AgentWorld's Best Model Succeeds Only 52% of the Time

Featured

• FuseReg: Regularizing Layer Fusion Mitigates the Reconstruction-Generation Gap in Representation Autoencoders score 9
  入选 HF Daily Papers; HF 热度: 115 upvotes (+4); 有代码实现
• Block Sparse Attention with Log-Linear Complexity score 8
  入选 HF Daily Papers; HF 热度: 20 upvotes (+4); 关键词(2): scaling, reasoning
• InternW0-$Δ$: A World Action Model Bridging Predictive Dynamics and Actions with 20K+ Hours of Open Data score 7
  入选 HF Daily Papers; HF 热度: 17 upvotes (+3); 关键词(3): distillation, open-source, open source
• Game Arena: Strategic LLM Evaluation in Competitive Environments score 7
  机构: DeepMind; 入选 HF Daily Papers; HF 热度: 4 upvotes (+1)
• AgentWorld: Benchmarking Long-Horizon Collaboration of Multi-agent LLMs score 6
  入选 HF Daily Papers; HF 热度: 6 upvotes (+2); 关键词(1): open-source
• Enhancing Photogrammetric Digital Surface Models with Pretrained Diffusion Models and Multimodal Conditioning score 5
  入选 HF Daily Papers; HF 热度: 8 upvotes (+2)
• Softmax Reparameterization for Output-Head Quantization score 5
  入选 HF Daily Papers; HF 热度: 2 upvotes (+1); 关键词(5): scaling, compression, quantization, latency, post-training

Also Worth Noting

• Externalized CPDAG Summaries Improve LLM Causal Deduction score 4
  关键词(1): reasoning; 顶会接收: NeurIPS
• Bayesian Optimization with Fisher Information Geometry: Gradient Bounds and Trust-Region Methods score 4
  关键词(1): scaling; 顶会接收: NeurIPS
• Can Linguistic Reasoning Vectors Enhance Multimodal Reasoning Ability? score 4
  关键词(4): scaling, lightweight, reasoning, vision-language; 顶会接收: NeurIPS
• Who Says What: Symbolic Trimodal Binding Mechanisms in Audio-Visual LLMs score 4
  关键词(3): lightweight, fine-tuning, reasoning; 顶会接收: NeurIPS
• Highlight-Then-Summarize: Learning to Compress Evidence for Long-Context Understanding score 4
  机构: Baidu; 关键词(2): reasoning, open-source
• Compress What You See, Not What You Say: Anchored Context Distillation for Latent-Observation Software Engineering Agents score 4
  机构: Peking University; 关键词(4): compression, distillation, serving, throughput
• SatNav: A Scalable Benchmark for Long-Horizon UAV Vision-Language Navigation from Satellite Imagery score 4
  关键词(3): scaling, reasoning, vision-language; 顶会接收: NeurIPS
• GraphWrit3R: End-to-End 3D Scene Graph Writing score 4
  机构: Stanford; 关键词(2): deployment, open-source
• Mutable Transcripts: Mitigating Context Pollution through Editable Conversation State score 3
  顶会接收: NeurIPS
• Nonparametric In-Context Learning under Growing Geometric Complexity: Minimax Optimality and Local Geometry-Adaptivity of Transformers score 3
  顶会接收: NeurIPS
• PriceBench: A Diagnostic Benchmark for Price, Quality, and Brand Preferences in LLM Booking Agents score 3
  顶会接收: EMNLP
• Beyond Empirical Support: Structured Outlier Generation via Sinkhorn Optimal Transport score 3
  顶会接收: NeurIPS
• Online Learning via Learned Latent Bayesian Tracking score 3
  顶会接收: NeurIPS