Daily Papers of 2026-02-23

  1. Does Your Reasoning Model Implicitly Know When to Stop Thinking? 253 upvotes, #1 of 2026-02-23
  2. VESPO: Variational Sequence-Level Soft Policy Optimization for Stable Off-Policy LLM Training 215 upvotes, #2 of 2026-02-23
  3. Generated Reality: Human-centric World Simulation using Interactive Video Generation with Hand and Camera Control 30 upvotes, #3 of 2026-02-23
  4. EgoPush: Learning End-to-End Egocentric Multi-Object Rearrangement for Mobile Robots 22 upvotes, #4 of 2026-02-23
  5. Spanning the Visual Analogy Space with a Weight Basis of LoRAs 14 upvotes, #5 of 2026-02-23
  6. DeepVision-103K: A Visually Diverse, Broad-Coverage, and Verifiable Mathematical Dataset for Multimodal Reasoning 10 upvotes, #6 of 2026-02-23
  7. Decoding as Optimisation on the Probability Simplex: From Top-K to Top-P (Nucleus) to Best-of-K Samplers 10 upvotes, #6 of 2026-02-23
  8. VidEoMT: Your ViT is Secretly Also a Video Segmentation Model 6 upvotes, #8 of 2026-02-23
  9. SARAH: Spatially Aware Real-time Agentic Humans 6 upvotes, #8 of 2026-02-23
  10. Avey-B 3 upvotes, #10 of 2026-02-23
  11. Adam Improves Muon: Adaptive Moment Estimation with Orthogonalized Momentum 3 upvotes, #10 of 2026-02-23
  12. Sink-Aware Pruning for Diffusion Language Models 3 upvotes, #10 of 2026-02-23
  13. Rubrics as an Attack Surface: Stealthy Preference Drift in LLM Judges 2 upvotes, #13 of 2026-02-23
  14. Selective Training for Large Vision Language Models via Visual Information Gain 2 upvotes, #13 of 2026-02-23
  15. 4RC: 4D Reconstruction via Conditional Querying Anytime and Anywhere 1 upvotes, #15 of 2026-02-23
  16. Whom to Query for What: Adaptive Group Elicitation via Multi-Turn LLM Interactions 1 upvotes, #15 of 2026-02-23
  17. ReIn: Conversational Error Recovery with Reasoning Inception 1 upvotes, #15 of 2026-02-23
  18. Learning Smooth Time-Varying Linear Policies with an Action Jacobian Penalty 1 upvotes, #15 of 2026-02-23

Data: hysts-bot-data/daily-papers-stats and the Daily Papers API. Open data: tardellirs/paper-pulse-data. Sister project: Model Pulse, the download history of every model on the Hub.