Mean OSWorld Task Score Rises From 33.0% to 40.8%

Featured

• Hierarchical Continuous Diffusion Language Models score 10
  入选 HF Daily Papers; HF 热度: 78 upvotes (+4); 有代码实现; 关键词(1): reasoning
• AutoGUIWorld: Image Generators as Visual World Models for GUI Agent score 10
  入选 HF Daily Papers; HF 热度: 48 upvotes (+4); 有代码实现; 关键词(1): fine-tuning
• Scaling and Distilling Text Embeddings for Better Diffusibility score 10
  入选 HF Daily Papers; HF 热度: 41 upvotes (+4); 有代码实现; 关键词(1): scaling
• Beyond Memory: Harnessing Long-Horizon Agents with Explicit Belief States score 10
  入选 HF Daily Papers; HF 热度: 78 upvotes (+4); 有代码实现; 关键词(2): scaling, compression
• 4Director: Controlling Video World Models with Rigid 3D Geometry score 10
  入选 HF Daily Papers; HF 热度: 27 upvotes (+4); 有代码实现; 关键词(1): production
• ActiveSaddler: Automated Curriculum Learning for Agent Harness Optimization score 7
  入选 HF Daily Papers; HF 热度: 55 upvotes (+4)
• RPTune: Learned Context Curation for LLM Catalog Search score 8
  入选 HF Daily Papers; HF 热度: 20 upvotes (+4); 关键词(1): post-training
• ROWBench: Do Video Models Render What the Program Specifies? score 9
  入选 HF Daily Papers; HF 热度: 58 upvotes (+4); 有代码实现
• Sharpening Tax in Post-Training score 10
  入选 HF Daily Papers; HF 热度: 78 upvotes (+4); 有代码实现; 关键词(4): post-training, agentic, tool use, coding

Also Worth Noting

• OmniSeek: Native Tool Integration for Multi-turn Audio-Visual Reasoning score 7
  入选 HF Daily Papers; HF 热度: 11 upvotes (+3); 关键词(3): agentic, tool use, reasoning
• FORALL-LEAN-AGENT for Auditable Reasoning in Formal Mathematics and Software Verification score 4
  关键词(2): coding, reasoning; 顶会接收: NeurIPS
• Precision over Scale: A Polish-Silesian Benchmark and a Translation System Outperforming Open-Source and Commercial Models score 4
  关键词(1): open-source; 顶会接收: EMNLP
• Learning Rate Transfer for Hybrid Transformer-SSM Architectures score 4
  关键词(3): scaling, production, mamba; 顶会接收: NeurIPS
• Mixture-Trained Merging for Unified Multi-Objective Models score 4
  关键词(1): post-training; 顶会接收: NeurIPS
• Not All Is Lost: Repairing Lossy User Preference States of Personalization Encoders score 4
  关键词(2): compression, finetuning; 顶会接收: NeurIPS
• ARCCS: An Automated Regulatory Compliance Checking System score 4
  关键词(2): agentic, open-source; 顶会接收: EMNLP
• Decision Titan: Test-Time Training for Long-Term Memory in Offline Reinforcement Learning score 4
  关键词(1): scaling; 顶会接收: ICML
• Before It Fades: Reinforcing Temporal Representations at Inference Time in VideoLLMs score 4
  关键词(1): reasoning; 顶会接收: NeurIPS
• ATI-VLA: Action-Centric Predictive Vision-Language-Action Models via Actionable Alignment Then Adaptive Injection score 4
  关键词(2): lightweight, vision-language; 顶会接收: NeurIPS
• World Motion Models: Flexible Sequence Modeling of SE(3) Trajectories score 4
  关键词(1): robotics; 顶会接收: NeurIPS
• Wasserstein Gradient Flows and Forward-Only Diffusion Are Not Enough for Multimodal Sampling score 4
  关键词(1): scaling; 顶会接收: NeurIPS
• KaliBench: A Fine-Grained Benchmark for Cybersecurity Tool Use on Kali Linux with Runtime-Free Verifiable Rewards score 10
  入选 HF Daily Papers; HF 热度: 4 upvotes (+1); 有代码实现; 关键词(4): fine-tuning, MoE, agentic, tool use; 顶会接收: NeurIPS
• Block Optimism for Nonstationary Bandits with Latent Linear Dynamics score 3
  顶会接收: NeurIPS
• It Takes Workflows to Evolve Better Workflows score 3
  入选 HF Daily Papers