Daily Papers of 2026-08-06

  1. Recursive Synthesis for Long-Horizon Terminal Tasks 239 upvotes, #1 of 2026-08-06
  2. ABSeeker: Training Long-Horizon Search Agents via Answer-Backtracked Credit Assignment 65 upvotes, #2 of 2026-08-06
  3. Towards Physics of Multimodal Pretraining: Knowledge Flow, Modality Synergy, Early Unification, and Recipes 59 upvotes, #3 of 2026-08-06
  4. ToolArtist: Tool-Using Unified Multimodal Models for Agentic Image Generation 58 upvotes, #4 of 2026-08-06
  5. The Personalization Mirage: How LLMs Fabricate User Profiles, and Why Self-Monitoring Misleads 40 upvotes, #5 of 2026-08-06
  6. HelloWorld: Enabling Socially Interactive Characters in Video World Models 38 upvotes, #6 of 2026-08-06
  7. OneDayAgent: Towards a Long-Horizon Harness for Autonomous Agents 35 upvotes, #7 of 2026-08-06
  8. GDPevo: Evaluating Agent Self-Evolution on Real Business Tasks 27 upvotes, #8 of 2026-08-06
  9. Toward Skill-Native LLMs: Skill Entropy for Benchmarking and Training Long-Horizon Reasoning 27 upvotes, #8 of 2026-08-06
  10. K-EXAONE 2.0 Technical Report 25 upvotes, #10 of 2026-08-06
  11. Ego2Robot: Scalable Robot Data Synthesis from Egocentric Human Data 24 upvotes, #11 of 2026-08-06
  12. When Teachers Mislead: Spurious-Signal-Aware On-Policy Distillation 23 upvotes, #12 of 2026-08-06
  13. NOLLI: A Difficulty-Calibrated Puzzle Benchmark for Diagnosing the English-Korean Performance Gap 23 upvotes, #12 of 2026-08-06
  14. AVE-Compass: Towards Holistic Evaluation for Audio-Video Editing Abilities 18 upvotes, #14 of 2026-08-06
  15. Distill Where You Fail: Recovering Learning Signals of Negative RL-Groups from Adaptive Teacher Guidance 16 upvotes, #15 of 2026-08-06
  16. When Memory Lies: An Empirical Study of Spatial Memory Staleness in VLM Agents 16 upvotes, #15 of 2026-08-06
  17. Lossless Tensor Compression as Program Synthesis 15 upvotes, #17 of 2026-08-06
  18. SKILL-KD: Contrastive Skill Distillation for LLM Agents 14 upvotes, #18 of 2026-08-06
  19. FocusMem: Factorizing Content, Readout, and Trust in Latent GUI Memory 14 upvotes, #18 of 2026-08-06
  20. WorldCycle: Self-Verifiable Reinforcement Learning for Long-Horizon Video World Models 13 upvotes, #20 of 2026-08-06
  21. Poly-OPD: Heterogeneous Multi-Teacher On-Policy Distillation for Capability-Selectable Flow Models 12 upvotes, #21 of 2026-08-06
  22. OPD-V: Visual On-Policy Self-Distillation with Modality Balance 12 upvotes, #21 of 2026-08-06
  23. FinanceHarness: Autonomous Financial Deep Research Framework 11 upvotes, #23 of 2026-08-06
  24. Self-Evolving Coding Agents 8 upvotes, #24 of 2026-08-06
  25. UniWorld-View: Large-Baseline View Synthesis via Video Diffusion Models 8 upvotes, #24 of 2026-08-06
  26. Consistency-Driven Co-Evolution for Self-Supervised Cross-Representation Learning 8 upvotes, #24 of 2026-08-06
  27. BridgeVLA++: A Data-Efficient, Generalizable, and Memory-Augmented Vision-Language-Action Framework for 3D Manipulation 8 upvotes, #24 of 2026-08-06
  28. Agent Against Agent: An Agentic System for Automatic Prompt Injection Red Teaming 8 upvotes, #24 of 2026-08-06
  29. TriGlue: a Biology-Inspired Generative Model for Generating Molecular Glue-Induced Ternary Complex 6 upvotes, #29 of 2026-08-06
  30. What AI Red-Team Evaluations Can and Cannot Prove 5 upvotes, #30 of 2026-08-06
  31. DRIFT: Derailing Denoising Trajectories of Flow-Matching VLAs with Adversarial Patch Attack 4 upvotes, #31 of 2026-08-06
  32. Resume Means Resume: A Machine-Checked Conformance Contract for Checkpoint, Interrupt, and Resume Semantics in Workflow Persistence Layers 4 upvotes, #31 of 2026-08-06
  33. SIGNPOST-Bench: Benchmarking Text-Vision Conflict Resolution in Multimodal Large Language Models 3 upvotes, #33 of 2026-08-06
  34. Helping Music Co-Creation Agents 'Listen' Well: Hierarchical Self-Supervised World Models for Understanding and Generation 2 upvotes, #34 of 2026-08-06

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.