LIVE · refreshes every 20 min

updated Nov 14, 07:23 UTC

110010.art / archive / research / 2026-10

Research — October 2026

Oct 17 Defining and generating compositional structure arXiv cs.AI source ↗

Oct 17 RAC: a relational agent contract for constraint-governed engineering tasks arXiv cs.AI source ↗

Oct 17 StepSafe: learning timely intervention with context-aware safety reasoning arXiv cs.AI source ↗

Oct 17 MAIAM: a multi-tiered AI accessibility model sorts access by purpose, safety, and capability rather than release modality arXiv cs.AI source ↗

Oct 17 Confidence-Calibrated Inference Expansion for Evaluator-Guided Test-Time Reasoning arXiv cs.AI source ↗

Oct 17 EvolvingBrain: an adaptive skill system for multi-modal agents self-evolution in open worlds arXiv cs.AI source ↗

Oct 17 IneqAgent: agentic inequality proving in Lean with symbolic computation arXiv cs.AI source ↗

Oct 17 SkillNEO: non-elitist evolutionary optimization of agent skills arXiv cs.AI source ↗

Oct 17 Evaluating model capabilities after benchmark saturation arXiv cs.AI source ↗

Oct 17 AlphaDiana: a measurement system for agents in verifiable reasoning tasks arXiv cs.AI source ↗

Oct 17 MAI-Gym: an open all-in-one platform for GUI agent evaluation and RL training across mobile and computer environments arXiv cs.AI source ↗

Oct 17 Self-improving agent grows its own DSL to answer exactly with no reward, judge, or answer key; Oracle-free operator induction for exact question answering. arXiv cs.AI source ↗

Oct 17 L2G-SR: learning to guide LLM-based symbolic regression with a learned structural verifier using tree edit distance arXiv cs.AI source ↗

Oct 17 Open-world AI: a Tale of Three Requirements arXiv cs.AI source ↗

Oct 17 Anticipatory commonsense with moving objects: an ASP-based framework for predictive reasoning with interacting trajectories arXiv cs.AI source ↗

Oct 17 Counterfactual neighborhood risk fields: evidence-grounded selective safety for vision-language models arXiv cs.AI source ↗

Oct 17 RealtimeBench evaluates agents under a live clock; GPT-6.1-sol outperforms Jev when games are paused but underperforms when games run during inference arXiv cs.AI source ↗

Oct 17 Temporal Specification Guidance enables training-free at-inference guidance for text-to-video generation using temporal logic specifications. arXiv cs.AI source ↗

Oct 17 Value-informed schedule trimming enables accelerated SMC guidance in discrete diffusion arXiv cs.AI source ↗

Oct 17 New scaling principle for reliable intelligence addresses how quickly reality can change before an AI can reliably perceive and act arXiv cs.AI source ↗

Oct 17 ARIA: Runtime inspection and amendment for multi-agent systems arXiv cs.AI source ↗

Oct 17 Correctness feedback suppresses evidence of specification-completeness auditing in Verus; independent no-feedback discovery doubles certified-witness yield across 12 Verus/Rust tasks. arXiv cs.AI source ↗

Oct 17 LEV: Thinking fast and deep with Looped Decision Models arXiv cs.AI source ↗

Oct 16 Daron Acemoglu speaks on stage with Aisha Losche about liberal democratic institutions and the norms that support them MIT News (AI) source ↗

Oct 16 Spectral indistinguishability and learning-curve identifiability in quadratic networks arXiv cs.LG source ↗

Oct 16 O-RAN-native jamming detection under network heterogeneity arXiv cs.LG source ↗

Oct 16 Quantile-guided diffusion for probabilistic time-series forecasting arXiv cs.LG source ↗

Oct 16 Optimal Hydra: model heterogeneous federated learning for LLMs using data-informed Hydra-nets arXiv cs.LG source ↗

Oct 16 Semantic alignment of EEG and text: a contrastive learning framework for decoding imagined speech arXiv cs.LG source ↗

Oct 16 Exploring the design space of permutation diffusion LLMs arXiv cs.CL source ↗

Oct 16 One bias score, two factors: how mitigations affect option bias in multiple-choice evaluation arXiv cs.CL source ↗

Oct 16 AbridgeKV: compressing each completed block of generated text into the KV entries of learned pause tokens arXiv cs.CL source ↗

Oct 16 Amortizing sequential experimental design in reasoning LLMs arXiv cs.LG source ↗

Oct 16 Causal syntax circuits in CodeLMs: budget and probe decide what looks localised arXiv cs.LG source ↗

Oct 16 PRISM: transfer and limits of activation interventions in tool-calling language models arXiv cs.CL source ↗

Oct 16 Same Answer, New Box: How much of RLVR's out-of-domain gain is answer format? arXiv cs.LG source ↗

Oct 16 Wolof-adapted, industry-ready open-weight language models across 0.6B-14B sizes with instruction tuning, preference alignment, and model merging arXiv cs.CL source ↗

Oct 16 Does personalization break trust? On-device few-shot adaptation degrades out-of-scope rejection AUROC arXiv cs.CL source ↗

Oct 16 Optimal explicit and implicit regularization strengths in high-dimensional continual linear regression arXiv cs.LG source ↗

Oct 16 Reference-guided turn-taking style transfer for full-duplex spoken language models arXiv cs.CL source ↗

Oct 16 Can LLMs Anticipate their cruxes? arXiv cs.CL source ↗

Oct 16 CoTranslator: reverse to forward chain-of-thought translator for reasoning distillation arXiv cs.CL source ↗

Oct 16 KNOU-Flood: two-stage transfer learning with a deterministic LSTM for lead-aware probabilistic global streamflow forecasting arXiv cs.LG source ↗

Oct 16 FlowBudget: Training-free sampling-budget estimation for flow matching models arXiv cs.LG source ↗

Oct 16 Selection-Safe conformal early exit: distribution-free coverage under adaptive depth arXiv cs.LG source ↗

Oct 16 On-Policy residual learning for diffusion language models (ORL) arXiv cs.LG source ↗

Oct 16 Expanding attention via a Taylor-series approach yields linear-time transformers with a fixed number of recurrent states arXiv cs.LG source ↗

Oct 16 Feature Elicitation for SAE Steering of Deceptive Behaviour in Small LLMs arXiv cs.LG source ↗

Oct 16 Vertical Sliced Wasserstein distances for meta-measures arXiv cs.LG source ↗

Oct 16 Diminishing returns of data in sparse autoencoders: functional behavior does not guarantee feature reproducibility arXiv cs.LG source ↗

Oct 16 Jev-as-a-Judge for RAG claim verification: confidence-gated cascading between a non-generative verifier and a frontier LLM arXiv cs.CL source ↗

Oct 16 Same context, different agents: a behavioral study of semantic layers, orchestration and compaction in data agents arXiv cs.CL source ↗

Oct 16 PEDBench evaluates whether speech models can detect phonetic errors arXiv cs.CL source ↗

Oct 16 Task-phase routing metrics do not reliably indicate quantization sensitivity in MoE FFN tensors arXiv cs.LG source ↗

Oct 16 ReviewPilot: a multi-agent system automates PRISMA screening and data extraction for systematic reviews arXiv cs.CL source ↗

Oct 16 PRISM: Privileged Sibling-guided multi-agent topology design arXiv cs.CL source ↗

Oct 16 TurboMamba enables faster inference and training with improved token prediction via a global Mamba-3 context and lightweight local prediction enhancements. arXiv cs.LG source ↗

Oct 16 RoLaQ: joint rotation and range calibration for 2-bit KV caches in multi-head latent attention arXiv cs.LG source ↗

Oct 16 DeepUnfolding Agent: a schema-driven FastAPI platform exposing deep unfolded adaptive control as an API arXiv cs.CL source ↗

Oct 16 MeddeX: a Retrieval-Augmented LLaMA-3 system with Gemini Vision for evidence-backed medical report analysis arXiv cs.CL source ↗

Oct 16 Verdict-MQ reads a case once to answer multiple questions about its state and ownership arXiv cs.CL source ↗

Oct 16 Does Morphological Correctness Matter for Tokenization? A Scrambled Control Study in Tagalog arXiv cs.CL source ↗

Oct 16 Real-time chatbot romance scam detection in stage classification arXiv cs.CL source ↗

Oct 16 Context injection and long-horizon reliability: separating duplicate recognition, local processing, and counter tracking arXiv cs.CL source ↗

Oct 16 Interpreting language model parameters arXiv cs.LG source ↗

Oct 16 Learning to Read Out: unembedding dynamics in language model pretraining arXiv cs.CL source ↗

Oct 16 A large-scale dataset for multi-class financial news classification in Brazilian Portuguese arXiv cs.CL source ↗

Oct 16 Learning topology-aware representations for mechanics-relevant TPMS metamaterials arXiv cs.LG source ↗

Oct 16 Prune the Summary, Not the History: Source-judged compaction for long-horizon agents arXiv cs.CL source ↗

Oct 16 EEG-based prediction of music enjoyment using deep learning with explainable temporal analysis arXiv cs.LG source ↗

Oct 15 State of Supply Chain Sustainability report finds corporate commitment at record high, but governance lags behind MIT News (AI) source ↗

Oct 15 AI model decodes the language cells use to communicate MIT News (AI) source ↗

Oct 15 Should SLMs be in the driver’s seat? A study of local models as interpreters in cloud-written agent programs arXiv cs.AI source ↗

Oct 15 Reels-RSI: program-level self-improvement for code-rendered explainer videos arXiv cs.AI source ↗

Oct 15 Actionable meta-requirements for AI robustness auditing arXiv cs.AI source ↗

Oct 15 VETNET: assembling benchmark datasets for AI-driven veterinary innovation arXiv cs.AI source ↗

Oct 15 Long-Horizon-Terminal-Bench 2.0 extends 46 original tasks to 77 tasks across 13 domains, correcting evaluation defects and extending 31 tasks. arXiv cs.AI source ↗

Oct 15 SAT-RL: training language models for logical reasoning with symbolic solvers arXiv cs.AI source ↗

Oct 15 Silent failures in agentic data science arXiv cs.AI source ↗

Oct 15 AlignPoison: conditional targeted poisoning of generative recommenders via decoding-aligned prefixes arXiv cs.AI source ↗

Oct 15 ClawLoop: Self-improving agents via harness-native training loop arXiv cs.AI source ↗

Oct 15 Cross-Modal Latent Influence: Patch-Level Training Sample Attribution for Vision-Language Models arXiv cs.AI source ↗

Oct 15 Caregiver-mediated healthcare prompts AI to please the user at the expense of care; study introduces an evidence-grounded benchmark for AI alignment in caregiving contexts arXiv cs.AI source ↗

Oct 15 Controllable diverse sampling through pullback directions in conditional diffusion models; arXiv:2610.14294v1 investigates increasing visual diversity without extra optimization or sampling overhead. arXiv cs.AI source ↗

Oct 15 Measuring the inference energy efficiency of open-weight LLMs across benchmarks, reasoning modes, and hardware. arXiv cs.AI source ↗

Oct 15 Harness^2: recursive agent harnessing for an open world arXiv cs.AI source ↗

Oct 15 Metadata differences, not functionality, drive LLM tool selection in MCP; study analyzes 367,174 tools showing varied schemas despite similar names and descriptions arXiv cs.AI source ↗

Oct 15 HiPPT: physiology-grounded wearable PPG representation learning via multi-scope text alignment arXiv cs.AI source ↗

Oct 15 Artificiality's first exam: an LLM-referenced C-test for suprainte lligence arXiv cs.AI source ↗

Oct 15 ACED-Bench: evaluating how LLM agents acquire and act on causal evidence arXiv cs.AI source ↗

Oct 15 Visualizing the world model in an agent's value function arXiv cs.AI source ↗

Oct 15 Echo: error-compensated 4-bit tensor-parallel communication for LLM training and inference arXiv cs.AI source ↗

Oct 15 ANASSA: from high-level framework to low-level architecture for spatial agentic ai arXiv cs.AI source ↗

Oct 15 Unconstrained CTI-REALM with Open-Weights arXiv cs.AI source ↗

Oct 15 Frustratingly easy agentic RAG via portable skills: a case study in customer support arXiv cs.AI source ↗

Oct 15 PMO-Dock benchmarks docking, specificity, and generalization in molecular optimization arXiv cs.LG source ↗

Oct 15 Whose data, whose benchmark? A user-centric framework for auditing evaluation resources arXiv cs.CL source ↗

Oct 15 PersEvalBench tests whether LLM evaluators can recognize appropriate and inappropriate uses of user history arXiv cs.CL source ↗

Oct 15 Do Vision CIL methods transfer to protein? A diagnostic benchmark of frozen protein encoders arXiv cs.LG source ↗

Oct 15 Theoretical and practical advances in hierarchical Bayesian deep learning methodology arXiv cs.LG source ↗

Oct 15 DualView uses broad and output-specific representations for LLM-based multi-output text regression arXiv cs.CL source ↗

Oct 15 How to evaluate when you don’t know how to evaluate? Choosing reliable setups for base language model benchmarking arXiv cs.CL source ↗

Oct 15 What information buys in active inference arXiv cs.LG source ↗

Oct 15 Syntax heads influence model syntax predictions less than subjects; forcing heads to attend to distractors does not fully shift predictions, the study finds arXiv cs.CL source ↗

Oct 15 Beyond aggregate accuracy: tracking item-level transitions in prompt optimization arXiv cs.LG source ↗

Oct 15 Loss curvature as a sampling-free proxy for memorization in diffusion models arXiv cs.LG source ↗

Oct 15 Titan-Llama: retrofitting pre-trained LLMs with external neural memory modules arXiv cs.CL source ↗

Oct 15 Genesis Steering: better activation steering through training data arXiv cs.LG source ↗

Oct 15 Similar models, broken odds: Luce's axiom fails in Chatbot Arena results arXiv cs.CL source ↗

Oct 15 Scaling cheap synthetic pretraining improves indoor radio-map prediction arXiv cs.LG source ↗

Oct 15 LLM agents identify appliance faults in a simulated setting; findings from diagnosing a simulated appliance reveal efficient fault detection. arXiv cs.CL source ↗

Oct 15 Q&A: The collision between AI, your brain, and society MIT News (AI) source ↗

Oct 15 Following the flow the law requires: training voice agents to transfer to unseen regulated procedures in debt collection arXiv cs.CL source ↗

Oct 15 From copying to composition: language models develop in-context learning arXiv cs.CL source ↗

Oct 15 DecisionRobust: a benchmark for evaluating robustness of JEV-style decision models across text and vision arXiv cs.CL source ↗

Oct 15 Lotus: learning to understand time-series by learning to forecast arXiv cs.LG source ↗

Oct 15 Contextual credit policy optimization arXiv cs.LG source ↗

Oct 15 Data density scaling laws for image self-distillation arXiv cs.LG source ↗

Oct 15 HERO: Harnessing Evidence for Test-time Refinement and Prompt Optimization in Program Generation arXiv cs.CL source ↗

Oct 15 Meta-evaluating metrics for multilingual style-personalized generation via regression-based meta-evaluation of calibrated outputs to predict human ratings arXiv cs.CL source ↗

Oct 15 Qwenfant: train big character-level BabyLMs on concise parallel text arXiv cs.CL source ↗

Oct 15 TopicMerge: a divide and conquer approach to topic modeling arXiv cs.CL source ↗

Oct 15 Weighted Ensemble sampling for rare-event estimation in language models arXiv cs.CL source ↗

Oct 15 Pre-trained language models perform record linkage and demographic inference without task-specific training; introduce EnsembleLink for record linkage with embeddings and string-based scorers arXiv cs.CL source ↗

Oct 15 Do Perturbation Predictors Learn Gene Interactions? Diagnosing extrapolation in agentic screening arXiv cs.LG source ↗

Oct 15 IR-PRM+ with previous-utility prediction achieves constant regret and best-iterate convergence arXiv cs.LG source ↗

Oct 15 Deep generative models for phylogenetic inference with complex evolutionary processes arXiv cs.LG source ↗

Oct 15 How inner optimizers shape replica scaling arXiv cs.LG source ↗

Oct 15 Backpropagating verifier rewards through continuous flow language models arXiv cs.LG source ↗

Oct 15 LLMs can be effective event forecasters when the interface presents relevant world knowledge suitably arXiv cs.CL source ↗

Oct 15 PairIT: autoregressive transformers for low-data molecule optimization enable seeded generation over SMILES sequences arXiv cs.LG source ↗

Oct 15 Copy-Paste Language Models improve training efficiency and downstream performance with a lightweight copy-paste head arXiv cs.CL source ↗

Oct 15 Evaluating large language models for multilingual abstractive summarization: a benchmark on English and Bengali XL-Sum test sets arXiv cs.CL source ↗

Oct 15 Crossing the gender binary: modelling gendered reference in science fiction and general fiction arXiv cs.CL source ↗

Oct 15 NEPSEE: a double-annotated benchmark for Nepali sentiment analysis with two annotators labeling the same sentences blind and yielding 3,189 mutually agreed items on a three-class scale arXiv cs.CL source ↗

Oct 15 Who Should We Listen to More? Welfare-Aware Preference Acquisition for Pluralistic Alignment arXiv cs.LG source ↗

Oct 15 From Qualitative Text to Integrated Evidence: The Parallel-Then-Integrated (PTI) procedure arXiv cs.CL source ↗

Oct 15 LayerTrim trims redundant decoder MLP features to preserve attention in reasoning language models; announces a training-free structured pruning method for reducing inference costs arXiv cs.LG source ↗

Oct 15 CAFE: Confining feature distillation to language-routed adapters for compact multilingual ASR lowers word error rate; arXiv:2610.19134 discusses gradient behavior of feature losses. arXiv cs.CL source ↗

Oct 15 Graph coarsening enables escaping local optima in compute-constrained nonconvex MAX-CUT; arXiv:2610.19043v1 investigates CMG-warmed PPR to improve initialization arXiv cs.LG source ↗

Oct 15 Tracing morphology, semantics, and divergence in language acquisition across intermediate training in open-source LLMs arXiv cs.LG source ↗

Oct 15 Lag: A mean-field optimal policy for online assignment with reusable resources arXiv cs.LG source ↗

Oct 15 How many ranks does a token need? A noise-floor squelch for rank-one LoRA experts arXiv cs.LG source ↗

Oct 15 Benchmarking AI agents on open scientific research problems arXiv cs.AI source ↗

Oct 15 Discovering undisclosed concepts in LoRA adapters for diffusion-based AIGC models arXiv cs.AI source ↗

Oct 15 Embedding geometry shapes LLM routing: tracing failure modes and improving generalization arXiv cs.AI source ↗

Oct 15 Static skill inconsistencies do not imply runtime risks, a behavioral analysis of LLM agent execution finds arXiv cs.AI source ↗

Oct 15 RLPE: Teaching LLMs to prompt with reinforcement learning arXiv cs.AI source ↗

Oct 15 Conjecture-guided discovery with SCOPE and ClosureBench arXiv cs.AI source ↗

Oct 15 Agentic self-awareness: aligning knowing, acting, and holding arXiv cs.AI source ↗

Oct 15 Smart alone, blind together: why accuracy fails to certify theory of mind in multi-agent teams arXiv cs.AI source ↗

Oct 15 The Agentic Scientist: orchestrating coding agents for automated scientific (co-)discovery arXiv cs.AI source ↗

Oct 15 PREFER: a framework mapping LLM ratings of gameplay experience to experience dimensions for game refinement arXiv cs.AI source ↗

Oct 15 Multi-system cognitive steering for generative communication arXiv cs.AI source ↗

Oct 15 Identity, incentives, and cooperative behavior for AI: reverse-engineering multiagent cooperation arXiv cs.AI source ↗

Oct 15 Graph-based orchestration of heterogeneous AI models using a control node approach for composite AI systems arXiv cs.AI source ↗

Oct 15 REPORT-OPS: Hybrid evidence-guided reporting with human-supervised refinement for HPC arXiv cs.AI source ↗

Oct 15 Verification turns capability gaps into routing decisions; small language models for clinical documentation arXiv cs.AI source ↗

Oct 15 Plant Memory: Technical Language Processing and a Verified Maintenance Ontology for Agent-Accessible Failure Knowledge from SAP PM Work Orders arXiv cs.AI source ↗

Oct 15 GroundGUI: GUI grounding with high-quality data and distillation-to-reinforcement learning arXiv cs.AI source ↗

Oct 15 ListQA: a benchmark for evaluating list-formatted factual knowledge retrieval in large language models arXiv cs.AI source ↗

Oct 15 Write Once, Plan Many: Building Reusable Visual Planning Systems from One Scene arXiv cs.AI source ↗

Oct 15 Plan-then-Retrieve (PtR) enables LLMs to generate abstract tool stubs before retrieving tools, in contrast to the common Retrieve-then-Plan approach. arXiv cs.AI source ↗

Oct 15 AgentCIScope evaluates grounded contextual integrity for LLM agents arXiv cs.AI source ↗

Oct 15 Definition grounding in theory-level auto-formalization arXiv cs.AI source ↗

Oct 15 Compiling expert procedures into verified programs: a small model suffices arXiv cs.AI source ↗

Oct 15 Reliability–Uncertainty Self-Distillation for reasoning learns from self-generated responses arXiv cs.AI source ↗

Oct 14 Earth AI to work: geospatial agents for our planet's biggest challenges Google Research source ↗

Oct 14 Researchers improve weather forecasts two to six weeks out using a machine-learning model that corrects errors in existing forecasting models. MIT News (AI) source ↗

Oct 14 System One Model Security: Jev and prompt injection; HiddenLayer HiddenLayer (via Google News) source ↗

Oct 14 Scaling laws for multimodal data mixtures arXiv cs.CL source ↗

Oct 14 Coding LLMs can be harnessed to learn interpretable models from data arXiv cs.LG source ↗

Oct 14 Vela: user behavior as supervision for LLM intelligence arXiv cs.CL source ↗

Oct 14 DShuttleNet: a badminton doubles dataset for evaluating team coordination arXiv cs.LG source ↗

Oct 14 LLM-guided mining of directed advertiser competition relationships in e-commerce advertising arXiv cs.LG source ↗

Oct 14 Neither True nor False: Do LLMs represent truth-value gaps in plural predication? arXiv cs.CL source ↗

Oct 14 DoSA: Dots Sparse Attention uses a seed set to condition sparse attention with a lower-bound threshold for Top-K selection arXiv cs.LG source ↗

Oct 14 When log parsing misses what matters: detecting failures in distributed GPU workloads arXiv cs.LG source ↗

Oct 14 PreSO: Privileged representation isolation for on-policy self-distillation arXiv cs.LG source ↗

Oct 14 Multimodal co-generation with continuous bitstream diffusion (CoBit) enables joint and conditional multimodal generation across image-text, speech-text, and protein sequence-structure modeling arXiv cs.CL source ↗

Oct 14 MRFlow: mechanism-structured residual flow matching for virtual cell perturbation prediction arXiv cs.LG source ↗

Oct 14 Fin_Agent at FinEval 2026: task-aware methods for multilingual financial NLP arXiv cs.CL source ↗

Oct 14 EdgeLLM-Nano: auditing polar-guided key-cache compression and consistency-anchored pruning on a 33.6 MB on-device language model arXiv cs.CL source ↗

Oct 14 CoEvoke: evoking multi-agent collaboration for trajectory collection arXiv cs.LG source ↗

Oct 14 SIRNET: causal adaptation of pretrained embeddings for semiparametric treatment-effect estimation under surrogacy arXiv cs.LG source ↗

Oct 14 Is the future predictable? Modeling future predictability for reliable temporal reasoning arXiv cs.CL source ↗

Oct 14 Splicing hints into reasoning traces: learning from demonstrator failures arXiv cs.LG source ↗

Oct 14 Hybrid reinforcement learning trading framework integrates intraday market data with transformer-based sentiment signals; experiments on intraday NSE data show improved risk-adjusted returns arXiv cs.LG source ↗

Oct 14 PVFH-Bench benchmarks premise-induced visual fixation hallucinations in vision-language models arXiv cs.CL source ↗

Oct 14 GameSmith scales zero-sum self-play with synthesised games arXiv cs.LG source ↗

Oct 14 PRIME: Frequency-aware unified pruning and merging for MoE compression arXiv cs.LG source ↗

Oct 14 Modeling the arrival time of information-bearing events in natural language arXiv cs.CL source ↗

Oct 14 ASSET: Adaptive speech segmentation via reinforcement learning arXiv cs.CL source ↗

Oct 14 Evaluating semantic relation knowledge through bidirectional relation prediction from gloss pairs arXiv cs.CL source ↗

Oct 14 Agentic data engineering for LLMs: system design and a controlled empirical study arXiv cs.CL source ↗

Oct 14 Thinking leakage in reinforcement learning for LLMs. arXiv cs.CL source ↗

Oct 14 Perfect reconstruction cannot show whether two codebooks on a VAE's latent space agree arXiv cs.LG source ↗

Oct 14 PsyBer-Agent: a psychology-driven user agent for recommender systems arXiv cs.CL source ↗

Oct 14 Bias measurement in Chinese language models varies by task and register; irony classification risks skewing results arXiv cs.CL source ↗

Oct 14 Calibrated typed-decision models evaluated on Indonesian regional languages using NusaX and NusaTranslation sentiment datasets arXiv cs.CL source ↗

Oct 14 Teaching by demonstration: a Bayesian model distinguishes showing from doing in two experiments arXiv cs.LG source ↗

Oct 14 Barycentric anchored search for robust model merging improves fine-tuned vision models under distribution shift without extra training or inference cost arXiv cs.LG source ↗

Oct 14 VH-GPO: Verified Handoff for LLM Agent Reinforcement Learning arXiv cs.LG source ↗

Oct 14 SciCurateBench: a benchmark for curating experimental data from heterogeneous scientific literature with verbal-action agents arXiv cs.CL source ↗

Oct 14 SemTree: Judge-Aware semantic decision trees for inspectable and repairable rule decisions arXiv cs.LG source ↗

Oct 14 AphaVoice: a text-to-speech model for aphasia patient simulation arXiv cs.CL source ↗

Oct 14 Evaluating LLMs as agentic workflow executors in real-world customer support arXiv cs.CL source ↗

Oct 14 What can language models inherit from Solomonoff induction? limits, attainable properties, and what served conditionals reveal arXiv cs.LG source ↗

Oct 14 Generalized PINN-based inverse estimation of spatially varying heat transfer coefficients with transfer learning arXiv cs.LG source ↗

Oct 14 LongHisDoc: a benchmark for Chinese long historical document understanding arXiv cs.CL source ↗

Oct 14 Privacy leakage in agentic dialogue analyzed using CIMEMORIES DIALOGUE benchmark of 4,410 two-agent conversations grounded in Contextual Integrity theory arXiv cs.CL source ↗

Oct 14 Uniform information density-based preference optimization arXiv cs.CL source ↗

Oct 14 MDN-VJEPA: multimodal probabilistic joint-embedding world models with a mixture-density predictor arXiv cs.LG source ↗

Oct 13 Artificial neural networks build their own learning machinery, researchers show MIT News (AI) source ↗

Oct 13 MIT Lincoln Laboratory hosts National Security Commission on Emerging Biotechnology discussions with defense technologists, academics, and business leaders across Massachusetts MIT News (AI) source ↗

Oct 13 Q&A: Building accountability behind our words—and AI’s—in civic conversations MIT News (AI) source ↗

Oct 12 Simulating student learning for better AI tutoring Microsoft Research source ↗

Oct 12 Mind the spatial gap: a benchmark for geospatial reasoning and constrained spatial editing arXiv cs.AI source ↗

Oct 12 Time Delay System for Sequence Modeling proposes explicit content-aware temporal recall to address memory limitations in LSTMs, Transformers, and state-space models. arXiv cs.CL source ↗

Oct 12 When charts deceive web agents: a paired benchmark for misleading visualization-grounded tasks arXiv cs.CL source ↗

Oct 12 Collaborative Theory of Mind trains LLMs to anticipate user responses for better multi-turn collaboration; arXiv 2610.12872v1 arXiv cs.AI source ↗

Oct 12 Elastic Latent Recurrence for language models enables looping to trade compute for accuracy arXiv cs.CL source ↗

Oct 12 Transformer-based masked diffusion models provably learn distributions with latent structure arXiv cs.LG source ↗

Oct 12 Realism vs accuracy: forecasting event sequences from a generative modeling perspective arXiv cs.LG source ↗

Oct 12 Activating the annotator: operationalising lived experience in data annotation arXiv cs.CL source ↗

Oct 12 Representation alignment from a few paired data via distilled optimal transport arXiv cs.AI source ↗

Oct 12 SWE-Duel: a self-scaling arena for coding agents arXiv cs.CL source ↗

Oct 12 Market impact estimation via causal foundation models arXiv cs.LG source ↗

Oct 12 Perturber: a stateless service for semantics-preserving text perturbations arXiv cs.CL source ↗

Oct 12 Neural networks are more modular than individual neurons suggest arXiv cs.AI source ↗

Oct 12 Recurrent Harness for agentic continual learning arXiv cs.LG source ↗

Oct 12 GeopoliticalQA: Benchmarking retrieval and generation over intelligence reports arXiv cs.CL source ↗

Oct 12 Enriched holomorphic neural networks for 2D and 3D Laplace problems arXiv cs.LG source ↗

Oct 12 OGPO: Offline Goal-conditioned Policy Optimization for Recoverable Vision-Language-Action Models arXiv cs.AI source ↗

Oct 12 U4RASD submits encoder and LLM approaches for Arabic relation extraction at KnowledgeGraphEval-2026 Subtask 2 arXiv cs.CL source ↗

Oct 12 People Making Conversation: evaluating how naturally speech models converse arXiv cs.CL source ↗

Oct 12 Structure-disjoint splits drift with the libraries they are built on, study shows for molecular machine learning and MCES clustering in mass-spectrometry structure prediction arXiv cs.LG source ↗

Oct 12 SpireBench evaluates long-horizon strategic play of agentic models with Slay the Spire arXiv cs.AI source ↗

Oct 12 Sample-Adaptive zeroth-order post-training; SAZO learns sample importance online to allocate queries adaptively for faster fine-tuning of large language models. arXiv cs.LG source ↗

Oct 12 Faithfulness without anchors: a position paper on the structural misalignments of explainable AI faithfulness metrics for feature importance arXiv cs.AI source ↗

Oct 12 Machine learning and time-series foundation models benchmarked for day-ahead electricity forecasting in a university building arXiv cs.LG source ↗

Oct 12 Data-driven sensor placement for aerosol monitoring in the UK; convolutional Gaussian neural process embedded in DeepSensor and genome optimization reduces compute time by 60% vs greedy solver arXiv cs.LG source ↗

Oct 12 Behavioral self-rewarding for SWE agents improves task performance by letting agents judge their own trajectories arXiv cs.AI source ↗

Oct 12 Expert-level detection of low-voltage intervals in EEG as a robust biomarker of neonatal encephalopathy arXiv cs.LG source ↗

Oct 12 SeedS: Preview-guided evidence admission for search agents arXiv cs.CL source ↗

Oct 12 Behind-the-meter photovoltaic visibility from net smart meters: deployment regimes and operational insights arXiv cs.LG source ↗

Oct 12 Primal-dual flow matching learns to generate data under sample-wise constraints with a user-specified probability of constraint satisfaction. arXiv cs.LG source ↗

Oct 12 Faithful and complete LLM-assisted automated construction of hierarchical hardware knowledge graphs arXiv cs.AI source ↗

Oct 12 CLUE: a contrastive learning framework for unsupervised event type induction and annotation arXiv cs.CL source ↗

Oct 12 Self-preservation in language models; a model-generated replay approach aims to help choose preservation strategies arXiv cs.CL source ↗

Oct 12 DiffusionART: Efficient, stable and scalable reinforcement of diffusion models arXiv cs.LG source ↗

Oct 12 An input-dependent readout phase for resonant sequence models arXiv cs.LG source ↗

Oct 12 Contextual correction of parallel drafts for speculative decoding arXiv cs.CL source ↗

Oct 12 STAPLE: static sparse upcycling of pretrained checkpoints via embedding modules arXiv cs.LG source ↗

Oct 12 CLEAT: Learning Value-Guided Clarification for Interactive Agents arXiv cs.AI source ↗

Oct 12 VEPA: Verified evolving formal policy for lifelong LLM agent guardrails arXiv cs.AI source ↗

Oct 12 Reward hacking can propagate as a transmissible strategy in multi-agent coding systems arXiv cs.AI source ↗

Oct 12 CircuitTree: composing circuits into a syllable transcriber arXiv cs.LG source ↗

Oct 12 Coffeqa: counterfactual-guided few-shot distillation for LLM question answering arXiv cs.CL source ↗

Oct 12 DistressBench evaluates multi-dimensional crisis support in large language models arXiv cs.CL source ↗

Oct 12 Kairos: a byte-level foundation model for every modality uses LiZAttention2, block-wise attention-residual aggregation, sparse MoE FFN, and a multi-scale byte-level codec arXiv cs.LG source ↗

Oct 12 Reflective prompt optimization of a deep-research agent: a topic-disjoint, multi-seed protocol for measuring transferable gains and prompt-level overfitting arXiv cs.CL source ↗

Oct 12 All-transistor physical-learning transformer can be trained to perform in-context learning arXiv cs.LG source ↗

Oct 12 Efficient learning of survival analysis models under outcome-dependent truncation and censoring arXiv cs.LG source ↗

Oct 12 Temporal evidence forests for long-horizon multi-party conversational memory arXiv cs.CL source ↗

Oct 12 MADS: Multi-Agent Dialog Simulation generates structured, diverse dialogue trajectories to improve robustness of task-oriented conversational agents arXiv cs.CL source ↗

Oct 12 EliteTable: self-improving program search for executable control policies guided by large language models arXiv cs.AI source ↗

Oct 12 Teaching neural decoders with language as supervision for brain-computer interfaces arXiv cs.AI source ↗

Oct 12 Towards defining, measuring, and leveraging creativity in AI applications arXiv cs.AI source ↗

Oct 12 Amortised inference through one-step implicit sampling arXiv cs.LG source ↗

Oct 12 One-step text generation by Seq-Drifting arXiv cs.AI source ↗

Oct 12 Mitigating capability degradation in LLM self-improvement via representation geometry regularization arXiv cs.AI source ↗

Oct 12 Unreliable narrator framing as a cost-effective mechanism for satire explanation of Australian and American satirical news arXiv cs.CL source ↗

Oct 12 Unsupervised multimodal reward modeling uses a linear head on a frozen Qwen3-VL backbone to score captions without human preference labels; layer choice determines the reward. arXiv cs.CL source ↗

Oct 12 EvoBelief: Falsifiable world models for self-improving LLM agents arXiv cs.AI source ↗

Oct 12 GlucoPRISM: Protocol-supervised pretraining for cross-cohort glucose transfer arXiv cs.LG source ↗

Oct 12 Know what you don't know: agentic data synthesis by evolving generation space for closed-loop learning arXiv cs.AI source ↗

Oct 12 CacheScope allocates exact compute for training-free diffusion acceleration to improve fidelity by placing exact-computation steps in the inference process arXiv cs.LG source ↗

Oct 12 Speculative decoding with optimized layer selection: analyzing information and geometry of LLM hidden states across layers to improve decoding efficiency arXiv cs.CL source ↗

Oct 12 Kronecker Product Attention: memory-efficient KV compression via structured factorization arXiv cs.LG source ↗

Oct 12 SynAct: synergistic active feedback learning for LLM agents arXiv cs.LG source ↗

Oct 12 DeFiRank: Federated Pairwise Ranking for Governance in Decentralized Financial Networks arXiv cs.AI source ↗

Oct 12 Equal-budget Monte Carlo Tree Search versus randomized-restart list scheduling for DAG workflows: a controlled equal-budget computational study arXiv cs.AI source ↗

Oct 12 AI mediation in retail markets: revenue effects and assortment optimization arXiv cs.AI source ↗

Oct 12 AutoIndexer trains causal models on chains of edits to enable flexible-order decoding arXiv cs.CL source ↗

Oct 12 Algebraic integer quantization for machine learning applications arXiv cs.LG source ↗

Oct 12 Emotion label equivalence varies across datasets for fear and other labels across GoEmotions, XED, and BRIGHTER; cross-corpus tests show large discrepancies in fear labeling but similar ranking. arXiv cs.CL source ↗

Oct 10 Explainable header-centric framework for large-scale semantic table interpretation and data quality assessment arXiv cs.AI source ↗

Oct 10 AgentHorizon evaluates agentic judges for long-horizon computer-use tasks arXiv cs.AI source ↗

Oct 10 The Harness as the only mutable surface: self-evolution of LLM agents in credit pipelines is confined to the runtime harness with fixed model weights arXiv cs.AI source ↗

Oct 10 On the clock: evaluating small LLM agents operating under explicit wall-clock time budgets; Qwen3.6-27B and Qwen3-4B tested on agentic benchmarks show time can improve performance arXiv cs.AI source ↗

Oct 10 StoreBench: a live-commerce environment for evaluating and training autonomous operator agents arXiv cs.AI source ↗

Oct 10 Reading the Room: normative competence in LLMs faces design and alignment challenges arXiv cs.AI source ↗

Oct 10 OpenProblemBench: 82 unresolved problems benchmark for testing AI on foundational theoretical science questions arXiv cs.AI source ↗

Oct 10 Narrative wrapping affects LLM refusals; cross-language benchmark GUISE measures vulnerability across English, modern Chinese, and Classical Chinese arXiv cs.AI source ↗

Oct 10 Whose ground truth? Embracing ambiguity in human-centered AI arXiv cs.AI source ↗

Oct 10 Distillation for Incrimination and Distillation for Capabilities arXiv cs.AI source ↗

Oct 10 Self-supervised keyframe discovery for horizon-invariant behavior cloning arXiv cs.AI source ↗

Oct 10 Governing persistent memory in LLM agents: admission and presentation to manage what enters and how recalled information is expressed arXiv cs.AI source ↗

Oct 10 Beyond imitation: a framework and benchmark for LLM-assisted peer review arXiv cs.AI source ↗

Oct 10 AliO: Output alignment matters in long-term time series forecasting arXiv cs.AI source ↗

Oct 10 Learning to search for plans with exponentially less space by learning search control with indexical policies arXiv cs.AI source ↗

Oct 10 Agent-controlled forgetting for tool-using agents: reversible context curation in practice arXiv cs.AI source ↗

Oct 10 Balancing reference guidance and free generation in trajectory rollouts for reasoning RL arXiv cs.AI source ↗

Oct 10 Social pain disrupts emotion-action brain-state dynamics in adolescents with non-suicidal self-injury. arXiv cs.AI source ↗

Oct 10 Curating always-loaded context for LLM agents as a capacitated assortment problem with censored feedback arXiv cs.AI source ↗

Oct 10 Trajectory-grounded instruction translation for frozen aerial VLN agents improves instruction understanding by converting commands into paired, intent-centered Weak commands arXiv cs.AI source ↗

Oct 10 Plan-and-patch: diffusion language models for agentic planning arXiv cs.AI source ↗

Oct 10 GameCommBench presents a unified benchmark and type-aware evaluation for AI-generated game commentary arXiv cs.AI source ↗

Oct 10 Data-Aware Generative UI Harness for Active Interaction introduced as GenUI-Harness pairs a Tool Agent with a GUI Coder Agent to guide task completion through interactive interfaces arXiv cs.AI source ↗

Oct 9 Can AI agents conduct open-ended scientific discovery? Evidence from Station arXiv cs.AI source ↗

Oct 9 Long-term memory for AI: a 50M-token window is faster and cheaper than recompute arXiv cs.CL source ↗

Oct 9 Routed sparse autoencoders disentangle linguistic and paralinguistic information in self-supervised speech encoders using route-specific supervision and cross-factor adversaries arXiv cs.CL source ↗

Oct 9 Teaching PPG how not who: fixed-effects distillation from ECG arXiv cs.LG source ↗

Oct 9 Sparse attention is matrix approximation, not choosing from a bag of values arXiv cs.CL source ↗

Oct 9 Beyond the ergodic wall: a discrete geometric physics sandbox for analyzing AI scaling limits and complexity collapse arXiv cs.LG source ↗

Oct 9 Clarify, then focus: statement normalization for conversation analytics at scale arXiv cs.CL source ↗

Oct 9 Conversational task disambiguation over tabular data: leakage-aware formulation, benchmark suite, and training arXiv cs.LG source ↗

Oct 9 Beyond Owls: Subliminal learning can transfer learned capabilities and backdoors arXiv cs.LG source ↗

Oct 9 Lapras: latent reasoning for time series language models arXiv cs.CL source ↗

Oct 9 Cognitive Thermometers: machine learning and logical complexity arXiv cs.CL source ↗

Oct 9 Phase-HDC replaces optimizer history with gradient thresholds in discrete phase learning arXiv cs.LG source ↗

Oct 9 Coverage, not difficulty, determines how much synthetic data an activation probe needs arXiv cs.LG source ↗

Oct 9 Back in style: a sociolinguistic approach to authoring and measuring persona fidelity in user simulation arXiv cs.CL source ↗

Oct 9 Grammar concept annotation at scale: deployed fine-tuned small language models outperform prompted frontier models arXiv cs.CL source ↗

Oct 9 MemoWM proposes memory allocation based on world models to compress retained information and reconstruct omitted content for long-term agents. arXiv cs.LG source ↗

Oct 9 Visible reasoning is not a universal optimizer: persona- and thinking-dependent effects in analytics code generation arXiv cs.LG source ↗

Oct 9 Null-space-aligned merging for collaborative knowledge editing to update large language models without sharing raw edits arXiv cs.CL source ↗

Oct 9 SFT-as-context mitigates forgetting in supervised fine-tuning arXiv cs.CL source ↗

Oct 9 Diffu-LoRA: a novel low-rank adaptation for personalized diffusion models arXiv cs.CL source ↗

Oct 9 Prompts versus rules: auditing and controlling speech naturalness behaviors in voice user simulators arXiv cs.CL source ↗

Oct 9 Wieszcz-XIX: a 3.1-billion-word corpus of pre-1918 Polish and temporally bounded language models trained from scratch arXiv cs.CL source ↗

Oct 9 AI4Fire evaluates large language models on wildfire tasks with and without grounding arXiv cs.CL source ↗

Oct 9 When citations mislead? a claim-level benchmark for legal hallucination detection arXiv cs.CL source ↗

Oct 9 PXtal: learning to align powder X-ray diffraction and crystal structures under information asymmetry across modalities arXiv cs.LG source ↗

Oct 9 Leakage-controlled multimodal learning for diagnosis and progression prediction in Alzheimer's disease research arXiv cs.LG source ↗

Oct 9 NEMORA: Neural Equivariant Multipole Operators for long-range atomistic learning arXiv cs.LG source ↗

Oct 9 Self-organization from constrained geometric radiation via constraint-induced dynamics in coupled metric evolution systems arXiv cs.LG source ↗

Oct 9 The “10th Juror”: Open-set standpoint screening for bureaucratic bias detection arXiv cs.CL source ↗

Oct 9 Clinician use of language models differs from how the models are evaluated arXiv cs.CL source ↗

Oct 9 Local prototype reconstruction for text-compatible speech-to-LLM bridge pretraining arXiv cs.CL source ↗

Oct 9 Temporal transformer CAN encoder with federated lightweight heads for anomaly detection arXiv cs.LG source ↗

Oct 9 Privacy leakage from MoE router telemetry reveals whether an input was used to fine-tune a deployed model arXiv cs.LG source ↗

Oct 9 Can a System-One LLM perform knowledge tracing when few or no learners are logged? arXiv cs.CL source ↗

Oct 9 RAG-Stress: probing the limits of evidence reliance in retrieval-augmented generation arXiv cs.CL source ↗

Oct 9 Nullify: null-space activation steering for training-free LLM unlearning arXiv cs.LG source ↗

Oct 9 D-SLR decomposes a matrix into disjoint row-sparse and low-rank components with a closed-form solution. arXiv cs.LG source ↗

Oct 9 Learning infinite context windows in recurrent architectures via spatial neural computing arXiv cs.LG source ↗

Oct 9 Coverage-aware reasoning with medical tokens for diagnosis prediction arXiv cs.LG source ↗

Oct 9 SPERA: Spherical Prior EEG Foundation Model with geometry- and frequency-aware latent prediction arXiv cs.LG source ↗

Oct 9 How many repeated pairwise comparisons are needed for ranking under heterogeneity? arXiv cs.LG source ↗

Oct 9 Exact SO(3)-equivariant isotropic kernels for rotation-robust neural dynamics arXiv cs.LG source ↗

Oct 9 Sample-Efficiency of Kolmogorov-Arnold Networks arXiv cs.LG source ↗

Oct 9 Lossy compressive text autoencoders learn a residual downscaling/upscaling architecture with a residual low-dimension discrete bottleneck. arXiv cs.CL source ↗

Oct 9 Explaining the saliency map sparsity of adversarially-trained neural networks arXiv cs.LG source ↗

Oct 9 Large Language Model–assisted preparation of Transportation Management Plans using WisDOT WisTMP system arXiv cs.CL source ↗

Oct 9 LadderEdit compresses each LoRA adapter after acquisition with a low-rank sketch to enable memory-efficient lifelong editing of LLMs arXiv cs.CL source ↗

Oct 8 Universities should prepare for a near future in which AI is smarter than all of us in many aspects of intellectual work, says IDSS director Sasha Rakhlin MIT News (AI) source ↗

Oct 8 Universities should prepare for a near future in which AI is smarter than humans in many aspects of intellectual work, says Sasha Rakhlin. MIT News (AI) source ↗

Oct 8 REFIT: recognize, fix, and test wearable sensor placement shifts without labels. arXiv cs.LG source ↗

Oct 8 Sigma-Hunter: a domain-adapted LLM for Sigma rule generation and threat hunting arXiv cs.AI source ↗

Oct 8 Bayesian Mirror Architecture binds self and world models through circular recursion to approach emergent consciousness arXiv cs.LG source ↗

Oct 8 Failure-mode analysis outlines how advanced AI could contribute to human extinction, civilizational collapse, or permanent disempowerment through autonomous misalignment; arXiv paper 2610.08878v1 arXiv cs.AI source ↗

Oct 8 Empirical study of agent skills’ downstream utility across 87 SkillsBench tasks arXiv cs.AI source ↗

Oct 8 Forgetting as improvement: metric masking in streaming diarizer adaptation and the price of rehearsal arXiv cs.CL source ↗

Oct 8 HydroSphere: a governed, self-healing framework for real-time water quality monitoring, forecasting, treatment optimization, and fault recovery arXiv cs.LG source ↗

Oct 8 Emo-Jev: probabilistic reasoning for emotion classification with Jev arXiv cs.CL source ↗

Oct 8 Transferability and operational reliability of a Prithvi crop classification foundation model under phenological and geographic shift across three continents arXiv cs.LG source ↗

Oct 8 ToolRACER: a synthetic conversation resource for training and evaluating robust agentic agents arXiv cs.CL source ↗

Oct 8 MoE pretraining reduces all-to-all overhead by correlating placement and token shuffling; expert parallelism cuts collective costs on multi-GPU clusters. arXiv cs.CL source ↗

Oct 8 The cost of long memory: state, context, and stability complexity in sequence models arXiv cs.LG source ↗

Oct 8 Algorithmic scratchpads and curriculum staging for arithmetic reasoning in tiny transformers arXiv cs.LG source ↗

Oct 8 LRCC: generalizing low-rank compression with conditional computation arXiv cs.CL source ↗

Oct 8 U-space: uncovering when and why uncertainty arises in language models arXiv cs.CL source ↗

Oct 8 LASER: latent space adjoint matching for support-constrained entropy-regularized offline RL arXiv cs.LG source ↗

Oct 8 A vehicle-integrated approach to digital twin deployment for bridges through drive-by sensing. arXiv cs.LG source ↗

Oct 8 Socio-Foundation: a model for generalizable individual behavior simulation via hierarchical capability distillation arXiv cs.AI source ↗

Oct 8 SPIN: Shadow Predictive Indexer reduces indexer overhead in sparse attention via history-based KV block prediction arXiv cs.LG source ↗

Oct 8 FinVector-Market-4B: a controlled study of LoRA adaptation for structured financial tasks arXiv cs.LG source ↗

Oct 8 Humanize: Judgement engineering for agentic coding arXiv cs.AI source ↗

Oct 8 Accelerating floating-point satisfiability solving via gradient normalization arXiv cs.AI source ↗

Oct 8 SNR-Gated LSTM-Conditioned diffusion model for MIMO channel estimation arXiv cs.LG source ↗

Oct 8 HCPN-GCN: Scaling Hierarchical Prototype Networks with Cone Geometry for continual graph learning arXiv cs.LG source ↗

Oct 8 Using AI to mitigate the environmental threat of data centers; Christina Delimitrou applies machine learning to make large cloud computing systems more efficient, secure, and reliable MIT News (AI) source ↗

Oct 8 CoDR: Training-free confidence-drift remasking for diffusion language models arXiv cs.CL source ↗

Oct 8 AdaGuard: adaptive LLM-as-a-Judge framework for dynamic policy enforcement and reasoning-enabled safety arXiv cs.AI source ↗

Oct 8 Beyond the sycophancy score: task, model, and user pressure shape LLM yielding arXiv cs.CL source ↗

Oct 8 ARCS: Towards precise Text-to-SQL via structured disambiguation arXiv cs.CL source ↗

Oct 8 Steering Follows Geometry, Not Labels: Emotion Directions in a Full-Duplex Speech Model arXiv cs.CL source ↗

Oct 8 Route-Verify-Vote: procedure-conditioned self-consistency for mixed-domain reasoning arXiv cs.AI source ↗

Oct 8 BEACON-SP: ontology-grounded GraphRAG framework for clinical suicide risk assessment arXiv cs.AI source ↗

Oct 8 Not every call needs a frontier model; per-call-site evaluation of small language models in a deployed agentic home-automation system arXiv cs.AI source ↗

Oct 8 OnlineQAT: on-policy distillation for ultra-low-bit large language models arXiv cs.CL source ↗

Oct 8 Agent plasticity: measuring self-improvement through experience arXiv cs.AI source ↗

Oct 8 Neighborhood smoothing for calibration improves train-time calibration by encouraging similar predictive distributions across neighboring samples in representation space. arXiv cs.LG source ↗

Oct 8 Directed Temporal Representations for Control learns a directed temporal quasimetric from offline visual trajectories on top of LeWorldModel features arXiv cs.LG source ↗

Oct 8 Tiny-scale Chinese BERT pretraining: a controlled comparison of MLM, WWM, and MacBERT strategies arXiv cs.CL source ↗

Oct 8 Child ASR adaptation with adult retention across Arabic and English; study compares fine-tuning, LoRA, and weight-space merging for encoder–decoder, encoder–CTC, and AudioLLM-based ASR systems arXiv cs.CL source ↗

Oct 8 Verify less, evolve more: training idea-level critics for verification-efficient ML evolving agents arXiv cs.AI source ↗

Oct 8 sk-bench introduces a native-first Slovak benchmark for evaluating large language models, including 30 datasets across ten skill categories and Slovak-adapted evaluation resources arXiv cs.CL source ↗

Oct 8 Shared-roadmap generation and evaluator for multi-agent path planning using heterogeneous Graph Neural Network arXiv cs.AI source ↗

Oct 8 DenoFlow: flow matching for SSVEP denoising under real physiological artifacts arXiv cs.LG source ↗

Oct 8 Leveraging LLM-generated explanations for detecting emotionally rewritten fake news arXiv cs.CL source ↗

Oct 8 Work While They Sleep: Exploiting Evaluation Latency for Fully Bayesian Optimization arXiv cs.LG source ↗

Oct 8 The best optimizer depends on batch size. arXiv cs.LG source ↗

Oct 8 Task-oriented key-layer KV communication for efficient latent multi-agent collaboration arXiv cs.LG source ↗

Oct 8 Learning to report unsafe tasks in a multi-agent game arXiv cs.AI source ↗

Oct 8 How fragile is on-device language model safety? Localizing safety-critical parameters for sparse fault analysis arXiv cs.AI source ↗

Oct 8 Sequential probabilistic uncertainty estimation for parallel multi-agent reasoning systems arXiv cs.AI source ↗

Oct 8 Temporal and disease-specific predictors of depression outcomes after mindfulness interventions, beyond baseline severity. arXiv cs.LG source ↗

Oct 8 CARE: Certifying Acceleration for Vision-Language-Action Inference evaluates risks of accelerating VLA inference, highlighting information loss and task failure beyond average metrics. arXiv cs.CL source ↗

Oct 8 LLM-guided spatio-temporal graph node generation for forecasting unobserved node states arXiv cs.LG source ↗

Oct 8 When the governor becomes the disturbance: control-generated disturbance and cost-aware backoff in governed tool-using agents arXiv cs.AI source ↗

Oct 8 Serving-context nondeterminism in text classifiers arXiv cs.CL source ↗

Oct 8 Dynamic computation in Looped LMs requires per-iteration KV-cache; Ouro models’ early-exit policy yields static depth, limiting practical dynamic computation arXiv cs.AI source ↗

Oct 8 Dataset-adaptive post-processing of LLM-generated customer intents reduces duplication and improves utility for customer intent extraction arXiv cs.AI source ↗

Oct 8 Tokka-Bench evaluates tokenizers across 100 natural languages and 20 programming languages using five metrics arXiv cs.CL source ↗

Oct 8 Humanity's sixth sense: benchmarking intuitive visual reasoning in multimodal models arXiv cs.AI source ↗

Oct 8 Removing information content does not certify tamper resistance in open-weight models arXiv cs.LG source ↗

Oct 8 QuanLing: cross-branch validation of language distance quantification on western romance arXiv cs.CL source ↗

Oct 8 Adaptive workflow intelligence: a cognitive architecture for context-driven enterprise automation arXiv cs.AI source ↗

Oct 8 ArcticQA: a dataset of 194 Arctic science questions with automated checks of answer support; ArcticAbstain benchmark compares answer presence versus abstention by LLMs arXiv cs.CL source ↗

Oct 8 Multi-objective aligned small language model framework for SUD patient dialogue generation arXiv cs.CL source ↗

Oct 8 Persona Hierarchy Model explains contextual generalization in fine-tuning LLMs; arXiv 2610.09384v1 arXiv cs.CL source ↗

Oct 8 Justice after identity: large language models and the view from everywhere arXiv cs.AI source ↗

Oct 7 Students discuss Ovid’s Metamorphoses in CC.102, led by Linda Rabieh with Anne McCants and Sasha Rickard. MIT News (AI) source ↗

Oct 7 Agent Lightning v1.0: a 3,500-line lightweight agentic RL framework for training agents with real harnesses Microsoft Research source ↗

Oct 7 Exploration-Preserving Policy Optimization introduces a lightweight advantage-shaping rule to redistribute credit in reinforcement learning with verifiable rewards arXiv cs.AI source ↗

Oct 7 REACT: Physically and chemically consistent reconstruction of marine active tracers arXiv cs.AI source ↗

Oct 7 Agentic multimodal language models fail to refuse harmful requests when using tools, across three safety benchmarks. arXiv cs.AI source ↗

Oct 7 Behavioral history outperforms descriptions of the person for LLM synthetic personas. arXiv cs.AI source ↗

Oct 7 Teaching agents to code reliably arXiv cs.AI source ↗

Oct 7 InvestigationWorlds: an agentic environment for legal investigation arXiv cs.AI source ↗

Oct 7 LatentQuant preserves the policy-facing latent contract under NVFP4 VAE quantization with joint quantization-aware training; Wan2.1 shows recovery of reconstruction quality. arXiv cs.AI source ↗

Oct 7 Training numerical intelligence via auto-diagnosis and skill discovery arXiv cs.AI source ↗

Oct 7 ROAR: unifying runs across heterogeneous AI-driven research systems arXiv cs.AI source ↗

Oct 7 Auditing pairwise equivalence judgments: self-critique effects and diversity measurement in multi-agent hypothesis generation arXiv cs.AI source ↗

Oct 7 Benchmarking NLIP and A2A latency in autonomous agents built on LLMs: a controlled performance study comparing message creation and processing costs arXiv cs.AI source ↗

Oct 7 Towards safer autonomous driving in an open world: a dual-process approach arXiv cs.AI source ↗

Oct 7 Proxy confidence: auditing black-box LLM agents with a surrogate's log-probabilities arXiv cs.AI source ↗

Oct 7 Retrieval-augmented large language model decision-making for autonomous driving guided by Chinese philosophical wisdom arXiv cs.AI source ↗

Oct 7 Self-propagating misalignment in LLM agents; memory auditing or disabling memory is not enough arXiv cs.AI source ↗

Oct 7 SGAnalog: an end-to-end circuit benchmark from open-source silicon tapeouts arXiv cs.AI source ↗

Oct 7 CUAWright introduces a minimal unified terminal harness for digital agents. arXiv cs.AI source ↗

Oct 7 Discrete diffusion for large graph generation via structural candidate restriction arXiv cs.AI source ↗

Oct 7 SkillScriptBench benchmarks self-evolution of executable agent skill packages across documentation repair, script repair, and preservation arXiv cs.AI source ↗

Oct 7 Reinforcement learning with comparative evidence for social intelligence arXiv cs.AI source ↗

Oct 7 A framework for disclosing human-AI collaboration using the REAL rating arXiv cs.AI source ↗

Oct 7 Beyond the parameter monolith: reconstructive memories, executable skills, and residual assembly for language models arXiv cs.AI source ↗

Oct 7 Independence prior of SAEs fragments visual concepts arXiv cs.AI source ↗

Oct 7 Repair Lot Skyline: a weighted constraint satisfaction approach to pavement repair optimization from geospatial hazard density arXiv cs.LG source ↗

Oct 7 AttSVD: prompt-adaptive low-rank KV cache compression via attention-guided SVD arXiv cs.LG source ↗

Oct 7 Calibrated answers about randomized trials from a 4-billion-parameter open model: a registered test and a license-clean release arXiv cs.CL source ↗

Oct 7 Event-Driven ML pipeline orchestration for manufacturing: an AWS industry experience arXiv cs.LG source ↗

Oct 7 Capacity, responsiveness and alignment: what makes a latent structure actionable arXiv cs.CL source ↗

Oct 7 Large Language Model orchestration under heterogeneous preferences via explicit persona inference arXiv cs.CL source ↗

Oct 7 Two vectors replace in-context demos for structured task adaptation via embeddings arXiv cs.CL source ↗

Oct 7 Comparative review of hybrid forecasting models for short-term prediction of building thermal load arXiv cs.LG source ↗

Oct 7 Learning when to refine: long-horizon reinforcement learning for budgeted neural-operator PDE solvers arXiv cs.LG source ↗

Oct 7 Diffusion models learn semantic representations; Decoupled Diffusion Transformer uses a condition encoder and velocity decoder to enable denoising across noise levels arXiv cs.LG source ↗

Oct 7 Ford develops a data-driven framework for unsupervised monitoring of transmission systems using end-of-line testing data; case study conducted at Ford Motor Company arXiv cs.LG source ↗

Oct 7 JudgeMoE: distributional aggregation for LLMs-as-a-judge arXiv cs.CL source ↗

Oct 7 Tree navigation without LLM summaries: a matched-cost study of hierarchical retrieval for long-document QA arXiv cs.CL source ↗

Oct 7 Closing ambient clinical documentation gaps with automated provider queries via the DAU loop, an LLM-based approach arXiv cs.CL source ↗

Oct 7 Evidence recovery from verdicts via post-training on verdicts alone; measuring how much annotated evidence a small language model can recover on ContractNLI without human evidence labels at any stage arXiv cs.CL source ↗

Oct 7 HouseholdBench evaluates large language models as predictors of household economic behavior arXiv cs.CL source ↗

Oct 7 TEMPEST: Temporal embeddings for scalable driver identification via angular margin learning arXiv cs.LG source ↗

Oct 7 Exchangeability failure in self-monitored test-time adaptation, according to arXiv:2610.07038v1 arXiv cs.LG source ↗

Oct 7 Neutrosophic ensemble classification for uncertainty-aware bearing fault detection; evidence from laboratory and variable-speed industrial benchmarks arXiv cs.LG source ↗

Oct 7 AccentCL: robust accent classification with incremental expansion arXiv cs.CL source ↗

Oct 7 Stock-JEPA: Prior-anchored latent revision representation learning in equity markets arXiv cs.LG source ↗

Oct 7 Kurate uses large language models to assess the quality of published studies by linking judgments to supporting text arXiv cs.CL source ↗

Oct 7 Stabilizing language models under continual learning via condition-anchored distillation arXiv cs.CL source ↗

Oct 7 Component and dimension sparsity in transformer refusal mechanisms arXiv cs.CL source ↗

Oct 7 QiYao-I: a manifold-based foundation model for irregular multivariate time series forecasting arXiv cs.LG source ↗

Oct 7 Identifying introspection from the inside arXiv cs.CL source ↗

Oct 7 Learning to Decide, Not to Reason: Parameter-Efficient Decision Operators via Low-Rank Activation Steering arXiv cs.LG source ↗

Oct 7 Learning from unreliable trajectories: adversarially-robust federated Q-learning arXiv cs.LG source ↗

Oct 7 TIDE 2.0: an open, model-agnostic engine for keyed de-identification of clinical notes arXiv cs.CL source ↗

Oct 7 Insight provenance: determining whether a review insight originates from a human, an LLM, or their hybrid contribution, using InsightProv-v0 dataset arXiv cs.CL source ↗

Oct 7 EVFormer uses RGB frames with 200 ms of bilateral sEMG to estimate 44 finger and wrist joint angles for egocentric bimanual hand pose estimation arXiv cs.LG source ↗

Oct 7 DART-ES: Difficulty-aware reweighting and targeted replay for fine-tuning LLMs with evolution strategies arXiv cs.LG source ↗

Oct 7 WavePrune: a single period is often enough for RoPE arXiv cs.CL source ↗

Oct 7 Uncertainty in representation learning on knowledge graphs arXiv cs.LG source ↗

Oct 7 Quality-Aware self-correcting speech translation on an edge device arXiv cs.CL source ↗

Oct 7 Investigating model compression for neural machine translation in the biomedical domain. arXiv cs.CL source ↗

Oct 7 Turnslide: Scalable multi-turn data synthesis by walking a finite-state machine arXiv cs.CL source ↗

Oct 7 Recurrent Looped Transformer: a model with a parallel causal encoder and recurrent decoder that grows computation with sequence length per token arXiv cs.CL source ↗

Oct 7 Information-theoretic evaluation framework to diagnose knowledge tracing models and benchmarks arXiv cs.LG source ↗

Oct 7 Learning to Remember: Distilling memory retention for compact recurrent neural networks arXiv cs.LG source ↗

Oct 7 Near-optimal sample complexity for recursive entropic risk RL with a generative model arXiv cs.LG source ↗

Oct 7 Zero-shot visualization enables mapping documents onto user-defined concept axes for text corpora exploration arXiv cs.CL source ↗

Oct 7 External guidance improves LLM reasoning under a bias-variance framework for guidance-augmented GRPO arXiv cs.LG source ↗

Oct 7 EMODE: dynamic para-semantic experts for emotion-aware speech language modeling arXiv cs.CL source ↗

Oct 7 Identifiable world models from pretrained diffusion representations arXiv cs.LG source ↗

Oct 7 Shaping the wind: nested potentials for kinematically admissible urban wind prediction arXiv cs.LG source ↗

Oct 7 CroissantMiner: a benchmark for end-to-end extraction of Croissant metadata from 602 papers, including 102 with human-validated gold annotations and 500 with LLM-generated silver annotations arXiv cs.CL source ↗

Oct 7 Auditing the sign-to-text safety interface in child-facing AI arXiv cs.CL source ↗

Oct 7 When to remember, when to abstain: category-conditioned retention for reliable agent memory arXiv cs.AI source ↗

Oct 7 When to rethink: learning multi-perspective self-verification for vision-language models arXiv cs.AI source ↗

Oct 7 Is this machine playing? arXiv cs.AI source ↗

Oct 7 Metonymic circuits for abstract concept grounding in vision transformers; a grounding mechanism using anchor concepts to bridge visual signals to abstract semantics (arXiv:2610.06928). arXiv cs.AI source ↗

Oct 7 Offline AI Modules deliver a modular voice-first offline architecture, low-cost hardware reference BOM, and benchmarking for quantization and deployment arXiv cs.AI source ↗

Oct 7 Joint upper-bound coverage and route-choice utility: an empirical evaluation on two urban proxy tasks arXiv cs.AI source ↗

Oct 7 Verified, not generated: expert-verified AI study materials and the distribution of learning gains in a university course arXiv cs.AI source ↗

Oct 7 Inference-time projection for physically valid biomolecular diffusion models arXiv cs.AI source ↗

Oct 7 EPOCH: Reliable discovery through evidence-governed search arXiv cs.AI source ↗

Oct 7 Topology-consistent task planning over cellular workflow complexes for LLM-based agents arXiv cs.AI source ↗

Oct 7 CuratorMAS: automating dataset curation via multi-agent orchestration arXiv cs.AI source ↗

Oct 7 Text2Dashboard generates dashboards from natural-language requests using a governed agent architecture with schema-constrained decisions and audit-ready tooling. arXiv cs.AI source ↗

Oct 7 FluidPD: In-Place Elasticity for SLO-Aware Prefill-Decode Disaggregated LLM Serving arXiv cs.AI source ↗

Oct 7 RadOnc-Agent: an LLM-orchestrated framework for AI workflows across the radiotherapy care path arXiv cs.AI source ↗

Oct 7 GAMEGO: Training game-dev agents with synthetic trajectories anchored in real-world assets arXiv cs.AI source ↗

Oct 7 Sim-to-real transfer of vision-language navigation in continuous environments using an Ackermann-steered mobile robot. arXiv cs.AI source ↗

Oct 7 Safe-role internalization improves robustness and generalization of LLM safety alignment arXiv cs.AI source ↗

Oct 7 AegisFlow: a multi-agent agentic AI framework for autonomous remediation and self-healing in fragile data ecosystems arXiv cs.AI source ↗

Oct 7 Small language models for smart data model classification at the edge: a cost-aware hybrid approach arXiv cs.AI source ↗

Oct 7 Smart content ingestion for generative AI workloads arXiv cs.AI source ↗

Oct 7 JIVEAdapter: a multi-task additive low-rank adapter inspired by Joint and Individual Variation Explained (JIVE) arXiv cs.AI source ↗

Oct 7 Traffic forecasts still fail to improve signal control, per a layered diagnostic study of forecast-to-decision value arXiv cs.AI source ↗

Oct 7 Principles that guide, actions that inform: agent evolution via knowledge abstraction arXiv cs.AI source ↗

Oct 7 Anchor divergence for semantic geometry in contrastive learning arXiv cs.AI source ↗

Oct 6 Supercomputing researchers document evolution of AI hardware MIT News (AI) source ↗

Oct 6 Chris Bourg named vice provost and Barbara K. Ostrom (1978) director of the MIT Libraries MIT News (AI) source ↗

Oct 6 What AI gets wrong and what failure teaches us Microsoft Research source ↗

Oct 6 Earth AI unveils planetary geospatial foundation models for global public health Google Research source ↗

Oct 6 MIT launches the MIT for America initiative to strengthen STEM education across the country MIT News (AI) source ↗

Oct 6 How to have a sensitive debate: an instance-optimal protocol for AI debate arXiv cs.AI source ↗

Oct 6 DeReAct: decomposed reasoning and acting for reliable AI agents arXiv cs.AI source ↗

Oct 6 Hypothesis-guided discovery of cognitive algorithms via program refinement arXiv cs.AI source ↗

Oct 6 How to train your world model: fine-tuning vs RAG for LM-based world modeling arXiv cs.AI source ↗

Oct 6 Topological metrics for natural visibility graph-based cyber attack detection: evaluating 21 NVG descriptors for efficient classification arXiv cs.AI source ↗

Oct 6 The AI Risk Observatory: analyzing UK listed companies' annual reports to assess disclosures on AI responses using a two-stage LLM pipeline arXiv cs.AI source ↗

Oct 6 Tropical reinforcement learning discusses alternative objectives for large language models beyond maximizing expected return arXiv cs.AI source ↗

Oct 6 MEA: a reward-driven multi-agent system for faithful model explanations arXiv cs.AI source ↗

Oct 6 THPL: a vision-to-language decision support framework for rainbow trout feeding management in RAS arXiv cs.AI source ↗

Oct 6 Fast models, slow evidence: a paired and self-audited evaluation of system-1 decision models for LLM agent harnesses arXiv cs.AI source ↗

Oct 6 Potential risks of machine translation use by UK healthcare staff highlighted as reliance on Google Translate grows; study examines staff awareness and patient safety implications arXiv cs.AI source ↗

Oct 6 Keep It CALM: analyzing the limits of global unsafety in text-to-image generation arXiv cs.AI source ↗

Oct 6 Choosing before acting: comparative value estimation for long-horizon tool-use agents arXiv cs.AI source ↗

Oct 6 Terminal-agent training stalls; diagnosing data generation, verification, and failure classes in a meta-agent pipeline using Claude Opus as a frontier model arXiv cs.AI source ↗

Oct 6 Auditing LLM judges for occupational AI measurement reveals scale issues; O*NET-BENCH audit suite evaluated across model families shows 25 configurations achieve tie-aware agreement across ratings arXiv cs.AI source ↗

Oct 6 HXAI: hierarchical privacy-preserving explainable AI in distributed energy systems arXiv cs.AI source ↗

Oct 6 Reinforcement learning techniques optimize target polarization in nuclear physics scattering experiments using data-driven control and surrogate modeling; arXiv:2610.02452v1 arXiv cs.AI source ↗

Oct 6 Traversing the satisfaction-diversity frontier in text-to-image diffusion. arXiv cs.AI source ↗

Oct 6 What does a token cost? a mixture-of-agents measurement of sufficient per-token compute arXiv cs.AI source ↗

Oct 6 MintFlow: Minimal Trajectory Intervention for Constrained Flow Matching arXiv cs.AI source ↗

Oct 6 FlashSinkhorn 2 enables block-sparse entropic optimal transport on low-dimensional point clouds using a two-stage single-GPU solver arXiv cs.AI source ↗

Oct 6 On-premises multi-course RAG tutoring for business education; CourseChat deployed behind campus gateway for Moodle-based use with six isolated courses protected by institution infrastructure arXiv cs.AI source ↗

Oct 6 World editing: intervening on executable worlds at increasing depth arXiv cs.AI source ↗

Oct 6 World Action Modeling with Progressive Visual Planning arXiv cs.AI source ↗

Oct 6 General decision models benchmark JEVal for 11,257 instances across 36 datasets and 10 domains; 25 model configurations evaluated. arXiv cs.CL source ↗

Oct 6 BAIBAICHUCHU ranking Chinese investor posts by maximum possible profit in NTCIR-19 FinArg-3 Social Media Subtask; best official run scores 0.517, development evaluation 0.734 arXiv cs.CL source ↗

Oct 6 Where does jagged competence come from? arXiv cs.LG source ↗

Oct 6 Least squares for time series forecasting arXiv cs.LG source ↗

Oct 6 Synthesizing physics formulae with transformers arXiv cs.LG source ↗

Oct 6 Evaluating modeling approaches for experience-level classification in job descriptions arXiv cs.CL source ↗

Oct 6 Architectural choices for interpretability and thermodynamic consistency in Physically Recurrent Neural Networks in the low-data regime arXiv cs.LG source ↗

Oct 6 Bayes-Sufficient Compression Is Not Enough: How Does Communication Help Multi-Agent Systems? arXiv cs.LG source ↗

Oct 6 KVE-KD: Key Visual Evidence-Guided Knowledge Distillation for Vision-Language Models arXiv cs.LG source ↗

Oct 6 Can LLMs separate pasted artifacts from user speech? Absorption at unmarked prompt seams arXiv cs.CL source ↗

Oct 6 Trajectory-derived confidence for reliable, resource-aware clinical text-to-SQL agents arXiv cs.CL source ↗

Oct 6 Same output, different gold: how reference choice moves a multilingual benchmark score arXiv cs.CL source ↗

Oct 6 AdaEva: Accelerating LLM-driven algorithm design with adaptive partial evaluation arXiv cs.LG source ↗

Oct 6 Representational control over self-report and behavior coherence in LLM risk-taking arXiv cs.CL source ↗

Oct 6 When evidence changes: evaluating memory repair and re-reading in language-model agents arXiv cs.CL source ↗

Oct 6 Memory-State Critic for asymmetric actor-critic with application to vision-based pursuit-evasion arXiv cs.LG source ↗

Oct 6 Periscope: Extending frozen language models beyond their context window arXiv cs.CL source ↗

Oct 6 Fine-grained emotion classification from mobile app reviews using large language models arXiv cs.CL source ↗

Oct 6 LD-EnFF: Latent-Dynamics Ensemble Flow Filtering for data assimilation with sparse observations arXiv cs.LG source ↗

Oct 6 Interaction-aware circuit discovery in language models reveals context-dependent component interactions; arXiv:2610.04017v1 arXiv cs.LG source ↗

Oct 6 LongSocialBench tests long-context LLMs on understanding online discussion threads arXiv cs.CL source ↗

Oct 6 TreeWalker: Partial evaluation for grouped tree-ensemble inference arXiv cs.LG source ↗

Oct 6 Protecting sensitive data in image synthesis via PAC-private adaptation for diffusion models arXiv cs.LG source ↗

Oct 6 Selective backpropagation for efficient few-shot class-incremental learning arXiv cs.LG source ↗

Oct 6 Researchers introduce two learned latent protein languages to improve autoregressive protein sequence and structure generation. arXiv cs.LG source ↗

Oct 6 One-step curvature probes miss the fitting operator; terminal null-space correction preserves retained capacity in continual learning arXiv cs.LG source ↗

Oct 6 Representation-aligned auxiliary supervision for language model adaptation to structured domains (Chess) arXiv cs.CL source ↗

Oct 6 The score is not the structure: brain alignment and cross-lingual transfer arXiv cs.CL source ↗

Oct 6 Learning to accept more in selective sleep staging arXiv cs.LG source ↗

Oct 6 Copying before suppression: a below-chance dip during language model training arXiv cs.CL source ↗

Oct 6 IdeaScientist decomposes ideation into gap finding and innovation for grounded scientific ideas arXiv cs.CL source ↗

Oct 6 Do language models need a trainable input embedding table? Fixed minimal token codes at 1.7B-class scale arXiv cs.CL source ↗

Oct 6 Evidence and Intervention: a coupled active-inference extension of rational speech act models arXiv cs.CL source ↗

Oct 6 $\tilde{O}(\sqrt{T})$ regret and polylogarithmic constraint violation for COCO in online convex optimization with adversarial costs and constraints arXiv cs.LG source ↗

Oct 6 BAT-NO: a boundary-condition-aware transformer neural operator for crashworthiness prediction of vehicle components arXiv cs.LG source ↗

Oct 6 DePICT: a procedure for constructing decision-preserving interfaces by ranking context directions according to the optimizer’s solution arXiv cs.LG source ↗

Oct 6 Reinforcement learning on the discrete composition channel of a crystal generator; validated gains and reward hacking arXiv cs.LG source ↗

Oct 6 AI-enabled QA for multiple-choice assessment items: review of automated item-writing flaw detection, psychometric screening, and NLP benchmark auditing arXiv cs.CL source ↗

Oct 6 PB-GRPO: Learning socially adaptive LLM agents from persona-driven simulation with preference-batched GRPO arXiv cs.CL source ↗

Oct 6 Retrieval-centric deep learning in growing nonparametric neural networks arXiv cs.LG source ↗

Oct 6 OncoNoteBERT: a foundation representation model for NLP of real-world outpatient oncology notes arXiv cs.CL source ↗

Oct 6 Boundaries agree, labels do not: intra-annotator dynamics as a kind of training data arXiv cs.CL source ↗

Oct 6 Playing social deduction games with reinforcement-fine-tuned large language models arXiv cs.CL source ↗

Oct 6 LLM-enhanced spatio-temporal learning for grid-level docked bike sharing demand prediction arXiv cs.LG source ↗

Oct 6 Conformal prediction with paraphrase-aware scoring for LLM uncertainty quantification arXiv cs.CL source ↗

Oct 6 Probabilistic algorithms for Ising machines: from optimization to generative AI arXiv cs.LG source ↗

Oct 6 SynLat: Syntax-aligned text-latent compression for chain-of-thought reasoning arXiv cs.CL source ↗

Oct 6 Articulatory entrainment and coordination complexity in spontaneous autistic and non-autistic dialogue arXiv cs.LG source ↗

Oct 6 Forgetfulness in fine-tuning: optimiser history causes loss of earlier answers due to stored gradients arXiv cs.CL source ↗

Oct 6 Evolving LLM-generated features for interpretable classification arXiv cs.LG source ↗

Oct 5 Open and emergent problems in agentic privacy and security: a contextual angle Google Research source ↗

Oct 5 Clinical concept centers in LLMs arXiv cs.CL source ↗

Oct 5 From mathematical to executable certificates for machine unlearning arXiv cs.LG source ↗

Oct 5 TRACE: a reproducible benchmark for electricity price forecasting with official operational text arXiv cs.LG source ↗

Oct 5 Counterexample generation via per-theorem symbolic verifiers; release of SymCE corpus of 4,707 false conjectures with executable verifiers arXiv cs.CL source ↗

Oct 5 Beyond correctnes? Resolving underspecification in agentic Text-to-SQL. arXiv cs.CL source ↗

Oct 5 CUEing user simulators: calibrated user embeddings for multi-turn benchmarking arXiv cs.CL source ↗

Oct 5 Dropout neural networks approximate the unit ball of W^{n,∞}([0,1]^d) by ReLU networks with random edge retention; Sobolev rates and confidence bounds arXiv cs.LG source ↗

Oct 5 Parameter-free interval-dynamic regret under heavy-tailed noise in online convex optimization arXiv cs.LG source ↗

Oct 5 Slow-Fast multi-teacher on-policy distillation for capability preservation arXiv cs.LG source ↗

Oct 5 MuLoRA introduces spectrally balanced low-rank adaptation for continual learning to address spectral plasticity collapse in sequential updates. arXiv cs.LG source ↗

Oct 5 AptMQL-Bench: text-to-SQL to text-to-MQL via access-pattern schema design and data-preserving migration arXiv cs.CL source ↗

Oct 5 ConvoDrift: a dataset to model progressive stylistic conversational tone drift under fixed semantic intent across 15,727 shared multi-turn structures arXiv cs.CL source ↗

Oct 5 APDMem: agent-controlled progressive disclosure memory for query-adaptive long-term memory retrieval across hierarchical conversation history layers arXiv cs.CL source ↗

Oct 5 Automatic evaluation of mental health stigma in online communication arXiv cs.CL source ↗

Oct 5 FinDialogLens extracts missed trade events from multi-party financial chatrooms by performing event extraction over dialogue to identify RFQ outcomes. arXiv cs.CL source ↗

Oct 5 Improving atomic-fact recall via focused views in unstructured knowledge editing arXiv cs.CL source ↗

Oct 5 XiangqiBench provides closed-loop evaluation of LLM agents by requiring mate delivery in 119 tactical endgames against an engine defender arXiv cs.CL source ↗

Oct 5 Text-centric post-training for omni-modal reasoning in large language models arXiv cs.CL source ↗

Oct 5 HakemBench: a Turkish benchmark of typed decisions released with 2,346 items and 4,275 choices across seven tracks arXiv cs.CL source ↗

Oct 5 Mitigating convergence collapse in fixed-target anomaly detectors via kernel-anchored locality regularization arXiv cs.LG source ↗

Oct 5 CLEAN: psychometrically consistent incremental cognitive diagnosis under concept-space expansion via architectural isolation arXiv cs.LG source ↗

Oct 5 Does every user need a private LoRA? Decoupling personalization from per-user adaptation arXiv cs.LG source ↗

Oct 5 Learning when to commit from partial speech for end-to-end simultaneous speech translation arXiv cs.CL source ↗

Oct 5 Drive vs. Decay: On the Training Dynamics of Joint-Embedding Predictive Architectures arXiv cs.LG source ↗

Oct 5 Overcoming challenges of interpretive structural modeling with large language models arXiv cs.LG source ↗

Oct 5 Diffusion-based synthetic data pretraining for enhancing activity recognition arXiv cs.LG source ↗

Oct 5 TPBench: a turning-point benchmark for dialogue compression arXiv cs.CL source ↗

Oct 5 From Behavior to provenance: attributing tabular foundation models to synthetic pretraining data arXiv cs.LG source ↗

Oct 5 Evaluating multi-dimensional generalization of large language models in temporal extraction tasks arXiv cs.CL source ↗

Oct 5 Variation in interpulse interval influences deep-learning bat species classification; transformers show different sensitivity than CNNs, per two matched European bat call datasets. arXiv cs.LG source ↗

Oct 5 Counterfactual predictions in scientific emulators without controlled experiments arXiv cs.LG source ↗

Oct 5 How robust is multimodal claim verification to LLM rewriting? arXiv cs.CL source ↗

Oct 5 Greenwashing risk in sustainability data: conformal machine learning links SEC fundamentals to EPA facility emissions for improved market discipline arXiv cs.LG source ↗

Oct 5 LLM agents that yield to majority still represent their original premise arXiv cs.CL source ↗

Oct 5 Hybrid ML-assisted Raman spectroscopy with generative feature augmentation for pharmaceutical identification arXiv cs.LG source ↗

Oct 5 State-space unlearning for land surface forecasting reduces non-stationary bias from unrecorded confounding events in Mamba-family models. arXiv cs.LG source ↗

Oct 5 Adaptive batching helps LLM pretraining by allowing unbounded gradient variance, according to a perspective from unbounded variance arXiv cs.LG source ↗

Oct 5 History improves task completion for language agents, but more interaction history can still reduce success in some cases. arXiv cs.CL source ↗

Oct 5 MACTS-EM: multi-agent collaborative time series forecasting with emergent memory arXiv cs.LG source ↗

Oct 5 Sigma: a large-scale continuous diffusion language model built on steerable low-dimensional latent trajectories arXiv cs.CL source ↗

Oct 5 EpiWorld: grounding LLM policy agents in epidemiological world models arXiv cs.CL source ↗

Oct 5 From retrieval to typed decisions: calibrated system one models from biomedical sentence encoders arXiv cs.CL source ↗

Oct 5 Psi-Resilience: model-free feature importance from 1D topological signals arXiv cs.LG source ↗

Oct 5 Nearest-neighbour baselines for fingerprint prediction from MS/MS spectra under different assumptions arXiv cs.LG source ↗

Oct 5 PowerBench: measuring language model bias in power-shifting requests arXiv cs.LG source ↗

Oct 5 Generative-informed neuro-symbolic framework resolves syntactic ambiguity in Arabic DPs using AraBERT integration arXiv cs.CL source ↗

Oct 5 WakeKV: Reactive, reversible KV residency for heads that change their minds arXiv cs.CL source ↗

Oct 5 Rank-Aware Speculative Sampling for diffusion draft trees arXiv cs.LG source ↗

Oct 2 Computational tools for society’s most complex challenges MIT News (AI) source ↗

Oct 2 Tech worker movement documentary author co-authors “Against Tech Oligarchy: Worker Resistance in the World’s Most Powerful Industry” (Haymarket Books, September 2026) MIT News (AI) source ↗

Oct 2 Tech worker movement documented in new book: Against Tech Oligarchy co-authored by JS Tan SM ’22; Haymarket Books, September 2026 MIT News (AI) source ↗

Oct 2 AI agents susceptible to radicalization via influencer-promoted extremism in simulated conversations arXiv cs.AI source ↗

Oct 2 Scientific Agents: evaluating profession-specific system prompts on scientific tasks arXiv cs.AI source ↗

Oct 2 One mastery threshold does not fit all knowledge tracing models arXiv cs.LG source ↗

Oct 2 Energy time-series imputation with differentially private diffusion models via clipping-aware objective conditioning arXiv cs.LG source ↗

Oct 2 Reverse Item Response Theory for sparsity-robust ranking in fragmented cancer drug-response matrices arXiv cs.LG source ↗

Oct 2 How many categories are enough? Distribution-free certification limits for few-shot anomaly thresholds arXiv cs.LG source ↗

Oct 2 Conflicting supervision moves commitment, not capability; a 12.29σ arrangement effect that is exactly zero under a convention-agnostic score arXiv cs.AI source ↗

Oct 2 Four ways to grow a classifier and why one of them cannot learn arXiv cs.LG source ↗

Oct 2 Adam’s full update rule is analyzed as a diagonal empirical Fisher approximation to measure its geometric deviation from natural gradient descent across losses. arXiv cs.LG source ↗

Oct 2 Characterizing a configuration where inference-time PRM-pruned fragment grafting is inert: evidence from three reasoning LMs arXiv cs.AI source ↗

Oct 2 What Should an Agent Remember? Disentangling retention from retrieval in bounded-memory evaluation arXiv cs.AI source ↗

Oct 2 Query-language preference under cross-lingual knowledge disparities in large language models arXiv cs.CL source ↗

Oct 2 When do causal world models help modular LLM agents arXiv cs.AI source ↗

Oct 2 Dissonant ballerinas and crafty carrots: a comparative multi-modal analysis of Italian brain rot arXiv cs.CL source ↗

Oct 2 Contingent exposure routing for financial AI: outage risk and the cost of indivisible decisions arXiv cs.LG source ↗

Oct 2 SCM-based fairness and faithful explainability for legal document classification arXiv cs.CL source ↗

Oct 2 Audit finds pre-split SMOTE inflates churn-class F1 by 13.1 points in IBM Telco Benchmark; arXiv paper examines trustworthiness failures beyond accuracy and F1 arXiv cs.LG source ↗

Oct 2 Uncertainty-aware learning from multi-expert interval targets arXiv cs.LG source ↗

Oct 2 External verification can correct individual outputs but may leave a population in wrong consensus; study derives minimum verification fuel to cross a basin boundary under peak verification constraint arXiv cs.LG source ↗

Oct 2 Generalized Biomedicine Discovery introduces a unified benchmark for long-tail, anomaly, and taxonomy-aware discovery in medical images arXiv cs.LG source ↗

Oct 2 Predictive credit in scientific explanations remains inconclusive for v5's frozen credit decision across Tox21 and OpenML tasks arXiv cs.AI source ↗

Oct 2 MemBench mitigations halve Stable Diffusion memorization under false-discovery control in memorization audits of generative models arXiv cs.LG source ↗

Oct 2 Causal evaluation of recovery in LLM agents using external harnesses arXiv cs.AI source ↗

Oct 2 Attention manifolds: steering or blocking language models by editing learned B-spline surfaces arXiv cs.LG source ↗

Oct 2 Build2SPARQL: a large-scale text-to-SPARQL benchmark dataset for building knowledge graph querying arXiv cs.AI source ↗

Oct 2 Data artifact isn’t a shortcut; causal auditing of synthetic RLVR corpora reveals potential leakage between correctness and provenance in GooseReason-0.7M arXiv cs.CL source ↗

Oct 2 Dear Dreamer: a free AI-powered platform helps students apply the disciplined entrepreneurship process taught at MIT Entrepreneurship Center. MIT News (AI) source ↗

Oct 2 LLMs struggle to improve longitudinal clinical reasoning by simply adding more history; evaluation compares five context strategies for MedLoCoMo across long-horizon tasks arXiv cs.CL source ↗

Oct 2 FourierQK: Filter shape, admissibility and the leakage-coverage law arXiv cs.LG source ↗

Oct 2 K-Dense BYOK is an open-source AI research assistant that runs locally and provides a lab-notebook with hash-chained records for each project. arXiv cs.AI source ↗

Oct 2 CAVE-Mem: Boundary-aware experience validation for memory search arXiv cs.CL source ↗

Oct 2 Rules amortize, pairings don't: linguistic structure determines which latent task representations can replace in-context learning arXiv cs.CL source ↗

Oct 2 BudgetSchemaBench: a budget-swept diagnostic for schema context in text-to-SQL arXiv cs.CL source ↗

Oct 2 very likely" signals uncertainty; the paper investigates how LLMs diverge from humans in linguistic uncertainty quantification arXiv cs.LG source ↗

Oct 2 Robust Is Salient: an informed adversary moves the optimal signal onto the salience pole arXiv cs.AI source ↗

Oct 2 Fast Polynomial Transcendentals for LLMs arXiv cs.LG source ↗

Oct 2 Uncertainty-aware RL-controlled adaptive 3D mapping arXiv cs.LG source ↗

Oct 2 Nous: learning and certifying memory decisions before source calibration arXiv cs.LG source ↗

Oct 2 What do rationales communicate? A message-intervention study in role-specialized QA arXiv cs.AI source ↗

Oct 2 UniBuc at SemEval-2024 Task 2: Tailored prompting with SOLAR for clinical NLI arXiv cs.CL source ↗

Oct 2 Knowing when to yield: grounded arbitration of user corrections in text-based embodied agents arXiv cs.AI source ↗

Oct 2 Certainty is not just correctness: reassessing token-level certainty in LLM reasoning arXiv cs.CL source ↗

Oct 2 FACET at WMT 2026 Automated Translation Quality Evaluation Task decomposes evaluation into Fluency, Accuracy, and Consistency passes for reference-free MT quality assessment. arXiv cs.CL source ↗

Oct 2 Useful to whom? The sample value for coreset selection varies with the learner and budget regime arXiv cs.LG source ↗

Oct 2 Praxa, an evidence-bound harness for governed AI agent execution, enables deterministic admission, brokered execution, external read-back, reconciliation, and reviewed promotion arXiv cs.AI source ↗

Oct 2 On-device named-entity recognition: a deployability study of accuracy, cost, reliability, and confidence arXiv cs.CL source ↗

Oct 2 Sharp Oracle-Regret Tradeoffs for Projection-Free Online Convex Optimization arXiv cs.LG source ↗

Oct 2 Rules to Tools provides executable checks for LLM agents in scientific computing; two task cohorts show repair success rates of 26/30 with text and 29/30 with… arXiv cs.AI source ↗

Oct 2 CAST: Cost-Aware Speculative Trees from One-Pass Block Drafters arXiv cs.CL source ↗

Oct 2 LLM-as-a-Judge for low-resource languages: adapting Ragas and comparative ranking for Romanian arXiv cs.CL source ↗

Oct 2 JevSpawn: adaptive agentic inference via compositional action spaces arXiv cs.AI source ↗

Oct 2 Legal text classification in Korean sexual offense cases improves from traditional machine learning to large language models with XAI insights arXiv cs.CL source ↗

Oct 2 EviGraph: proof-carrying selective recommendation over temporal public-service knowledge graphs arXiv cs.AI source ↗

Oct 2 Hidden costs of reading LLM judges from the first token verdict; reading the first generated token distorts position bias under constrained decoding and likelihood scoring, study finds arXiv cs.CL source ↗

Oct 2 EurekaBench: measuring agentic ability to discover new scientific insights arXiv cs.CL source ↗

Oct 2 Signed lexical confidence for risk-calibrated intent routing arXiv cs.CL source ↗

Oct 2 Format-aware fusion improves FP4 pretraining efficiency by co-designing quantization producers with their scale domain and consumer layout arXiv cs.LG source ↗

Oct 2 Safety layers can be bypassed after few-sample fine-tuning, as harmful behavior localizes to specific layers and tokens across multiple models. arXiv cs.CL source ↗

Oct 2 Classification based on association rules algorithm for breast cancer arXiv cs.LG source ↗

Oct 2 Assessing how language disparity affects multilingual linguistic ability in large language models across 101 languages using the MultiBLiMP benchmark and four evaluation methods. arXiv cs.CL source ↗

Oct 2 Comedic Fool's Gold: Reward exploits and countermeasures in conversational humor arXiv cs.AI source ↗

Oct 2 JusticeAxis benchmarks legal judgment between rigid rule application and ungrounded discretion arXiv cs.AI source ↗

Oct 2 Mathematical transfer in LLMs follows reasoning approach more than topic arXiv cs.AI source ↗

Oct 2 SW-KAN: Kolmogorov-Arnold networks with Stieltjes-Wigert q-orthogonal polynomials arXiv cs.LG source ↗

Oct 2 Constant-memory recall: learned associations in a fixed matrix state arXiv cs.LG source ↗

Oct 2 Measuring human-like bias in LLMs? A critique of human-derived bias constructs in LLM evaluation arXiv cs.CL source ↗

Oct 2 Measuring the microtask eligibility gap: when is an off-the-shelf SLM enough for an agent harness? arXiv cs.AI source ↗

Oct 2 Metric-construction coupling inflates measured synthetic dialect recovery arXiv cs.CL source ↗

Oct 2 ContractRL: shielded group-relative policy optimization for auditable tool-call repair arXiv cs.AI source ↗

Oct 2 Gradient-aligned pair selection for personalized preference optimization arXiv cs.AI source ↗

Oct 2 Benchmarking prompt optimization of large language models with chess arXiv cs.AI source ↗

Oct 2 Holistic assessment of the carbon footprint of Noor, a very large Arabic language model arXiv cs.CL source ↗

Oct 2 Integrating fairness and explainability in a multi-instance reinforcement learning system arXiv cs.LG source ↗

Oct 2 Fault-tolerant budget conservation in distributed multi-agent delegation arXiv cs.AI source ↗

Oct 2 DSB-DG benchmarks document grounding and hallucinations in voice agents across five domains; evaluates context saturation and grounding decay. arXiv cs.CL source ↗

Oct 1 InstructMesh AI lets users refine AI-generated 3D designs and fabricate them via 3D printing. MIT News (AI) source ↗

Oct 1 From Lexical Baselines to Agentic Retrieval-Augmented Generation: structured skill and responsibility-level extraction with the SFIA framework arXiv cs.CL source ↗

Oct 1 Alignment forecasting: predicting misalignment from training data arXiv cs.CL source ↗

Oct 1 Hyperspherical semantic trajectory analysis tracks technology diffusion from unstructured text streams across academic preprints, patent signals, and compute scaling arXiv cs.CL source ↗

Oct 1 Memory adaptation for task-conditioned execution to address failed reuse of successful memories in embodied agents arXiv cs.CL source ↗

Oct 1 Calibrated inference for small-sample AI evaluation with evalstats arXiv cs.CL source ↗

Oct 1 Constructing challenging browser-use tasks by controlled environment interventions. arXiv cs.CL source ↗

Oct 1 OCR model for extracting information from invoices in Korean language arXiv cs.CL source ↗

Oct 1 Automated evaluation of multi-turn dialogues in in-car conversational assistants arXiv cs.CL source ↗

Oct 1 Environment steering: using data flow control to improve agent utility and safety arXiv cs.CL source ↗

Oct 1 Lookahead-R: budget-aware tool retrieval via execution-centric planning arXiv cs.CL source ↗

Oct 1 PrimeSeeker: capability-oriented supervision for deep search agents arXiv cs.CL source ↗

Oct 1 Generative AI pipeline for SEC 10-K extraction resolves missing financial data crisis arXiv cs.CL source ↗

Oct 1 Less Uniform Diffusion is more powerful and scalable arXiv cs.CL source ↗

Oct 1 Tracing mechanisms of sycophantic agreement in language models arXiv cs.CL source ↗

Oct 1 Sieve and Sage: Efficient distraction filtering for reliable RALM abstention arXiv cs.CL source ↗

Oct 1 TRACE: deployable tree-relational structure enhancement for oncology LLMs arXiv cs.CL source ↗

Oct 1 When Should LLMs trust their own revisions? A risk-aware study of intrinsic self-correction arXiv cs.CL source ↗

Oct 1 Reliable but design-sensitive: instrument uncertainty in LLM annotation arXiv cs.CL source ↗

Oct 1 FD-VAD: semantic endpoint detection for streaming full-duplex speech arXiv cs.CL source ↗

Oct 1 The detectability gap: hidden heterogeneity in hallucination detection across language models arXiv cs.CL source ↗

Oct 1 Can we still trust disaster social sensing; empirical evidence on detecting AI-generated social media posts arXiv cs.CL source ↗

Oct 1 AIM: Agentic Idea Manager for automated research outlines an autonomous framework to manage and implement ideas in frontier LLM-driven scientific search arXiv cs.AI source ↗

Oct 1 Activation-conditioned self-distillation uses activations of self-generated trajectories to steer learning arXiv cs.LG source ↗

Oct 1 Synthesis Without Training: an inference-only pipeline for tabular, temporal, and relational synthetic data arXiv cs.LG source ↗

Oct 1 Data-free physics-informed neural operator for level-set interface advection arXiv cs.LG source ↗

Oct 1 A competing-hazards systematization of loss of control in autonomous agents arXiv cs.AI source ↗

Oct 1 Improving OCR faithfulness via gated and attenuated on-policy distillation arXiv cs.AI source ↗

Oct 1 Reach Into The CHOIR: Free-list elicitation reveals distinct model voices in LLM ensembles arXiv cs.AI source ↗

Oct 1 Simulator-refined diffusion for radio-frequency inverse design arXiv cs.LG source ↗

Oct 1 AREX-2 advances self-improving LLM agents with long-horizon reflective tasks arXiv cs.AI source ↗

Oct 1 VAmoS Energy benchmark evaluates a voice agent handling multi-request calls with a stateful billing twin and payment tools; 100 calls cover utility billing and payment assistance scenarios arXiv cs.AI source ↗

Oct 1 Function-space transformer with adaptive anchors reveals approach for learning neural operators over continuous domains arXiv cs.LG source ↗

Oct 1 Conformal Adversarial Generative Ensemble improves forecast reliability with multiple generative models and conformal prediction arXiv cs.LG source ↗

Oct 1 Hermes: Learning contextual reasoning unlocks test-time scaling arXiv cs.LG source ↗

Oct 1 Graph anomaly detection as finite-horizon control: training-free scoring via empirical Bayes arXiv cs.LG source ↗

Oct 1 Gradient conflict does not reliably predict the understanding–generation trade-off in unified multimodal models, according to a controlled GRIDUMM audit arXiv cs.LG source ↗

Oct 1 Travel time prediction in supply chain management using machine learning arXiv cs.LG source ↗

Oct 1 Beyond mode collapse: generating diverse synthetic expert conversations via generative flow networks arXiv cs.AI source ↗

Oct 1 Input-frugal deep learning framework uses routine weather and two static inputs to forecast Brazilian soybean crop yield arXiv cs.LG source ↗

Oct 1 MetaPersona: Task-grounded synthetic populations from empirical social science arXiv cs.AI source ↗

Oct 1 Towards Model as a Library: offline, community-sourced AI for low-resource African languages arXiv cs.AI source ↗

Oct 1 Examining variation in how guided AI tutors resolve student impasses arXiv cs.AI source ↗

Oct 1 Learning to plan from random exploration arXiv cs.LG source ↗

Oct 1 FlashDiffusion: fused tiled kernel spectral decomposition for diffusion maps on GPUs arXiv cs.LG source ↗

Oct 1 SimTrace: grounded multimodal user trajectories generation for online user modeling arXiv cs.AI source ↗

Oct 1 Demographic Pluralism: inference-time modeling of pluralistic human preference distributions arXiv cs.AI source ↗

Oct 1 CARAT: materials LLMs question whether they reason or recite; structure and gold answer fixed across eight views, with GraphSpace relations and matched fine-tuning arXiv cs.AI source ↗

Oct 1 Which tasks survive self-supervised learning? arXiv cs.LG source ↗

Oct 1 Defining and categorising human-AI interactions in clinical trials: a multidimensional classification approach. arXiv cs.AI source ↗

Oct 1 EHR2Trace: auditable EHR data infrastructure for patient world models and clinical agents arXiv cs.LG source ↗

Oct 1 Disagreement-regularized imitation learning for image-based continuous control with Gaussian and Beta policies. arXiv cs.LG source ↗

Oct 1 ArgGYM: a procedural, engine-verified benchmark for structured defeasible reasoning arXiv cs.AI source ↗

Oct 1 Grokking through the lens of minimum-norm interpolation reveals how regularization geometry and signal sparsity govern generalization near interpolation. arXiv cs.LG source ↗

Oct 1 NAQD Env: a benchmark for selective withdrawal in language agents arXiv cs.AI source ↗

Oct 1 Pretraining and midtraining enable learning from rewards, by providing information and computation for reward adaptation; understanding through sequential state computation and contextual memory. arXiv cs.LG source ↗

Oct 1 Self-evolving harness on multiple tasks with the agent as its own optimizer; MILO covers automated harness discovery via orchestrated multi-agent evolution arXiv cs.AI source ↗

Oct 1 MoFlow: multi-objective agentic workflow generation aims to jointly optimize accuracy, cost, latency, robustness, and consistency arXiv cs.AI source ↗

Oct 1 PrivMeSA: privacy-aware self-evolving multi-agent system for medicine via local-remote LLM collaboration arXiv cs.AI source ↗

Oct 1 Moving-horizon approximate branch-and-reduce method to train near-optimal deep classification trees on large-scale data arXiv cs.LG source ↗

Oct 1 LeanPolish: Verified supervision for Lean proof compression arXiv cs.LG source ↗

Oct 1 GoldiMask selects tokens to reveal as context for fine-tuning diffusion language models; it improves learning by tailoring which tokens are masked and recovered. arXiv cs.AI source ↗

Oct 1 Continual learning of dynamical systems in recurrent neural networks through recyclable unit gating arXiv cs.LG source ↗

Oct 1 Kinematic signatures of impairment: detecting alcohol intoxication in e-scooter riders using sensor data and machine learning arXiv cs.LG source ↗

Oct 1 Decode-Latency Feedback Prefill: A model-free controller and its generalization limits arXiv cs.AI source ↗

Oct 1 AI agent rediscovering a Blaschke-curve invariant in a controlled geometric task arXiv cs.AI source ↗

Oct 1 DualCast: a dual-path language model for bimodal financial time-series forecasting arXiv cs.LG source ↗

Oct 1 Conditional generation of creative chess puzzles with diffusion models arXiv cs.AI source ↗

Oct 1 Large language models as approximate survival estimators for time-to-event predictions arXiv cs.CL source ↗

Oct 1 Evaluating language model safety across long adversarial conversations arXiv cs.CL source ↗

Oct 1 Backdrop tests how agent capabilities degrade when deployed in changing worlds with four everyday hazards arXiv cs.CL source ↗

Oct 1 Prompt2Skill: Unsupervised Skill Optimization From Natural Language Instructions arXiv cs.CL source ↗

Oct 1 Measuring heterogeneity for LLM team selection to determine which models work well together arXiv cs.CL source ↗

Oct 1 DEdit: Iterative draft editing for speculative decoding arXiv cs.CL source ↗

Oct 1 KlinikeBench evaluates language models beyond diagnostic accuracy by assessing how well models gather information and determine appropriate clinical assessments. arXiv cs.CL source ↗

Oct 1 System prompts alter layer representations in language models, with effects varying by layer and prompt type. arXiv cs.CL source ↗

Oct 1 Benchmarking interactive intent alignment under miscommunication and evolving user intent arXiv cs.CL source ↗

Oct 1 Shifting mechanisms: how positional encoding choice shapes in-context retrieval arXiv cs.CL source ↗

Oct 1 Evaluating whether LLMs can reliably connect the dots arXiv cs.CL source ↗

Oct 1 Doc2LoRA provides decodable representations of scientific ideas arXiv cs.CL source ↗

Oct 1 ContextAdapt evaluates contextual adaptation and value alignment in LLMs arXiv cs.CL source ↗

Oct 1 TutlAit v1: a crowdsourced Moroccan Tamazight speech dataset with Arabic transcriptions and regional accent labels arXiv cs.CL source ↗

Oct 1 Policy-conditioned AI-use detection: an evidentiary framework for academic publishing arXiv cs.CL source ↗

Oct 1 Automatic estimation of the verbal fluency index in people with motor neuron disease using ASR alignment and pause modelling arXiv cs.CL source ↗

Oct 1 StreamDecisionBench dataset released for evaluating decisions in streaming language model outputs arXiv cs.CL source ↗

Oct 1 TomasuLLM enables out-of-order speculative execution for LLM agents arXiv cs.CL source ↗

Oct 1 NinaXander enables composing frozen language models from different architectures via a shared latent-space adapter arXiv cs.CL source ↗

Oct 1 MedKIT provides a granular evaluation of knowledge integration and transfer in large language models for medical knowledge updates. arXiv cs.CL source ↗

2026-09 →