HF Daily Papers · 2026-10-14

HF Daily Papers 精选（15 篇，按社区票数排序）

  • [272▲] S-Space: 在多模态语言模型中探索空间工作区 — https://arxiv.org/abs/2610.13527
  • [247▲] GroundingPI: 面向物理智能的接地基础模型与视觉基元 — https://arxiv.org/abs/2609.39601
  • [215▲] Thinking Inertia: 大语言模型在被告知停止时仍持续思考 — https://arxiv.org/abs/2610.11765
  • [201▲] AgentSWE: 编码智能体能否构建你真正想要的智能体 — https://arxiv.org/abs/2610.13361
  • [133▲] Scaling Agentic Data for Long-Horizon Terminal Intelligence — https://arxiv.org/abs/2610.13166
  • [78▲] VIS-Ground: 基于上下文接地的视频交互式叙事 — https://arxiv.org/abs/2610.09326
  • [73▲] GroundAnything: 以极速协调并行解码与精确视觉接地 — https://arxiv.org/abs/2609.39600
  • [34▲] Multi-Stream Attention Residuals — https://arxiv.org/abs/2610.13630
  • [30▲] Relevance Does Not Imply Applicability: 面向个人 GUI 智能体的经验激活 — https://arxiv.org/abs/2609.33304
  • [26▲] Contextual Correction of Parallel Drafts for Speculative Decoding — https://arxiv.org/abs/2610.12570
  • [25▲] PR-OPD: 用于智能体强化学习的特权表示在线自蒸馏 — https://arxiv.org/abs/2609.36642
  • [21▲] Humanity's Sixth Sense: Benchmarking Intuitive Visual Reasoning in Multimodal Models — https://arxiv.org/abs/2610.08966
  • [20▲] CrossTimeEdit: 跨越十年的跨视角数据集与用于历史街景生成的奖励引导编辑 — https://arxiv.org/abs/2609.36616
  • [17▲] MetaKernelBench: 衡量超越代码的 GPU Kernel 知识迁移 — https://arxiv.org/abs/2610.05014
  • [17▲] When Do We Need On-Policy Distillation? Distilling on Offline Student Rollouts Is Often Better — https://arxiv.org/abs/2610.11291