RRSI五个分布外基准最高提升4.7分

重点关注

• RRSI: Regularized Recursive Self-Improvement of Agent Harnesses score 10
  入选 HF Daily Papers；HF 热度: 189 upvotes (+4)；有代码实现；关键词(2): agentic, coding
• WorldCrafter: Consistent Video World Model with Implicit 3D-aware Memory score 10
  入选 HF Daily Papers；HF 热度: 143 upvotes (+4)；有代码实现；关键词(1): distillation
• GameHorizon Suite: Multi-Horizon Data and Evaluation in Gameplay score 9
  入选 HF Daily Papers；HF 热度: 126 upvotes (+4)；有代码实现
• onPanda: Efficient Annotation of On-Policy Alignment Data for LLMs and Agents via Token-Level Correction score 9
  入选 HF Daily Papers；HF 热度: 45 upvotes (+4)；有代码实现
• Jev-Mem: System-One-Controlled Agentic Memory for Efficient AI Agents score 10
  入选 HF Daily Papers；HF 热度: 22 upvotes (+4)；有代码实现；关键词(4): lightweight, latency, agentic, reasoning
• Harness-Zero: Harness Distillation via Agent-as-Harness score 10
  入选 HF Daily Papers；HF 热度: 29 upvotes (+4)；有代码实现；关键词(4): distillation, deployment, fine-tuning, tool use
• Document Retrieval-Aware Chunking (D-RAC): Universal Retrieval-Aware Ingestion of Enterprise Documents via PDF Normalization and Multimodal Markdown Conversion score 8
  入选 HF Daily Papers；HF 热度: 61 upvotes (+4)；关键词(4): lightweight, retrieval-augmented, RAG, agentic
• VideoGen-Agent: Reinforcing Video Generation Agents score 8
  入选 HF Daily Papers；HF 热度: 57 upvotes (+4)；关键词(4): fine-tuning, agentic, tool use, text-to-video
• ACLArena: Agent Continue Learning in Multi-stage Post-training score 9
  入选 HF Daily Papers；HF 热度: 10 upvotes (+3)；有代码实现；关键词(6): distillation, deployment, fine-tuning, post-training, agentic
• Complex KDA: Understanding and Enhancing the Expressivity of Kimi Delta Attention score 8
  入选 HF Daily Papers；HF 热度: 7 upvotes (+2)；有代码实现；关键词(2): scaling, open source

也值得关注

• Measuring the Checker: Mutation Analysis for GPU-Kernel Benchmark Oracles score 5
  入选 HF Daily Papers；HF 热度: 5 upvotes (+2)
• On Mitigation of Subliminal Learning in Large Language Models score 4
  机构: Microsoft；关键词(2): distillation, fine-tuning
• Do LLMs Choose Like Humans? Using Cognitive Theory to Evaluate LLM Decision-Making score 4
  机构: Harvard；关键词(2): reasoning, open-source
• A Tutorial on Prompt Engineering: From Messy Thoughts to AI Workflows score 4
  机构: Cambridge；关键词(1): agentic
• Authority-Preserving Evaluation of Medical Vision-Language Assistants score 4
  机构: ETH Zurich；关键词(1): vision-language
• EvidenT: Building Trustworthy Enterprise Assistants through Evidence Groundedness and Traceability score 4
  机构: Carnegie Mellon；关键词(2): lightweight, RAG
• Beetle: A Bilingual Model Suite for Modelling Second-Language Processing score 4
  机构: MIT；关键词(2): pretraining, open-source
• Intervention, Not Shared Latents: Blocking Visual Shortcuts in Audio-Video Generation score 3
  机构: Tsinghua
• The Right Future for Action: Learning Action-Relevant Predictive States in World Action Models score 3
  机构: Alibaba
• Bridging Reconstruction and Generation: A Latent Distribution Perspective on Evaluation and Improvement score 3
  机构: University of Toronto
• You Can Tell Who's Asking: What the Web's Questions Are Made Of, and Where They Come From score 3
  机构: Google