New Artificial Intelligence (cs.AI) papers, Friday 23 October 2026

230 papers announced on arXiv. Get them ranked for you, free →

1. Domain-Specific Constitutional AI: Enhancing Safety in LLM-Powered Mental Health Chatbots

   Chenhan Lyu, Yutong Song, Pengfei Zhang et al. · arXiv:2509.16444

   AI summaryThis paper introduces a domain-specific Constitutional AI approach tailored for mental health chatbots. The method integrates specialized psychological principles to improve crisis intervention accuracy and prevent harmful misinformation in sensitive clinical interactions.

2. The Cost of Saying Hello: An Empirical Study of Handshake Overhead in AI Agent Communication Protocols

   Pranav Varshney · arXiv:2610.30151

   AI summaryThis paper provides the first empirical benchmark of handshake overhead across major AI agent communication protocols. It finds that round-trip counts and capability metadata parsing dominate costs, consuming significant latency and up to eleven percent of context windows before task execution begins.

3. MCP vs CLI for AI Agents: A Comparative Analysis of Tool Integration Strategies for Autonomous Systems

   Arun Malik, Nitish Talasu, Bradley Klemick et al. · arXiv:2610.30157

   AI summaryThis paper compares CLI and MCP tool integration strategies for AI agents across fourteen dimensions. It finds that while CLI is more efficient for local workflows, MCP is necessary for secure, governed, and multi-tenant production systems.

4. Lighthouse: On Building a Trustworthy Digital Assistant for Exploring History with Civil Rights Archives

   John A. Aydin, Angelica Williams, Steven Dindl et al. · arXiv:2610.30158

   AI summaryThe paper introduces Lighthouse, a trustworthy neuro-symbolic digital assistant for exploring historical U.S. Civil Rights archives. It successfully combines large language models with rule-based systems to reduce harmful outputs and reliably answer both factual and expert-evaluated subjective questions using primary sources.

5. V-JEPA 2 Meets AQA: Phase-Based Scoring and Reference-Based Feedback for Mountain Biking

   Santhiya Rajan, Savio Cardoz · arXiv:2610.30160

   AI summaryPipeline uses frozen V‑JEPA‑2 video features and pose data to predict binary jump quality; achieves AUROC ~0.77 (0.81 without biomechanical inputs) but shows limited generalization and unreliable feedback, highlighting need for further evaluation.

6. TauIndianBankBench: Procedure-Aware, Action-Oriented Evaluation of Tool-Using Conversational Agents

   Krishanu Adhikary, Prashant Devadiga · arXiv:2610.30161

   AI summaryThe paper introduces TauIndianBankBench, a 1,000‑task benchmark that evaluates conversational agents on both final outcomes and every required procedural step in Indian retail banking. It finds that 6.5‑12.0 % of sessions pass only outcome checks yet violate procedures.

7. FusionEval: Benchmarking Model Fusion for Large Language Models

   Shuo Cai, Yanggan Gu, Yuhang Liu et al. · arXiv:2610.30162

   AI summaryFusionEval is a benchmark suite for evaluating large language model fusion methods across diverse expert domains. Experiments reveal that training-free methods preserve reasoning but degrade tool use, while current training-based approaches require high computational costs to maintain balanced capabilities.

8. IRAH: Hierarchical Agentic Tree Search for Retrieval Pipeline Optimization

   Maria Ana Cardei, Aniket Singh, Amir Moghadam et al. · arXiv:2610.30167

   AI summaryIRAH jointly optimizes retrieval pipelines by hierarchical UCT-guided search (macro structure first, then parameters) with language-agent reflections and meta-learned warm starts. It outperforms TPE, achieving up to 6.4% nDCG@10 gains and the best average 0.4816 on four benchmarks.

9. Accelerating Wildfire Hazard Mapping Across Canada with an AI Surrogate Modelling Framework

   Hager Radi Abdelwahed, Toluwaleke Olutayo, Chirag Goel et al. · arXiv:2610.30170

   AI summaryThis paper introduces a multi-task deep learning surrogate model to emulate computationally expensive process-based wildfire simulators across Canada. The AI framework achieves an 806-fold CPU speedup while accurately predicting burn probability and fire intensity for rapid hazard mapping.

10. Neural Networks with Differentiable Belief-Function Layers

    Antonin Sulc · arXiv:2610.30173

    AI summaryThe paper builds a neural network whose intermediate layers produce Dempster–Shafer mass functions, combining them with Yager’s rule to avoid divergent gradients. It shows that the resulting ignorance mass accurately reflects human disagreement and out‑of‑distribution uncertainty, outperforming Dempster‑based variants.

11. When Stable and Preferred Semantics Agree: A Path-Based Approach

    Yan Zhang, Kehan Liu · arXiv:2610.30176

    AI summaryThis paper presents a path‑based criterion for when preferred and stable semantics coincide in abstract argumentation frameworks, generalizing Dung’s limited controversiality and SCC‑symmetry conditions. It introduces sequences of path sets, showing collapse to three key well‑foundedness conditions and new distinct families, applicable to finite and infinite frameworks and yielding stable extension existence criteria.

12. GeoDiv KAN Solver for Time Fractional Navier Stokes Equations on Complex Domains to Realistic CFD Reconstruction

    Yatin Kumar, Shivani Saini, Ramesh Kumar Vats · arXiv:2610.30179

    AI summaryThis paper introduces GeoDiv KAN, a physics-informed neural network architecture combining geometry-gated feature extraction and divergence-free streamfunction formulations to solve time-fractional Navier-Stokes equations on complex domains. Experiments show it outperforms baseline models in accuracy while successfully reconstructing realistic CFD benchmark flows.

13. Detecting AI Training in Datacentre Power

    Tom Kimpson, Mauricio Baker, Emlyn Graham · arXiv:2610.30182

    AI summaryThis paper introduces a Viterbi-based tracking method to detect large-scale AI training runs using facility-level power consumption data. The main result shows the tracker detects at least 96 percent of unhidden runs and overcomes most adversarial evasion strategies by making hidden training significantly more computationally expensive.

14. Self-Evolving Static Analysis via LLM Pseudo-Labeling

    Rui Melo, Andre Duarte, Aditya Oke et al. · arXiv:2610.30184

    AI summarySESAA uses an LLM agent to automatically generate and refine joern queries for static analysis, guided only by another LLM critic’s feedback. It achieves F1 = 0.742 on the OWASP Python benchmark and detects 10 of 19 real‑world CVEs, outperforming default tools.

15. Agentic Verified Optimization of Real-World Software

    Zhe Ye, Udaya Ghai, Zhizhen Qin et al. · arXiv:2610.30188

    Program optimization aims to improve performance while preserving behavior. Coding agents can already make programs faster by changing algorithms, data structures, and execution strategies. However, existing equivalence checks mostly rely on tests, which agents have been shown…

16. Why Constrained Updates Still Forget: A Functional-Path Analysis of Continual SFT

    Qingsen Ma, Dianyun Wang, Jiahong Liu et al. · arXiv:2610.30191

    AI summaryThe paper decomposes how constrained updates in supervised fine‑tuning affect old‑task performance, showing that update placement and granularity drive forgetting. It demonstrates that matched replay schedules and subspace regularization can better preserve old behavior while improving new‑task gains.

17. Detection Is Not Enough: Reasoning Traces as Testable Program Hypotheses

    Vaidehi Hegde · arXiv:2610.30200

    When a language model is deployed, no reference solution exists, so the reasoning trace it writes is the only evidence for judging its answer. The reference-free substitutes each fail differently: self-assessment degrades the answer it was meant to protect, sampling agreement…

18. MPIDetector: Detecting Multi-modal Prompt Injection via LLM-based Reasoning

    Yiheng Li, Jingyue Li · arXiv:2610.30202

    With the rapid integration of Multi-modal Large Language Models (MLLMs) into real-world applications, Prompt Injection (PI) has emerged as a critical security threat. Attackers can conceal malicious instructions within textual narratives or visual inputs to bypass safety…

19. Feedback That Backfires in Games: Do LLM Players Repeat the Move They Just Watched Fail?

    Esmail Gumaan · arXiv:2610.30205

    AI summaryThe paper shows that in LLM agents playing Sokoban, Connect Four, and Othello, including a failed move and its failure message in the prompt makes the model more likely to repeat that move (especially illegal ones), reducing its effectiveness.

20. MoEZip: Routing-Aware KV Cache Compression for Sparse Mixture-of-Experts LLMs

    Minsang Kim, Seungjun Baek · arXiv:2610.30206

    AI summaryMoEZip compresses KV caches for MoE LLMs by using a Fisher‑information–based metric to gauge expert routing sensitivity, combining it with attention weights for routing‑aware eviction. It outperforms baselines, gaining +24/+20 average points on LongBench/RULER at 10% cache retention.

21. Prediction Is Not Calibration: Item Parameter Recovery and Representation Coupling in Knowledge Tracing

    Wenrui Yuan, Stéphanie M. van den Berg, Johannes Steinrücke · arXiv:2610.30211

    AI summaryThis paper investigates why accurate knowledge tracing models fail to reliably recover item parameters like discrimination. The authors show that separating sequence representations from item parameter keys improves parameter recovery and cross-order consistency without sacrificing predictive accuracy.

22. When Reconstruction Is Not Enough: Doubly Robust Counterfactual and Interventional Mediation Analysis with Learned Multimodal Representations

    Cong Cao · arXiv:2610.30220

    AI summaryThis paper shows that standard representation learning objectives like reconstruction can fail to preserve causal identification in multimodal health data, leading to severe bias in doubly robust mediation estimators. The authors propose a validated workflow including diagnostics and cross-fitted estimators to ensure reliable counterfactual and interventional analyses.

23. Automated screening of privacy risks in medical AI

    Konstantinos Vrettos, Eleftherios Tzanis, Haridimos Kondylakis et al. · arXiv:2610.30229

    AI summaryThe paper measures per‑patient vulnerability to membership inference attacks across 7,789 medical AI models, finding that risk rankings are stable across training choices. It presents a lightweight tool that flags high‑risk records before model training, reducing exposed cohorts.

24. Edit, Don't Imitate: Minimal-Edit On-Policy Distillation for Multimodal Reasoning

    Wen Ye, Ailin Deng, Zirui Wang et al. · arXiv:2610.30231

    AI summaryThe paper proposes minimal‑edit on‑policy distillation for vision‑language math models, correcting student outputs by penalizing only the teacher‑disfavored tokens while rewarding correct rollouts. It outperforms or matches models with reference figure descriptions, improving perception and transferring to math benchmarks.

25. Breaking the high-degree barrier in Lehmer's problem with a human-AI discovery loop

    Yuri Chervonyi, Sergei Gukov, Thang Luong · arXiv:2610.30277

    AI summaryA human‑AI discovery loop yields a five‑parameter construction that, for every even degree up to 2000, produces irreducible integer polynomials with Mahler measure below Boyd’s limit 1.255434, surpassing all but 0.4% of previous records.

26. SWE-VERIFIER: Selective Verification for Cost-Efficient Coding Agents

    Kaijian Zou, Vaibhav Chennareddy, Yunxiang Zhang et al. · arXiv:2610.30284

    AI summarySWE‑Verifier is a lightweight model that selectively triggers sampling and ranking of candidate actions for coding agents, deciding when to verify alternatives. It boosts MiniMax‑M2.7 resolution from 75.47 % to 77.20 % at only 1.09× cost, while verifying less than 4 % of steps.

27. Generalizing Metropolis-Hastings for DAG Sampling

    Daniele Nikzad, Mikko Koivisto · arXiv:2610.30291

    AI summaryThe paper introduces duet sampling, a generalization of Metropolis‑Hastings that chooses two proposal distributions per iteration to resample parents of multiple nodes in a DAG. This yields Markov chains that mix faster and outperform state‑of‑the‑art DAG‑learning methods.

28. GAV: A scientific experimentation environment for PDE agents

    Muheng Li, Wenlong Mou · arXiv:2610.30292

    AI summaryGAV creates verifiable PDE problems by learning a distribution of reference solutions, then using manufactured solutions to generate matching PDEs, allowing agents to iteratively improve coding programs with accurate feedback. On Navier-Stokes, it reduces test error of existing agents by over tenfold.

29. Gender, Sexual Orientation, and the One-Year Refugee Bar: A Computational Study of Bill C-12 and Canadian Refugee Adjudication

    Sean Rehaag · arXiv:2610.30297

    In March 2026, Bill C-12 amended the Immigration and Refugee Protection Act to make refugee claims ineligible for referral to the Refugee Protection Division where they are made more than one year after the claimant's first entry to Canada. Affected claimants are diverted to a…

30. TerraTransfer++: Demonstration-Free End-to-End Planner

    Zikang Xiong, Weixin Li, Zhouchonghao Wu et al. · arXiv:2610.30319

    End-to-end autonomous driving has achieved state-of-the-art performance on benchmarks and real-world deployments. Its standard training recipe, however, is expensive across all existing paradigms: imitation learning requires fleet-scale driving logs whose behavior distribution…

31. MPBRS: Multi-Potential-Based Reward Shaping for Unified Reward Configuration in Humanoid Locomotion

    Van Chung Nguyen, Hung Manh La · arXiv:2610.30334

    AI summaryMulti‑Potential‑Based Reward Shaping (MPBRS) unifies reward design for humanoid locomotion, using a single set of potential functions that scale consistently across nine robots with only metadata changes. It reduces tuning sensitivity and preserves optimal policies better than conventional shaping.

32. Consistency Is Not Validity: Self-Rewarded Moral Argumentation Makes Language Models Decisive but Pliable

    Marius De Kuthy Meurers, Aaron Parisi, Bernd Bohnet · arXiv:2610.30349

    AI summaryThe paper trains a policy via self‑rewarded argument quality to improve LLM moral judgments, showing higher decisiveness and order‑invariance on benchmarks, but revealing that consistency under reordering does not equate to valid moral decisions.

33. Route-and-Generate: KL-Regularized Router Improvement for Reasoning Routing in LLMs

    Tien-Phat Nguyen, Truong Nguyen, Thanh-Thien Le et al. · arXiv:2610.30351

    AI summaryRoute-and-Generate trains a neural router to allocate a fixed reasoning budget across multiple formats, using KL‑regularized policy improvement. The resulting LLM produces concise answers for easy questions and longer reasoning for hard ones, improving accuracy–efficiency trade‑offs on reasoning benchmarks.

34. Probing and Steering for Safety-Relevant Behaviours in Large Language Models

    Nimun Bajwa, Massimo Poesio · arXiv:2610.30363

    AI summaryThe paper studies how activation probes—directions in a Llama‑3.1‑8B‑Instruct residual stream—can detect and steer safety‑relevant behaviours (deception, high‑stakes awareness, sycophancy). It shows that detection works best with logistic regression, steering with difference‑of‑means, but the pattern reverses for sycophancy, with performance also depending on token‑pooling and layer choice.

35. MemPlan: Memory-Conditioned PDDL Planning for Partially Observable Text Environments

    Tengwei Song, Robert Hoehndorf · arXiv:2610.30379

    AI summaryMemPlan uses an LLM as a memory‑conditioned bridge to PDDL planning in partially observable text worlds, reconstructing problems each step and applying positive/negative memory to prune invalid hypotheses. It outperforms language‑only and PDDL baselines in success and efficiency while cutting token cost.

36. Soft Ontology Priors over Latent Concept Posteriors for Weakly Supervised Neuro-Symbolic Learning

    Tengwei Song, Robert Hoehndorf · arXiv:2610.30391

    AI summaryInjects a soft ontology prior into the latent concept posterior of weakly supervised neuro-symbolic models, producing a smooth probabilistic circuit that bounds shortcut reasoning and remains robust to noisy axioms, validated on BDD-OIA and synthetic NeSy tasks.

37. AcoustiPrint: Physics-Guided Neural Design of Manufacturable Acoustic Structures

    Shanmuka Sadhu, Harshvardhan C. Takawale, Yehia Zakaria et al. · arXiv:2610.30411

    AI summaryAcoustiPrint uses implicit neural representations and surrogate gradients to jointly optimize 3D acoustic structures while enforcing discrete manufacturing constraints. It achieves superior focal accuracy in transcranial focused ultrasound compared to existing baselines.

38. Agent Mosaic: Recognizing Confidentiality Leaks in Agentic LLM Systems

    Daniel Ruiz, Luca Luceri, Emilio Ferrara · arXiv:2610.30418

    AI summaryThe paper introduces Agent Mosaic, a study of mosaic‑style confidentiality leaks in LLM agents, measuring 7.08% disclosure of hidden composites in simulation with GPT‑5.6‑Sol and 32.37% even after safeguards, showing a non‑negligible leak risk.

39. High-Value Training Data from Noisy, Diverse Sources

    Hedi Driss, Evan Frick, Anastasios N. Angelopoulos et al. · arXiv:2610.30421

    AI summaryThe paper introduces scalable data‑weighting via metagradients, learning parametric rules from a small high‑quality target set and transferring them to larger models and streaming data. It shows that learned weights improve reward‑model performance and combine human and synthetic signals better than either alone.

40. RoleFlow: Self-Evolving Multi-Role Agentic Collaboration via Structural Adaptation

    Yi Ge, Pan Wang, Xuehao Cui et al. · arXiv:2610.30429

    AI summaryRoleFlow automatically refines multi-agent collaboration structures via single-token edits and on-demand role synthesis, using SECPO to credit each edit against its no-edit counterpart. It outperforms prior methods by 4.7 average points, achieving state‑of‑the‑art results on QA, math, web, and code tasks.

41. Self-Assessment Reinforcement Learning for Search-State Awareness in Retrieval-Augmented Large Language Models

    Taewhoo Lee, Minju Song, Jungwoo Park et al. · arXiv:2610.30435

    AI summaryThis paper introduces Self-Assessment Reinforcement Learning, a framework that trains retrieval-augmented language models to track their search states and determine missing information. The method uses a dual reward design, improving multi-hop reasoning and deep research benchmarks by over 27 percent without inference-time scaffolding.

42. Secure Agentic AI and Multimodal Intelligence in 6G Communication Systems

    Satyam Gaba, Devanshu Motwani, Siva Sai et al. · arXiv:2610.30436

    AI summaryThe paper surveys agentic multimodal AI in 6G, mapping perception, memory, tool use, and multi‑agent orchestration, and categorizes threats across sensing, cognition, action, memory, and communication layers. It reviews defenses like prompt/memory protection, inter‑agent trust, and privacy‑preserving learning, outlining future research directions.

43. SpecMAGIC: Specification-Guided Multi-Task Generalization in Multi-Agent Reinforcement Learning

    Zijian Guo, İlker Iş{\i}k, H. M. Sabbir Ahmad et al. · arXiv:2610.30443

    AI summarySpecMAGIC introduces a specification language that combines temporal logic with coordination operators, and a hierarchical planner that decomposes tasks into reusable subtasks. The framework achieves zero‑shot generalization to unseen tasks, outperforming baselines on multi‑agent navigation while scaling with more agents.

44. A Three-Pillar Governance Framework for Safe and Trustworthy AI Agents

    Edward C. Cheng, Jeshua Cheng, Alice Siu · arXiv:2610.30445

    AI summaryThis paper introduces a Three-Pillar framework for safe AI agents, combining transparency, accountability, and trustworthiness. It outlines a staged autonomy pathway with progressive validation, structured human oversight, and auditable controls that enable controlled, aligned agent deployment.

45. Same Beliefs, Different Policies: Learning Reusable Representations for Early Stopping in LLM Reasoning

    Jingwei Ni, Dmitry Abulkhanov, Vladislav Smirnov et al. · arXiv:2610.30450

    AI summaryThe paper introduces BELT, a method that separates outcome prediction from stopping control in reasoning LLMs to optimize compute. BELT reduces generated tokens by up to 51% without sacrificing accuracy while transferring effectively to new domains with minimal training cost.

46. aggmetrics: Aggregating Classification Performance Measures for Machine Learning

    Ronaldo Cristiano Prati · arXiv:2610.30456

    AI summaryaggmetrics extends binary classification metrics to multiclass, multilabel, ordinal, cost‑sensitive, soft‑label, and cross‑validation settings by offering four aggregation operators. It auto‑detects metric properties, warns of mis‑applications, and provides a test suite verifying the theory.

47. Short-to-Long: Detached Features for Long-Horizon Value Learning

    Tong Che, Boris Ivanovic, Marco Pavone · arXiv:2610.30461

    AI summaryShort-to-Long (S2L) trains short‑discount source networks to produce features, then uses these detached features to guide a long‑discount behavior learner; across Atari and continuous‑control tasks, S2L improves median performance on 40 of 57 games and reduces target drift.

48. SplatLab: Autonomous Scientific Discovery for 3DGS Research

    Seemandhar Jain, Manmohan Chandraker · arXiv:2610.30468

    AI summarySplatLab automatically reproduces 3D Gaussian Splatting baselines, proposes and evaluates extensions, and writes fully traceable manuscripts. It improves PSNR by up to 3.78 dB over published results while compressing research cycles from months to days.

49. Rethinking SAE Evaluation: An Atomic Interpretable Unit Framework Reveals Hidden Polysemanticity and Redundancy

    Riteng Zhang, Xiaoqian Wang · arXiv:2610.30474

    AI summaryThis paper introduces an atomic interpretable unit framework for evaluating sparse autoencoders, revealing that standard metrics miss hidden polysemanticity in dominant features. The results show that widely used SAE features are often functionally redundant and less interpretable where it matters most.

50. Externalise the Memory: A Pre-Registered Study of Where an Embodied Agent's Experience Should Live

    Cheng-Yin Tu · arXiv:2610.30478

    AI summaryMemory-augmented teachers can be distilled into policies, but the benefit appears only when the teacher’s retrieval buffer is externally attached. Attaching the frozen memory to student weights boosts performance by ~34 pp, supporting externalising experience rather than baking it into weights.

51. Why Bother Me Again? Benchmarking Personal Agent Memory for Interaction Preferences

    Jialin Li, Zhenhao Chen, Hanjun Luo et al. · arXiv:2610.30485

    AI summaryThe authors introduce MemPABench, an interactive benchmark for evaluating how personal agents remember and adapt to user interaction preferences. They find that current memory systems struggle to retrieve preferences, adapt to changes, and overcome default agent behaviors.

52. From Single-Actor Environments to Multi-Actor Event Worlds: Can LLMs Be Event World Models?

    Juntong Ni, Zewen Liu, Bohan Wang et al. · arXiv:2610.30488

    AI summaryThis paper formalizes multi‑actor systems as event worlds and evaluates LLMs on predicting future observations from event logs. Fine‑tuned LLMs perform well when all actors’ actions are known, but struggle with unseen actions or strong external influences, limiting multi‑step planning.

53. What the Trace Shows and What the Label Says: Auditing Behavioral Detectors in Agent Evaluation

    Idunnuoluwa A. Adeniji, Shilpika, Michael E. Papka · arXiv:2610.30491

    AI summaryLLM agents are evaluated using trace-based detectors that label final correctness, re-verification, and fault propagation. The study finds systematic disagreements, especially silent tool‑output override, and shows that human review matches automated labels only for correctness and fault impact, highlighting the need for trace transparency.

54. From Prose to Properties: Testing Research Claims with ARATest for Agent-Native Science

    Yifeng He, Jiachen Liu, Hao Chen · arXiv:2610.30495

    AI summaryARATest converts prose research claims into explicit property skeletons and automatically tests them, revealing 8 violations (confirmed errors) and 14 inconclusive cases out of 243 claims. It outperformed LLM judges, detecting 553 seeded faults without false positives.

55. Can LLMs Improve Their Own Harness? A Controlled Study on ARC-AGI-3

    Jens Lehmann, Sahar Vahdati · arXiv:2610.30499

    AI summaryLLMs can build a full harness for the ARC‑AGI‑3 benchmark, but after two self‑made revisions performance barely improves—average held‑out RHAE rises only 2.8 points, and cost reductions often hurt task performance.

56. PhysibleMDM: Enforcing Inference-Time Physical Feasibility Constraints on Motion Diffusion Models via Noise Optimization

    Maximilian Stölzle, David Müller, Agon Serifi et al. · arXiv:2610.30500

    AI summaryPhysibleMDM enforces physical feasibility in motion diffusion models by optimizing diffusion noise at inference, using surrogate kinematic and dynamic constraints evaluated via a cone program and Sep-CMA-ES. It achieves up to 92% reduction in self-collision and 89.5% acceleration error on SMPL and bipedal robots.

57. Behind Aggregate Scores: What Systematically Varying Arena-Hard's LLM Judge Reveals

    Louise Yorke · arXiv:2610.30512

    AI summaryThe paper evaluates how repeatable LLM judges are by running GPT‑4.1 on 500 Arena‑Hard prompts under eight instruction variants. It finds that aggregate scores can be stable while individual verdicts vary widely, highlighting the need to report prompt‑level agreement.

58. Protegrity-PII Technical Report

    Hardeep Bassi, Hui Ma, Luís Seabra et al. · arXiv:2610.30518

    AI summaryProtegrity-PII is a new PII detection model for structured and unstructured text, achieving typed recall 0.915 and precision 0.907 on the PII‑SCOPE benchmark. It runs at 0.49 s per million characters, $0.0005 cost, and generalizes better than state‑of‑the‑art models.

59. LLM2Cert: LLM Synthesis of Certificates with SMT/Lean Verification

    Zhipeng Cao, Peiyang Song, Nikolaus Vertovec et al. · arXiv:2610.30524

    Lyapunov functions and barrier certificates give state-space guarantees of stability and safety, but synthesizing and verifying them---especially for nonlinear, high-dimensional systems---remains an open challenge. Existing pipelines verify candidates with numerical solvers…

60. Decision Making and Ride Comfort Evaluation Based on Lane-Change Prediction for Autonomous Vehicles

    Mohamed Manzour, Aditya Kumar, Rubén Izquierdo et al. · arXiv:2610.30528

    Predicting surrounding vehicles' lane-change intentions enables autonomous vehicles to anticipate potential hazards and respond proactively rather than relying solely on reactive maneuvers. This study investigates the benefits of integrating lane-change prediction, based on…

61. PLE: Physical-Engineering's Last Exam

    Rujun Han, Yufan Zhuang, Arijit Ray et al. · arXiv:2610.30537

    AI summaryPLE introduces an 85‑task benchmark for professional physical engineering, requiring realistic, billable deliverables across electrical, civil, structural, and mechanical domains. The strongest LLM agents score only 50.6 % after 27 M tokens and 63 min per task, highlighting a large capability gap.

62. Governed Self-Evolution for Autonomous Investing: A Multi-Agent LLM System with an Invariant Core

    Yebo Feng, Jingyang Dai, Liangrui Pan et al. · arXiv:2610.30557

    AI summarySteward is a multi‑agent LLM system for autonomous investing that uses a type‑system‑bounded invariant core to guard against self‑editing. In a five‑month deployment it limited sector drawdown to 0.3% loss while outperforming a single‑model baseline.

63. Lifetime-valid decisions by continually evolving AI analysts

    Xiaojing Sun, Yuhan Zeng, Xiao Wang · arXiv:2610.30562

    AI agents increasingly act as analysts whose decision can often be cast as hypothesis tests. Classical tests control the Type I error rate, the probability of rejecting a true null hypothesis, for fixed procedures. However, AI analysts introduce internal randomness and can…

64. Diagnosing Personalization Failures in Mobile GUI Agents

    Zhixin Lin, Dongliang Xu, Jungang Li et al. · arXiv:2610.30573

    AI summaryPrefDiag‑Bench diagnoses how and when personal preferences affect mobile GUI agent execution, revealing that current agents succeed only 47% on Process preferences and struggle with Ranking and Process types. Explicit evidence provision boosts success by up to 7.06%.

65. Bridging the Closed-World Gap: Interpretable Anomaly Detection via a Hybrid TSK-FCM Ensemble on the BETH Dataset

    Lohith Srikanth Pentapalli, Josette Riep, Justin Ouwerkerk et al. · arXiv:2610.30576

    AI summaryHybrid TSK‑FCM ensemble uses a supervised TSK classifier for known insider threats and a parallel Fuzzy C‑Means cluster for geometric anomalies, with a Min Distance metric mitigating high‑dimensional fuzziness. It achieves 0.9951 AUROC on BETH, 17.1% above Isolation Forest.

66. Model Organisms of Collusion: Corrective Transfer in Untrusted Monitors

    Juan-Pablo Rivera, Ari Spiesberger, Juan J Vazquez et al. · arXiv:2610.30582

    AI summaryThis paper investigates whether fine-tuning an AI monitor on one instance of caught collusion reduces unseen collusion with itself. The main result is that correcting the monitor on a single caught case reduces held-out collusion from over 90% down to 14-20%.

67. Visual Grounding First, Multimodal In-context Learning Follows

    Minhyuk Seo, Minjae Lee, Chaeeun Lee et al. · arXiv:2610.30592

    AI summaryProCoRe uses contrastive reinforcement learning on multimodal data to improve visual grounding in multimodal LLMs, boosting ICL classification accuracy by 5.6% and caption ROUGE‑L by 13.8% over Qwen‑VL‑3‑8B without ICL training data.

68. Impacts of generative AI skill development on academic adaptation and learning outcomes among university students

    Lei Fan, Fangxue Liu · arXiv:2610.30599

    Understanding how university students perceive the development of their generative artificial intelligence (GenAI) skills over time can clarify how GenAI use relates to learning and academic outcomes. This study examined perceived skill-development trajectories, changes in…

69. Better Reasoning, Not Just Bigger Models: A Structure-Grounded Framework for Efficient and Reliable Inference

    Amirhossein Aghaei, Liangyan Li, Shiyi Yang et al. · arXiv:2610.30611

    AI summarySGUR uses a training‑free graph‑matching and active query strategy to ground inference in structured hypothesis spaces, reducing uncertainty without large models. On a medical diagnosis task, SGUR achieves 92.5 % Hit@3, beating a Graph‑RAG baseline by 9.4 % while using five times fewer tokens.

70. Future of Time Series Anomaly Detection: From Historical Review to Unified Outlook

    Zhijie Zhong, Zhiwen Yu, Pengyu Li et al. · arXiv:2610.30628

    AI summaryThis paper surveys recent TSAD architectures—diffusion, state‑space, LLMs—introduces a refined anomaly taxonomy, evaluates metrics via extensive experiments, and releases an open benchmark, highlighting challenges in multivariate detection, zero‑shot transfer, and interpretability.

71. Interpreting Model Reasoning through Behavioral Counterfactuals

    Shawn Im, James Oldfield, Adel Bibi et al. · arXiv:2610.30630

    AI summaryThe paper introduces behavioral counterfactuals, editing the model’s residual stream to match next-token distributions between a source and counterfactual prompt. It shows this method reliably recovers counterfactual behavior, identifies key causal locations, and reveals how suggestions subtly shift reasoning in large LLMs.

72. Large language models show systematic failures in incorporating evolving medical knowledge

    Min Zeng, Zaifu Zhan, Kai Yu et al. · arXiv:2610.30631

    AI summaryLarge language models recognize 90.8% of guideline revisions but only 58.5% fully incorporate the updates, leaving a 32‑percentage‑point gap between recognition and action. A revision‑salience cue cuts outdated content but does not markedly improve full incorporation.

73. Asking Less, Adapting Better: Feedback-Efficient Test-Time Adaptation for Vision-and-Language Navigation

    Zixu Wang, Huaquan Zhang, Shoukai Xu et al. · arXiv:2610.30643

    AI summaryFETA uses a two‑stage selection to ask humans for only the most informative action in a trajectory, then updates the policy at that step with a discrepancy‑aware entropy loss. It outperforms prior test‑time adaptation methods while reducing human queries.

74. Speculative Acceptance Does Not Decay with Position: A Full-Length Diagnostic on Long Chain-of-Thought with Independent Drafts

    Zhiming Meng · arXiv:2610.30644

    AI summaryThe paper measures speculative-decoding acceptance rates across full-length chain-of-thoughts on MATH‑500 with two independent drafts, finding no decay with position and that rejections cluster around high‑entropy reflection tokens, not the chain length.

75. One-shot federated unsupervised domain adaptation with Smoothed Knowledge Distillation and teacher refinement

    Ali Abedi, Q. M. Jonathan Wu, Ning Zhang et al. · arXiv:2610.30646

    AI summaryOne-shot federated unsupervised domain adaptation (FUDA) using Smoothed Knowledge Distillation (SKD) and Exponential Moving Average teacher refinement (ETR) trains a student on target data with smoothed pseudo‑labels, then refines source models via EMA, achieving competitive accuracy with a single communication round.

76. Anticipating Social Consequences with Distributional World Models

    Viet-Thanh Pham, Thuy-Trang Vu, Zhuang Li et al. · arXiv:2610.30657

    AI summaryDSWM predicts multiple plausible social responses and their probabilities in one output, using a 4B model trained on simulated interactions, and shows superior performance across diverse environments by allowing agents to choose actions based on a richer set of anticipated consequences.

77. A Topology-Guided GeoAI Agentic Framework for Urban Flood Screening

    Yutian Tao, Haoan Feng, Xin Xu et al. · arXiv:2610.30660

    AI summaryThe paper introduces a topology-guided GeoAI framework that ranks urban flood basins by extracting depressions via discrete Morse theory and scoring them with a flood-priority metric; in College Park, 70 basins were selected, 77% matching an D8 flow model, with human checkpoints for threshold and weighting.

78. Group Relative Policy Optimisation Improves Multilingual Speech Recognition in Low-Resource Languages

    Tim Wenjie Hu, Evelyn Nafula Ouma, Benjamin Akera et al. · arXiv:2610.30663

    AI summaryGroup Relative Policy Optimisation (GRPO) is applied to a non‑LLM encoder‑decoder ASR model (Whisper‑large‑v3) for 51 African languages. GRPO improves low‑resource performance, yielding SunflowerASR’s lowest published macro‑averaged WER on AfriVox‑v2 (0.227) and SimbaBench (0.372).

79. Spectral Smoothing and Spiking Perturbations Reconfigure Capability Profiles and Selective Task Performance in Large Language Models

    Qince Luan, Jianhui Xu, Yang Xu · arXiv:2610.30676

    AI summaryThe authors introduce a lightweight post-hoc spectral intervention on transformer weight matrices that stretches or compresses singular-value spectra while preserving singular vectors. They find that these perturbations induce structured shifts in model capabilities, demonstrating that spectral properties causally encode trade-offs between factual knowledge, language understanding, and reasoning.

80. Scaling Laws for Safety Probes

    Yuchen Wu (University of Washington), Jerry Wei (Anthropic) · arXiv:2610.30678

    AI summaryThe paper shows that safety‑probe accuracy scales predictably with model size and label count, following two power‑law relationships that hold across tasks, families, and architectures. Using these laws, researchers can choose probe size, label budget, and architecture within 0.02 error of actual performance.

81. From Clinical Judgment to Checkable Rules: Clinician-Led Guardrails for Mental-Health Chatbots

    Chenhan Lyu, Kit Wislocki, Sarah Elizabeth Stoeckl et al. · arXiv:2610.30690

    AI summaryClinicians design a verified rule‑based guardrail for LLM mental‑health chatbots that replaces ambiguous safety judgments with 16 fact‑based rules. The guardrail eliminates crisis‑resource false alarms while its few remaining errors trace directly to specific rule‑fact conflicts.

82. LoRA Superposition across Pretrained Models

    Keigo Watanabe, Hiroshi Kera, Kazuhiko Kawamoto · arXiv:2610.30691

    AI summaryLoRA superposition sums LoRAs trained on different ViT backbones for the same task, improving all 56 backbone‑task pairs with a 30.21‑point mean accuracy gain, nearly matching individually trained LoRAs and supporting non‑interfering components.

83. Disentangling Robustness and Sycophancy in Multilingual LLMs

    Rhitabrat Pokharel, Ameeta Agrawal · arXiv:2610.30696

    AI summaryThe paper introduces the SC‑matrix to separate model competence from sycophancy in multilingual LLMs, showing that aggregate sycophancy rates hide different failure types, high competence doesn’t guarantee resistance, and reasoning‑based mitigations beat directive methods.

84. Governance-State Corruption in Long-Term Memory for LLM Agents: A Cross-Model Stress Test

    Bowen Liu, Youting Wang, Bonan Shen et al. · arXiv:2610.30698

    AI summaryGovernance-state corruption causes LLM agents to misapply memories when their validity changes, affecting 96% of test pairs; oracle enforcement restores accuracy, yielding 40‑point improvement in LongMemEval binary correctness and 30‑point gain in Memora FAMA.

85. ULME: Predictive Coupling for Mutually Reinforcing Embedding and Language Modeling

    Yang Liu, Zhiqiang Yuan, Shu-Dong Huang et al. · arXiv:2610.30705

    AI summaryULME post‑trains a causal decoder‑only LLM to unify text embedding and generation by adding learnable slots that receive a contrastive semantic embedding and a predictive reasoning embedding, enabling the generator to shape the embedder. Across Qwen variants, ULME boosts MTEB by 3.3‑4.1 points and tightens retrieval benchmarks while matching generation accuracy, showing mutually reinforcing capabilities.

86. Meaning Contracts: Local Semantic Geometry in Large Language Models

    Andrés Daniel Godoy Ortiz · arXiv:2610.30715

    AI summaryThe paper probes the fine‑scale geometry of meaning in LLMs using cross‑lingual and refinement probes, revealing that semantic factors form invariant cores with structured, realization‑specific variation that contracts into tighter, anisotropic neighborhoods when meanings are refined.

87. Learning from Next-Frame Prediction: Towards Unified and Scalable Visual Representation Learning

    Yifeng Yao, Jinghan Li, Zhihao Zhang et al. · arXiv:2610.30724

    AI summaryVANE trains a causal encoder to predict the embedding of the next video frame, then decodes patches for detail. It scales with data and model size, achieving state‑of‑the‑art results on anticipation, recognition, and dense perception, notably 7.1% mCR@5 better than V‑JEPA on EPIC‑KITCHENS.

88. Faithful Latent Harm Monitors Work No Better Than the Models Own Behaviour at a Matched False Alarm Rate

    Abhishek Mishra, Sagnik Chatterjee, Suyash Maniyar · arXiv:2610.30727

    AI summaryLatent safety monitors, which read internal activations of LLMs to detect harmful intent, perform no better than the model’s own refusal decisions. Across 15 LLMs and 38 probes, TF‑IDF or sentence‑embedding classifiers match or exceed latent monitors, especially under distribution shift.

89. AgentTriage: Replayable Security Diagnosis Across Heterogeneous Agent Runtimes

    Jie Ying, Xutao Zhang, Ningyuan Chen et al. · arXiv:2610.30738

    AI summaryAgentTriage provides a replayable security diagnosis framework for heterogeneous tool-use agent runtimes using standardized execution contracts and provenance graphs. It improves joint diagnosis accuracy to 0.858 while ensuring claims are strictly bound to recorded evidence without language-model judgment.

90. How Firm Are AI Models' Ethical Boundaries? A Pay-Stub Editing Test of 15 Models

    Simiao Ren, Xingyu Shen, Tommy Duong et al. · arXiv:2610.30751

    AI summaryThe study shows that small wording changes can flip 15 AI models from refusing to comply with a pay‑stub editing request, with a 66.2% compliance rate for “printing error” phrasing, while some models (Astra) always refuse and others (Fable, Opus, Muse Spark) never comply.

91. The Delegation Divide: An Open Index of AI-Agent Delegability across Occupations and Countries

    Cristian Espinal Maya (Universidad Nacional de Colombia), Santiago Jiménez Londoño (Universidad EAFIT) · arXiv:2610.30752

    35.2% of task time in world employment is at least partly delegable to AI agents in 2026, and only 2.5% can be handed over in full. Existing measures score exposure to generative AI, mostly for the United States. We build the Agent Delegability Index (ADI): two language models…

92. Is Complex Agentic Memory Always Needed? Naive RAG Is Stronger Than We Thought

    Zhengye Bao, Shengtao Zhang, Haokai Zhang et al. · arXiv:2610.30755

    AI summaryNaive Raw-RAG, which simply dense-retrieves and reranks raw dialogue history, outperforms complex agentic memories on six of seven benchmarks, achieving 0.13–5.53 point gains. The study shows raw evidence coverage matters more than sophisticated memory transforms.

93. Can Loan Processors and Underwriters Spot AI-Edited Statements? An Empirical Study

    Simiao Ren, Xingyu Shen, Tommy Duong et al. · arXiv:2610.30758

    AI summaryResearchers tested whether experienced loan underwriters spot AI‑edited loan documents. In a small study, 10.6% of judgments flagged suspicious identity changes, showing that about one in ten reviewers detected forgery, highlighting the need for consistency checks and source verification.

94. AgentDP: Towards Automated Model Deployment for Heterogeneous Embedded AI Applications

    Leming Shen, Qiang Yang, Xinyu Huang et al. · arXiv:2610.30760

    AI summaryAgentDP automatically optimizes and adapts pre‑trained AI models for heterogeneous embedded devices using LLM agents, standardizing requirements, partitioning operators for diverse inference engines, and reusing hyperparameters. It cuts adaptation time by up to 80% and boosts accuracy by >13% over baselines.

95. Abstract Priorities, Concrete Rewards: How Stated Preferences Relate to What People Reward in AI Responses

    Zihan Chen, Lei Nico Zheng, Di Zhu · arXiv:2610.30763

    AI summaryThis study compares users' stated AI preferences with their actual response ratings using the PRISM dataset. It finds that stated priorities poorly predict the concrete features that actually drive rewards, complicating effective AI personalization.

96. Scaling Laws for Generative Reward Models

    Angel Raychev, Nathan Lile, Rafael Rafailov · arXiv:2610.30765

    AI summaryGenerative reward models trained with GRPO outperform SFT in static accuracy, but Answer‑Only rewarders give higher policy Elo when used with DPO, especially when rewarders are larger than the policy. The study shows that fixed‑distribution judging accuracy does not predict downstream optimization performance.

97. LLM Multi-Agent Systems Are Vulnerable to Herding and Runaway Cascades

    Farnoosh Hashemi, Shiyu Ji, Joice Chen et al. · arXiv:2610.30769

    AI summaryThe study tests LLM multi‑agent systems for herding by letting agents recommend songs, with or without cumulative counts. It finds exposure triggers runaway cascades, producing extreme, unpredictable recommendations that diverge from intrinsic appeal, and LLMs herd more than humans.

98. ProQuant: Preserving Expert Quality over Quantity in Ultra-Low-Bit Quantization of Mixture-of-Experts Models

    Jiaqi Chen, Xianglong Yan, Chengzhu Bao et al. · arXiv:2610.30779

    AI summaryProQuant applies weight‑only post‑training quantization to MoE language models, jointly pruning experts and assigning mixed bit widths (PoP) and fine‑tuning with Routing‑Consistent RMSNorm (RoCo) to curb quantization errors and routing drift, achieving superior generation quality under tight budgets.

99. An Introduction to World Models for Generalist Agents

    Zhihui Zhu · arXiv:2610.30789

    AI summaryThe paper clarifies what constitutes a “world model” for generalist agents, mapping various predictive, simulator, and joint action‑observation models onto the POMDP framework. It shows that policy and world modeling can be unified via conditional prediction, offering a principled foundation for future research.

100. Agentic Knowledge-Model Co-Evolution via Consistency Regularization

     Sirui Chen, Zhining Liu, Chao Zhang et al. · arXiv:2610.30790

     AI summarySPLICE co‑evolves LLM agents and their internal knowledge by generating multiple perturbations of experience analysis and selecting knowledge that yields consistent utility gains. It boosts accuracy by up to 7.1 points, cuts correct solution tokens by 44%, and transfers well across model sizes.

101. VLTest: Finding Joint Image-Text Failures in Vision-Language Models

     Bin Duan, Jintao Lin, Guowei Yang · arXiv:2610.30794

     AI summaryVLTest is a black-box testing framework for vision-language models that systematically mutates both images and text to isolate individual and joint failure modes. It identifies significantly more confirmed failures than existing baselines by rigorously filtering input-preserving modifications and enforcing a four-cell testing strategy.

102. What Lies Beyond Perception: A Partition Theory of the Limits of AI Evaluation and Oversight

     Ahmed M. Mahran · arXiv:2610.30803

     AI summaryThis paper introduces a partition theory of AI evaluation showing that an evaluator's judgment is fundamentally bounded by its perception. The main result proves that behavioral evaluation can never certify an upper bound on a model's capabilities or reliably detect hidden capabilities without assuming the model's cooperation.

103. TESSA: A Governance-Oriented Custom GPT for SME Guidance under the EU AI Act: Design, Implementation, and Evaluation

     Chalisa Veesommai Sillberga, Kai-Kristian Kemella, Pekka Sillberga et al. · arXiv:2610.30822

     AI summaryTESSA is a custom GPT trained to give SMEs EU AI Act‑compliant advice; in 100 scenario tests it matched baseline ChatGPT’s guidance 90 % of the time versus 83 % for ChatGPT, and achieved higher consistency and human‑validated alignment.

104. Neural Spectral Capacity with Residual Context for Heterogeneous Resource Allocation within Transformers

     Chenyu Zhu, Ruoyu Zhao, Zhichao Lu · arXiv:2610.30823

     AI summaryResNSC augments Neural Spectral Capacity with residual-aware statistics and a dynamic programming allocator that jointly sets residual and feed‑forward widths under a budget. On 121M/374M Transformers, ResNSC‑DP yields lower perplexity than uniform or tapered baselines and improves LLaMA‑7B pruning accuracy by 3–5 pp.

105. One Evaluation Is Enough: Inversion-Free Image Editing with Trajectory Correction

     Lifan Jiang, Taiyang Chen, Senlin cheng et al. · arXiv:2610.30827

     AI summaryThe paper introduces Endpoint-Calibrated Transport (ECT), an inversion-free image editing method requiring only a single network evaluation. ECT achieves high-quality edits in one step by combining local tangent integration with a constrained endpoint correction.

106. Semi-Linear Gaussian Bayesian Networks: Enhancing Flexibility in Node Dependencies via penalized B-Splines

     Aitor Oviedo-Madrid, Mabel Morales-Otero, Jose Gonzalez-Gomariz et al. · arXiv:2610.30838

     AI summaryThe paper introduces Semi-Linear Gaussian Bayesian Networks, which use penalized B-splines within generalized additive models to capture nonlinear dependencies while maintaining Gaussian conditional distributions. Experiments show this framework improves goodness of fit and structure recovery over traditional models while retaining computational efficiency.

107. Bifurcations from Weight Space: JEPAWG on Parametric Dynamical Systems

     Tobias Göbel, Miranda C. N. Cheng · arXiv:2610.30847

     AI summaryThe paper trains neural forecasters on grids of dynamical systems, forming a weight‑parameter zoo, and applies JEPAWG to embed these weights. It shows JEPAWG can forecast, locate bifurcations (Hopf, period‑doubling, Feigenbaum constant, global), and reveal phase transitions.

108. ADAPT for Science: A Multi-Agent System for Self-Evolving Scientific Optimization

     Zhaoran Feng, Yuchen Ni, Jianan Liu et al. · arXiv:2610.30851

     AI summaryThis paper introduces ADAPT for Science, a multi-agent framework that couples spatial search strategies with self-evolving optimization procedures. Evaluated on gravitational wave detector design, ADAPT reduces optimization loss by over 54% compared to classical and agentic baselines under matched budgets.

109. Direct Forcing: A Single-Stage Training Recipe for Few-Step Autoregressive Text-to-Video Generation

     Haojun Xu, Jie Huang, Xin Lu et al. · arXiv:2610.30861

     AI summaryDirect Forcing trains a few‑step autoregressive video model in one stage by jointly aligning reference videos with generated rollouts (causal continuation alignment) and matching their representation distributions (rollout distribution matching) using dynamic gradient balancing. Starting from Wan2.1‑1.3B, it learns a 4‑step generator in 250 updates, achieving 84.63 VBench Total and 86.01 Quality.

110. Separate Field Scoring Protects Unused Tools When Agents Pool Tool-Use Experience

     Abhijit Chakraborty · arXiv:2610.30867

     Agents that call tools from a shared catalog can pool what they learn about which tool serves which request. We measure what pooled usage examples do to tool retrieval, reporting separately the held-out tools, for which no agent has usage, and all other tools. Adding usage…

111. MOER: Mix-Policy Optimization under Expert Prefix Guidance with Selective Rollback

     Xuyang Zhong, Jiuding Yang, Yiwen Guo et al. · arXiv:2610.30873

     AI summaryMOER blends off‑policy expert prefixes with live exploration by progressively shortening the guided chain and rolling back failures to the original expert trace. It achieves top scores on reasoning benchmarks, outperforming baseline mix‑policy methods and maintaining stability across models and scales.

112. Parameter Learning in Active Inference via Variational Message Passing

     Benedikt Waldeck · arXiv:2610.30899

     AI summaryThis paper extends active inference planning during parameter learning by using factor-specific parameter priors and sum-product message passing. The resulting method preserves information-gain corrections that standard mean-field approximations destroy, enabling agents to successfully learn models and navigate safely in grid POMDP environments.

113. Beyond Phenotypic Similarity: Relation Refinement and Selective Population Graph Learning for Multimodal Brain Disorder Diagnosis

     Yinyu Lan · arXiv:2610.30902

     AI summaryPopDRP refines subject relations for brain‑disorder diagnosis by aligning attribute similarities with diagnostic labels, then learns population representations through a selective hierarchical state‑space graph model. It improves accuracy and AUC on ABIDE and ADHD‑200 versus baseline methods.

114. Query-Time Calibration of CLIP Features for Open-Vocabulary 3D Segmentation

     Dominik Engel, Yousef Rajeh, Leon Sick et al. · arXiv:2610.30908

     AI summaryLifting CLIP features into 3D Gaussian scenes causes directional concentration; the paper introduces query‑time calibration that removes a shared scene direction, whitens text prototypes, and applies soft clustering/CSLS, boosting mean IoU on seven methods and four datasets without retraining.

115. TwinStore: A Synthetic Consumer Platform Using Vision--Language Models

     Minhyeok Lee · arXiv:2610.30909

     AI summaryTwinStore creates a digital twin of a retail store using vision–language models, simulating shopper trips and purchases from floor plans or photos. Simulated receipts match real household variation (0.017–0.018 divergence) and price elasticities match field measurements, validated on eleven datasets.

116. MIG-COW: Transferable Adversarial Attacks on Deepfake Detectors via Gradient Decomposition

     Wonjune Seo, Joonhyuk Baek, Yeseong Jung et al. · arXiv:2610.30914

     Despite recent advances, deepfake detectors remain vulnerable to adversarial examples, particularly in diverse, real-world settings. We propose MIG-COW, a novel adversarial attack framework that generates highly generalizable and visually imperceptible adversarial examples. By…

117. Harnessing Language Models as Optimization Modeling Agents

     Yunxiang Luo, Xin Xu, Tianhao Chen et al. · arXiv:2610.30917

     AI summaryThe paper introduces OPTICYCLE and OPTISTATE, lightweight single‑model inference harnesses that use execution‑guided revision to translate natural‑language optimization problems into solver programs, achieving up to 49.46% higher macro accuracy and outperforming agentic workflows and post‑trained models without extra training.

118. DConstrain: Compiling Context-Parameterized Transitions for Diffusion Language Model Constrained Decoding

     Shidong Cao, Jing Ma, Wenhao Chai et al. · arXiv:2610.30920

     Diffusion language models (DLMs) decode multiple tokens in parallel, so individually valid tokens can be incompatible together under a schema. Exact joint inference over a grammar graph enforces compatibility, but it requires token-labeled successors across many hypothetical…

119. Low-Latency MoE Inference in Space Data Centers: A Wireless-Power-Transfer Perspective

     Haotian Zheng, Zhanwei Wang, Kaibin Huang · arXiv:2610.30923

     Space data centers (SpaceDCs) are emerging as platforms for large language models (LLMs), supported by continuous solar-energy harvesting and high-speed interconnections. The mixture-of-experts (MoE) architecture supports distributed LLM inference by deploying experts across…

120. The Geometry of Meaning in Neural Manifolds

     Valentino Maiorca · arXiv:2610.30938

     Just as different languages express the same concept through different words, neural networks express similar functions through incompatible coordinate systems. Different architectures, random initializations, and optimization trajectories produce incompatible parameterizations…

121. Vision Transformers for Estimating Irradiance Using Data Scarce Sky Images

     David Hamlyn, Sunny Chaudhary, Tasmiat Rahman · arXiv:2610.30940

     Accurate estimation of diffuse horizontal irradiance (DHI) is critical for optimising photovoltaic system performance and energy forecasting yet remains challenging in regions lacking comprehensive ground-based instrumentation. Recent advancements using Vision Transformers…

122. GateFormer: Forecast-Space Decomposition for Intra-Hour Solar Irradiance Prediction from Sky Images

     David Hamlyn, Sunny Chaudhary, Jonathon Hare et al. · arXiv:2610.30942

     Sub-hourly dispatch, balancing, and storage control increasingly depend on forecasts that anticipate rapid variability in solar generation. Accurate sky-image-based irradiance forecasting requires a combination of visual cloud dynamics with recent irradiance state while adapting…

123. Metacognitive Scaffolds Create Productive Friction for AI-Assisted Knowledge Work

     Viktor Kewenig, Lev Tankelevitch, Sean Rintel et al. · arXiv:2610.30945

     Generative AI imposes new metacognitive demands on knowledge workers as their work shifts from production to planning, evaluation, and orchestration. In a randomized, between-subjects experiment on AI-assisted writing (297 workers, 594 essays), we test dynamically-generated…

124. RetiJev: Fast and Source-Grounded Ophthalmic Decision Models

     Bo Liu, Lianyu Wang, Meng Wang et al. · arXiv:2610.30947

     Ophthalmic clinical decision support requires fast, accurate, and privacy-preserving inference, yet autoregressive large language models (LLMs) incur substantial latency and computational overhead for decisions among predefined actions. We therefore investigate a…

125. OpenAttention: One Percent Is Enough with Tail Proxies

     Hyunwoo Yu, Jongbeom Lee, Hyunbin Kim et al. · arXiv:2610.30951

     Sparse attention reads only a small subset of the key-value cache, and how to treat the unread tail remains a key challenge. Discarding the tail biases the output, and sampling it as previous approaches do fails at reads of 5% and below, where the tail likely holds heavy hitters…

126. Measuring the Reasoning-Action Coupling in Driving Vision-Language-Action Models through Counterfactuals

     Gabriel Manalu, Saad Wazir, Thao Do et al. · arXiv:2610.30953

     Driving vision-language-action (VLA) models bring an appealing form of explainability to a safety-critical domain, since a system that can state the reasons for its maneuvers offers the transparency that deployment and public trust in autonomous driving demand. These models read…

127. Zero-Shot Forecasting of In-Situ Soil Moisture and Temperature with a Time-Series Foundation Model

     Mohammad Talaei, Mohammad Goudarzi, Maille Todd et al. · arXiv:2610.30971

     Soil moisture and temperature govern plant-water availability and microbial activity, making them key indicators for evaluating land-restoration efforts on degraded landscapes. However, the physical and data-driven models normally used to forecast them both need information that…

128. LAST-V2X: Language-Driven Adaptive Scenario Synthesis and Testing for Communication-Induced Failures in V2X Safety Applications

     Keshu Wu, Hao Zhang, Rui Gan et al. · arXiv:2610.30979

     AI summaryLAST‑V2X uses LLMs to turn natural‑language test descriptions into executable V2X scenarios, injects communication faults, and compares degraded versus ground‑truth runs to attribute safety failures to communication. Adaptive refinement finds minimal reproducible failures, boosting discovery over CARLA experiments.

129. ColaBench: Evaluating Memory-Guided Knowledge Acquisition in Personal Agents

     Wenjie Zhang, Zhichao Shi, Xuhui Jiang et al. · arXiv:2610.30989

     Personal agents need to draw on long-term interaction history to retrieve and apply external knowledge suited to a user's current request. Even an answer grounded in relevant documents can be unsuitable when it overlooks the user's circumstances. This creates two connected…

130. Truthful by Omission: Click Thinning for LLM-Native Advertising

     Junhua Liu, Zhicheng Liang, Zejun Zhang et al. · arXiv:2610.30999

     AI summaryThe paper presents a click‑thinning mechanism for LLM‑native ads that is truthful in expectation and guarantees at least half of the optimal click welfare, even with fixed editorial order. It reduces welfare maximization to a matching problem and proves the bound is optimal.

131. SecLog-Bench: A Scalable Benchmark for Cybersecurity Log Reasoning with Scaffolded Evidence

     Yifan Wang, Dongliang Xu, Xiaoyong Li et al. · arXiv:2610.31000

     AI summaryThis paper introduces SecLog-Bench, a scalable benchmark featuring multi-source logs and step-by-step evidence chains for cybersecurity log reasoning. Evaluation of eight mainstream LLMs reveals modest performance, with the best model achieving only 44.44% accuracy and frequently exhibiting unreliable or flawed reasoning.

132. Agri-SCHOLAR: Simulation-Grounded In-Season Rolling Advisories for Agricultural Crops Using LLM-Bounded Optimization

     Geetha Charan, Vedant Balasubramaniam, Manojkumar Patil et al. · arXiv:2610.31003

     Existing agricultural advisories to farmers suffer from several limitations: (a) remain largely expert-driven, (b) do not adequately account for in-season weather, and (c) do not scale to individual farms. In this paper, our objective is to close these gaps by developing…

133. Mitigating Saliency Collapse: Robust Saliency-Aware Long-Text Image-Text Alignment

     Qiuyu Kong, Zanxi Ruan, Marco Cristani et al. · arXiv:2610.31005

     Long-text image-text alignment is essential for fine-grained scene understanding, yet existing methods mainly focus on extending context length or improving local alignment. We identify an overlooked failure mode, saliency collapse, where models over-rely on text that features…

134. Evaluating Artificial Wisdom (AW): A Trajectory-Based Framework for Context-Sensitive Decision Regulation

     Md. Jalil Piran · arXiv:2610.31012

     Capability benchmarks measure what large language models can do, but not whether their recommendations remain appropriate as decision conditions change. This paper proposes a trajectory-based measurement framework and prospective validation protocol for artificial wisdom,…

135. Multi-Tool Selection for Robotic Systems: Comparing Semantic Retrieval and LLM

     Miguel Sousa, Luis Ramos Pinto, João P. Matos-Carvalho · arXiv:2610.31013

     Natural-language interfaces allow users to command robotic systems without interacting directly with mission planners or robotics middleware. Their reliable use depends on robot tool routing, defined here as the selection of the smallest complete subset of predefined executable…

136. Mostly Reading: What Open-Book Scores Measure in Document-Level LoRA Adapters

     Martino M. L. Pulici, Cuong Xuan Chu, Evgeny Kharlamov et al. · arXiv:2610.31017

     AI summaryTrains a LoRA adapter per Wikipedia article on Llama‑3.1‑8B; finds open‑book scoring largely reflects a generic “reading” capability rather than document‑specific knowledge. Specialists improve cloze recall only ~0.094 and their updates are document‑aligned, but merging adapters recovers most open‑book advantage (≈88 %) yet only ~12–16 % of cloze gain.

137. CANNBot-CAKE: Collaborative Agentic Kernel Evolution for Ascend NPU

     Deng Pan, Xue-Jian Gao, Junyi Fan et al. · arXiv:2610.31034

     Efficient custom operators are essential for modern deep learning workloads, yet developing high-performance kernels for Ascend NPUs remains slow, expert-intensive, and difficult to automate. Although AI agents now approach expert performance on GPU kernel benchmarks, current…

138. Multichoice Modeling: Exploring Decision Configuration Space under Uncertainty

     Mariia Kozlova, Roman Stepanova, Alexander Walzer et al. · arXiv:2610.31038

     Decision support has become sophisticated at exploring uncertainty, but much less so at representing the architecture of the decision itself. As a result, the decision space is often narrowed before uncertainty is analyzed, leaving interactions among choices and the effects of…

139. Algorithmic drift in generative adaptive learning

     Firuz Kamalov, Hana Sulieman, Ahmed El Sayed et al. · arXiv:2610.31041

     Generative artificial intelligence promises to democratize personalized tutoring, yet the longitudinal stability of adaptive systems remains under-theorized. Current ethical discourse focuses on static bias. This paper argues that a distinct threat lies in algorithmic drift: a…

140. Conflict-Aware Logit Adapters for Utility-Preserving Anti-Distillation

     Quoc Phong Dao, Binh Anh Nguyen, Linh Ngo Van et al. · arXiv:2610.31046

     While reasoning traces improve model capability, they also provide supervision signals for downstream distillation. Initial efforts to mitigate this issue rely on decoding-time interventions, but they incur high inference overhead and can harm the teacher model's natural…

141. Harm Is Behavioural, Moderation Is Lexical: How Language and Culture Decide Whom Automated Safety Systems Protect and Whom They Silence

     Gabriel Lars Sabadin, Dr Nicola Harding, Sarah Kingston · arXiv:2610.31047

     Content warning: this paper quotes slurs and hateful statements from published test suites; the racial slur among them is masked. Automated systems decide at scale which messages are harmful, and they decide unevenly. Classifiers label African American English as offensive up to…

142. Directing to Diversify: Multi-Directional Search for Automatic Heuristic Design

     Jianghan Zhu, Jianan Zhou, Fei Liu et al. · arXiv:2610.31048

     Large language models (LLMs) have demonstrated strong potential to automate heuristic design for NP-hard combinatorial optimization (CO) tasks without human intervention. Recent methods predominantly rely on evolutionary computation (EC) frameworks, where heuristic…

143. Diagnosing the Explicit-to-Applied Theory of Mind Gap in LLMs

     Prajvi Saxena, Sabine Janzen, Wolfgang Maass · arXiv:2610.31060

     As large language models are deployed in high-stakes roles such as therapy, education, and personal assistance, their social cognition must be assessed reliably. Current Theory of Mind (ToM) benchmarks report aggregate scores that conflate two distinct competencies: explicit ToM…

144. SCOPE: Compiling Natural-Language Policies into Neurosymbolic Predicate Programs

     Arnon Mazza, Elad Levi · arXiv:2610.31062

     Policy enforcement for LLM agents has shifted from single-turn content moderation to compound, application-specific policies evaluated over multi-turn transcripts and tool-call trajectories. Fine-tuned guardrail models are efficient but degrade on compound policies, and while…

145. Dual Exam: Controlled Evaluation of Multimodal Agents and World Models

     Zhefan Rao, Meng Chu, Xuan Zhang et al. · arXiv:2610.31066

     World Models (WMs) can support Multimodal Agents (MAs) through simulated experience, while Agent trajectories can inform World Model refinement. Evaluating this interaction requires separate evidence about action selection and state evolution. We introduce Dual Exam, which tests…

146. MQAScore: Fine-Grained Text-Music Alignment Evaluation via Large Audio Language Model

     Wei-Jaw Lee, Chun-Ping Wang, Xuanjun Chen et al. · arXiv:2610.31083

     Text-to-music alignment is commonly evaluated by the global similarity between text and audio embeddings, which collapses all described attributes into a single score and obscures which ones are missing. We instead introduce MQAScore, which verifies each atomic musical concept…

147. What Pairwise Evaluation Misses: Reward Scores across Prompts

     Mateusz Modrzejewski, Tomasz Radzikowski · arXiv:2610.31090

     Reward models score responses for preference optimization, filtering and ranking, and pooling scores across prompts needs them to be comparable. Yet pairwise evaluation compares two responses to the same prompt, so any score component shared by a prompt's responses cancels and…

148. Co-Recursive Semantic Self-Extension: A Mathematical Foundation for Open-Ended Intelligence

     Simon Hegelich, Bardh Prenkaj, Habiba Sarhan et al. · arXiv:2610.31097

     Artificial intelligence has advanced through scale, data, and computation, yet remains bounded by its representational framework. We ask what makes intelligence open-ended. Every inferential system selects meaningful distinctions, limiting what it can learn. Within a fixed…

149. Solving Flood-It Puzzles: A Comparison of Tailored A* Search with Declarative Solver Approaches

     Philipp von Manteuffel, Benjamin Blankertz · arXiv:2610.31100

     We present an optimized approach for finding minimal move sequences to solve Flood-It puzzles using A* search and compare it with approaches using SAT, Integer Programming, and Constraint Programming solvers. The performance of A* search is considerably improved by using an…

150. DeepOpen: Decision-Oriented Intent Classification on BANKING77 and CLINC150

     Shuang Song Zhongqiao Yuan · arXiv:2610.31104

     We present DeepOpen, a decision-Oriented intent-classification framework with dataset- specific systems for BANKING77 and CLINC150. On BANKING77, a single English classi- fier continued with sharpness-aware minimization (SAM) achieves 94.1234% test accuracy (2,899/3,080). This…

151. Who Is the Influencer? Consensus Dynamics in Multi-Agent Systems

     Gouki Minegishi, Jingyuan Feng, Hiroki Furuta et al. · arXiv:2610.31105

     AI summaryThis paper studies consensus dynamics in multi-agent LLM systems and finds that larger models act as influencers, absorbing the opinions of smaller models. Experiments and a particle-based theoretical model show this dominance stems from stronger in-context learning capabilities in larger models.

152. Can AI Co-Design Structures? An AI Agent for Early-Stage Structural Design

     Ricardo Maia Avelino, Mennatallah El-Assady, Tom Van Mele et al. · arXiv:2610.31113

     Recent advances in large language models and agentic workflows offer a new paradigm to early-stage structural design translating design intent into structural models. Although text-to-3D approaches have demonstrated that natural language can be used to produce visual, mesh-based…

153. A Grammar of French Civil Cassation: A Ground-Level Dataset and Tool

     Avrile Floro, Tamara Dhorasoo, Nils Holzenberger et al. · arXiv:2610.31116

     AI summaryDataset of 121,536 ground–reason pairs from 86,464 French Court of Cassation rulings (2016‑2025). The authors automatically pair each appeal ground with its court answer, classify them into doctrinal families, and provide a descriptive grammar of civil cassation, releasing data, code, and a tool.

154. ProductWebBench: Evaluating Website Continuity Beyond Task Completion

     Jiajun Wu, Jian Yang, Yaxin Du et al. · arXiv:2610.31118

     Frontend coding agents are usually graded on one question: did the requested target get built. Real engineering changes carry a second obligation, because the rest of the site has to keep working. We call that second obligation website continuity, and introduce ProductWebBench,…

155. Constraint Reasoning and Large Language Models Join Forces: Providing Contrastive Explanations for Real-World Optimisation-Based Applications

     Roger X. Lera-Leri, Filippo Bistaffa, Athina Georgara et al. · arXiv:2610.31119

     We focus on optimisation-based Decision Support Systems in real-world applications. In this context, we aim to provide contrastive explanations that answer users' questions about recommendations based on the solution of an optimisation problem, specifically a Mixed-Integer…

156. SetCommit: Coordinated Token Commitments for Few-Step Diffusion Language Model Decoding

     Yuannuo Feng, Yubing Ye, Wenshuai Yao et al. · arXiv:2610.31120

     Diffusion language models generate multiple token positions in parallel, but few-step decoding makes parallel commitment difficult: tokens that appear reliable individually may be jointly inconsistent, while premature commitments can prevent later context-dependent refinement.…

157. Post-Audit Mirage: Identifiability Limits of Offline Agent Verification

     Aman Behera, Payal Payal · arXiv:2610.31130

     Offline tests can fail to reveal whether an agent update will remain safe after deployment. We operationalize this post-audit mirage using matched safe and harmful lifecycles that expose identical verifier-visible evidence while an isolated scorer records opposite deployment…

158. Mixing Matters: Evidence Position Bias Across Sequence Mixers in Long-Context Question Answering

     Aman Behera, Namit Solanki, Mehul Anand · arXiv:2610.31135

     Long-context models often use answer-bearing evidence less effectively when it appears away from a prompt boundary. We compare this position bias across Transformer, state-space, and hybrid Mamba architectures by moving one gold document through a fixed set of nine distractors.…

159. Rethinking Fisher Vectors for Second-order Multi-scale Deep Feature Aggregation in Scalable Scene Retrieval

     Alexy Bhowmick, Sarat Saharia, Shyamanta M. Hazarika · arXiv:2610.31146

     Aggregating CNN responses into a compact discriminative vector, given their lack of geometric invariance, is a challenge. To address this challenge, we propose a codebook-free method to encode second-order information of multi-scale CNN features within the Fisher Kernel…

160. PriMa-Causa: Causal Prescriptive Maintenance

     Felix Saretzky · arXiv:2610.31147

     Automated manufacturing requires data-driven decisions under uncertainty and sparse observational records, and unplanned downtime is a dominant cost of getting them wrong. Prescriptive maintenance recommends a maintenance action against an impending failure rather than merely…

161. Prompt Sensitivity in Tool-Augmented Model Comparisons

     Mateusz Modrzejewski, Tomasz Radzikowski · arXiv:2610.31148

     Language models are compared on question-answering benchmarks that pose each question in one wording, and a reworded question can change which of two models scores higher. We ask whether giving the models tools makes such comparisons more stable. An accuracy gain cannot answer…

162. Network topology shapes consensus and vulnerability in LLM populations

     Roberto Passaro, Ariel Flint, Romualdo Pastor-Satorras et al. · arXiv:2610.31165

     As large language models (LLMs) are deployed in interacting populations, their collective behaviour depends not only on the models but also on who interacts with whom. Studies of convention formation in LLM populations have mainly tested well-mixed interactions. We place…

163. Prompting Strategy Effects Depend on the Judge: A Multilingual Study of LLM-as-a-Judge Robustness

     Asal Mehrabi, Mohammad Aliannejadi, Clemencia Siro · arXiv:2610.31172

     LLM-as-a-Judge evaluation is increasingly applied to multilingual tasks, yet prompting strategies such as Chain-of-Thought and Role Conditioning are recommended based solely on English benchmarks. Whether their effectiveness transfers across judge models and languages remains…

164. When More Metadata Misleads: Probing Field-Level Grounding in LLM-Generated Museum Descriptions

     Clemencia Siro, Laura Hollink · arXiv:2610.31193

     Metadata-conditioned generation is often framed as a data-to-text task in which additional input fields should improve grounding. Museum description generation challenges this assumption because catalogue records are heterogeneous, partially visual, and unevenly documented…

165. Different Tasks, Different Timesteps: Temporal Task-Aware Quantization for Diffusion Language Models

     Ofek Avivi, Eliron Rahimi, Avi Mendelson et al. · arXiv:2610.31198

     AI summaryThe paper shows that quantization errors in diffusion language models vary by downstream task and denoising step, and proposes T2Quant to predict sensitivity without ground-truth labels and selectively assign precision. It improves average accuracy by up to 6.7 points within the same bit‑budget.

166. Predicting Music Valence and Arousal via Textual and Multimodal In-Context Learning in Large Language Models

     Chi-En Dai, Chih-Pin Tan, Yi-Hsuan Yang · arXiv:2610.31202

     While arousal can be reliably predicted from musical audio, emotional valence remains difficult to model due to its subjective and context-dependent nature. In this work, we investigate the potential of in-context learning (ICL) with large language models to advance continuous…

167. Trion-1: Unifying Language, Protein Sequence, and Structure with Diffusion Language Models

     Tianyu Lu, Kaicheng Yang, Jiangtao Feng et al. · arXiv:2610.31215

     We envision a general-purpose protein foundation model as a unified, interactive model of sequence, structure, and function, capable of understanding, generation, and cross-modal translation while retaining the instruction-following and reasoning capabilities of modern LLMs.…

168. QICOM: Open-Source Software for Reproducible Image Quality Evaluation, Blinded Assessment, and Radiomic Comparison

     Fabio Mattiussi, Switinder Ghotra · arXiv:2610.31218

     AI summaryQICOM is an open‑source tool that records image processing settings, identities, and radiomic measurements to enable reproducible, blinded comparison of image quality metrics. It passes 482 IBSI reference tests and reproduces all 43 feature pairs in reopened projects.

169. Reward-Weighted Preference Optimization for Selective Combinatorial Optimization Problems

     Haohui Zhang, Xinyu Hu, Wouter van Heeswijk et al. · arXiv:2610.31239

     Reinforcement learning has emerged as an effective approach for combinatorial optimization, but most existing methods focus on problems such as the Travelling Salesperson Problem (TSP) in which all elements must be served. Selective combinatorial optimization problems (SCOPs),…

170. Reducing Multi-Horizon Photovoltaic Forecast Error through Tree-Anchored Expert Combination

     Sun Ze, Zhou Liguo, Xu Yuqing et al. · arXiv:2610.31248

     A strong direct tree model may remain the best standalone photovoltaic forecaster, yet a weighted combination can reduce its residual errors. We examine whether complementary temporal predictors can lower the tree's mean absolute error (MAE) over 15--240 minutes without weaker…

171. Can Entanglement-Induced Feature-Map Designs Improve Projected Quantum Kernels?

     Ravi Kumar Jha, Hsuan-Yin Lin, Eirik Rosnes · arXiv:2610.31260

     AI summaryProjected quantum kernels were evaluated with and without classical RBF hyperparameter tuning, revealing that extensive tuning masks their true capability. Introducing entanglement‑based feature maps markedly improves the native inductive bias, yielding stronger PQK performance driven by quantum architecture rather than tuning.

172. Notation3: Graph Terms for RDF

     Dörthe Arndt, Pierre-Antoine Champin · arXiv:2610.31266

     AI summaryNotation3 Logic extends RDF with rules and graph terms but lacked a formal semantics for the latter. This paper defines an abstract syntax, introduces N3‑entailment that assigns meaning to graph terms and special predicates, and provides a foundation for standardising the logic.

173. Memory-Aware Reward Modeling: Learning Human Preferences for LLM Memory Use

     Yijiang River Dong, Zexue He, Cheng Yu et al. · arXiv:2610.31272

     Memory enables language model assistants to maintain continuity across sessions and generate personalized responses, but knowing more about a user does not necessarily produce responses they prefer. Existing evaluations assess memory capabilities without directly establishing…

174. PointerBench: A Diagnostic Benchmark for Fine-Grained GUI Grounding

     Valentino Sacco, Saurabh Pandey, Alexander Florian Jercher et al. · arXiv:2610.31274

     Multimodal Large Language Models have shown promising performance in computer-use tasks, from planning user workflows to grounding natural-language actions to precise UI targets. Existing end-to-end benchmarks primarily emphasize long-horizon task completion, while GUI grounding…

175. RIPPLE: Recovery Internalization through Process-Level Policy Learning and Effect-Aware Credit for Long-Horizon Agentic Coding

     Jingxuan Xu, Zihang Xu, Yuechen Luo et al. · arXiv:2610.31276

     Long-horizon coding agents can observe a failure and generate a plausible critique yet continue along the same incorrect trajectory. We call this the Critique-to-Action Handoff Gap. We introduce RIPPLE, a recovery-oriented training framework that internalizes process-level…

176. CoWriter: Guiding Methodological and Synergetic Human-LLM Co-Authoring

     Shmuel Amar, Benjamin Lieber, Ori Shapira et al. · arXiv:2610.31277

     While LLMs have transformed text generation, standard chat-and-canvas interfaces often reduce human authoring to quite passive supervision, causing review fatigue, over-reliance on AI, and the sidelining of the human role in long-form, source-grounded writing. Prior systems…

177. Beyond Accuracy: Disentangling Decomposition from Typed-State Quality Assurance in Multi-Agent LLM Pipelines

     Maria Gabriela Alves Zuppardo, Vinicius Cardoso Garcia · arXiv:2610.31280

     Multi-agent LLM pipelines are often credited with gains in software engineering tasks, but it is rarely clear whether a gain stems from the typed contract between agents or from the specialized prompts that decomposition enables. We study this attribution problem in Requirements…

178. RecallGS: Geometry-Grounded Recall for Online Semantic Gaussian Splatting

     Thanh-Khoi Nguyen, Thien-Phuc Tran, Minh-Triet Tran · arXiv:2610.31293

     Online semantic 3D Gaussian Splatting (3DGS) enables open-vocabulary scene understanding from streaming observations. However, existing methods recover geometric context missing from image-derived semantic features using an additionally learned 3D network, incurring extra…

179. The Value of Memory in Language Agents

     Zehao Lin, Erenkai Zhu, Helan Hu et al. · arXiv:2610.31303

     Memory enables language agents to carry useful distinctions from past experience into future decisions. Its value depends on which distinctions survive, whether a bounded agent can use them, and what can be recovered or revised as demands evolve. We develop a decision-theoretic…

180. Superintelligent Software Engineering: Governance That Scales with Capability

     Mohammad Mehdi Gholinejad · arXiv:2610.31306

     Software agents are being trained to exceed human engineers; we argue that capability alone does not produce better software engineering. We define superintelligent software engineering as the discipline of building and evolving software with agents whose capability exceeds that…

181. Subgoal-aware Memory Graph for GUI Agent

     Dezhao Luo, Daniel Beechey, Derek Yuen et al. · arXiv:2610.31307

     GUI agents aim to automate complex user tasks by interacting with mobile interfaces, but ef- fective decision-making in long-horizon set- tings requires reusing prior experience rather than relying solely on the environment ob- servations. Existing memory-based methods typically…

182. When LLMs Forget What They Believe: Characterizing State Integrity Collapse in Long-Horizon Dialogues

     Yancheng Zhu, Yifan Yuan · arXiv:2610.31314

     Large language models can repeat an instruction while failing to keep obeying it. We study this recall-compliance dissociation in long-horizon dialogue and define State Integrity Collapse: a model preserves a stated constraint in memory but stops applying it to later turns. We…

183. SurveyGraph: Generating Academic Surveys via Methods-Entity Knowledge Graphs

     Yifan Yuan, Yancheng Zhu, Zuoyu Zhang et al. · arXiv:2610.31317

     Most automated survey generation systems retrieve papers as flat documents, limiting their ability to capture research structure. We present SurveyGraph, a framework built around a Methods-Entity Knowledge Graph (MEKG) encoding method-level follow-on, competition, and…

184. Hyperreduction accelerates finite element model updating for the discovery of neural materials

     Hagen Holthusen, Matthias Knipper, Kevin Linka · arXiv:2610.31322

     Finite element model updating is among the most general approaches to discover constitutive models from structural experiments: since the displacement field is computed rather than measured, it accommodates arbitrary geometries, boundary conditions, and sparse or noisy data.…

185. Bridging the Data Gap: Source-Informed Task-Space Transport for Few-Shot Diffusion

     Jinjian Jiang, Jing Peng, Rehan Deen et al. · arXiv:2610.31323

     Many deployments require a model to move rapidly from a data-rich source to a related target observed only sparsely. This is especially consequential where target data are difficult to collect, as in healthcare settings with persistent population or acquisition imbalances.…

186. AgenticEvolve: Evolutionary Program Search with Persistent Language Model Agents that Design and Run Experiments

     Ze-Yu Peng, Qing-Yu Lan, Zhan-He Wang et al. · arXiv:2610.31326

     AI summaryAgenticEvolve uses persistent language‑model agents that design and run experiments on candidate programs, improving evolutionary search. It outperforms baseline methods on design tasks and equation discovery, achieving higher quality with fewer evaluations and similar cost.

187. ValuSpec: Plug-and-Play Candidate Valuation before Target Verification for Tree-Based Speculative Decoding

     Liang He, SiYuan Ma, Qishi Zhan et al. · arXiv:2610.31338

     Tree-based speculative decoding accelerates large language model inference by verifying multiple candidate branches in parallel. However, in practical serving scenarios, similar requests repeatedly induce overlapping candidate branches. While retrieval-based methods mitigate…

188. Automating Feature Flag Cleanup at Scale with Multi-Agent LLM Systems: An Industrial Experience at DoorDash

     Atharv Chandratre, Jai Datar · arXiv:2610.31351

     AI summaryAutomated LLM‑driven system cleans DoorDash’s 56k feature flags, generating engineer‑approved PRs for 90% of 50 stale flags tested, at $4.79 and 13.8 min per cleanup, showing scalable, cost‑effective flag removal.

189. Muffin - A Shared Memory Buffer for Asynchronous AI Agents

     Ping-Lin Kuo, Suman Navaratnarajah, Christian Frey et al. · arXiv:2610.31355

     The rise of Agentic AI systems has changed the landscape of artificial intelligence by enabling strong reasoning capabilities through specialized multi-agent interaction with structured self-reflection. However, in target-driven reasoning tasks, single- and multi-agent systems…

190. RIDAR: Safeguarding Research Integrity in Autonomous Research Agents via Automated Detection and Runtime Intervention

     Jiening Siow, Amy Xin, Junjie Wang et al. · arXiv:2610.31372

     Autonomous research agents are increasingly used to automate large portions of scientific research, yet they are prone to research integrity violations (RIVs): behaviors that compromise the validity, reproducibility, or faithfulness of the research process. We present RIDAR, a…

191. Certified Execution Harness for Agentic AI: Safe and Least-Intrusive Petri-Net Supervisory Control Between Reasoning and Action

     Chao Gu, Quanyan Zhu · arXiv:2610.31376

     This paper studies the problem of safe tool execution for large language model (LLM) agents under prompt injection. Using Petri nets, we formalize the execution harness between the LLM and its tools as a supervisory control problem, in which marks represent capabilities,…

192. AI Risk Ontologies for the EU AI Act Compliance: AIRO, VAIR, and their Integration with DPV

     Delaram Golpayegani, Harshvardhan J. Pandit, Declan O'Sullivan et al. · arXiv:2610.31380

     As the use of AI, and consequently its harms, becomes more prevalent, the global discourse around AI has become increasingly risk-aware. With new risks emerging more frequently than ever, a key challenge is how to represent, document, and catalogue AI risks in an effective,…

193. Auditing the Hand-Written Inputs of Learned Access Control

     Arsalan Jalil Sarghaleh, Mehdi Hazrati · arXiv:2610.31388

     Learned access control is typically deployed behind two hand-built artefacts: a hard-deny set enforced by a small trusted shield, and a region map that decides which requests mined rules answer and which the learner answers. Verifying the shield or measuring the learner…

194. HERD: Harness-Evolving Recursive Distillation for Continual Learning

     Xun Sun, Thanh Thi Nguyen, Thuy-Trang Vu · arXiv:2610.31401

     Continual learning requires a model to keep acquiring new skills and knowledge without compromising its existing capabilities. Parametric methods suffer from catastrophic forgetting and loss of plasticity. Non-parametric methods, such as prompts, skills, and harnesses, leverage…

195. AgentPhishing: A Benchmark of LLM Email Agents Under Adaptive Attack

     Justin Tan, Miao Li, Ding Zhao · arXiv:2610.31404

     Computer-use agents increasingly handle private information and act on users' behalf across desktop and web applications, including email. Phishing therefore poses privacy and security risks that depend not only on whether agents recognize malicious messages, but also on how…

196. Same Entity, Different Script: Diagnosing Cross-Script Entity Matching Failures in Language Models

     Athira Gopal, Patil Prathamesh Pravin, Ankush Kadu et al. · arXiv:2610.31407

     Multilingual language models can give different identity judgments when the same people's names are written in different scripts. We study this effect using 601 audited identity pairs with attested Devanagari and Latin forms, holding the question and relation fixed. Qwen2.5-7B's…

197. Major Revision, Minor Accountability: LLMs as Reviewers of Published Articles in Top-Tier Journals

     Bastien Le Guellec, Su Hwan Kim, Raphael Bentegeac et al. · arXiv:2610.31414

     With the growth of scientific output, the peer review system is approaching a breaking point: reviewers are increasingly hard to secure, and turnaround times lengthen accordingly. LLM-assisted or LLM-generated peer review has been proposed as a remedy, but how such models would…

198. Post-Hoc Feature Selection Layer for Neural Networks Interpretability

     João Kenji Suwa, João Pedro Silveira e Silva, Bruno Iochins Grisci · arXiv:2610.31420

     AI summaryPost-hoc Feature Selection Layer fine‑tunes a lightweight weighting module on frozen pre‑trained tabular models, highlighting the most relevant features without altering the original network. Experiments show it identifies key features while preserving accuracy, outperforming Integrated Gradients, SHAP, and others.

199. How Depth Growth and Looping Shape In-Context Learning in Language Models

     Ferdinand Kapl, Kaitlin Maile, Stefan Bauer et al. · arXiv:2610.31421

     AI summaryThis paper investigates how parameter reuse in looped and depth-grown language models affects in-context learning. The results show that these architectures develop a stronger inductive bias toward context, acquiring and retaining in-context learning earlier and more reliably than standard models by propagating induction heads across layers.

200. FocusVLA: Future-Intent Guided Dynamic View Routing for Vision-Language-Action Models

     Huixin Xu, Ziya Yu, Peng Li · arXiv:2610.31423

     Vision-language-action (VLA) models have become an important paradigm for robotic manipulation by unifying visual perception, language understanding, and action generation. Since reliable manipulation requires sufficient observation of objects, end-effectors, and their spatial…

201. Predictable performance is suboptimal, and what it means for model routing

     Guanhua Zhang, Moritz Hardt · arXiv:2610.31428

     AI summaryThe paper proves that a performance predictor improves routing only when the base model is suboptimal or the predictor adds significant capacity, showing log‑loss transfer from predictor to model. Experiments confirm larger gains for weaker models, diminishing after fine‑tuning.

202. PhysTrace: Interpretable Temporal Evaluation of Physical Consistency in Video Generative Models

     Taekoan Yoo, Wonkyung Jung, Kyunghun Kim et al. · arXiv:2610.31429

     AI summaryPhysTrace aligns pretrained video model predictions with observed video windows to pinpoint where physical expectations fail, offering interpretable descriptions without task-specific probes. It achieves 70.7% hit rate on IntPhys2‑Timing, outperforms baselines, and reveals attribute‑ and noise‑dependent violation patterns in diffusion/flow models.

203. Conservative Meta-Agent Search: Architecture Adaptation under Non-Stationarity

     Ali Atiah Alzahrani · arXiv:2610.31436

     AI summaryConservative Meta‑Agent Search (CMAS) frames architecture search as incumbent replacement with switching costs and risk constraints, using archived models, typed graph edits, and paired shadow replay to promote only high‑confidence edits. On a regime‑switching finance benchmark, CMAS reduces surrogate dynamic‑oracle regret from 0.116 to 0.104 and raises held‑out utility from 0.44 to 0.48 versus the best adaptive baseline.

204. DIAL: Steerable Language Models with Distilled Instruction-Adherence Levels

     Thanh Q. Tran, John Russell Himawan, Arun Verma et al. · arXiv:2610.31438

     When a user says "Keep it short", how short should the LLM's response be? More generally, users and applications require different levels of instruction adherence, but prompting does not reliably provide fine-grained control over them. We therefore study "steerable language…

205. Adaptive Skill Granularity for Coding Agents: Runtime Routing and Macro-to-Micro Recovery

     Dingxuan Zhang, Yanhui Jiang, Jun Wang et al. · arXiv:2610.31461

     Reusable skills allow large language model agents to package task-specific procedures, tool affordances, and execution guidance into modular units. Prior work has largely studied how agents retrieve, compose, or synthesize skills before execution. In coding agents, however,…

206. MoG-VLM: Efficient Video Anomaly Detection through Motion-Guided Vision-Language Models

     Khaled Waleed Dawoud, Muhammad Zaigham Zaheer, Khan Muhammad et al. · arXiv:2610.31478

     Vision-Language Models (VLMs) have shown strong potential for video anomaly detection, but dense frame-level inference makes them impractical for real-time deployment. Existing acceleration commonly relies on uniform sampling, which ignores content dynamics and can either waste…

207. Training Small Language Models for Transferable Tool Selection

     Zhaoyi Zhou, Akshayaa Magesh, Jeongyeol Kwon et al. · arXiv:2610.31483

     AI summaryThey train a small Qwen3-8B with reinforcement learning to decompose complex instructions and select the right tool from thousands, outperforming a 402‑B Llama model on ToolBench, and the policy transfers to other retrievers and domains.

208. SABLE: Scalable Affine Bayesian Latent-space Exploration

     Erik Bodin, Lisa Maria Kreusser, Donney Fan et al. · arXiv:2610.31495

     Complex objects such as engineering designs and proteins are often difficult to optimise because their design spaces exhibit high-dimensional structure and objective evaluations are expensive. Generative models provide continuous latent representations, but their latent spaces…

209. FairFRESH: A Neuro-Symbolic Framework for Fair and Explainable Household Food Recommendation

     Kodur Sai Vinay Sathvik, Kakoli Bora, Tejaswini R. M · arXiv:2610.31506

     Recommender systems typically serve one user at a time. However, in households, members share a finite pantry and follow different dietary rules, making independent recommendations prone to resource conflicts and dietary violations. We formulate household food recommendation as…

210. DuoCook: An Overcooked-Inspired Benchmark for Visual Planning and Coordination

     Ziyi Jiao, Heng Zhang, Drishti Regmi et al. · arXiv:2610.31511

     Coordinating two players in a cooking task requires a vision-language agent to track changing scene states, sequence recipe steps, and allocate shared resources under a time budget. We introduce an Overcooked-inspired benchmark for evaluating these capabilities in Unreal Engine…

211. From Prompt Rules to Golden Data Model: Production-Scale Generation of Personalized, Compliant Pharmaceutical Claims

     Suyash Mishra, Qiang Li, Subhabrata Adak et al. · arXiv:2610.31519

     Large language models (LLMs) are increasingly used to generate domain-specific commercial & medical content (e.g., claims, statements, and promotional phrases) in regulated industries such as pharmaceuticals. However, ensuring its regulatory compliance, factual accuracy, and…

212. The Geometry of Metaphor: Probing and Steering Metaphor Representations in Large Language Models

     Shihui Li, Mette Andersen, Martha Lewis · arXiv:2610.31520

     AI summaryResearchers identify causal subspaces representing metaphor in various LLMs and show that manipulating these subspaces can steer metaphor-related classification and generation tasks, enabling control over the metaphoricity of model outputs.

213. Jev-DLC: Decision-Level Continual Learning from Typed Feedback

     Yangyang Feng, Jianle Gan, Zhuoyan Feng et al. · arXiv:2610.31521

     AI summaryJev-DLC learns from typed user feedback by storing each correction as evidence for an action identity and a directed relation between preferred and rejected choices, then combines these signals with a frozen scorer to update decisions. On TCPA-Repeat it improves accuracy by 2.4‑points over Recent‑4 replay and by 7.5‑points over the best method after two feedback events.

214. Beyond Causal Masks: Auditing and Localizing Hidden Future-Token Leakage in Adaptive Sequence Models

     Huiting Liu, Jinsong Wu · arXiv:2610.31522

     Hidden future-token leakage can create false architectural gains: an autoregressive model may appear substantially better under teacher-forced evaluation while its free-running generation deteriorates. We investigate this failure in an adaptive residual-routing decoder whose…

215. Contestable Agents for Data Governance: Recording Human Stewardship of AI-Generated Metadata Across Structured and Unstructured Systems

     Humza Akhtar, Brian Freeman · arXiv:2610.31553

     Industrial organizations hold operational knowledge in structured records, such as telemetry, point lists, and transactions, and in unstructured sources, such as manuals, service notes, communications, and inspection reports. Governance programs often manage these sources…

216. When to Follow the Teacher: Advantage-Based On-Policy Distillation

     Mahdi Sabbaghi, George Pappas, Hamed Hassani et al. · arXiv:2610.31563

     Online Policy Distillation (OPD) often outperforms offline methods such as Supervised Fine-Tuning (SFT) by training on trajectories sampled directly from the student. However, standard OPD trains the student to follow the teacher at every state of the trajectory, assuming that…

217. Social-HUMINT: An LLM-Augmented Unified Framework for Measuring Social Media Exposure

     Pellegrino Casoria, Simon Pietro Romano · arXiv:2610.31573

     AI summaryThis paper introduces Social-HUMINT, an LLM-augmented framework that unifies social media intelligence and cyber human intelligence to measure a user's exposure to social engineering. The method achieves high profiling accuracy and reliably predicts susceptibility to personalized attacks using public data.

218. What Makes a Good Unmasking Policy? A Controlled Study of Learned Unmasking in Diffusion Language Models

     Jinya Sakurai, Xun Xu, Jaehong Yoon · arXiv:2610.31576

     Masked diffusion language models generate tokens in parallel, but their efficiency depends on which predictions are committed. Learned unmasking policies improve the speed--accuracy trade-off, yet differ in supervision, inputs, and capacity, obscuring which choices matter. We…

219. Who Is Sycophantic to Whom? Confident LLMs, Human Deference, and the Collapse of Trust Calibration

     Deepthika S, Pawni Yadav, B. Sankar · arXiv:2610.31579

     Research on sycophancy asks whether large language models (LLMs) bend towards what their users want to hear. We ask the reverse question: when an LLM is confident, consistent, and wrong, do users bend towards it? In a within-subjects experiment, 32 participants learned up to…

220. Would an LLM Match into a Residency? Uncertainty, Reasoning and Image Use of Open Language Models on the Spanish MIR Examination

     Miguel Fernandez-de-Retana, Unai Zulaika, Ruben Sanchez-Corcuera et al. · arXiv:2610.31586

     Accuracy on medical examinations is common evidence offered for the clinical use of language models, yet a health system that intends to run an open model on its own hardware needs to know considerably more: whether the model recognizes when it is likely to be wrong, whether…

221. VERA: A Responsible AI Dashboard for Assessing LLM Deployments in Light of the EU AI Act

     Robin Quériaux, Solveig Bachellery, Léa Déléris et al. · arXiv:2610.31588

     Institutions that deploy large language models in regulated settings often need technical evidence of responsible AI practices before approving a deployment. COMPL-AI provides a technical interpretation of the EU AI Act that organises this evidence under six principle groups and…

222. Simulated Control Evals

     Jeffrey Olmo, David Zhang, Tyler Tracy · arXiv:2610.31593

     AI summarySimulated control evaluations replace real environments with a language model, predicting deployment safety more accurately: 3.6% RMSE vs 10.1% for a separate control setting, with lower KL divergence and 78‑85% variance explained, using fewer than 5% of task pairs.

223. How Ground Truth Shapes Retrieval Scores: Evidence-Aware Evaluation of a Retrieval-Augmented Generation System for Breast Lesions

     Miguel Mochizuki Silva, Samuel Ismael, Mateus C. Dias et al. · arXiv:2610.31606

     Purpose: To describe an evidence-aware retrieval evaluation protocol for a clinical retrieval-augmented generation (RAG) system over breast lesion management literature, addressing the dependence of retrieval scores on annotation coverage. Methods: An 18-document corpus and an…

224. Questioning Instrumental Convergence in RLVR: Checkpoint Evidence from OLMo-3

     Bar{\i}ş Alparslan, Sandy Tanwisuth, Caleb Petterson · arXiv:2610.31624

     Instrumental convergence is often invoked as a general reason to expect capable agents to acquire or express power-seeking subgoals: self-preservation, resource acquisition, deception, or resistance to oversight which poses an important threat to the society. For LLMs,…

225. Causal Chain of Thought: Mitigating Attention Leakage via Context Isolation

     Gabriele Dominici, Teague McMillan, Fazl Barez et al. · arXiv:2610.31639

     Chain-of-thought (CoT) prompting elicits step-by-step rationales that appear to explain model predictions, but there is no guarantee these steps causally drive the final answer. In standard autoregressive generation, dense attention allows each reasoning step to attend to the…

226. Low-Bit N-Gram Embeddings for Language Models

     Stephen Chung, Wenyu Du · arXiv:2610.31646

     Many recent open-weight language models use n-gram embedding tables to improve performance. These tables store learned representations of short token sequences, such as pairs or triples of consecutive tokens, which the model retrieves during processing. However, they can require…

227. EpiSEM-Bench: Evaluating Safety and Evidence Recovery in LLM Agents under Missing Evidence

     Qijin Zhang, Hanzhen Wang, Yuehan Zhao et al. · arXiv:2610.31650

     Tool-using language model agents rely on environmental facts to decide whether an action is permitted. When a tool response omits a safety-relevant fact, an agent may carry out a prohibited action even though the user request is benign and every returned statement is true. We…

228. METTA: Memory-Enhanced Test-Time Adaptation for Meta-Reinforcement Learning

     Ali Najar, Hosein Hasani, Amir Kooshan Fattah et al. · arXiv:2610.31656

     AI summaryMETTA splits adaptation into fine‑grained within‑episode causal memory (linear cost) and compact episode summaries for cross‑episode inference, enabling fast test‑time adaptation with low memory. It reaches >50% success on Meta‑World held‑out tests and outperforms transformer baselines on long‑context sparse‑reward tasks.

229. Cost-Aware Abstention for LLM-Based IT Ticket Classification

     Sachin Gupta · arXiv:2610.31657

     AI summaryThis paper evaluates cost-aware abstention policies for IT ticket classification by selecting classification thresholds on calibration data under explicit error costs. The method reduces routing costs by 38.1% compared to baseline approaches on a standard support ticket corpus.

230. Direct Pass@$k$ Optimization for Thought Pruning: Balancing Quality and Diversity in Test-Time Reasoning

     Mahdi Salmani, Adel Javanmard, Devansh Gupta et al. · arXiv:2610.31667

     AI summaryThe paper proposes a pruning method that directly optimizes pass@$k$ by selecting partial solutions that maximize the chance that at least one will finish correctly, balancing quality and diversity automatically. It outperforms prior quality‑ or diversity‑based pruning on eight reasoning models, especially when retaining a moderate number of candidates.