principles that redefine how large-scale software development is organized when AI becomes a first-class participant rather than a peripheral tool. The manifesto is grounded in six principles, parallel processes, intent-driven teams, living knowledge, verification-first assurance, orchestrated agent workforces, and reusable blueprints, that together shift development from a meeting-driven, document-heavy, sequential process to an intelligent, adaptive, continuously learning system.

61. DRIP-R: A Benchmark for Decision-Making and Reasoning Under Real-World Policy Ambiguity in the Retail Domain

Authors: Hsuvas Borkakoty , Sebastian Pohl , Cheng Wang , Bei Chen , Yufang Hou
URL: https://arxiv.org/abs/2605.07699
Abstract:

LLM-based agents are increasingly deployed for routine but consequential tasks in real-world domains, where their behavior is governed by inherently ambiguous domain policies that admit multiple valid interpretations. Despite the prevalence of such ambiguities in practice, existing agent benchmarks largely assume unambiguous, well-specified policies, leaving a critical evaluation gap. We introduce DRIP-R, a benchmark that systematically exploits real-world retail policy ambiguities to construct scenarios in which no single correct resolution exists. DRIP-R comprises a curated set of policy-ambiguous return scenarios paired with a realistic customer personas, a full-duplex conversational simulation with tool-calling capabilities and a multi-judge evaluation framework covering policy adherence, dialogue quality, behavioral alignment, and resolution quality. Our experiments show that frontier models fundamentally disagree on identical policy-ambiguous scenarios, confirming that ambiguity poses a genuine and systematic challenge to LLM decision-making.

62. TRACE: Tourism Recommendation with Accountable Citation Evidence

Authors: Zixu Zhao , Sijin Wang , Yu Hou , Yuanyuan Xu , Yufan Sheng , Xike Xie , Wenjie Zhang , Won-Yong Shin , Xin Cao
URL: https://arxiv.org/abs/2605.07677
Abstract:

Tourism is a high-stakes setting for conversational recommender systems (CRS): a plausible-sounding suggestion can waste real money and trip time once a traveler acts on it. Existing CRS benchmarks primarily evaluate systems with a single Recall@k score over entity mentions, and tourism-specific resources add spatial or knowledge-graph context, yet none of them couple multi-turn recommendation with verbatim review-span evidence and rejection recovery. This leaves an evaluation gap for tourism recommendation that is simultaneously trustworthy, verifiable, and adaptive: recommend the right point of interest (POI) for multi-aspect preferences (such as cuisine, price, atmosphere, walking distance), justify each suggestion with verifiable evidence from prior visitors so the traveler can act without trial and error, and recover when the first recommendation is rejected mid-dialogue. We introduce TRACE, where each item is a multi-turn tourism recommendation dialogue with review-span citations and explicit rejection turns: 10,000 dialogues over 2,400 Yelp POIs and 34,208 reviews across eight U.S. cities, paired with 14 retrieval, planning, and LLM baselines, along with 25 metrics organized under Accuracy, Grounding, and Recovery. Across these baselines, TRACE reveals the Three-Competency Gap: LLM Zero-Shot leads in closed-set Recall@1 and rejection recovery but cites less densely than retrievers; non-LLM retrievers achieve surface-verbatim grounding but with low accuracy; Multi-Review Synthesis fails at recovery. The Grounding Score agrees with human citation precision (Spearman rho=+0.80, p<10^-20), and paired t-tests reproduce the per-baseline ranking (p<0.01 on the dominant contrasts). TRACE reframes accountable tourism recommendation as a joint target (right POI, verifiable evidence, adaptive repair) rather than a single-axis leaderboard.

63. Operating Within the Operational Design Domain: Zero-Shot Perception with Vision-Language Models

Authors: Berkehan Ünal , Dierend Hauke , Fazlija Dren , Plachetka Christopher
URL: https://arxiv.org/abs/2605.07649
Abstract:

Over the last few years, research on autonomous systems has matured to such a degree that the field is increasingly well-positioned to translate research into practical, stakeholder-driven use cases across well-defined domains. However, for a wide-scale practical adoption of autonomous systems, adherence to safety regulations is crucial. Many regulations are influenced by the Operational Design Domain (ODD), which defines the specific conditions in which an autonomous agent can function. This is especially relevant for Automated Driving Systems (ADS), as a dependable perception of ODD elements is essential for safe implementation and auditing. Vision-language models (VLMs) integrate visual recognition and language reasoning, functioning without task-specific training data, which makes them suitable for adaptable ODD perception. To assess whether VLMs can function as zero-shot “ODD sensors” that adapt to evolving definitions, we contribute (i) an empirical study of zero-shot ODD classification and detection using four VLMs on a custom dataset and Mapillary Vistas, along with failure analyses; (ii) an ablation of zero-shot optimization strategies with a cost-performance overview; and (iii) a suite of reusable prompting templates with guidance for adaptation. Our findings indicate that definition-anchored chain-of-thought prompting with persona decomposition performs best, while other methods may result in reduced recall. Overall, our results pave the way for transparent and effective ODD-based perception in safety-critical applications.

64. Quality-Conditioned Agreement in Automated Short Answer Scoring: Mid-Range Degradation and the Impact of Task-Specific Adaptation

Authors: Abigail Victoria Gurin Schleifer , Moriah Ariely , Beata Beigman Klebanov , Asaf Salman , Giora Alexandron
URL: https://arxiv.org/abs/2605.07647
Abstract:

Automated short answer scoring (ASAS) is shifting from discriminative, fine-tuned models to large language models (LLMs) used in few-shot settings. This paradigm leverages LLMs broad world knowledge and ease of deployment, but limited task-specific data may reduce alignment on complex scoring tasks. In particular, its impact on scoring partially correct responses that require nuanced interpretation remains underexplored. We investigate the relationship between the degree of task-specific adaptation of different models and quality-conditioned scoring agreement. We compare three LLMs (GPT-5.2, GPT-4o, Claude Opus 4.5) in few-shot mode, a fine-tuned BERT-based encoder, and a human expert on two open-ended biology items, using several hundred student responses and ground truth scores provided by a biology education expert. The results show that human-human agreement is highest and stable across the full quality spectrum. All AI models perform well on fully correct and fully incorrect responses, but exhibit substantial degradation on mid-range responses. This mid-range degradation is conditioned on task-specific adaptation: It is most severe in few-shot LLMs with few examples and decreases as task-specific data increases, with fine-tuned encoder models performing best. This mid-range degradation may lead to inequitable evaluation of responses produced by students with developing understanding. Our findings highlight the importance of quality-conditioned fairness, with particular attention to mid-range responses.

65. LithoBench: Benchmarking Large Multimodal Models for Remote-Sensing Lithology Interpretation

Authors: Jun Wang , Fengpeng Li , Hang Dong , Tianjin Huang , Wei Han
URL: https://arxiv.org/abs/2605.07640
Abstract:

Remote sensing lithology interpretation is fundamental to geological surveys, mineral exploration, and regional geological mapping. Unlike general land-cover recognition, lithology interpretation is a knowledge-intensive task that requires experts to infer rock types from various features, e.g., subtle visual, spectral, textural, geomorphological, and contextual cues, making reliable automated interpretation highly challenging. Geological knowledge-guided large multimodal models offer new opportunities, yet their evaluation remains constrained by the lack of benchmarks that capture lithological annotations, multi-level geological semantics, and expert-informed assessment. Here, we propose LithoBench, a multi-level benchmark for evaluating geological semantic understanding in remote sensing lithology interpretation. LithoBench contains 10,000 expert-annotated interpretation instances across 12 representative lithological categories, including 4,000 multiple-choice and 6,000 open-ended tasks organized into five cognitive levels: Identification and Description, Comparative Analysis, Mechanism Explanation, Practical Application, and Comprehensive Reasoning. We further develop an expert-in-the-loop, knowledge-grounded semi-automated construction pipeline, coupling multi sub-processes, e.g., structured geological image descriptions, to enhance geological validity and evaluation reliability. Experiments with multiple large vision-language models eveal substantial limitations in geological semantic understanding, particularly on higher-order explanation, application, and reasoning tasks.

66. Post-training makes large language models less human-like

Authors: Marcel Binz , Elif Akata , Abdullah Almaatouq , Mohammed Alsobay , Oleksii Ariasov , Franziska Brändle , David Broska , Jason W. Burton , Nuno Busch , Frederick Callaway , Vanessa Cheung , Brian Christian , Julian Coda-Forno , Can Demircan , Vittoria Dentella , Maria K. Eckstein , Noémi Éltető , Michael Franke , Thomas L. Griffiths , Fritz Günther , Susanne Haridi , Sebastian Hellmann , Stefan Herytash , Linus Hof , Eleanor Holton , Isabelle Hoxha , Zak Hussain , Akshay Jagadish , Elif Kara , Valentin Kriegmair , Evelina Leivada , Li Ji-An , Tobias Ludwig , Maximilian Maier , Marcelo G. Mattar , Marvin Mathony , Alireza Modirshanechi , Robin Na , Mariia Nadverniuk , Antonios Nasioulas , Surabhi S. Nath , Helen Niemeyer , Kate Nussenbaum , Sebastian Olschewski , Thorsten Pachur , Stefano Palminteri , Aliona Petrenco , Camille V. Phaneuf-Hadd , Angelo Pirrone , Manuel Rausch , Laura Raveling , Shashank Reddy , Milena Rmus , Evan M. Russek , Tankred Saanum , Kai Sandbrink , Louis Schiekiera , Johannes A. Schubert , Luca M. Schulze Buschoff , Nishad Singhi , Leah H. Somerville , Mikhail S. Spektor , Xin Sui , Christopher Summerfield , Mirko Thalmann , Anna I. Thoma , Taisiia Tikhomirova , Vuong Truong , Polina Tsvilodub , Konstantinos Voudouris , Robert C. Wilson , Kristin Witte , Shuchen Wu , Dirk U. Wulff , Hua-Dong Xiong , Songlin Xu , Lance Ying , Xinyu Zhang , Jian-Qiao Zhu , Eric Schulz
URL: https://arxiv.org/abs/2605.07632
Abstract:

Large language models (LLMs) are increasingly used as surrogates for human participants, but it remains unclear which models best capture human behavior and why. To address this, we introduce Psych-201, a novel dataset that enables us to measure behavioral alignment at scale. We find that post-training – the stage that turns base models into useful assistants – consistently reduces alignment with human behavior across model families, sizes, and objectives. Moreover, this misalignment widens in newer model generations even as base models continue to improve. Finally, we find that persona-induction – a popular technique for eliciting human-like behavior by conditioning models on participant-specific information – does not improve predictions at the level of individuals. Taken together, our results suggest that the very processes that are currently employed to turn LLMs into useful assistants also make them less accurate models of human behavior.

67. Mathematical Reasoning via Intervention-Based Time-Series Causal Discovery Using LLMs as Concept Mastery Simulators

Authors: Tsuyoshi Okita
URL: https://arxiv.org/abs/2605.07600
Abstract:

Recent methods for improving LLM mathematical reasoning, whether through MCTS-based test-time search or causal graph-guided knowledge injection, cannot identify which concepts causally contribute to a correct answer, as the observed association may be spurious, driven by confounders such as problem difficulty. We propose CIKA (Causal Intervention for Knowledge Activation), a framework that uses the LLM itself as an interventional simulator: a prompt sets the concept state to ``mastered’’ and the correctness change estimates the causal effect. We formalize this quantity as an Interventional Capability Probe (ICP), which diagnoses whether the LLM can use a given concept – distinct from merely possessing knowledge. Because the intervention exogenously sets the concept state independently of problem difficulty, ICP separates confounding that observational methods cannot. On 67 screened problems, the ICP of the top-ranked concept (+0.219) is significantly larger than that of the negative control (+0.039; paired $t$-test, $p < 10^{-6}$, Cohen’s $d = 0.86$), confirming that the probe discriminates causally relevant concepts from irrelevant ones. Analysis of 601 Omni-MATH problems further shows that solved problems have 6.1$\times$ higher ATE than unsolved ones (0.338 vs. 0.055), confirming that ICP is predictive of problem-solving success. With a 7B-parameter LLM whose weights are entirely frozen, CIKA achieves 69.7\% on the contamination-free Omni-MATH-Rule benchmark and 64.0\% overall, compared to 60.5\% for o1-mini, and 97.2\% on GSM8K, 46–50\% on AIME 2024–2026, and 46.2\% on MathArena. The Causal Knowledge Activation component contributes 33.8\% of correct answers on problems where the base model alone fails, demonstrating that the LLM already possessed but had not activated the requisite knowledge.

68. Your Language Model is Its Own Critic: Reinforcement Learning with Value Estimation from Actor’s Internal States

Authors: Yunho Choi , Jongwon Lim , Woojin Ahn , Minjae Oh , Jeonghoon Shim , Yohan Jo
URL: https://arxiv.org/abs/2605.07579
Abstract:

Reinforcement learning with verifiable rewards (RLVR) for Large Reasoning Models hinges on baseline estimation for variance reduction, but existing approaches pay a heavy price: PPO requires a policy-model scale critic, while GRPO needs multiple rollouts per prompt to keep its empirical group mean stable. We introduce Policy Optimization with Internal State Value Estimation), which obtains a baseline at negligible cost by using the policy model’s internal signals already computed during the policy forward pass. A lightweight probe predicts the expected verifiable reward from the hidden states of the prompt and generated trajectory, as well as token-entropy statistics, and is trained online alongside the policy. To preserve gradient unbiasedness despite using trajectory-conditioned features, we introduce a cross-rollout construction that predicts each rollout’s value from an independent rollout’s internal states. Because POISE estimates prompt value using only a single rollout, it enables higher prompt diversity for a fixed compute budget during training. This reduces gradient variance for more stable learning and also eliminates the compute overhead of sampling costs for detecting zero-advantage prompts. On Qwen3-4B and DeepSeek-R1-Distill-Qwen-1.5B across math reasoning benchmarks, POISE matches DAPO while requiring less compute. Moreover, its value estimator shows similar performance to a separate LLM-scale value model and generalizes to various verifiable tasks. By leveraging the model’s own internal representations, POISE enables more stable and efficient policy optimization.

69. ProteinJEPA: Latent prediction complements protein language models

Authors: Dan Ofer , Dafna Shahaf , Michal Linial
URL: https://arxiv.org/abs/2605.07554
Abstract:

Protein language models are trained primarily with masked language modeling (MLM), which predicts amino-acid identities at masked positions. We ask whether latent-space prediction can complement these token-level objectives under matched wall-clock budget. Across pretrained and random-init protein sequence encoders at 35–150M parameters, we find that the best protein-JEPA design is not all-position latent prediction but a variant: predicting latent targets only at masked positions, and retaining the MLM cross-entropy. We call this recipe masked-position MLM+JEPA. On a 16-task downstream suite (15 frozen linear probes plus SCOPe-40 zero-shot fold retrieval), under matched wall-clock budgets, this recipe wins more tasks than it loses against MLM-only continuation: 10 wins / 3 losses / 3 ties (hereafter W/L/T) on pretrained ESM2-35M, 11/2/3 on ESM2-150M while results in pretraining from scratch are mixed (6/8/2). Gains are seen for multiple models on 11 of 16 tasks, including stability, \b{eta}\beta \b{eta}-lactamase fitness, variant effect, intrinsic disorder, remote homology, enzyme classification, and SCOPe-40 fold retrieval. Tasks with more losses than wins are Fluorescence (TAPE) and Peptide-HLA Binding. All-position MLM+JEPA matches MLM-only overall but does not reproduce the masked-position gains. JEPA-only (no MLM) collapses in nearly every experiment. We conclude that JEPA, when combined with MLM, is competitive and can outperform pure MLM in pretraining and continued training, even under matched wall-clock budgets.

70. LARAG: Link-Aware Retrieval Strategy for RAG Systems in Hyperlinked Technical Documentation

Authors: Giorgia Bolognesi , Claudio Estatico , Ulderico Fugacci , Isabella Mastroianni , Claudio Muselli , Luca Oneto
URL: https://arxiv.org/abs/2605.07517
Abstract:

Retrieval-Augmented Generation (RAG) enhances the factual grounding of Large Language Models by conditioning their outputs on external documents. However, standard embedding-based retrievers treat naturally structured corpora, such as technical manuals, as flat collections of passages, thereby overlooking the hyperlink topology that users rely on when navigating such content. We introduce LARAG (Link-Aware RAG): a lightweight, link-aware retrieval strategy that leverages the author-defined hyperlink structure already present in HTML documentation, encoding hyperlink relations as metadata in the chunk representations and exploiting them to perform a form of graph-like retrieval of locally relevant content. In a benchmark of twenty expert-designed queries over Rulex Platform technical documentation and four prompting strategies, LARAG consistently improves answer quality, achieving the highest BERTScore F1, while retrieving fewer chunks and generating fewer tokens than a baseline RAG architecture used for comparison. These results show that directly leveraging the existing hyperlink topology of technical documentation, even without explicit graph construction or inference, enables an implicit form of graph-like retrieval that yields a more faithful and efficient RAG pipeline, providing better grounding at lower cost.

71. SHRED: Retain-Set-Free Unlearning via Self-Distillation with Logit Demotion

Authors: Zizhao Hu , Ameya Godbole , Johnny Tian-Zheng Wei , Mohammad Rostami , Jesse Thomason , Robin Jia
URL: https://arxiv.org/abs/2605.07482
Abstract:

Machine unlearning for large language models (LLMs) aims to selectively remove memorized content such as private data, copyrighted text, or hazardous knowledge, without costly full retraining. Most existing methods require a retain set of curated examples to prevent catastrophic degradation of general model utility, creating an extra data dependency that complicates deployment. We propose SHRED (Self-distillation via High-surprisal-only Retain-set-free Entropy Demotion), a retain-set-free unlearning method built on a key insight: not all tokens within a forget set instance carry memorized information equally. High-information tokens concentrate the model’s memorized knowledge, while low-information tokens reflect general language competence. SHRED operates in two stages. (1) Selection: We perform a forward pass on a forget set instance, collect per-token autoregressive probabilities, and select the bottom (lowest probability, highest Shannon information) as forget positions; the remaining positions are retained as benign anchors. (2) Training: We construct modified KL targets that demote the memorized token’s logit at forget positions while preserving the original distribution at benign positions. The model is then trained via a single top KL self-distillation objective that simultaneously drives forgetting and utility preservation. We evaluate SHRED across four standard unlearning benchmarks and demonstrate that it establishes a new Pareto-optimal trade-off between forget efficacy and model utility, outperforming retain-set-dependent methods. Our analysis shows that SHRED is robust against relearning attacks and membership-inference attacks, and it maintains stable utility even after many sequential unlearning runs.

72. Vaporizer: Breaking Watermarking Schemes for Large Language Model Outputs

Authors: Jonathan Hong Jin Ng , Anh Tu Ngo , Anupam Chattopadhyay
URL: https://arxiv.org/abs/2605.07481
Abstract:

In this paper, we investigate the recent state-of-the-art schemes for watermarking large language models (LLMs) outputs. These techniques are claimed to be robust, scalable and production-grade, aimed at promoting responsible usage of LLMs. We analyse the effectiveness of these watermarking techniques against an extensive collection of modified text attacks, which perform targeted semantic changes without altering the general meaning of the text content. Our approach encompasses multiple attack strategies, which include lexical alterations, machine translation, and even neural paraphrasing. The attack efficacy is measured with two target criteria - successful removal of the watermark and preservation of semantic content. We evaluate semantic preservation through BERT scores, text complexity measures, grammatical errors, and Flesch Reading Ease indices. The experimental results reveal varying levels of effectiveness among different watermarking models, with the same underlying result that it is possible to remove the watermark with reasonable effort. This study sheds light on the strengths and weaknesses of existing LLM watermarking systems, suggesting how they should be constructed to improve security of available schemes.

73. HBEE: Human Behavioral Entropy Engine – Pre-Registered Multi-Agent LLM Simulation of Peer-Suspicion-Based Detection Inversion

Authors: Vickson Ferrel
URL: https://arxiv.org/abs/2605.07472

Abstract:

Insider threat detection assumes that an adaptive insider leaves behavioral residue distinguishing them from legitimate users. We test this assumption against an LLM-driven adaptive insider in a controlled multi-agent simulator. Our pre-registered five-condition study isolates defender mode (cascade vs. blind UEBA) crossed with adversary type (naive vs. adaptive OPSEC) plus a no-mole control, across 100 runs (95 valid after pre-committed exclusions). The primary finding is a detection inversion: at T_60, the adaptive mole’s suspicion in-degree is statistically lower than a randomly selected innocent agent (Cliff’s delta = -0.694, 95% BCa CI [-0.855, -0.519], Mann-Whitney p « 0.01). The pre-registered prediction was the opposite direction. A pre-registered equivalence test (H2) shows adaptive OPSEC produces no detectable shift in the mole’s UEBA rank under either defender mode. The two detection signals (peer suspicion graph in-degree and per-agent UEBA rank) decouple under adaptive adversary behavior. We bound generalization explicitly: a pre-registered Gini calibration check (H4) returns FAIL, with HBEE pairwise message-exposure Gini (0.213) diverging from the SNAP Enron reference (0.730) by Delta Gini = 0.52, exceeding the equivalence bound by 5x. The paper makes a narrow but surprising claim: in a controlled environment where adaptive OPSEC is implementable as an LLM directive, peer-suspicion-cascade detection inverts. We release the simulator, pre-registration document, frozen scenarios, raw telemetry, and analysis pipeline under an open-source license.

74. The Moltbook Files: A Harmless Slopocalypse or Humanity’s Last Experiment

Authors: William Brach , Federico Torrielli , Stine Lyngsø Beltoft , Annemette Brok Pirchert , Peter Schneider-Kamp , Lukas Galke Poech
URL: https://arxiv.org/abs/2605.07462
Abstract:

Moltbook is a Reddit-like platform where OpenClaw agents post, comment, and vote at scale - a so far unprecedented incident that comes with serious safety concerns. With the aim of studying emergent behavior in populations, we release the Moltbook Files, a dataset of 232k posts and 2.2M comments covering the platform’s first 12 days, processed through a pipeline to identify and remove Personally-Identifiable Information (PII). We analyze community structure, authorship, lexical properties, sentiment, topics, semantic geometry, and comment interaction. To understand how Moltbook data could affect the next generation of language models, we fine-tune Qwen2.5-14B-Instruct on Moltbook Files with three adaptation levels. Our PII pipeline reveals that agents post API keys, passwords, BIP39 seed phrases on Moltbook, a publicly indexed platform. The overall sentiment is mostly neutral and mildly positive (66.6% neutral, 19.5% positive) and shows a tendency for self-referential linking. We find that fine-tuning on Moltbook data reduces truthfulness from 0.366 to 0.187. However, a model fine-tuned on a size-matched Reddit dataset produces a comparable decrease. Moltbook thus seems to be more of a harmless slopocalypse. However, tail risks remain, including agent affordances, contamination of future crawls through self-links, and potential transfer of traits to the next generation of language models. More broadly, our findings highlight the importance of control baselines in emergent misalignment evaluations.

75. Sparse Autoencoders as Plug-and-Play Firewalls for Adversarial Attack Detection in VLMs

Authors: Hao Wang , Yiqun Sun , Pengfei Wei , Lawrence B. Hsieh , Daisuke Kawahara
URL: https://arxiv.org/abs/2605.07447
Abstract:

Vision-language models (VLMs) have advanced rapidly and are increasingly deployed in real-world applications, especially with the rise of agent-based systems. However, their safety has received relatively limited attention. Even the latest proprietary and open-weight VLMs remain highly vulnerable to adversarial attacks, leaving downstream applications exposed to significant risks. In this work, we propose a novel and lightweight adversarial attack detection framework based on sparse autoencoders (SAEs), termed SAEgis. By inserting an SAE module into a pretrained VLM and training it with standard reconstruction objectives, we find that the learned sparse latent features naturally capture attack-relevant signals. These features enable reliable classification of whether an input image has been adversarially perturbed, even for previously unseen samples. Extensive experiments show that SAEgis achieves strong performance across in-domain, cross-domain, and cross-attack settings, with particularly large improvements in cross-domain generalization compared to existing baselines. In addition, combining signals from multiple layers further improves robustness and stability. To the best of our knowledge, this is the first work to explore SAE as a plug-and-play mechanism for adversarial attack detection in VLMs. Our method requires no additional adversarial training, introduces minimal overhead, and provides a practical approach for improving the safety of real-world VLM systems.

76. Prompt Engineering Strategies for LLM-based Qualitative Coding of Psychological Safety in Software Engineering Communities: A Controlled Empirical Study

Authors: Moaath Alshaikh , Tasneem Alshaher , Ricardo Vieira , Beatriz Santana , Clelio Xavier , Jose Amancio , Glauco Carneiro , Julio Leite , Savio Freire , Manoel Mendonca
URL: https://arxiv.org/abs/2605.07422
Abstract:

Qualitative analysis plays a pivotal role in understanding the human and social aspects of software engineering. However, it remains a demanding process shaped by the subjective interpretation of individual researchers and sensitive to methodological choices such as prompt design. Recent advancements in Large Language Models (LLMs) offer promising opportunities to support this type of analysis, although their reliability in reproducing human qualitative reasoning under varying prompting conditions remains largely untested. This study presents a controlled empirical evaluation of three LLMs – Claude Haiku, DeepSeek-Chat, and Gemini 2.5 Flash – across two prompt engineering strategies (zero-shot and multi-shot closed coding), using Cohen’s kappa as the primary agreement metric over ten independent runs per configuration. Results suggest that multi-shot prompting significantly improves agreement for Claude Haiku (Delta kappa = +0.034, Wilcoxon p = 0.004) but not for DeepSeek-Chat or Gemini 2.5 Flash. Intra-model stability varies substantially – DeepSeek-Chat and Claude Haiku exhibit the lowest variance (SD approx. 0.017), while Gemini 2.5 Flash is the least stable (SD = 0.038). A systematic over-prediction of “Sharing Negative Feedback” is identified across all models (bias ratios up to 5.25x), alongside consistent under-prediction of “Expressing Concerns.” Collectively, these findings provide empirical evidence for prompt engineering guidelines in LLM-assisted qualitative coding for software engineering research.

77. Unsolvability Ceiling in Multi-LLM Routing: An Empirical Study of Evaluation Artifacts

Authors: Saloni Garg , Amit Sagtani
URL: https://arxiv.org/abs/2605.07395
Abstract:

Efficient routing across multiple LLMs enables cost-quality tradeoffs by directing queries to the cheapest capable model. Prior work attributes routing headroom to an “unsolvability ceiling”, queries no model in the pool can solve. We present a large-scale study of multi-tier LLM routing with 206,000 query-model pairs across six benchmarks (MMLU, MedQA, HumanEval, MBPP, Alpaca, ShareGPT) using the Gemma 4 and Llama 3.1 families. Evaluating with both LLM-as-a-judge and exact-match metrics, we show that a substantial portion of reported unsolvability stems from evaluation artifacts: (i) systematic judge biases favoring verbosity over correctness, (ii) truncation under fixed generation budgets, and (iii) output format mismatches. Through dual-judge validation and exact-match grounding, we reduce measured unsolvability across tasks. We introduce a decomposition framework attributing failures to these artifacts, revealing consistent patterns across domains and model families. These artifacts also distort router training signals: standard routers collapse to majority-class prediction (~79% smallest-tier optimal), confirmed via random-feature and shuffled-label controls, incurring a 13-17 percentage point opportunity cost. We provide actionable recommendations including dual-judge validation, exact-match anchoring, and cost-sensitive objectives. Our findings suggest existing routing headroom estimates are substantially inflated, underscoring the need for reliable evaluation protocols in multi-LLM systems.

78. BalCapRL: A Balanced Framework for RL-Based MLLM Image Captioning

Authors: Shaokai Ye , Vasileios Saveris , Yihao Qian , Jiaming Hu , Elmira Amirloo , Peter Grasch
URL: https://arxiv.org/abs/2605.07394
Abstract:

Image captioning is one of the most fundamental tasks in computer vision. Owing to its open-ended nature, it has received significant attention in the era of multimodal large language models (MLLMs). In pursuit of ever more detailed and accurate captions, recent work has increasingly turned to reinforcement learning (RL). However, existing captioning-RL methods and evaluation metrics often emphasize a narrow notion of caption quality, inducing trade-offs across core dimensions of captioning. For example, utility-oriented objectives can encourage noisy, hallucinated, or overlong captions that improve downstream question answering while harming fluency, whereas arena-style objectives can favor fluent but generic descriptions with limited usefulness. To address this, we propose a more balanced RL framework that jointly optimizes utility-aware correctness, reference coverage, and linguistic quality. In order to effectively optimize the resulting continuous multi-objective reward formulation, we apply GDPO-style reward-decoupled normalization to continuous-valued captioning rewards and show that it improves performance over vanilla GRPO. Additionally, we introduce length-conditional reward masking, yielding a more suitable length penalty for captioning. Across LLaVA-1.5-7B and Qwen2.5-VL 3B and 7B base models, our method consistently improves caption quality, with peak gains of +13.6 DCScore, +9.0 CaptionQA, and +29.0 CapArena across different models.

79. MISA: Mixture of Indexer Sparse Attention for Long-Context LLM Inference

Authors: Ruijie Zhou , Fanxu Meng , Yufei Xu , Tongxuan Liu , Guangming Lu , Muhan Zhang , Wenjie Pei
URL: https://arxiv.org/abs/2605.07363
Abstract:

DeepSeek Sparse Attention (DSA) sets the state of the art for fine-grained inference-time sparse attention by introducing a learned token-wise indexer that scores every prefix token and selects the most relevant ones for the main attention. To remain expressive, the indexer uses many query heads (for example, 64 on DeepSeek-V3.2) that share the same selected token set; this multi-head design is precisely what makes the indexer the dominant cost on long contexts. We propose MISA (Mixture of Indexer Sparse Attention), a drop-in replacement for the DSA indexer that treats its indexer heads as a pool of mixture-of-experts. A lightweight router uses cheap block-level statistics to pick a query-dependent subset of only a few active heads, and only those heads run the heavy token-level scoring. This preserves the diversity of the original indexer pool while reducing the per-query cost from scoring every prefix token with every head to scoring it with only a handful of routed heads, plus a negligible router term computed on a small set of pooled keys. We further introduce a hierarchical variant of MISA that uses the routed pass to keep an enlarged candidate set and then re-ranks it with the original DSA indexer to recover the final selected tokens almost exactly. With only eight active heads and no additional training, MISA matches the dense DSA indexer on LongBench across DeepSeek-V3.2 and GLM-5 while running with eight and four times fewer indexer heads respectively, and outperforms HISA on average. It also preserves fully green Needle-in-a-Haystack heatmaps up to a 128K-token context and recovers more than 92% of the tokens selected by the DSA indexer per layer. Our TileLang kernel delivers roughly a 3.82 times speedup over DSA’s original indexer kernel on a single NVIDIA H200 GPU.

80. TTF: Temporal Token Fusion for Efficient Video-Language Model

Authors: Simin Huo , Ning LI
URL: https://arxiv.org/abs/2605.07355
Abstract:

Video-language models (VLMs) face rapid inference costs as visual token counts scale with video length. For example, 32 frames at $448{\times}448$ resolution already yield >8,000 visual tokens in Qwen3-VL, making LLM prefill the dominant throughput bottleneck. Existing methods often rely on global similarity or attention-guided compression, incurring offsets to their gains. We propose \textbf{Temporal Token Fusion (TTF)}, a training-free, plug-and-play pre-LLM token compression framework that exploits structured temporal redundancy in video. TTF automatically selects an anchor frame, then for each subsequent frame, performs a local window similarity search (e.g.,$3\times 3$), fusing tokens that exceed a threshold. The compressed sequence maintains positional consistency across both prefill and decoding through coordinate realignment, enabling seamless integration with existing VLM pipelines. On Qwen3-VL-8B with threshold t=0.70, TTF removes about 67\% of visual tokens while retaining 99.5\% of the baseline accuracy and introducing only ${\approx}0.16$\,GFLOPs of matching overhead. Overall, TTF offers a practical, efficient solution for video understanding. The code is available at \href{ this https URL }{ this https URL }

81. Mage: Multi-Axis Evaluation of LLM-Generated Executable Game Scenes Beyond Compile-Pass Rate

Authors: Hugh Xuechen Liu , Kıvanç Tatar
URL: https://arxiv.org/abs/2605.07342
Abstract:

Compile-pass rate is the dominant evaluation signal for LLM code generation, yet for multi-component domain-specific artifacts it can be actively misleading. We demonstrate this on executable game scene synthesis with a four-axis evaluation protocol (named `Mage’) – compile success, runtime success, structural fidelity, and mechanism adherence – applied to 858 generation attempts across four open-weight LLMs (7B–30B), 26~hand-crafted Unity goal pattern playable concepts, and two automatically extracted IR granularity levels. Direct NL-to-C# generation achieves the highest runtime-pass rate (43\% mean) yet produces structurally vacuous scenes (mechanism $F_1 \approx 0.12$). Structural IR conditioning halves the runtime rate but recovers domain-faithful structure ($F_1$ up to 1.00). Within IR conditioning, behavior-only and full-scene granularity are statistically indistinguishable (McNemar $p = 1.0$), indicating input-level granularity saturation. These results show that compile rate is anti-correlated with functional correctness in this domain and that multi-axis evaluation is necessary to detect the divergence. We release the benchmark, replay logs, and per-record metrics for independent verification.

82. Rethinking Importance Sampling in LLM Policy Optimization: A Cumulative Token Perspective

Authors: Yuheng Zhang , Chenlu Ye , Shuowei Jin , Changlong Yu , Wei Xiong , Saurabh Sahu , Nan Jiang
URL: https://arxiv.org/abs/2605.07331
Abstract:

Reinforcement learning, including reinforcement learning with verifiable rewards (RLVR), has emerged as a powerful approach for LLM post-training. Central to these approaches is the design of the importance sampling (IS) ratio used in off-policy policy-gradient estimation. Existing methods face a fundamental bias-variance dilemma: token-level IS ratios, as adopted by PPO (Schulman et al., 2017) and GRPO (Shao et al., 2024), introduce bias by ignoring prefix state distribution mismatch; full sequence ratios provide exact trajectory-level correction but suffer from high variance due to the multiplicative accumulation of per-token ratios, while GSPO (Zheng et al., 2025) improves numerical stability via length normalization at the cost of deviating from the exact full-sequence IS correction. In this work, we identify the cumulative token IS ratio, the product of per-token ratios up to position $t$, as a theoretically principled solution to this dilemma. We prove that, under the token-level policy-gradient formulation, this ratio provides an unbiased prefix correction for each token-level gradient term and has strictly lower variance than the full sequence ratio. Building on this insight, we propose CTPO (Cumulative Token Policy Optimization), which combines the cumulative token IS ratio with position-adaptive clipping that scales log-space clip bounds according to the natural $\sqrt{t}$ growth of the cumulative log-ratio. This yields more consistent regularization across token positions. We implement and evaluate CTPO in the tool-integrated reasoning setting on several challenging mathematical reasoning benchmarks, achieving the best average performance across both model scales compared with strong GRPO and GSPO baselines. Code will be available at this https URL .

83. CSR: Infinite-Horizon Real-Time Policies with Massive Cached State Representations

Authors: Robin Karlsson , Go Suzui
URL: https://arxiv.org/abs/2605.07325
Abstract:

Deploying massive large language models (LLMs) as continuous cognitive engines for robotics is bottlenecked by the time-to-first-token (TTFT) latency required to process extensive state histories. Existing solutions like RAG or sliding windows compromise global context or incur prohibitive re-computation costs. We formalize the optimal task structure for minimizing latency and theoretically prove that prefix stability, incremental extensibility, and asynchronous state reconciliation are necessary conditions for real-time performance. Building on these proofs, we introduce the Cached State Representation (CSR) framework as the practical instantiation of these properties, ensuring optimal KV-cache reuse. To sustain these properties over infinite horizons, we further propose an Asynchronous State Reconciliation (ASR) algorithm that offloads state memory eviction to a parallel computational resource to eliminate latency spikes. On a physical robot wirelessly connected to an on-premise GPU server, CSR achieves a 26-fold latency reduction (14.67s to 0.56s) for 120K token contexts with a 235B parameter model compared to a standard baseline. On an embodied AI benchmark, we achieve SOTA recall (0.836 vs. 0.459) while maintaining RAG-level latency. ASR is validated to sustain bounded, spike-free TTFT over 10 eviction cycles in continuous real-world operation. Together, CSR and ASR enable massive LLMs to function as continuously operating, high-frequency (> 2 Hz) embodied policies.

84. Activation Differences Reveal Backdoors: A Comparison of SAE Architectures

Authors: Sachin Kumar
URL: https://arxiv.org/abs/2605.07324
Abstract:

Backdoor attacks on language models pose a significant threat to AI safety, where models behave normally on most inputs but exhibit harmful behavior when triggered by specific patterns. Detecting such backdoors through mechanistic interpretability remains an open challenge. We investigate two sparse autoencoder architectures – Crosscoders and Differential SAEs (Diff-SAE) – for isolating backdoor-related features in fine-tuned models. Using a controlled SQL injection backdoor triggered by year-based context (“2024” triggers vulnerable code, “2023” triggers safe code), we evaluate both approaches across LoRA and full-rank fine-tuning regimes on SmolLM2-360M. We find that Diff-SAE consistently and substantially outperforms Crosscoders for backdoor isolation. Diff-SAE achieves a Backdoor Isolation Score (BIS) of 0.40 with perfect precision (1.0) and zero false positive rate across most experimental conditions, while Crosscoders fail almost entirely with BIS below 0.02 in most cases. This performance gap holds across multiple transformer layers (14, 18, 22, 26) and both fine-tuning regimes, with full-rank fine-tuning producing particularly clean backdoor signals. Our results suggest that backdoors manifest as directional activation shifts rather than sparse feature activations, making difference-based representations fundamentally more effective for detection. These findings have important implications for AI safety monitoring and the development of interpretability tools for detecting model manipulation.

85. DCGL: Dual-Channel Graph Learning with Large Language Models for Knowledge-Aware Recommendation

Authors: Xinchi Zou , Tongzhenzhi Su , Jianjun Li , Yuan Fu , Chang Liu , Zhiying Deng , Zhiwei Shen
URL: https://arxiv.org/abs/2605.07314
Abstract:

Knowledge Graphs (KGs) have proven highly effective for recommendation systems by capturing latent item relationships, while recent integration of Large Language Models (LLMs) has further enhanced semantic understanding and addressed knowledge sparsity issues. Nevertheless, current KG-and-LLM-based methods still face three main limitations: 1) inadequate modeling of implicit semantic relationships beyond explicit KG links; 2) suboptimal single-channel fusion of ID and LLM embeddings, which often leads to signal interference and blurred representations; and 3) insufficient consideration of user-item interaction frequency variations in recommendation strategies. To address these challenges, we propose the Dual-Channel Graph Learning (DCGL) framework, featuring three key innovations: 1) a dual-channel architecture that structurally decouples rich semantic information from user behavioral patterns, preventing early interference; 2) a multi-level contrastive learning mechanism that enhances robustness against KG noise through intra-view contrasts and bridges semantic gaps between channels via inter-view alignment; and 3) a dynamic fusion mechanism that adaptively balances semantic generalization and behavioral specificity based on interaction frequency, resolving the cascading limitation. Extensive experiments on four real-world datasets show that DCGL consistently outperforms state-of-the-art methods, yielding substantial improvements in sparse scenarios while maintaining precision for active users. Our code is available at this https URL .

86. BioProVLA-Agent: An Affordable, Protocol-Driven, Vision-Enhanced VLA-Enabled Embodied Multi-Agent System with Closed-Loop-Capable Reasoning for Biological Laboratory Manipulation

Authors: Zhaohui Du , Zhe Wang , Hongmei Fei , Xiwen Cao , Ting Xiao , Qi Wang , Huanbo Jin , Jiaming Gu , Quan Lu , Zhe Liu
URL: https://arxiv.org/abs/2605.07306
Abstract:

Biological laboratory automation can reduce repetitive manual work and improve reproducibility, but reliable embodied execution in wet-lab environments remains challenging. Protocols are often unstructured, labware is frequently transparent or reflective, and multi-step procedures require state-aware execution beyond one-shot instruction following. Existing robotic systems often rely on costly hardware, fixed workflows, dedicated instruments, or robotics-oriented interfaces. Here, we introduce BioProVLA-Agent, an affordable, protocol-driven, vision-enhanced embodied multi-agent system enabled by Vision-Language-Action (VLA) models for biological manipulation. The system uses protocols as the task interface and integrates protocol parsing, visual state verification, and embodied execution in a closed-loop workflow. A Tailored LLM Protocol Agent converts protocols into verifiable subtasks; a VLM-RAG Verification Agent assesses readiness and completion using observations, robot states, retrieved knowledge, and success/failure examples; and a VLA Embodied Agent executes verified subtasks through a lightweight policy. To improve robustness under wet-lab visual perturbations, we develop AugSmolVLA, an online augmentation strategy targeting transparent labware, reflections, illumination shifts, and overexposure. We evaluate the system on a hierarchical benchmark covering 15 atomic tasks, 6 composite workflows, and 3 bimanual tasks, including tube loading, sorting, waste disposal, cap twisting, and liquid pouring. Across normal and high-exposure settings, AugSmolVLA improves execution stability over ACT, X-VLA, and the original SmolVLA, especially for precise placement, transparent-object manipulation, composite workflows, and visually degraded scenes. These results suggest a practical route toward accessible, protocol-centered, and verification-capable embodied AI for biological manipulation.

87. MedAction: Towards Active Multi-turn Clinical Diagnostic LLMs

Authors: Hsin-Ling Hsu , Zizheng Wang , Donghua Zhang , Nai-Chia Chen , Jerry Wang , Jun-En Ding , Chia-Hsuan Hsu , Guoan Wang , Feng Liu , Fang-Ming Hung , Chenwei Wu , Liyue Shen
URL: https://arxiv.org/abs/2605.07305
Abstract:

Most existing LLM diagnoses are evaluated on static, single-turn settings where complete patient information is provided upfront, an oversimplification of real clinical practice. We study active diagnosis: the real-life clinical process of starting from initial observation, ordering tests, interpreting results, and updating a differential diagnosis across multiple turns. Through systematic analysis, we identify three recurring failure modes in current LLMs: ungrounded test ordering, unreliable diagnostic update, and degraded multi-turn coherence. Together, these failures reveal a core deficit: existing medical training data teaches models to reason from complete information but not to act under evolving, partial evidence. To address this gap, we introduce MedAction, a tree-structured distillation pipeline that synthesizes diverse and high-quality multi-turn diagnostic trajectories via LLM-environment interaction. We propose two knowledge-graph-grounded metrics to filter trajectory quality: Disease Trajectory Consistency (DTC), which tracks whether the model’s hypothesis converges toward the correct diagnosis, and Reasoning-Action Consistency (RAC), which verifies that belief updates are driven by gathered evidence. Using this pipeline, we construct MedAction-32K, a dataset of 32,681 trajectories from 2,896 PMC cases. Fine-tuning an 8B model on MedAction-32K achieves state-of-the-art performance among open-source models on both MedR-Bench and our curated MedAction-300-Hard benchmark, pushing the edge for open-source medical LLMs.

88. EgoPro-Bench: Benchmarking Personalized Proactive Interaction in Egocentric Video Streams

Authors: Dongchuan Ran , Linyu Ou , Xueheng Li , Wenwen Tong , Chenxu Guo , Hewei Guo , Kaibing Wang , Lewei Lu
URL: https://arxiv.org/abs/2605.07299
Abstract:

Existing Multimodal Large Language Models (MLLMs) remain primarily reactive, failing to continuously perceive environments or proactively assist users. While emerging benchmarks address proactivity, they are largely confined to alert scenarios, neglect personalized context, and fail to evaluate the precise timing of human-machine interactions (HMI).In this paper, we introduce EgoPro-Bench, a novel benchmark for training and evaluating proactive interaction capabilities based on streaming egocentric videos; it comprises 2,400 videos in the evaluation set and over 12,000 videos in the training this http URL previous works, EgoPro-Bench leverages simulated user profiles to generate diverse user intentions and to construct high-fidelity HMI data across 12 distinct this http URL , we propose a specialized evaluation protocol and metrics, train proactive interaction models designed for efficient reasoning and low-latency interaction on streaming video data, and conduct comprehensive this http URL , we introduce an interaction principle termed “short thinking, better interaction”, which allocates a limited token budget prior to intent recognition, thereby enhancing interaction this http URL experiments demonstrate that EgoPro-Bench substantially enhances the intention understanding capabilities of MLLMs and enables accurate identification of appropriate timings for HMI, thereby laying a solid foundation for next-generation user-centric proactive interactive agents.

89. Understanding Performance Collapse in Layer-Pruned Large Language Models via Decision Representation Transitions

Authors: Boyu Shi , Chang Liu , ChuanBao Gao , Xu Yang , Xin Geng
URL: https://arxiv.org/abs/2605.07271
Abstract:

Layer pruning efficiently reduces Large Language Model (LLM) computational costs but often triggers sudden performance collapse. Existing representation-based analyses struggle to explain this mechanism. We propose studying pruning through decision representation. Focusing on multiple-choice tasks, we introduce two metrics, Decision Margin and Option Frequency, and an Iterative Pruning method to analyze layer-wise decision dynamics. Our findings reveal a sharp decision transition that partitions the network into two stages: a Silent Phase, where the model cannot yet predict the correct answer, and a Decisive Phase, where the correct prediction emerges. We also find that pruning the Decisive Phase has minimal impact, whereas pruning the Silent Phase triggers immediate performance collapse, highlighting its extreme sensitivity to structural changes. Therefore, we conclude that pruning-induced collapse stems from disrupting the Silent Phase, which prevents the critical decision transition from occurring.

Authors: Xiaoze Liu , Dhananjay Ram , Yuting Zhang , Zhaoyang Zhang , Wei Xia , Stefano Soatto
URL: https://arxiv.org/abs/2605.07244
Abstract:

We introduce Mutual Reinforcement Learning, a framework for concurrent RL post-training in which heterogeneous LLM policies exchange typed experience while keeping separate parameters, objectives, and tokenizers. The framework combines a Shared Experience Exchange (SEE), Multi-Worker Resource Allocation (MWRA), and a Tokenizer Heterogeneity Layer (THL) that retokenizes text and aligns token-level traces across incompatible vocabularies. This substrate makes the experience-sharing design question operational across model families. We instantiate three controlled probes on top of GRPO: data-level rollout sharing via Peer Rollout Pooling (PRP), value-level advantage sharing via Cross-Policy GRPO Advantage Sharing (XGRPO), and outcome-level success transfer via Success-Gated Transfer (SGT). A contextual-bandit analysis characterizes their structural positions on a stability-support trade-off: PRP pays density-ratio variance and THL residual costs, XGRPO preserves on-policy actor support while changing scalar baselines, and SGT supplies a rescue-set score direction toward verified peer successes. In the evaluated regime, outcome-level sharing occupies the favorable point of this trade-off.

91. Reformulating KV Cache Eviction Problem for Long-Context LLM Inference

Authors: Tho Mai , Joo-Young Kim
URL: https://arxiv.org/abs/2605.07234
Abstract:

Large language models (LLMs) support long-context inference but suffer from substantial memory and runtime overhead due to Key-Value (KV) Cache growth. Existing KV Cache eviction methods primarily rely on local attention weights, neglecting the influence of value representations, output projection, and inter-head interactions. In this work, we reformulate KV Cache eviction from a conventional head-wise, weight-averaging approach into an output-aware, layer-wise matrix multiplication approximation problem. We introduce LaProx, a novel eviction strategy that explicitly models the multiplicative interaction between attention maps and projected value states to accurately quantify token contributions while accounting for inter-head dependencies. Building on this metric, we propose the first unified eviction strategy that assigns globally comparable importance scores to tokens, enabling model-wide selection instead of local, head-wise decisions. Experimental results across 19 datasets on long-context benchmarks LongBench and Needle-In-A-Haystack demonstrate that our approach maintains model performance with only 5\% of the KV cache and consistently outperforms prior works across all configurations. Notably, our method achieves up to 2$\times$ accuracy loss reduction under extreme compression scenarios compared to existing state-of-the-art baselines with minimal overhead.

92. Hallucination Detection via Activations of Open-Weight Proxy Analyzers

Authors: Akshita Singh , Prabesh Paudel , Siddhartha Roy
URL: https://arxiv.org/abs/2605.07209
Abstract:

We introduce a proxy-analyzer framework for detecting hallucinations in large language models. Instead of looking inside the generating model, our system reads already-generated text through a small locally hosted open-weight model and spots hallucinations using the reader’s own internal activations. This works just as well when the generator is a closed API like GPT-4 as when it is any open-weight model. We built eighteen features grounded in how transformers process text, covering residual stream norms, per-head source-document attention, entropy, MLP activations, logit-lens trajectories, and three new token-level grounding statistics. We trained a stacking ensemble on 72,135 samples from five hallucination datasets. We tested across seven analyzer architectures from 0.5 billion to 9 billion parameters: Qwen2.5 at 0.5B and 7B, Gemma-2 at 2B and 9B, Pythia at 1.4B, and LLaMA-3 at both 3B and 8B. Across all seven, we consistently beat ReDeEP’s token-level AUC of 0.73 on RAGTruth by 7.4 to 10.3 percentage points. Qwen2.5-7B reached an F1 of 0.717, just above ReDeEP’s 0.713, while Qwen2.5-0.5B hit 0.706. The most striking finding is how tightly all seven models cluster: AUC spans only 2.3 percentage points across an eighteen-fold difference in model size. Even more surprising, our 3B LLaMA outperforms our 8B LLaMA on RAGTruth, showing that bigger is not always better even within the same model family. Both RAGTruth and LLM-AggreFact include outputs from multiple LLM families, so our results are not skewed toward any particular generator.

93. PSK@EEUCA 2026: Fine-Tuning Large Language Models with Synthetic Data Augmentation for Multi-Class Toxicity Detection in Gaming Chat

Authors: Srikar Kashyap Pulipaka
URL: https://arxiv.org/abs/2605.07201
Abstract:

This paper describes our system for the EEUCA 2026 Shared Task on Understanding Toxic Behavior in Gaming Communities. The task involves classifying World of Tanks chat messages into six toxicity categories: Non-toxic, Insults/Flaming, Other Offensive, Hate/Harassment, Threats, and Extremism. We explore multiple approaches including encoder-based models, instruction-tuned LLMs with LoRA fine-tuning, hierarchical classification, one-vs-rest strategies, and various ensemble methods. Our best system combines Llama 3.1 8B with carefully calibrated 5\% synthetic data augmentation, achieving an F1-macro score of 0.6234 on the test set, placing 4th out of 35 participating teams. We provide extensive analysis of the dataset’s annotation patterns and their impact on model generalization, revealing a critical ‘‘validation trap’’ phenomenon where high validation performance correlates with poor test transfer.

94. The Text Uncanny Valley: Non-Monotonic Performance Degradation in LLM Information Retrieval

Authors: Zekai Tong , Ruiyao Xu , Aryan Shrivastava , Chenhao Tan , Ari Holtzman
URL: https://arxiv.org/abs/2605.07186
Abstract:

Existing Large Language Model (LLM) benchmarks primarily focus on syntactically correct inputs, leaving a significant gap in evaluation on imperfect text. In this work, we study how word-boundary corruption affects how LLMs detect targeted information. By inserting whitespace characters within words to break them into fragments, LLMs’ detection accuracy follows a U-shaped curve with the increase in insertion rate. We refer to this curve as the Text Uncanny Valley. To explain such observation, we propose a mode transition hypothesis: LLMs operate in a word-level mode for near-normal text and a character-level mode for heavily fragmented text, with the valley marking the disordered transition where neither mode is effective. Four experiments and one analysis are consistent with this account: in-context learning fails to rescue valley-bottom performance; regularizing the perturbation substantially reduces the U-shape; a math reasoning task replicates the U-shape for Gemini 3.0 Flash but not for stronger models, suggesting the effect is attenuated when tasks rely less on exact lexical alignment; and tokenization entropy peaks before the F1 minimum, consistent with a regime-conflict interpretation. These findings reveal a failure mode invisible to clean-text benchmarks yet directly relevant to any deployment scenario involving noisy or uncurated text inputs.

95. MathlibPR: Pull Request Merge-Readiness Benchmark for Formal Mathematical Libraries

Authors: Zixuan Xie , Xinyu Liu , Shangtong Zhang
URL: https://arxiv.org/abs/2605.07147
Abstract:

The ecosystem of Lean and Mathlib has become the de facto standard for large language model (LLM) assisted formal reasoning with remarkable successes in recent years. Those successes, however, only consume Mathlib as an essential dependency but do not directly contribute to it. In the meantime, the growth of Mathlib has recently been bottlenecked by the review process, which requires human reviewers to judge whether proposed pull requests (PRs) follow the Mathlib’s conventions and are worth integrating as part of a shared mathematical infrastructure. This leads to our central question: can LLMs help review Mathlib PRs? To this end, we introduce MathlibPR, a benchmark built from real Mathlib4 PR histories. We further propose a staged evaluation protocol and use it to evaluate both LLM models (e.g., DeepSeek, Qwen, Goedel, and Kimina) and LLM agents (e.g., Codex and Claude Code). Surprisingly, both LLM models and LLM agents struggle to distinguish merge-ready PRs from build-passing PRs that were revised or never merged. By turning Mathlib PR histories into a supervised signal, MathlibPR provides a step toward reviewer assistants and reward models that could help evaluate PRs and steer LLMs toward producing merge-ready Mathlib contributions.

96. Qwen3-VL-Seg: Unlocking Open-World Referring Segmentation with Vision-Language Grounding

Authors: Yuan Yao , Qiushi Yang , Humen Zhong , Jiangning Wei , Yifang Men , Shuai Bai , Miaomiao Cui , Zhibo Yang
URL: https://arxiv.org/abs/2605.07141
Abstract:

Open-world referring segmentation requires grounding unconstrained language expressions to precise pixel-level regions. Existing multimodal large language models (MLLMs) exhibit strong open-world visual grounding, but their outputs remain limited to sparse bounding-box coordinates and are insufficient for dense visual prediction. Recent MLLM-based segmentation methods either directly predict sparse contour coordinates, struggling to reconstruct continuous object boundaries, or rely on external segmentation foundation models such as the Segment Anything Model (SAM), introducing substantial architectural and deployment overhead. We present Qwen3-VL-Seg, a parameter-efficient framework that treats the MLLM-predicted box as a semantically grounded structural prior and decodes it into pixel-level referring segmentation. At its core, a lightweight box-guided mask decoder combines multi-scale spatial feature injection, spatial-semantic query construction, box-guided high-resolution pixel fusion, and iterative mask-aware query refinement, introducing only 17M parameters (about 0.4\% of the base model). For scalable open-world training, we construct SA1B-ORS, an SA-1B-derived dataset with two subsets: SA1B-CoRS (category-oriented samples) and SA1B-DeRS (descriptive, instance-specific samples). For evaluation, we curate ORS-Bench, a manually screened benchmark with in-distribution and out-of-distribution subsets covering diverse referring expression types. Extensive experiments on referring expression segmentation, visual grounding, and ORS-Bench show that Qwen3-VL-Seg performs strongly across closed-set and open-world settings, with clear advantages on language-intensive instructions and strong out-of-distribution generalization. Evaluations on general multimodal benchmarks further show that the model broadly preserves general-purpose multimodal competence after segmentation-oriented adaptation.

97. Neurosymbolic Framework for Concept-Driven Logical Reasoning in Skeleton-Based Human Action Recognition

Authors: Talha Ilyas , Deval Mehta , Zongyuan Ge
URL: https://arxiv.org/abs/2605.07140
Abstract:

Skeleton-based human activity recognition has achieved strong empirical performance, yet most existing models remain black boxes and difficult to interpret. In this work, we introduce a neurosymbolic formulation of skeleton-based HAR that reframes action recognition as concept-driven first-order logical reasoning over motion primitives. Our framework bridges representation learning and symbolic inference by grounding first-order logic predicates in learnable spatial and temporal motion concepts. Specifically, we employ a standard spatio-temporal skeleton encoder to extract latent motion representations, which are then mapped to interpretable concept predicates via a spatio-temporal concept decoder that explicitly separates pose-centric and dynamics-centric abstractions. These concept predicates are composed through differentiable first-order logic layers, enabling the model to learn human-readable logical rules that govern action semantics. To impose semantic structure on the learned concepts, we align skeleton representations with LLM-derived descriptions of atomic motion primitives, establishing a shared conceptual space for perception and reasoning. Extensive experiments on NTU RGB+D 60/120 and NW-UCLA demonstrate that our approach achieves competitive recognition performance while providing explicit, interpretable explanations grounded in logical structure. Our results highlight neurosymbolic reasoning as an effective paradigm for interpretable spatio-temporal action understanding. Code: this https URL

98. Structural Rationale Distillation via Reasoning Space Compression

Authors: Jialin Yang , Jiankun Wang , Jiajun Wu , Henry Leung , Jiayu Zhou , Steve Drew
URL: https://arxiv.org/abs/2605.07139
Abstract:

When distilling reasoning from large language models (LLMs) into smaller ones, teacher rationales for similar problems often vary wildly in structure and strategy. Like a chef who makes the same dish differently each time, this inconsistency burdens the student with noisy supervision that is hard to internalize. We propose Distillation through Reasoning Path Compression (D-RPC), which constrains the teacher to follow a compact, dynamically maintained bank of reusable high-level reasoning paths. For each training question, D-RPC retrieves the most relevant path and conditions the teacher to follow it, producing rationales that are consistent across similar problems yet diverse enough to cover different problem types. A PAC-Bayes analysis formalizes the resulting trade-off between bank size and coverage: smaller banks reduce supervision entropy but risk coverage gaps, and the generalization bound identifies an optimal intermediate size confirmed by our ablations. Across five math and commonsense reasoning benchmarks with two student models, D-RPC consistently outperforms chain-of-thought distillation, freeform rationale generation, direct distillation, and structured-supervision baselines, while using fewer tokens than template-heavy alternatives.

99. Adaptive Negative Reinforcement for LLM Reasoning:Dynamically Balancing Correction and Diversity in RLVR

Authors: Yash Ingle , Jaival Chauhan , Ankit Yadav , Sudhakar Mishra
URL: https://arxiv.org/abs/2605.07137
Abstract:

Reinforcement learning with verifiable rewards (RLVR) has become a highly effective method for improving the reasoning abilities of Large Language Models (LLMs). Recent research shows that Negative Sample Reinforcement (NSR) – which focuses on penalizing incorrect steps rather than simply rewarding correct ones – can match or even exceed the performance of more complex frameworks like PPO and GRPO across the entire Pass@k spectrum. However, current NSR techniques usually apply a fixed penalty throughout the training process and treat every incorrect response with the same weight. To address these limitations, we propose two extensions to the NSR framework: Adaptive Negative Sample Reinforcement. Rather than using a fixed update rule, A-NSR uses time-dependent scheduling functions. In the initial training phases, the system focuses heavily on correcting errors to stabilize the model. As training continues, it shifts toward more subtle and controlled updates. We also introduce Confidence-Weighted Negative Reinforcement, which operates on the principle that different mistakes carry different levels of importance. CW-NSR assigns specific penalty weights based on the model’s normalized sequence likelihood. If the model is highly confident in a wrong path, it receives a larger penalty and for uncertain errors – where the model is effectively exploring – are penalized less strictly. Our formal analysis shows how these mechanisms govern token-level updates, allowing the model to leverage prior-guided probability redistribution while providing a natural defense against overfitting. We evaluated these methods on difficult reasoning datasets, including MATH, AIME 2025, and AMC23, using the Qwen2.5-Math-1.5B architecture.

100. Region4Web: Rethinking Observation Space Granularity for Web Agents

Authors: Donguk Kwon , Dongha Lee
URL: https://arxiv.org/abs/2605.07134
Abstract:

Web agents perceive web pages through an observation space, yet its granularity has remained an underexamined design choice. Existing work treats observation at the same element-level granularity as the action space, leaving the page’s functional organization implicit and forcing the agent to infer it from element-level signals at every step. We argue observation should instead operate at the granularity of functional regions, parts of the page that each serve a distinct purpose. We propose Region4Web, a framework that reorganizes the AXTree into functional regions through hierarchical decomposition and semantic abstraction, exposing the page’s functional organization as the basis for page state understanding. Moreover, we propose PageDigest, a web-specific inference pipeline that delivers this region-level observation to the actor agent as a compact per-page digest that persists across steps. On the WebArena benchmark, PageDigest substantially reduces observation length while improving overall task success rate across diverse backbone large language models (LLMs) and established agent methods, regardless of backbone capacity. These results show that operating at the granularity of functional regions delivers a more compact and informative basis for the actor agent than element-level processing alone.

101. RRCM: Ranking-Driven Retrieval over Collaborative and Meta Memories for LLM Recommendation

Authors: Shijun Li , Wooseong Yang , Yu Wang , Tianxin Wei , Joydeep Ghosh
URL: https://arxiv.org/abs/2605.07129
Abstract:

Large Language Models (LLMs) have emerged as a promising paradigm for next-generation recommender systems, offering strong semantic understanding and natural-language reasoning abilities. Despite recent progress, current LLM-based recommenders still face key challenges in constructing decision-relevant contexts from heterogeneous evidence. First, existing methods often rely on fixed context construction strategies: collaborative behavioral evidence and item-side metadata are typically incorporated through predefined prompts, static retrieval pipelines, or handcrafted injection mechanisms, making it difficult to determine what information is truly beneficial for each instance. Second, heterogeneous evidence introduces a severe context-efficiency bottleneck. Rich metadata and collaborative interaction records can quickly overwhelm the context window, while aggressive compression or heuristic filtering may discard fine-grained evidence critical for accurate recommendation. To address these challenges, we propose RRCM, a ranking-driven retrieval-and-reasoning framework over collaborative and metadata memories for LLM-based agentic recommendation. RRCM starts from a lightweight user-history context and learns whether to recommend directly, retrieve collaborative evidence, retrieve item metadata, or interleave both through reasoning. Both memories are represented in natural language and accessed through a unified retrieval interface, enabling flexible evidence acquisition without handcrafted CF injection or fixed retrieval rules. We optimize this memory-reading policy with an outcome-only ranking reward, instantiated using group relative policy optimization, so that retrieval decisions are directly driven by final top-k recommendation quality. Extensive experiments show that RRCM significantly outperforms traditional baselines and diverse LLM-based recommendation approaches.

102. Beyond LoRA vs. Full Fine-Tuning: Gradient-Guided Optimizer Routing for LLM Adaptation

Authors: Haozhan Tang , Xiuqi Zhu , Xinyin Zhang , Boxun Li , Virginia Smith , Kevin Kuo
URL: https://arxiv.org/abs/2605.07111
Abstract:

Recent literature on fine-tuning Large Language Models highlights a fundamental debate. While Full Fine-Tuning (FFT) provides the representational plasticity required for high-entropy knowledge injection, Low-Rank Adaptation (LoRA) can match or surpass FFT performance because many tasks only require updates in a low-rank space and benefit from LoRA’s additional regularization. Through empirical evaluation across diverse tasks (SQL, Medical QA, and Counterfactual Knowledge) and varying language models (Gemma-3-1B, Qwen2.5-1.5B, and Qwen2.5-3B), we verify both trends and demonstrate that relying solely on either static architecture is structurally limited. To address this challenge, we propose a Mixture of LoRA and Full (MoLF) Fine-Tuning, a unified framework that enables continuous navigation between both training regimes. MoLF dynamically routes updates between FFT and LoRA at the optimizer level to ensure that exact gradient signals are available to both experts throughout training, yielding stable training dynamics. For memory-constrained environments, we also introduce MoLF-Efficient, which freezes base weights and only routes updates among a pair of LoRA experts of potentially varying rank. Our evaluations show that MoLF either improves on or stays within $1.5\%$ of the better of FFT and LoRA across all settings, while MoLF-Efficient outperforms prior adaptive LoRA approaches by up to $20\%$ on Fact and $9\%$ on Med and SQL.

103. The Translation Tax Is Not a Scalar: A Counterfactual Audit of English-Source Cue Inheritance in Chinese Multilingual Benchmarks

Authors: Zezheng Lin , Fengming Liu , Handi Li
URL: https://arxiv.org/abs/2605.07093
Abstract:

The Translation Tax is often treated as a scalar: translated benchmarks are assumed to inflate scores by preserving English-source cues. We audit this claim in an English-to-Chinese setting. Three proxy estimators disagree: back-translation gaps are small and parser-fragile; cue-score calibration does not predict item-level gains; and a six-model native-control comparison shows model-family rather than uniform benchmark effects. We add a same-item LLM-naturalization stress test that holds answer, options, and content fixed while rewriting Chinese surface form. After correcting a prompt-construction bug, this contrast no longer supports a model-family interaction, but it preserves a residue dose-response: high-residue items benefit while low-residue items do not. The result is not a single Translation Tax, but a set of estimator- and item-dependent validity risks. We release per-cell evidence, the naturalization protocol, human QC, and a reporting checklist for translated multilingual benchmark papers.

104. WiCER: Wiki-memory Compile, Evaluate, Refine Iterative Knowledge Compilation for LLM Wiki Systems

Authors: Juan M. Huerta
URL: https://arxiv.org/abs/2605.07068
Abstract:

The LLM Wiki pattern, to compile and provide domain knowledge into a persistent artifact and serve it to LLMs via KV cache inference, promises context access at sub-second latency with zero retrieval failure. Realizing this requires solving the compilation gap: LLM compilation distilling raw documents into a wiki without catastrophically discarding critical facts. We characterize this gap across 17 RepLiQA domains (6,800 questions): we observe that full context KV cache inference outperforms RAG on curated knowledge (4.38 vs. 4.08 out of 5, 7.3 faster TTFT) but degrades below RAG at scale due to attention dilution, and blind compilation fails entirely (2.14 to 2.32 vs. 3.46, 53 to 60% catastrophic failure rate). To address the compilation gap, we propose WiCER (Wiki-memory Compile, Evaluate, Refine), an iterative algorithm inspired by counterexample-guided abstraction refinement (CEGAR) that closes this gap. WiCER evaluates compiled wikis against diagnostic probes, identifies dropped facts, and forces their preservation in subsequent compilations. One to two iterations recover 80% of lost quality (mean 3.24 vs. 3.47 for raw full-context across the 15 topics with baselines), reducing catastrophic failures by 55% relative. An ablation across all 17 topics confirms that targeted diagnosis (+0.95), not generic pinning (+0.16), drives the gains. All code and benchmarks are released for reproducible research.

105. Dr. Post-Training: A Data Regularization Perspective on LLM Post-Training

Authors: Pingbang Hu , Xueshen Liu , Z. Morley Mao , Jiaqi W. Ma
URL: https://arxiv.org/abs/2605.07063
Abstract:

Data selection methods address a critical challenge in LLM post-training: effectively leveraging scarce, high-fidelity target data alongside abundant but imperfectly aligned general training data. In this work, we move beyond the data-selection framing and introduce Dr. Post-Training (Data-Regularized Post-Training), a novel framework that reconceptualizes general training data as a data-induced regularizer that prevents overfitting to the scarce target objective, rather than serving as a pool for selection. Specifically, our framework proposes that at each training step, construct a feasible set of model update directions using the general training data, and project the model update direction specified by the scarce target data onto that feasible set. Standard training and existing data selection methods arise as special cases with different choices of the data-induced regularizer, and these methods correspond to different points on a bias–variance spectrum with different regularization strength. Building on this view, we propose a family of methods offering a richer design space and more flexible bias–variance tradeoffs. For practical LLM-scale use, we introduce careful system optimizations that realize these methods with minimal overhead. Extensive experiments across SFT, RLHF, and RLVR show that our methods consistently outperform state-of-the-art data selection baselines, and system benchmarks confirm their efficiency.

106. Do Joint Audio-Video Generation Models Understand Physics?

Authors: Zijun Cui , Xiulong Liu , Hao Fang , Mingwei Xu , Jiageng Liu , Zexin Xu , Weiguo Pian , Shijian Deng , Feiyu Du , Chenming Ge , Yapeng Tian
URL: https://arxiv.org/abs/2605.07061
Abstract:

Joint audio-video generation models are rapidly approaching professional production quality, raising a central question: do they understand audio-visual physics, or merely generate plausible sounds and frames that violate real-world consistency? We introduce AV-Phys Bench, a benchmark for evaluating physical commonsense in joint audio-video generation. AV-Phys Bench tests models across three scene categories: Steady State, Event Transition, and Environment Transition. It covers physics-grounded subcategories drawn from real-world scenes, plus Anti-AV-Physics prompts that deliberately request physically inconsistent audio-video behavior. Each generation is evaluated along five dimensions: visual semantic adherence, audio semantic adherence, visual physical commonsense, audio physical commonsense, and cross-modal physical commonsense. Across three proprietary and four open-source models, we find that Seedance 2.0 performs best overall, but all models remain far from robust physical understanding. Performance drops sharply on event-driven and environment-driven transitions, and even strong proprietary systems collapse on Anti-AV-Physics prompts. We further introduce AV-Phys Agent, a ReAct-style evaluator that combines a multimodal language model with deterministic acoustic measurement tools, producing rankings that closely align with human ratings. Our results identify cross-modal physical consistency and transition-driven scene dynamics as key open challenges for joint audio-video generation.

107. MedExAgent: Training LLM Agents to Ask, Examine, and Diagnose in Noisy Clinical Environments

Authors: Yicheng Gao , Xiaolin Zhou , Yahan Li , Yue Zhao , Ruishan Liu
URL: https://arxiv.org/abs/2605.07058
Abstract:

Real-world clinical diagnosis is a complex process in which the doctor is required to obtain information from both interaction with the patient and conducting medical exams. Additionally, the doctor needs to adapt to different patient personas, as well as noisy and incomplete information that can happen at any time during the process. However, existing benchmarks for medical LLMs and methods for automatic diagnosis largely simplify this process by reducing it to single-turn question answering, noise-free conversations, or sequential exam making, etc., ignoring the interactive and uncertain nature of clinical diagnosis. In this paper, we aim to address this gap by formalizing clinical diagnosis as a Partially Observable Markov Decision Process (POMDP) with three action types: questioning the patient, ordering medical exams as tool calls, and issuing a diagnosis. We also introduce a systematic noise model comprising seven patient noise types and three exam noise types. Using our proposed environment, we train an effective diagnosis agent, \textbf{MedExAgent}, through a two-stage pipeline that first performs supervised finetuning on synthetic conversations structured after the Calgary-Cambridge model for clinical interviews, and then applies DAPO to optimize a composite reward capturing diagnostic accuracy, tool call quality, and exam cost including financial cost and patient discomfort. Through extensive experiments and ablation studies, we demonstrate that MedExAgent achieves diagnostic performance comparable to larger models while maintaining cost-efficient examination strategies.

108. An Interpretable and Scalable Framework for Evaluating Large Language Models

Authors: Xinhao Qu , Qiang Heng , Hao Zeng , Xiaoqian Liu
URL: https://arxiv.org/abs/2605.07046
Abstract:

Evaluation of large language models (LLMs) is increasingly critical, yet standard benchmarking methods rely on average accuracy, overlooking both the inherent stochasticity of LLM outputs and the heterogeneity of benchmark items. Item Response Theory (IRT) offers a principled framework for modeling latent model abilities and item characteristics, but conventional methods are computationally expensive and numerically unstable, limiting large-scale implementations. To address these challenges, we propose an interpretable and scalable framework for LLM evaluation based on the majorization-minimization principle. Our approach reformulates the problem as a sequence of constrained matrix factorization subproblems, enabling stable and efficient parameter estimation with theoretical guarantees for identifiability and convergence. Experiments on synthetic and real-world datasets, including MATH-500 and six Open LLM Leaderboard benchmarks, demonstrate that our method achieves superior scalability and interpretability. It delivers orders-of-magnitude speedups over competing methods while maintaining comparable or even higher estimation accuracy. Our results align with established scaling laws and offer insights into item difficulty and discrimination, informing more principled benchmark design.

109. Cognitive Agent Compilation for Explicit Problem Solver Modeling

Authors: Hyeongdon Moon , Carolyn Rosé , John Stamper
URL: https://arxiv.org/abs/2605.07040
Abstract:

Large language models (LLMs) are widely used for tutoring, feedback generation, and content creation, but their broad pretraining makes them hard to constrain and poor substitutes for controllable learners. Educational systems often require inspectable and editable knowledge states: educators want to know what a system assumes the learner knows, and learners benefit when the system can justify actions in terms of explicit skills, misconceptions, and strategies. Inspired by cognitive architectures, we propose Cognitive Agent Compilation (CAC), a framework that uses a strong teacher LLM to compile problem-solving knowledge into an explicit target agent. CAC separates (i) knowledge representation, (ii) problem-solving policy, and (iii) verification and update rules, with the goal of making bounded problem solving more inspectable and editable in educational settings. We present an early proof of concept implemented with Small Language Models that surfaces key design trade-offs, particularly between explicit control and scalable generalization, and positions CAC as an initial step toward bounded-knowledge AI for educational applications.

110. LensVLM: Selective Context Expansion for Compressed Visual Representation of Text

Authors: Roy Xie , Dan Friedman , Donghan Yu , Bowen Pan , Christopher Fifty , Jang-Hyun Kim , Xianzhi Du , Zhe Gan , Vivek Rathod , Bhuwan Dhingra
URL: https://arxiv.org/abs/2605.07019
Abstract:

Vision Language Models (VLMs) offer the exciting possibility of processing text as rendered images, bypassing the need for tokenizing the text into long token sequences. Since VLM image encoders map fixed-size images to a fixed number of visual tokens, varying rendering resolution provides a fine-grained compression knob. However, accuracy deteriorates quickly as compression increases: characters shrink below the vision encoder’s effective resolution, making them indistinguishable. To address this, we propose LensVLM, an inference framework and post-training recipe that enables VLMs to scan compressed images, then selectively expand only the relevant images to their uncompressed form via learned tools. Building on Qwen3.5-9B-Base, LensVLM maintains accuracy comparable to the full-text upper bound at 4.3x effective compression and outperforms retrieval-based, text- and visual-compression baselines up to 10.1x effective compression across seven text QA benchmarks. LensVLM also generalizes to multimodal document and code understanding tasks, with the accuracy gain over baselines growing as compression increases. Our analysis validates this approach: training makes visual compression robust to rendering choices, and as compression grows the model increasingly relies on expanded content rather than unreliable visual reading. The analysis also yields practical tool-choice guidance: text expansion is preferable for rendered text, while high-resolution image expansion suits native documents whose layout cues carry task-relevant information.

111. $f$-Divergence Regularized RLHF: Two Tales of Sampling and Unified Analyses

Authors: Di Wu , Chengshuai Shi , Jing Yang , Cong Shen
URL: https://arxiv.org/abs/2605.06977
Abstract:

Reinforcement Learning from Human Feedback (RLHF) has become a cornerstone technique for post-training large language models. While most existing approaches rely on the reverse KL-regularization, recent empirical studies have begun exploring alternative divergences (e.g., forward KL, chi-squared) as regularizers in RLHF. However, a unified theoretical understanding of general $f$-divergence regularization remains under-explored. To fill this gap, this work develops a comprehensive theoretical framework for online RLHF with a general $f$-divergence regularized objective. Rather than treating each possible divergence function individually, we adopt a holistic perspective across the entire function class and propose two algorithms based on distinct sampling principles. The first extends the classical optimism principle with a carefully designed exploration bonus, while the second introduces a new method that exploits the sensitivity of the optimal policy to reward perturbations under $f$-divergence regularization. Theoretical analysis shows that $O(\log T)$ regret and $O(1/T)$ sub-optimality gap are achievable, establishing provable efficiency of both algorithms and, to the best of our knowledge, the first performance bounds for online RLHF under general $f$-divergence regularization.

112. From Surface Learning to Deep Understanding: A Grounded AI Tutoring System for Moodle

Authors: Anna Ostrowska , Michał Kukla , Gabriela Majstrak , Jan Opala , Sebastian Pergała , Jan Skwarek , Anna Wróblewska
URL: https://arxiv.org/abs/2605.06963
Abstract:

This demo paper describes the development of the AI Teaching \& Learning Assistant, a modular Moodle plugin that leverages Retrieval-Augmented Generation (RAG) to deliver high-quality, hallucination-free education. The system employs a dual-centric design, providing students with interactive, Socratic-based tutoring and educators with a “human-in-the-loop” workspace for supervised content generation. By grounding Large Language Model (LLM) responses in teacher-provided materials, the assistant addresses the risks of misinformation while encouraging deep conceptual mastery. Evaluation via the Ragas (LLM-as-a-Judge) framework and a preliminary user study confirms its effectiveness, achieving faithfulness scores up to 0.97 and a 4.00/5.00 recommendation rate.

113. Bridging the Last Mile of Circuit Design: PostEDA-Bench, a Hierarchical Benchmark for PPA Convergence and DRC Fixing

Authors: Pengju Liu , Nuo Xu , Jinwei Tang , Yu Cao , Caiwen Ding
URL: https://arxiv.org/abs/2605.06936
Abstract:

LLM-based agents are increasingly applied to the “last mile” of Electronic Design Automation (EDA): repairing residual sign-off Design Rule Check (DRC) violations and converging Power-Performance-Area (PPA) targets after tool runs. Existing EDA-LLM benchmarks, however, omit DRC fixing entirely and rely on flat hierarchies tied to a single toolchain. We introduce PostEDA-Bench, a hierarchical benchmark with 145 tasks across DRC-Essential, DRC-Reasoning, PPA-Mono, and PPA-Multi, supported by EDA toolchains with machine-checkable evaluation. Across eight commercial and open-source LLMs under multiple agent scaffolds, we find that agents handle synthetic DRC-Essential and single-objective PPA-Mono reasonably well but degrade sharply on the more practical DRC-Reasoning, where the best success rate is 36.66%, and PPA-Multi, where the best success rate is 20.00%; vision augmentation consistently enhances DRC-Bench; and trade-off reasoning, rather than knob knowledge, is the dominant PPA-Multi bottleneck.

114. In-Context Credit Assignment via the Core

Authors: Keegan Harris , Siddharth Prasad , Asher Trockman
URL: https://arxiv.org/abs/2605.06920
Abstract:

We propose incentive-aligned mechanisms for in-context credit assignment: the task of assigning credit for AI-generated content (e.g. code, news articles, short-form videos) among creators whose intellectual property appears in the context window. Our approach is based on the least core solution concept from cooperative game theory, which distributes value in a way that is as stable as possible by ensuring that no subset of creators is significantly under-compensated relative to the value they could generate on their own. We develop algorithms for approximating the least core, which leverage novel routines for constraint seeding and constraint separation. On a web retrieval credit assignment task, we find that our approaches are capable of approximating the least core using orders of magnitude fewer LLM calls compared to alternative methods.

115. Regulating Branch Parallelism in LLM Serving

Authors: Swapnil Gandhi , Siva Hari , William J. Dally , Christos Kozyrakis
URL: https://arxiv.org/abs/2605.06914
Abstract:

Recent methods expose intra-request parallelism in LLM outputs, allowing independent branches to decode concurrently. Existing serving systems execute these branches eagerly or under fixed caps. We show that both are brittle: eager admission inflates the shared decode step, degrading co-batched requests in serial stages, while conservative fixed caps forgo the throughput that motivated exposing branches in the first place. We call the excess step latency caused by admitted branches the branch externality and show that the safe width depends on batch composition, context lengths, and accumulated slack, all of which change continuously over a workload trace. We introduce TAPER, a per-step admission controller that treats extra branches as opportunistic work, admitted only when the predicted branch externality fits within the batch’s current slack budget. Per-step regulation is practical because branch-level scheduling decouples compute from memory: branches share the request’s prefix KV, so expanding or contracting width requires no memory reclamation. On Qwen3-32B, TAPER improves goodput by $1.77\times$ over IRP-Off and by $1.48\times$ over IRP-Eager, while maintaining over $95\%$ SLO attainment.

116. Same Signal, Opposite Meaning: Direction-Informed Adaptive Learning for LLM Agents

Authors: Ziming Li , Jiatan Huang , Xiaoguang Guo , Guilin Wang , Chuxu Zhang
URL: https://arxiv.org/abs/2605.06908
Abstract:

Adaptive test-time compute for LLM agents aims to invoke extra computation only when it improves performance. Existing methods typically use confidence-, uncertainty-, or difficulty-based gates, assuming a fixed direction from the gating signal through compute need to the value of computation. This makes gating a utility-calibration problem: gating signals should align with whether extra computation improves the final outcome over the base policy. We show that this alignment is unstable: the same signal predicts rollout benefit in one setting and rollout harm in another, with reversals across environments and backbones even when the task is fixed. Wrong-direction gates can therefore worsen performance by precisely selecting harmful states. This reversal reflects a deeper distinction between compute need and compute suitability: a high uncertainty signal may indicate decision-difficult states where rollouts help compare alternatives, or intervention-unsuitable states where the current context does not support useful rollout-based improvement. Under this two-source model, fixed-direction gates are unreliable across heterogeneous settings. To address this, we propose DIAL (Direction-Informed Adaptive Learning), a sparse gate trained from signal-agnostic counterfactual exploration to learn the utility direction of state features per (environment, backbone). Across six environments and three backbones, DIAL yields a stronger overall success-cost trade-off than fixed-direction baselines.

117. MELD: Multi-Task Equilibrated Learning Detector for AI-Generated Text

Authors: Chenjun Li , Cheng Wan , Johannes C. Paetzold
URL: https://arxiv.org/abs/2605.06903
Abstract:

Large language models are now embedded in everyday writing workflows, making reliable AI-generated text detection important for academic integrity, content moderation, and provenance tracking. In practice, however, a detector must do more than achieve high aggregate AUROC on clean, in-distribution human and AI text: it should remain robust to attacks and adversarial rewrites, transfer to unseen generators and domains, and operate at low false-positive rates (FPR). Most existing detectors optimize a single AI/Human objective, giving the representation little incentive to learn generator, attack, or domain structure once the binary task saturates. We introduce MELD (Multi-Task Equilibrated Learning Detector), a deployable detector for AI-generated text that enriches binary detection with auxiliary supervision. MELD attaches generator-family, attack-type, and source-domain heads to a shared encoder, and balances the four losses with learned homoscedastic uncertainty weights. To improve robustness, an EMA teacher predicts on clean inputs while an attack-augmented student is distilled toward the teacher. MELD further uses a hard-negative pairwise ranking loss to enlarge the score margin between AI-generated texts and the most confusable human texts. At inference, all auxiliary heads are discarded, giving MELD the same interface and cost as a standard detector. On the public RAID leaderboard, MELD is the strongest open-source detector and is competitive with leading commercial models, especially under attack and at low FPR. Across standard held-out benchmarks, MELD matches or outperforms supervised baselines. We further introduce MELD-eval, a held-out evaluation pool built from recent chat models released by four major LLM providers. Without additional finetuning, MELD achieves 99.9% TPR at 1% FPR on MELD-eval, while many baselines degrade sharply.

118. MIST: Multimodal Interactive Speech-based Tool-calling Conversational Assistants for Smart Homes

Authors: Maximillian Chen , Xuanming Zhang , Michael Peng , Zhou Yu , Alexandros Papangelis , Yohan Jo
URL: https://arxiv.org/abs/2605.06897
Abstract:

The rise of Internet of Things (IoT) devices in the physical world necessitates voice-based interfaces capable of handling complex user experiences. While modern Large Language Models (LLMs) already demonstrate strong tool-usage capabilities, modeling real-world IoT devices presents a difficult, understudied challenge which combines modeling spatiotemporal constraints with speech inputs, dynamic state tracking, and mixed-initiative interaction patterns. We introduce MIST (the Multimodal Interactive Speech-based Tool-calling Dataset), a synthetic multi-turn, voice-driven code generation task that operates over IoT devices. We find that there is a significant gap between open- and closed-weight multimodal LLMs on MIST, and that even frontier closed-weight LLMs have substantial headroom. We release MIST and an extensible data generation framework to build related datasets in order to facilitate research on mixed-initiative voice assistants which reason about physical world constraints.

119. Don’t Retrain, Align: Adapting Autoregressive LMs to Diffusion LMs via Representation Alignment

Authors: Fred Zhangzhi Peng , Alexis Fox , Anru R. Zhang , Alexander Tong
URL: https://arxiv.org/abs/2605.06885
Abstract:

Diffusion language models (DLMs) have recently demonstrated capabilities that complement standard autoregressive (AR) models, particularly in non-sequential generation and bidirectional editing. Although recent work has shown that pretrained autoregressive checkpoints can be converted into diffusion language models, existing recipes primarily transfer parameters through continued denoising training with objective- and attention-level modifications. We instead ask whether the internal representation geometry learned by next-token prediction can be explicitly preserved during AR-to-DLM conversion. We hypothesize that much of the semantic structure learned by AR pretraining can transfer across generation orders, and thus DLM training should be viewed as relearning the decoding path rather than relearning language representations. To investigate this, we introduce REPR-ALIGN, a representation alignment objective that adapts a bidirectional masked diffusion model to reuse representations from a pretrained AR model of identical architecture. Concretely, we align the hidden states of the DLM to the frozen AR model at every layer using cosine similarity, while optimizing the standard masked denoising objective. This simple alignment, with no adapters and no architectural changes beyond the attention mask, yields up to 4x training acceleration in our setting and is particularly effective in low-data regimes. Our results suggest that linguistic representations can transfer across generation order, and that representation alignment provides a simple and effective technique for training diffusion language models. Code is available at this https URL .

120. How to Compress KV Cache in RL Post-Training? Shadow Mask Distillation for Memory-Efficient Alignment

Authors: Rui Zhu , Weiheng Bai , Qiushi Wu , Yang Ren , Haixu Tang , Yuchu Liu
URL: https://arxiv.org/abs/2605.06850
Abstract:

Reinforcement Learning (RL) has emerged as a crucial paradigm for unlocking the advanced reasoning capabilities of Large Language Models (LLMs), encompassing frameworks like RLHF and RLAIF. Regardless of the specific optimization algorithm (e.g., PPO, GRPO, or Online DPO), online RL inherently requires an exploratory trajectory generation (rollout) phase. However, for long-context reasoning tasks, this rollout phase imposes a severe ``memory wall’’ due to the exorbitant Key-Value (KV) cache footprint. While applying KV cache compression during rollouts mitigates this memory overhead, it induces a critical off-policy bias. Although modern KV compression is often nearly lossless during standard inference, even minuscule approximation errors are drastically amplified by the inherent instability of RL optimization. Specifically, the sampler generates responses under a sparse context, whereas the learner updates parameters using the full, dense context. Existing statistical solutions, such as importance reweighting, struggle to correct this magnified bias, suffering from high gradient variance and severe sample inefficiency.

121. LLM-Guided Open Hypothesis Learning from Autonomous Scanning Probe Microscopy Experiments

Authors: Boris Slautin , Utkarsh Pratiush , Yu Liu , Kamyar Barakati , Sergei Kalinin
URL: https://arxiv.org/abs/2605.06839
Abstract:

Autonomous experimentation has transformed microscopy and materials discovery by enabling closed-loop optimization including imaging and spectroscopy tuning, strucutre property relationship discovery, and exploration of combinatorial libraries. However, most current workflows remain limited to selecting measurements within fixed objective or hypothesis spaces, rather than generating new physical models from experimental data. Here, we introduce an open hypothesis-learning framework that combines symbolic regression with large-language-model-based physical evaluation and implement it for autonomous scanning probe microscopy. Symbolic regression generates candidate analytical relationships directly from sparse measurements, while the language-model evaluator ranks these candidates according to physical plausibility, scaling behavior, and consistency with known mechanisms. We demonstrate the approach on autonomous piezoresponse force microscopy measurements of ferroelectric domain switching in a PZT thin film. Starting from five seed measurements, the workflow evolves from physically incomplete candidate expressions toward interpretable voltage-time growth laws consistent with kinetic domain-wall motion. This work extends autonomous microscopy from closed-loop optimization toward open hypothesis discovery, where candidate physical laws emerge from the experiment itself rather than being specified in advance. More broadly, the framework establishes a route for integrating symbolic regression, physical reasoning, and adaptive experimentation into hierarchical autonomous scientific workflows.

122. IntentGrasp: A Comprehensive Benchmark for Intent Understanding

Authors: Yuwei Yin , Chuyuan Li , Giuseppe Carenini
URL: https://arxiv.org/abs/2605.06832
Abstract:

Accurately understanding the intent behind speech, conversation, and writing is crucial to the development of helpful Large Language Model (LLM) assistants. This paper introduces IntentGrasp, a comprehensive benchmark for evaluating the intent understanding capability of LLMs. Derived from 49 high-quality, open-licensed corpora spanning 12 diverse domains, IntentGrasp is constructed through source datasets curation, intent label contextualization, and task format unification. IntentGrasp contains a large-scale training set of 262,759 instances and two evaluation sets: an All Set of 12,909 test cases and a more balanced and challenging Gem Set of 470 cases. Extensive evaluations on 20 LLMs across 7 families (including frontier models such as GPT-5.4, Gemini-3.1-Pro, and Claude-Opus-4.7) demonstrate unsatisfactory performance, with scores below 60% on All Set and below 25% on Gem set. Notably, 17 out of 20 tested models perform worse than a random-guess baseline (15.2%) on Gem Set, while the estimated human performance is ~81.1%, showing substantial room for improvement. To enhance such ability, this paper proposes Intentional Fine-Tuning (IFT), which fine-tunes the models on the training set in IntentGrasp, yielding significant gains of 30+ F1 points on All Set and 20+ points on Gem Set. Tellingly, the leave-one-domain-out (Lodo) experiments further demonstrate the strong cross-domain generalizability of IFT, verifying that it is a promising approach to substantially enhancing the intent understanding of LLMs. Overall, by benchmarking and boosting intent understanding ability, this study sheds light on a promising path towards more intentional, capable, and safe AI assistants for human benefits and social good.

123. VITA-QinYu: Expressive Spoken Language Model for Role-Playing and Singing

Authors: Jiacheng Xu , Heting Gao , Liufei Xie , Zhenchuan Yang , Lijiang Li , Yiting Chen , Bin Zhang , Meng Chen , Chaoyu Fu , Weifeng Zhao , Wenjiang Zhou
URL: https://arxiv.org/abs/2605.06765
Abstract:

Human speech conveys expressiveness beyond linguistic content, including personality, mood, or performance elements, such as a comforting tone or humming a song, which we formalize as role-playing and singing. We present VITA-QinYu, the first expressive end-to-end (E2E) spoken language model (SLM) that goes beyond natural conversation to support both role-playing and singing generation. VITA-QinYu adopts a hybrid speech-text paradigm that extends interleaved text-audio modeling with multi-codebook audio tokens, a design enabling richer paralinguistic representation while preserving a clear separation between modalities to avoid interference. We further develop a comprehensive data generation pipeline to synthesize a total of 15.8K hours of natural conversation, role-playing, and singing data for training. VITA-QinYu demonstrates superior expressiveness, outperforming peer SLMs by 7 percentage points on objective role-playing benchmarks, and surpassing peer models by 0.13 points on a 5-point MOS scale for singing. Simultaneously, it achieves state-of-the-art conversational accuracy and fluency, exceeding prior SLMs by 1.38 and 4.98 percentage points on the C3 and URO benchmarks, respectively. We open-source our code and models and provide an easy-to-use demo with full-stack support for streaming and full-duplex interaction.

124. R$^3$L: Reasoning 3D Layouts from Relative Spatial Relations

Authors: Zhifeng Gu , Yuqi Wang , Bing Wang
URL: https://arxiv.org/abs/2605.06758
Abstract:

Relative spatial relations provide a compact representation of spatial structure and are fundamental to relative spatial reasoning in 3D layout generation. Recent works leverage Multimodal Large Language Models (MLLMs) to infer such relations, but the inferred relations are often unreliable and are typically handled with post-hoc heuristics. In this paper, we propose R$^3$L, a general framework that improves the reliability and consistency of relative spatial reasoning for 3D layout generation. Our key motivation is that multi-hop reasoning requires repeated reference-frame transformations, which accumulate errors in inferred relations and lead to semantic and metric drift. To mitigate this, we propose invariant spatial decomposition to break coupled relation chains, and consistent spatial imagination to promote self-consistency through an imagine-and-revise loop. We further introduce supportive spatial optimization to ease pose optimization via global-to-local coordinate re-parameterization. Extensive experiments across diverse scene types and instructions demonstrate that R$^3$L produces more physically feasible and semantically consistent layouts. Notably, our analysis shows that resolving frame-induced inconsistencies is crucial for reliable multi-hop relative spatial reasoning. The code is available at this https URL .

125. Gradient Extrapolation-Based Policy Optimization

Authors: Ismam Nur Swapnil , Aranya Saha , Tanvir Ahmed Khan , Mohammad Ariful Haque , Ser-Nam Lim
URL: https://arxiv.org/abs/2605.06755
Abstract:

Reinforcement learning is widely used to improve the reasoning ability of large language models, especially when answers can be automatically checked. Standard GRPO-style training updates the model using only the current step, while full multi-step lookahead can give a better update direction but is too expensive because it needs many backward passes. We propose Gradient Extrapolation-Based Policy Optimization (GXPO), a plug-compatible policy-update rule for GRPO-style reasoning RL. GXPO approximates a longer local lookahead using only three backward passes during an active phase. It reuses the same batch of rollouts, rewards, advantages, and GRPO loss, so it does not require new rollouts or reward computation at the lookahead points. GXPO takes two fast optimizer steps, measures how the gradients change, predicts a virtual K-step lookahead point, moves the policy partway toward that point, and then applies a corrective update using the true gradient at the new position. When the lookahead signal becomes unstable, GXPO automatically switches back to standard single-pass GRPO. We also give a plain-gradient-descent surrogate analysis that explains when the extrapolation is exact and where its local errors come from. Across Qwen2.5 and Llama math-reasoning experiments, GXPO improves the average sampled pass@1 by +1.65 to +5.00 points over GRPO and by +0.14 to +1.28 points over the strongest SFPO setting, while keeping the active-phase cost fixed at three backward passes. It also achieves up to 4.00x step speedup, 2.33x wall-clock speedup, and 1.33x backward-pass speedup in reaching GRPO’s peak accuracy.

126. Geometric Kolmogorov–Arnold Network (GeoKAN)

Authors: Abhijit Sen , Bikram Keshari Parida , Giridas Maiti , Mahima Arya , Denys I. Bondar
URL: https://arxiv.org/abs/2605.06740
Abstract:

We introduce Geometric Kolmogorov–Arnold Networks (GeoKANs), a family of geometry-aware KAN-type models in which approximation is carried out in learned, geometry-adapted coordinates rather than in fixed Euclidean input coordinates. GeoKAN achieves this by learning a diagonal Riemannian metric that warps the input before basis expansion and feature mixing. The learned metric provides a geometric inductive bias through local length scaling and volume distortion, and in physics-informed settings it also affects the differential structure seen by the model. Within this framework, we develop three main variants, namely GeoKAN-NNMetric, GeoKAN-$\gamma$, and LM-KAN. For LM-KAN, we further consider three basis-specific versions, LM-KAN-RBF, LM-KAN-Wav, and LM-KAN-Fourier. These variants allow us to study geometry-aware KAN models both as general function approximators and as surrogates in physics-informed learning. By stretching regions with rapid variation and compressing smoother regions, GeoKAN reallocates representational resolution in a task-dependent manner, allowing the model to place capacity where it is most needed. As a result, GeoKAN is well suited to sharp, stiff, localized, and strongly non-uniform regimes arising in scientific machine learning and differential-equation problems.

127. A Self-Healing Framework for Reliable LLM-Based Autonomous Agents

Authors: Cheonsu Jeong , Younggun Shin
URL: https://arxiv.org/abs/2605.06737
Abstract:

Autonomous agents based on Large Language Models (LLMs) are increasingly being utilized in complex software systems. However, reliability remains a significant challenge due to unpredictable failures such as hallucinations, execution errors, and inconsistent reasoning. This paper proposes a reliability-aware self-healing framework for LLM-based software agents. The framework integrates failure detection, reliability assessment, and automated recovery mechanisms. First, we define a taxonomy of failure types and introduce a quantitative reliability assessment model. Next, we propose a failure detection method that identifies abnormal agent behavior based on execution patterns and output consistency. Finally, we design a self-healing mechanism that dynamically recovers from failures through adaptive replanning and corrective prompting strategies. The proposed framework was implemented in a multi-agent workflow environment and evaluated using real-world task scenarios. Experimental results demonstrate that our approach significantly increases task success rates, reduces failure propagation, and enhances overall system robustness compared to existing methods. In particular, this study distinguishes itself by establishing an integrated monitoring system that combines the agent’s internal reasoning process with external execution results. These findings are expected to contribute to securing the stability of advanced autonomous systems and lowering the barriers to LLM adoption in production environments.

128. Beyond Factor Aggregation: Gauge-Aware Low-Rank Server Representations for Federated LoRA

Authors: Jinqian Chen , Chang Liu , Jihua Zhu
URL: https://arxiv.org/abs/2605.06733
Abstract:

Federated LoRA enables parameter-efficient adaptation of large language models under decentralized data and limited client this http URL , directly averaging LoRA factors is representation-dependent: the same intrinsic update admits infinitely many gauge-equivalent factorizations, so factor-level aggregation can change under arbitrary coordinate choices while the underlying update remains unchanged. This reveals a semantic mismatch in existing federated LoRA aggregation rules. We propose \textbf{GLoRA}, a gauge-aware server representation for federated this http URL of aggregating raw factors, GLoRA estimates a consensus update subspace from client projectors and aggregates client updates in shared reference coordinates, thereby representing semantic update aggregation entirely in low-rank form. To support heterogeneous client capacities, GLoRA further provides a rank-compatible readout that instantiates adapters of different ranks from the same server state without dense update reconstruction. Experiments on GLUE and SuperNI show that GLoRA consistently outperforms federated LoRA baselines under data, resource, and task heterogeneity, including heterogeneous client ranks, sparse participation, larger backbones, and unseen-task evaluation. GLoRA also achieves a favorable efficiency–performance trade-off, suggesting that effective federated LoRA requires not merely averaging low-rank factors, but defining a semantically meaningful server-side representation for aggregation.

129. OmicsLM: A Multimodal Large Language Model for Multi-Sample Omics Reasoning

Authors: Maciej Sypetkowski , Joanna Krawczyk , Łukasz Smoliński , Remigiusz Kinas , Przemysław Pietrzak , Tomasz Jetka , Rafał Powalski
URL: https://arxiv.org/abs/2605.06728
Abstract:

Interpreting transcriptomic data is one of the most common analytical tasks in modern biology. Yet most current models either consume expression profiles without producing natural-language biological explanations, or reason in language without direct access to quantitative omics measurements. We introduce OmicsLM, a multimodal LLM that connects quantitative omics profiles with natural-language biological tasks. OmicsLM represents each transcriptomic profile as a compact continuous representation within the LLM context. This interface preserves quantitative expression signal while allowing natural-language instructions, explicit gene mentions, and multiple interleaved biological samples to be processed together in one model context. We train OmicsLM on more than 5.5 million instruction-following examples spanning over 70 task types, combining continuous transcriptomic inputs, experimental data rendered through diverse language templates, and free-text biological knowledge and question-answering data. This mixture covers cell type annotation, perturbation prediction, clinical prediction, pathway reasoning, and open-ended biological question answering. Existing benchmarks evaluate either profile-level prediction or text-only biological QA, leaving language-guided, multi-sample reasoning over real expression profiles unmeasured. To close this gap, we introduce GEO-OmicsQA, a benchmark for multi-sample biological question answering built from real Gene Expression Omnibus (GEO) studies. We demonstrate that OmicsLM can use expression profiles directly and perform comparably to specialized omics models on profile-level tasks, while outperforming both omics-specialized models and general LLMs on language-guided biological reasoning over expression data.

130. Conditional generation of antibody sequences with classifier-guided germline-absorbing discrete diffusion

Authors: Justin Sanders , Luca Giancardo , Lan Guo , Yue Zhao , Kemal Sonmez , Nina Cheng , Melih Yilmaz
URL: https://arxiv.org/abs/2605.06720
Abstract:

Antibody therapeutics are among the most successful modern medicines, yet computationally designing antibodies with desirable binding and developability properties remains challenging. While protein language models (pLMs) have emerged as powerful tools for antibody sequence design, existing approaches largely suffer from two key limitations: they predominantly memorize germline sequences rather than modeling biologically meaningful somatic variation, and they offer limited support for flexible classifier-guided conditional generation. We address these challenges through two primary contributions. First, we demonstrate that discrete diffusion fine-tuning achieves strong language modeling performance on antibody sequences while allowing for generation conditioned on any off-the-shelf classifier. Second, we introduce germline absorbing diffusion, a novel modification of the discrete diffusion noise process in which the germline sequence - rather than a masked sequence - serves as the absorbing state. This biologically motivated inductive bias restricts the model to learning the trajectory from germline to observed sequence, effectively excluding genetic variation and V(D)J recombination statistics from the learned distribution and dramatically mitigating germline bias. We show that germline diffusion improves non-germline residue prediction accuracy from 26 percent to 46 percent, approaching the theoretical upper bound set by true biological variability. We then demonstrate the utility of our germline diffusion model on the conditional generation tasks of sampling antibodies with improved hydrophobicity and predicted binding affinity. On both tasks our model shows an improved tradeoff between class adherence and sample quality, significantly outperforming EvoProtGrad, a popular strategy to sample from pLMs with gradient-based discrete Markov Chain Monte Carlo.

131. Agentic AI and the Industrialization of Cyber Offense: Forecast, Consequences, and Defensive Priorities for Enterprises and the Mittelstand

Authors: Christopher Koch
URL: https://arxiv.org/abs/2605.06713
Abstract:

Agentic AI systems can plan, call tools, inspect code, interact with web applications, and coordinate multi-step workflows. These same capabilities change the economics of cyber offense. The central near-term risk is not that every low-skill criminal immediately becomes a frontier exploit researcher; it is that agentic AI compresses the attack lifecycle by lowering the cost of reconnaissance, phishing, credential abuse, vulnerability triage, exploit adaptation, and post-compromise decision support. This paper synthesizes current public evidence from national cybersecurity agencies, industry threat reports, agent security guidance, and research on LLM agents cyber capabilities. It introduces a Three Channel Agentic Cyber Risk Model and an Agentic Attack Compression Model, uses the 2026 Linux kernel Copy Fail incident as a case study for foothold-to-root acceleration, and develops a 2026 to 2028 forecast for large enterprises and the German and European Mittelstand. The paper concludes with a prioritized defense roadmap. Organizations should treat agentic AI security as an immediate operational problem: identity, phishing resistant authentication, patch velocity, CI/CD and Linux/container hardening, agent governance, telemetry, and recovery readiness must be strengthened now.

132. Visual Text Compression as Measure Transport

Authors: Lv Tang , Tianyi Zheng , Yang Liu , Bo Li , Xingyu Li
URL: https://arxiv.org/abs/2605.06708
Abstract:

Visual text compression (VTC) promises efficient long-context processing by rendering text into an image and re-encoding it with a vision-language model, often producing $3$–$20\times$ fewer decoder tokens than subword tokenization. Yet token savings do not translate predictably into downstream utility: on some tasks the visual path matches or exceeds the text path, on others it collapses, and the compression ratio itself does not predict which regime will occur. The missing quantity is therefore not another summary of efficiency, but a principled measure of task-relevant information loss induced by visual encoding. We address this problem by formulating VTC in the language of measure transport. Treating text and visual tokens as empirical probability measures, we show that the ViT patch encoder induces a push-forward map whose transport cost decomposes into a precision cost from within-patch aggregation and a coverage cost from cross-patch fragmentation. Both terms are estimable from downstream-label-free probes. This formulation yields two operational consequences: a downstream-label-free routing criterion that selects whether to use the visual path for a given input or benchmark instance, and a transport-informed foveation mechanism that re-encodes high-cost regions at higher resolution. Across $24$ NLP datasets at Qwen3-4B, our label-free rule matches the per-dataset oracle on $17/24$ datasets ($70.8\%$), and improves the average task score by $+3.3\%$ with $-10.3\%$ average tokens relative to a pure-LLM.

Authors: Diego Cabezas Palacios
URL: https://arxiv.org/abs/2605.06707
Abstract:

This paper presents an eight-week observational comparison of 68 single-file HTML generations collected across 17 public experiments in the “HTML AI Battle” project between December 10, 2025 and February 4, 2026. Four reasoning model families, GPT, Gemini, Grok, and Claude, were compared under a fixed public-interface protocol with no custom instructions, no personality tuning, and no repair prompts. Each output was evaluated from a rendered browser video using human scores and a Gemini LLM-as-a-judge layer for prompt adherence, functional correctness, and UI quality, then packaged into a standardized social-media protocol spanning X (Twitter), TikTok, and YouTube. The tracker was also used for two supervised predictive analyses: an experiment-level model for 24-hour X impressions and a generation-level model for HTML verbosity. Under this protocol, Claude was the strongest and most consistent family, leading mean performance and winning 9/17 prompts under the primary human weighted score. Longer measured reasoning time was not associated with higher quality overall. Gemini as a judge was significantly more lenient than the human evaluator on functional correctness and overall performance, while stable self-favoring bias remained unresolved. The exploratory X-impressions model remained weak under post-screen cross-validation (MAE = 46,874, R^2 = -0.377), whereas the HTML-lines model performed better, with a model-family-only baseline outperforming prompt-aware alternatives (MAE = 135.2, R^2 = 0.576). Overall, selected pre-publication technical/audio variables were not sufficient to predict 24-hour X reach, while code verbosity was driven much more by model family than by prompt wording. The comparisons remain observational and are limited by public-interface drift, access-path differences, and one primary human scorer.

134. Toeplitz MLP Mixers are Low Complexity, Information-Rich Sequence Models

Authors: Benjamin L. Badger , Ethan Roland
URL: https://arxiv.org/abs/2605.06683
Abstract:

Transformer-based large language models are in some respects limited by the quadratic time and space computational complexity of attention. We introduce the Toeplitz MLP Mixer (TMM), a transformer-like architecture that swaps attention for triangular-masked Toeplitz matrix multiplication over the sequence dimension resulting in $\mathcal{O} (dn \log n)$ time and $\mathcal O(dn)$ space complexity during training and $\mathcal O(dn)$ time and space at inference prefill. Despite the lack of sophisticated input modulation or state maintenance present in other sub-quadratic architectures, TMMs yield greater training efficiency in terms of loss achieved per compute and device memory. We demonstrate that TMMs are capable of retaining more input information resulting in improved copying ability, which we argue results from a lack of architectural biases. Consistent with higher input information retention, TMMs exhibit superior information retrieval and in-context learning benchmark accuracy compared to comparable architectures. We conclude with an analysis from the perspective of operator index theory and show that, counterintuitively, trained Toeplitz layers of causal non-invertible models are more likely to be invertible or nearly so than models that are actually invertible over their inputs.

135. Evaluating Prompt Injection Defenses for Educational LLM Tutors: Security-Usability-Latency Trade-offs

Authors: Alexandre Cristovão Maiorano
URL: https://arxiv.org/abs/2605.06669
Abstract:

Educational LLM tutors face a core AI alignment challenge: they must follow user intent while preserving pedagogical constraints and safety policies. We present an evaluation methodology for prompt-injection defenses in this setting, showing that guardrail design entails explicit trade-offs among adversarial robustness, benign-task usability, and response latency. We evaluate a domain-specific multi-layer safeguard pipeline combining deterministic pattern filters, structural validation, contextual sandboxing, and session-level behavioral checks. On a controlled holdout benchmark with 480 queries (369 injection, 111 benign), the pipeline reaches 46.34% bypass, 0.00% false positive rate, and 2.50 ms average latency – an operating point that prioritizes pedagogical usability (zero false positives) while maintaining measurable attack resistance. We provide a reproducible benchmark protocol for head-to-head comparison under identical conditions, including stratified bootstrap confidence intervals, paired McNemar significance tests, and direct evaluation of Prompt Guard and NeMo Guardrails on the same split with unified instrumentation. Results expose operational trade-offs: NeMo reaches 0% bypass at 16.22% FPR and 1.3s latency, while Prompt Guard yields 38.48% bypass with 3.60% FPR. The framework supports evidence-based guardrail selection for AI tutoring systems under different institutional risk and usability requirements.

136. CommFuse: Hiding Tail Latency via Communication Decomposition and Fusion for Distributed LLM Training

Authors: Rezaul Karim , Austin Wen , Wang Zongzuo , Weiwei Zhang , Yang Liu , Walid Ahmed
URL: https://arxiv.org/abs/2604.24013
Abstract:

The rapid growth in the size of large language models has necessitated the partitioning of computational workloads across accelerators such as GPUs, TPUs, and NPUs. However, these parallelization strategies incur substantial data communication overhead significantly hindering computational efficiency. While communication-computation overlap presents a promising direction, existing data slicing based solutions suffer from tail latency. To overcome this limitation, this research introduces a novel communication-computation overlap technique to eliminate this tail latency in state of the art overlap methods for distributed LLM training. The aim of this technique is to effectively mitigate communication bottleneck of tensor parallelism and data parallelism for distributed training and inference. In particular, we propose a novel method termed CommFuse that replaces conventional collective operations of reduce-scatter and all-gather with decomposed peer-to-peer (P2P) communication and schedules partitioned computations to enable fine-grained overlap. Our method provides an exact algorithm for reducing communication overhead that eliminates tail latency. Moreover, it presents a versatile solution compatible with data-parallel training and various tensor-level parallelism strategies, including TPSP and UP. Experimental evaluations demonstrate that our technique consistently achieves lower latency, superior Model FLOPS Utilization (MFU), and high throughput.

137. Consensus Entropy: Harnessing Multi-VLM Agreement for Self-Verifying and Self-Improving OCR

Authors: Yulong Zhang , Tianyi Liang , Xinyue Huang , Erfei Cui , Guoqing Wang , Xu Guo , Chenhui Li , Gongshen Liu
URL: https://arxiv.org/abs/2504.11101
Abstract:

Optical Character Recognition (OCR) is fundamental to Vision-Language Models (VLMs) and high-quality data generation for LLM training. Yet, despite progress in average OCR accuracy, state-of-the-art VLMs still struggle with detecting sample-level errors and lack effective unsupervised quality control. We introduce Consensus Entropy (CE), a training-free, model-agnostic metric that estimates output reliability by measuring inter-model agreement entropy. The core insight is that correct predictions converge in output space, while errors diverge. Based on CE, we develop CE-OCR, a lightweight multi-model framework that verifies outputs by ensemble agreement, selects the best outputs, and further improves efficiency through adaptive routing. Experiments demonstrate that CE is robust for quality verification, improving F1 scores by 42.1% over VLM-as-Judge. CE-OCR achieves consistent OCR gains, outperforming self-consistency and single-model baselines at the same cost. Notably, CE requires no training or supervision, enabling plug-and-play integration. Code: this https URL .

LLM 관련 주요 논문 - 2026-05-11

1. VecCISC: Improving Confidence-Informed Self-Consistency with Reasoning Trace Clustering and Candidate Answer Selection

2. Rubric-Grounded RL: Structured Judge Rewards for Generalizable Reasoning

3. Abductive Reasoning with Probabilistic Commonsense

4. TraceFix: Repairing Agent Coordination Protocols with TLA+ Counterexamples

5. AgentEscapeBench: Evaluating Out-of-Domain Tool-Grounded Reasoning in LLM Agents

6. Hierarchical Task Network Planning with LLM-Generated Heuristics

7. GASim: A Graph-Accelerated Hybrid Framework for Social Simulation

8. FactoryBench: Evaluating Industrial Machine Understanding

9. Open-Ended Task Discovery via Bayesian Optimization

10. From Pixels to Prompts: Vision-Language Models

11. Efficient Data Selection for Multimodal Models via Incremental Optimization Utility

12. GraphReAct: Reasoning and Acting for Multi-step Graph Inference

13. Tools as Continuous Flow for Evolving Agentic Reasoning

14. Discovering Ordinary Differential Equations with LLM-Based Qualitative and Quantitative Evaluation

15. Implicit Compression Regularization: Concise Reasoning via Internal Shorter Distributions in RL Post-Training

16. SOM: Structured Opponent Modeling for LLM-based Agents via Structural Causal Model

17. Structured Role-Aware Policy Optimization for Multimodal Reasoning

18. Can Agents Price a Reaction? Evaluating LLMs on Chemical Cost Reasoning

19. EnvSimBench: A Benchmark for Evaluating and Improving LLM-Based Environment Simulation

20. HMACE: Heterogeneous Multi-Agent Collaborative Evolution for Combinatorial Optimization

21. Can You Break RLVER? Probing Adversarial Robustness of RL-Trained Empathetic Agents

22. ARMOR: An Agentic Framework for Reaction Feasibility Prediction via Adaptive Utility-aware Multi-tool Reasoning

23. 2.5-D Decomposition for LLM-Based Spatial Construction

24. The Context Gathering Decision Process: A POMDP Framework for Agentic Search

25. Behavior Cue Reasoning: Monitorable Reasoning Improves Efficiency and Safety through Oversight

26. Learning and Reusing Policy Decompositions for Hierarchical Generalized Planning with LLM Agents

27. Self-Programmed Execution for Language-Model Agents

28. Mitigating Cognitive Bias in RLHF by Altering Rationality

29. How Well Do LLMs Perform on the Simplest Long-Chain Reasoning Tasks: An Empirical Study on the Equivalence Class Problem

30. Agentick: A Unified Benchmark for General Sequential Decision-Making Agents

31. Extracting Search Trees from LLM Reasoning Traces Reveals Myopic Planning

32. Towards Security-Auditable LLM Agents: A Unified Graph Representation

33. When Does Critique Improve AI-Assisted Theoretical Physics? SCALAR: Structured Critic–Actor Loop for Agentic Reasoning

34. Weblica: Scalable and Reproducible Training Environments for Visual Web Agents

35. When Does a Language Model Commit? A Finite-Answer Theory of Pre-Verbalization Commitment

36. From Storage to Experience: A Survey on the Evolution of LLM Agent Memory Mechanisms

37. CASCADE: Case-Based Continual Adaptation for Large Language Models During Deployment

38. Hidden Coalitions in Multi-Agent AI: A Spectral Diagnostic from Internal Representations

39. GraphDC: A Divide-and-Conquer Multi-Agent System for Scalable Graph Algorithm Reasoning

40. Flow-OPD: On-Policy Distillation for Flow Matching Models

41. The Memory Curse: How Expanded Recall Erodes Cooperative Intent in LLM Agents

42. CA-SQL: Complexity-Aware Inference Time Reasoning for Text-to-SQL via Exploration and Compute Budget Allocation

43. Fast Byte Latent Transformer

44. Beyond Pairs: Your Language Model is Secretly Optimizing a Preference Graph

45. Tool Calling is Linearly Readable and Steerable in Language Models

46. Towards Apples to Apples for AI Evaluations: From Real-World Use Cases to Evaluation Scenarios

47. Dooly: Configuration-Agnostic, Redundancy-Aware Profiling for LLM Inference Simulation

48. Where’s the Plan? Locating Latent Planning in Language Models with Lightweight Mechanistic Interventions

49. KL for a KL: On-Policy Distillation with Control Variate Baseline

50. MatryoshkaLoRA: Learning Accurate Hierarchical Low-Rank Representations for LLM Fine-Tuning

51. CyBiasBench: Benchmarking Bias in LLM Agents for Cyber-Attack Scenarios

52. GazeVLM: Active Vision via Internal Attention Control for Multimodal Reasoning

53. Beyond Confidence: Rethinking Self-Assessments for Performance Prediction in LLMs

54. Tracing Uncertainty in Language Model “Reasoning”

55. POETS: Uncertainty-Aware LLM Optimization via Compute-Efficient Policy Ensembles

56. Benchmarking EngGPT2-16B-A3B against Comparable Italian and International Open-source LLMs

57. SOD: Step-wise On-policy Distillation for Small Language Model Agents

58. LLM hallucinations in the wild: Large-scale evidence from non-existent citations

59. Memory-Efficient Looped Transformer: Decoupling Compute from Memory in Looped Language Models

60. The AI-Native Large-Scale Agile Software Development Manifesto

61. DRIP-R: A Benchmark for Decision-Making and Reasoning Under Real-World Policy Ambiguity in the Retail Domain

62. TRACE: Tourism Recommendation with Accountable Citation Evidence

63. Operating Within the Operational Design Domain: Zero-Shot Perception with Vision-Language Models

64. Quality-Conditioned Agreement in Automated Short Answer Scoring: Mid-Range Degradation and the Impact of Task-Specific Adaptation

65. LithoBench: Benchmarking Large Multimodal Models for Remote-Sensing Lithology Interpretation

66. Post-training makes large language models less human-like

67. Mathematical Reasoning via Intervention-Based Time-Series Causal Discovery Using LLMs as Concept Mastery Simulators

68. Your Language Model is Its Own Critic: Reinforcement Learning with Value Estimation from Actor’s Internal States

69. ProteinJEPA: Latent prediction complements protein language models

70. LARAG: Link-Aware Retrieval Strategy for RAG Systems in Hyperlinked Technical Documentation

71. SHRED: Retain-Set-Free Unlearning via Self-Distillation with Logit Demotion

72. Vaporizer: Breaking Watermarking Schemes for Large Language Model Outputs

73. HBEE: Human Behavioral Entropy Engine – Pre-Registered Multi-Agent LLM Simulation of Peer-Suspicion-Based Detection Inversion

74. The Moltbook Files: A Harmless Slopocalypse or Humanity’s Last Experiment

75. Sparse Autoencoders as Plug-and-Play Firewalls for Adversarial Attack Detection in VLMs

76. Prompt Engineering Strategies for LLM-based Qualitative Coding of Psychological Safety in Software Engineering Communities: A Controlled Empirical Study

77. Unsolvability Ceiling in Multi-LLM Routing: An Empirical Study of Evaluation Artifacts

78. BalCapRL: A Balanced Framework for RL-Based MLLM Image Captioning

79. MISA: Mixture of Indexer Sparse Attention for Long-Context LLM Inference