https://amarcu.github.io/ai-paper-digest/ A daily, machine-summarized digest of new AI research papers from arXiv. en Wed, 30 Sep 2026 07:59:24 +0000 https://amarcu.github.io/ai-paper-digest/archive/2026-09-30.html https://amarcu.github.io/ai-paper-digest/archive/2026-09-30.html Wed, 30 Sep 2026 07:00:00 +0000 1776 new AI papers on arXiv. Leading entries: Omni-IO Skills: Harnessing Your Agent Omni-Native; Raven: The Harness of Harnesses for Composable Agentic Intelligence; ROSS: Relearning from Self-Generated Rollouts through Selective Supervision; LongCat-DeepResearch Technical Report; SAKI: Maximal-Coupling-Routed Teacher Supervision for On-Policy Distillation https://amarcu.github.io/ai-paper-digest/archive/2026-09-29.html https://amarcu.github.io/ai-paper-digest/archive/2026-09-29.html Tue, 29 Sep 2026 07:00:00 +0000 1810 new AI papers on arXiv. Leading entries: TraceDance: An Automated System for Building Agent Behavior Benchmarks from Real-World Agent Deployment Traces; Learning to Learn from Context: Synthetic Training from Perturbed Public Documents; CompoWorld: Compositional Environment Scaling for General Agents; YuE2: Unifying Symbolic and Audio Music Generation at Frontier Quality; Skill2Env: Capability-Oriented Environment Synthesis from Skills for General Agents https://amarcu.github.io/ai-paper-digest/archive/2026-09-28.html https://amarcu.github.io/ai-paper-digest/archive/2026-09-28.html Mon, 28 Sep 2026 07:00:00 +0000 404 new AI papers on arXiv. Leading entries: Cosine Similarity Is Not Evidence: Measuring the Noise Floor of Interpretability Transfer Under Quantization; NeuralCert: certified computational discovery of extremal mathematical constructions; Stealth Apart, Harm Together: Skill Cascading Attacks on Skill-Based Agent Systems; Recursive Self-Improvement via On-Policy Distillation for Reasoning; Analyzing and Mitigating Cost-Inefficient Behaviors in Coding Agents https://amarcu.github.io/ai-paper-digest/archive/2026-09-25.html https://amarcu.github.io/ai-paper-digest/archive/2026-09-25.html Fri, 25 Sep 2026 07:00:00 +0000 487 new AI papers on arXiv. Leading entries: Reward Hacking Challenges Oversight of Autonomous Research Agents; Training Object Permanence in World Models; Rufus-Air: An Open LLM Post-Training Recipe; IterSynth: Rethinking Deep Search Agents via Role-Decoupled Iterative Synthesis; AgentKernel: The Trust-Native Agentic Operating System https://amarcu.github.io/ai-paper-digest/archive/2026-09-24.html https://amarcu.github.io/ai-paper-digest/archive/2026-09-24.html Thu, 24 Sep 2026 07:00:00 +0000 528 new AI papers on arXiv. Leading entries: Lean Pool: An AI-Maintained Archive of Formalized Mathematics; From Decorative to Load-Bearing: Task Difficulty Shapes the Causal Role of Chain-of-Thought; Evaluating Coding Agents on Kernel Exploit Generation; Optimizing the Score, Losing Sight of the Task: Reward Hacking Across Weights, Selection, and Prompts; PACT: From Credit Assignment to Critic Alignment https://amarcu.github.io/ai-paper-digest/archive/2026-09-23.html https://amarcu.github.io/ai-paper-digest/archive/2026-09-23.html Wed, 23 Sep 2026 07:00:00 +0000 443 new AI papers on arXiv. Leading entries: FrontierMath Erd\H{o}s; Ovis-Embedding: Pushing the Frontiers of Universal Omni-Modal Embeddings; From Pattern Recognizers to Personalized Companions: A Survey of Large Language Models in Mental Health; Lean Pool: An AI-Maintained Archive of Formalized Mathematics; Geometric and Semantic Coupling for Interaction Understanding in 3D Scenes https://amarcu.github.io/ai-paper-digest/archive/2026-09-22.html https://amarcu.github.io/ai-paper-digest/archive/2026-09-22.html Tue, 22 Sep 2026 07:00:00 +0000 709 new AI papers on arXiv. Leading entries: Measuring the Checker: Mutation Analysis for GPU-Kernel Benchmark Oracles; Deep Persona: A Psychologically Grounded Architecture and Evaluation Framework for Role-Playing Agents and Simulations; OmniEdu: Open Foundation Models for Learning and Teaching; One to More, More to One: Category-Aware Iterative Expert Training for Software Engineering Agents; Why Do Video Diffusion Models Violate Physics? Unveiling the Flaws in Attention Mechanisms https://amarcu.github.io/ai-paper-digest/archive/2026-09-21.html https://amarcu.github.io/ai-paper-digest/archive/2026-09-21.html Mon, 21 Sep 2026 07:00:00 +0000 298 new AI papers on arXiv. Leading entries: Training-Adaptive Convolutional Sparse Coding via Information Bottleneck for Robust Visual Representation; Recursive Language Models Generalize Out of Domain; When AI Reviews Train AI Reviewers: Scientific-Judgment Collapse and Mitigation; SWE-Proof: Can Language Models Resolve Real-World Issues with Machine-Checked Proofs?; Calibrating Teacher--Student Discrepancy for On-Policy Distillation https://amarcu.github.io/ai-paper-digest/archive/2026-09-18.html https://amarcu.github.io/ai-paper-digest/archive/2026-09-18.html Fri, 18 Sep 2026 07:00:00 +0000 414 new AI papers on arXiv. Leading entries: When2Think: Learning Difficulty-Aware Length Control for Efficient Hybrid Reasoning Models; DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression; Stress-testing Alignment Midtraining; SoL-Pi: Recursively Scaling Auto-Research Loops for Efficient Agent Harness; What Does Privileged Information Add to On-Policy Self-Distillation? https://amarcu.github.io/ai-paper-digest/archive/2026-09-17.html https://amarcu.github.io/ai-paper-digest/archive/2026-09-17.html Thu, 17 Sep 2026 07:00:00 +0000 422 new AI papers on arXiv. Leading entries: EvolveTrade: Experience-Driven Policy Refinement for Self-Evolving LLM Trading Agents; Confidence Comes from Experience: Experiential Confidence Estimation from Reasoning to Agents; Zing-0.5: Toward Playable Worlds with Real-Time Joint Action and Text Control; Gaze as Evidence for Common Grounding: A Cross-Corpus Analysis of MapTask and MUNDEX; Agora: Git as Shared Memory for Collective AutoResearch https://amarcu.github.io/ai-paper-digest/archive/2026-09-16.html https://amarcu.github.io/ai-paper-digest/archive/2026-09-16.html Wed, 16 Sep 2026 07:00:00 +0000 675 new AI papers on arXiv. Leading entries: Drift-Constrained Optimization: Only Direction Matters in Fine-Tuning Instruct Models; How Many Thoughts Can a Vector Hold? The Capacity of Reasoning by Superposition; Convergent Emergence of In-Context Learning Across Modalities; Shallow Beliefs: Synthetic document finetuning does not inoculate against emergent misalignment from reward hacking; The Router Within: Eliciting Native Skill Routing from a Frozen LLM https://amarcu.github.io/ai-paper-digest/archive/2026-09-15.html https://amarcu.github.io/ai-paper-digest/archive/2026-09-15.html Tue, 15 Sep 2026 07:00:00 +0000 763 new AI papers on arXiv. Leading entries: LLaDA-UI: Bringing Block-wise Diffusion to Vision-Language GUI Agents; ZGCM-1: A Fully Open and Extremely Efficient Foundation Model for Math and Agentic Search; Dream-RSI: Recursive Self-Improvement through Evolving Worlds; Omni-Streaming Thinking; HazardAuditor: From Executable Threats to Safer Computer-Use Agents https://amarcu.github.io/ai-paper-digest/archive/2026-09-14.html https://amarcu.github.io/ai-paper-digest/archive/2026-09-14.html Mon, 14 Sep 2026 07:00:00 +0000 313 new AI papers on arXiv. Leading entries: Occamy-1.0: Open Pareto-frontier 35B Intelligence for Co-work; Harness or Model? Isolating the Harness Effect in Agentic Coding with a Contamination-Controlled Private Suite; Fixed State, Long Reach: What a Constant-Size Cache Buys Block Diffusion at Scale; The Cost of Compression: A Rate-Distortion Limit on Factual Hallucination; Breaking the Token Ceiling: Distilling Smaller, Stronger Byte Models https://amarcu.github.io/ai-paper-digest/archive/2026-09-11.html https://amarcu.github.io/ai-paper-digest/archive/2026-09-11.html Fri, 11 Sep 2026 07:00:00 +0000 382 new AI papers on arXiv. Leading entries: Do Agents Know When They Succeed? Calibrating Agent Confidence from Internal Representations; MetroLLM-Bench: Evaluating Language Models as Transit Kiosk Runtimes; NCP-ArchPreview Technical Report: Moving towards Latent Space Language Models through Next Concept Prediction; Story Imprinting: AI Assistants Absorb Traits from Human Characters They Resemble; EGGROLL, Unrolled: Understanding and Improving Low-Rank Evolution Strategies at Scale https://amarcu.github.io/ai-paper-digest/archive/2026-09-10.html https://amarcu.github.io/ai-paper-digest/archive/2026-09-10.html Thu, 10 Sep 2026 07:00:00 +0000 323 new AI papers on arXiv. Leading entries: World-Time Compute with Verified Code World Models; AgenticGen: Reward-Guided Agentic Video Generation for Advertising; Scores Alone Do Not Prove Discovery: The Discovery Certification Protocol for Auditing AI Research Agents; Do Agents Know When They Succeed? Calibrating Agent Confidence from Internal Representations; Stable Answers, Unfinished Reasoning: Why Self-Consensus Is Not a Safe Early-Exit Signal https://amarcu.github.io/ai-paper-digest/archive/2026-09-09.html https://amarcu.github.io/ai-paper-digest/archive/2026-09-09.html Wed, 09 Sep 2026 07:00:00 +0000 957 new AI papers on arXiv. Leading entries: Steering Geometry: Validating Human Value Geometry in LLM Steering Space; Reason Through the Latent! Making Latent Visual Reasoning Necessary; Online Draft Co-Training for Speculative Decoding in Large-Scale, Long-Context RL Post-Training; Kalman Delta Networks: Uncertainty-aware Associative Memory; NeoHorse-1: Towards Recursive Self-Improvement via Agentic Post-Training with Routing Harness https://amarcu.github.io/ai-paper-digest/archive/2026-09-07.html https://amarcu.github.io/ai-paper-digest/archive/2026-09-07.html Mon, 07 Sep 2026 07:00:00 +0000 354 new AI papers on arXiv. Leading entries: Iris: Climbing to the Search Frontier; When Quantization Breaks Memory: Recurrent-State Write-Back in Low-Precision Temporal Inference; MaxKernel: Agentic Kernel Generation for TPUs; Distilled Continuous Diffusion Language Models Can Write Code in Few Steps---or One; Extremely Sparse Supervision Incentivizes Reasoning Ability https://amarcu.github.io/ai-paper-digest/archive/2026-09-04.html https://amarcu.github.io/ai-paper-digest/archive/2026-09-04.html Fri, 04 Sep 2026 07:00:00 +0000 357 new AI papers on arXiv. Leading entries: PACE: Towards Surfacing Hidden Conflicts in User Requests; Random Attention: Rethinking KV Cache Eviction for Efficient Reasoning; LLaDA-Image: Building Strong Image Generators with Fully Open Training Recipes; Editable Visual Design; CORE: Improving Compositional Reasoning in MLLM Embedding via Reranker Distillation https://amarcu.github.io/ai-paper-digest/archive/2026-09-03.html https://amarcu.github.io/ai-paper-digest/archive/2026-09-03.html Thu, 03 Sep 2026 07:00:00 +0000 318 new AI papers on arXiv. Leading entries: EvalDetectBench: A Benchmark for Measuring Evaluation Awareness in Frontier Language Models; ExecRetrieval: Measuring the Functional-Correctness Gap in Code-Embedding Retrieval; CRISP: Cliff-awaRe Input-adaptive Sparse Prefilling with Structural-Mass-Motivated Routing; AI agents reshape consensus formation in human groups; PaperCompiler: Faithful Paper-to-Code Generation via Repository-Level Specification Compilation https://amarcu.github.io/ai-paper-digest/archive/2026-09-02.html https://amarcu.github.io/ai-paper-digest/archive/2026-09-02.html Wed, 02 Sep 2026 07:00:00 +0000 505 new AI papers on arXiv. Leading entries: UI-Venus-2 Technical Report; Safin-1: Safety from Within through Memory-Native State Evolution; EM^2Mem: Event-Centric Multimodal Memory for Large Language Models; Control-Data Flow Separation: Stable Prompt Optimization in Multi-Agent LLMs; DiagEvo: Diagnosis-Guided Self-Evolution via Hierarchical Error Memory https://amarcu.github.io/ai-paper-digest/archive/2026-09-01.html https://amarcu.github.io/ai-paper-digest/archive/2026-09-01.html Tue, 01 Sep 2026 07:00:00 +0000 769 new AI papers on arXiv. Leading entries: Verification-Aware Training for Speculative Decoding; CAST: Critique-Aware Supervision for Training Reliable Long-Horizon Tool-Calling Agents; PaperBanana-Interact: Scientific Diagram Refinement with Multi-Turn Human Feedback; On the Design of Qwen3.8-Next Architecture: Evaluation, Efficiency, and Training Stability; Lies We Can See: Joint Verbal and Non-Verbal Deception by VLM Agents in Embodied Social Interactions https://amarcu.github.io/ai-paper-digest/archive/2026-08-31.html https://amarcu.github.io/ai-paper-digest/archive/2026-08-31.html Mon, 31 Aug 2026 07:00:00 +0000 305 new AI papers on arXiv. Leading entries: Quantization-Triggered Backdoors in Language Models: Cross-Quantizer Transferability and the Validation--Deployment Gap; Beyond Search-Imitation: Prior-Directed Exploration for Searchless Chess; Fast Weight Attention for Continual Learning; Rubric-to-Code Credit Assignment for Reinforcement Learning; The Approximation Rank of Softmax Attention: Sharp Geometric Laws and Robust Interaction Dimension https://amarcu.github.io/ai-paper-digest/archive/2026-08-28.html https://amarcu.github.io/ai-paper-digest/archive/2026-08-28.html Fri, 28 Aug 2026 07:00:00 +0000 397 new AI papers on arXiv. Leading entries: GameWAM: A World Action Model for Video Games; Same Model, Different Harness: Different Coding-Agent Results; PILOT in the Loop: Live Self-Improvement for Long-Horizon Agents; J-Zero: Unified Challenger--Solver--Judge Co-Evolution from Zero Data; Squeezing More from Limited Data with Recursive Transformers https://amarcu.github.io/ai-paper-digest/archive/2026-08-27.html https://amarcu.github.io/ai-paper-digest/archive/2026-08-27.html Thu, 27 Aug 2026 07:00:00 +0000 490 new AI papers on arXiv. Leading entries: Are Android GUI Agents Robust Against Runtime Anomalies? AnTrap: Evaluating Agents in Dynamic Adversarial Environments; The Handoff Tax: Continuing Non-Native Trajectories in LLM Agents; Demystifying Reinforcement Learning Post-Training of Language Models; FrontierChallenge: Evaluating Scientific Workflow Completion; D$^3$-MOPD: Adaptive Dynamic Domain ScheDuling for Efficient Multi-Teacher Distillation https://amarcu.github.io/ai-paper-digest/archive/2026-08-26.html https://amarcu.github.io/ai-paper-digest/archive/2026-08-26.html Wed, 26 Aug 2026 07:00:00 +0000 381 new AI papers on arXiv. Leading entries: AI Agents Push Humans Out of the Loop; Autonomous Mathematical Discovery in an Open-World Multi-Agent Environment; Serving Masked Diffusion LLMs: Characterization and Design Principles from Real Hardware; WeMM-Embedding: WeChat Multi-Modal Embedding Technical Report; On-policy Distillation with Verifiable Reward https://amarcu.github.io/ai-paper-digest/archive/2026-08-25.html https://amarcu.github.io/ai-paper-digest/archive/2026-08-25.html Tue, 25 Aug 2026 07:00:00 +0000 666 new AI papers on arXiv. Leading entries: GameXpert-Bench: How Far Are Coding Agents from Expert Game Development?; RAG Collapse: LLM Responses Collapse When Retrieved Documents Are Self-Authored; The Compaction Cliff in Long-Running AI Agent Memory; Same Agent, Different Answers: A Repeat-Aware Audit of Corpus-Induced Answer Churn in Retrieval-Augmented QA; Better Retrieval, Worse Robustness:How Multi-hop RAG Amplifies Upstream ASR Errors https://amarcu.github.io/ai-paper-digest/archive/2026-08-24.html https://amarcu.github.io/ai-paper-digest/archive/2026-08-24.html Mon, 24 Aug 2026 07:00:00 +0000 310 new AI papers on arXiv. Leading entries: Hadith computational science in the age of large language models: a critical narrative review; Peer-Voted LLM-Agent Stress Tests Find Feed-Induced Lexical Convergence but No Reliable Matched-Exposure Advantage for Distributed Sources; Stored in Optimizer State, Valued by Later Training: A Causal Account of Subliminal Trait Transfer; Open-Weight Masked Introspection: Measuring What Language Models Can Report About Their Own Computation; FlavourBench: Ranking Frontier Language Models with Executable Culinary Ground Truth https://amarcu.github.io/ai-paper-digest/archive/2026-08-20.html https://amarcu.github.io/ai-paper-digest/archive/2026-08-20.html Thu, 20 Aug 2026 07:00:00 +0000 372 new AI papers on arXiv. Leading entries: Co-RL: Unsupervised Reasoning Emerges from Diverse Cohort in Multi-agent RL; Debate Training Reduces Reward Hacking in RLAIF; Entity tracking emerges in sub-billion parameter language models and exceeds human performance in naturalistic narratives; Accurate Decoding of Natural Sentences from Non-Invasive Brain Recordings; Temporal Multi-Signal Fusion for Token-Level Hallucination Detection https://amarcu.github.io/ai-paper-digest/archive/2026-08-19.html https://amarcu.github.io/ai-paper-digest/archive/2026-08-19.html Wed, 19 Aug 2026 07:00:00 +0000 589 new AI papers on arXiv. Leading entries: Discrete Diffusion Language Models Are Training-Free Multi-Label Classifiers; Pushing the Limits of High-Resolution Weather Forecasting through Data Scaling; Unifying Graph Neural Networks Through a Common Layer Equation; Cross-Model Memory Transfer via Target-Side Reader Adaptation; DiSCO: Defending text-to-image generation through distribution-guided contrastive prompt optimization https://amarcu.github.io/ai-paper-digest/archive/2026-08-17.html https://amarcu.github.io/ai-paper-digest/archive/2026-08-17.html Mon, 17 Aug 2026 07:00:00 +0000 284 new AI papers on arXiv. Leading entries: MobileMem: Learning from a Year of Mobile Experiences; Your Probabilistic JEPA Is Secretly a Hidden Markov Model: A State-Space Interpretation of Joint-Embedding Predictive Learning; Second Thought: Reasoning in Parallel as LLM Agents Act and Observe; From BERT to Frontier Agents: Eight Years of Language-Model Progress, the Collapse of the Capability-Cost Curve, and the Rise of Task-Targeted Models; Explanation Multiplicity: Circuit-Level Interpretability Evidence Does Not Survive Defensible Analytic Variation