Capability Watch

Papers

Every frontier-AI research paper we’ve surfaced from arXiv, newest first — the raw capability signals Frontier Watch tracks. Scroll to load older ones.

2,841 papers tracked · ← dashboard

  1. 1PerceptRL: Making Visual Perception Rewardable in MLLMs from Answers to Interactions
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-27
  2. 2QD-PINN:A Physics-Informed Neural Network Approach to Bayesian Quickest Detection in High-Dimensional Diffusions
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-27
  3. 3Follow the Gradient: Tracing Safety Degradation Back to Training Data
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-27
  4. 4From Interaction to Capability: Task Advancement as State and Supervision for GUI Agents
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-27
  5. 5Teach reality, not ethics: a drop-in continued-pretraining corpus installs coupled human-biosphere-AI welfare dispositions across model families
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-27
  6. 6Strategyproof Deep Reinforcement Learning for Online Resource Allocation
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-27
  7. 7A note on data-efficient finetuning of vision foundation models
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-27
  8. 8Orthogonal and Parametrized Multi-Attribute Steering of LLMs
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-27
  9. 9The Structure Gap: When More Data Cannot Replace Structured Reasoning
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-27
  10. 10GTFormer: A Syntax-Free Graph Network Alternative to Transformers for 8192$\times$ Length Extrapolation in Linear Time
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-27
  11. 11Which Hand Was Late? Diagnosis for Robot Failure Recovery in Bimanual Tool Use
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-27
  12. 12When, Where, and What to Correct: Human-Guided Adaptive Residual Correction for Robot Policies
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-27
  13. 13Collective Dynamics of Self-Reinforcement: How Cross-Task Coupling Drives Policy Collapse in Unsupervised RLVR
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-27
  14. 14Gym-Anything-RT: Scaling Real-Time Environments for Vision-Language Agents
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-27
  15. 15TeamGym: A Gym for Team Decision-Making by AI Agents with Distributed Information
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-27
  16. 16Timestral
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-27
  17. 17Understanding the limits of latent world models
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-27
  18. 18GAS-PFN: Graph-Aware Spatiotemporal Prior-Data Fitted Networks for Zero-Shot Traffic Forecasting
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-27
  19. 19EnvSched: Environment-Aware Scheduling for Rollout-Efficient Agent Reinforcement Learning
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-27
  20. 20VideoLVR: Latent Visual Reasoning over Grounded Video Evidence
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-27
  21. 21Rare, Relevant, and Still Wrong: How Memory Fails LLM Group Decisions
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-27
  22. 22What Improves During Pretraining? Tracking Knowledge Acquisition with Point-in-Time Models
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-27
  23. 23Metacognitive Decoding: Efficient Inference for Diffusion Language Agents
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-27
  24. 24Multi-State Linear Attention with Generalized Second-Order Routing
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-27
  25. 25MuonGS: Pretraining LLMs with Relaxed Weight-Norm Control
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-27
  26. 26Scaling Laws for Tool-Use Agents via Controllable Environment Synthesis
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-27
  27. 27Can LLMs Uncover Hidden Physical Laws through Free Exploration?
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-27
  28. 28GEPA-Curator: Curated Learning for Self Improving Agents
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-27
  29. 29Attention as Lattice Gauge Theory: Relations Interact Only Through Wilson Loops
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-27
  30. 30Heads of Security: Learned Attention Gates Improve Prompt Injection Robustness
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-27
  31. 31When Should We Intervene at Test Time?
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-27
  32. 32ThinkWorld: World Imagination Reinforces Chain-of-Dynamics Reasoning
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-27
  33. 33TrySee: Modular Perception-Enhanced VLM Grounding for Closed-Loop Robot Task Planning in Specialized Domains
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-26
  34. 34EvoRouter: Cost-Efficient Multi-Agent Routing via Insight-Guided On-Policy Distillation
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-26
  35. 35Not One Belief: Answer Payloads, Persistence, and Dissociated Confidence Signals in Conversational Question Answering
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-26
  36. 36MiRL: Error-Targeted Reinforcement Learning for Correcting Terminal Agent Misbehaviors
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-26
  37. 37From Self-Confidence to Statistical Accuracy: Scaling Human-in-the-Loop Evaluation with Direct and LLM-Assisted Auditing
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-26
  38. 38Triage: Skip, Skim, or Refine Attention Tiles in Video Diffusion Transformers
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-26
  39. 39Quantitative Generalization Bounds for Multi-Layer Transformers: A Few Long Strings can Beat Many Short Ones
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-26
  40. 40Audit Where Hacks Pay: Budgeted Strong-Verifier Audits Against Reward Hacking in GRPO
    arXiv cs.AI / cs.LG / cs.CL (recent) · 2026-10-26