十组护栏的基准分数范围相差23倍

重点关注

• CoworkEnv: Synthesizing Training Environments for Long-Horizon Cowork Agents score 10
  入选 HF Daily Papers；HF 热度: 38 upvotes (+4)；有代码实现；关键词(1): fine-tuning
• MT Forcing: Towards Stable and Consistent Multi-Turn Image Editing with Autoregressive Diffusion score 10
  入选 HF Daily Papers；HF 热度: 36 upvotes (+4)；有代码实现；关键词(1): distillation
• FineSplat: Fine Patches, Local Groups, Detailed Splats score 8
  入选 HF Daily Papers；HF 热度: 15 upvotes (+3)；有代码实现
• StopLoss: Teaching Agents to Predict Failure and Stop Early score 8
  入选 HF Daily Papers；HF 热度: 6 upvotes (+2)；有代码实现；关键词(1): coding
• Think Visually Across Space and Time: Grounded Reasoning for Video Segmentation score 7
  入选 HF Daily Papers；HF 热度: 18 upvotes (+3)；关键词(2): fine-tuning, reasoning
• World Model Interpretability: From Decoding Representations to Understanding Latent Worlds score 7
  入选 HF Daily Papers；HF 热度: 7 upvotes (+2)；有代码实现
• Policy Is Not Capability: Some Safety Benchmarks Measure Which Definition of "Unsafe" You Trained On score 6
  入选 HF Daily Papers；HF 热度: 8 upvotes (+2)；关键词(1): deployment

也值得关注

• Meta Inverse Prompting for Video Generative Models score 4
  关键词(2): production, text-to-video；顶会接收: NeurIPS
• SteadyThought: Mitigating LLM Under-Thinking via Thought-Level Preference Optimization score 4
  关键词(1): reasoning；顶会接收: NeurIPS
• MEMORA: A Memory-Enhanced Multimodal Committee Reranking Agent for Reasoning-Intensive Retrieval score 4
  关键词(2): agentic, reasoning；顶会接收: EMNLP
• Aligning Cultural Behaviors in Arabic Language Models via Preferences score 4
  关键词(1): post-training；顶会接收: EMNLP
• Paloma: Phase-Conditioned Residual Modulation for Time Series Forecasting score 4
  关键词(1): scaling；顶会接收: NeurIPS
• When Form Changes but Logic Doesn't: Building Logic-invariant LLMs through Logical Structure score 4
  关键词(1): reasoning；顶会接收: NeurIPS
• RARE: Reinforcement Learning with Adaptive Rubric Evolution for Open-Ended Generation score 4
  机构: Huawei；关键词(1): production
• Towards Human-AI Complementarity for AI Safety score 3
  机构: Cambridge
• UtoMe: Observation-Uncertainty-Guided Token Merging for Weather Foundation Models score 3
  顶会接收: NeurIPS
• Auto-MD-bench: Evaluating Autoresearch Agents for Real-World Medical Tasks score 3
  机构: Amazon
• MemBubble: Benchmarking How Agent Memory Shapes LLM Response Diversity score 3
  机构: Huawei
• Holo4D: Holistic 4D Reconstruction as Geometric Control for Video Diffusion score 3
  顶会接收: NeurIPS
• MCSplat: Multi-View Photometric and Geometric Consistent Feed-Forward Gaussian Splatting for Driving Scenes score 3
  顶会接收: NeurIPS