ChatPaper

AI 論文熱榜

近 7 天全球 AI 研究者都在按讚的論文，白話解讀

  1. 1

    Repo-To-Skill：將 GitHub 儲存庫蒸餾為 AI4AI 技能

    🔥 4942026-09-02深入閱讀arXiv 原文Repo-To-Skill: Distilling GitHub Repositories Into AI4AI Skills
  2. 2

    StudentSim：訓練基於LLM的學生模擬器

    🔥 4582026-09-01深入閱讀arXiv 原文StudentSim: Training LLM-based Student Simulators
  3. 3

    Qwen-Drive-1.0：朝向自動駕駛視覺語言基礎模型的初步探索

    🔥 3372026-08-31深入閱讀arXiv 原文Qwen-Drive-1.0: An Initial Step towards a Vision-Language Foundation Model for Autonomous Driving
  4. 4

    HarnessDev：大型語言模型能否創建並演化自己的智能體框架？

    🔥 2242026-09-01深入閱讀arXiv 原文HarnessDev: Can LLMs Create and Evolve Their Own Agent Harness?
  5. 5

    Terminal-Universe：將代理軌跡轉化為可擴展的終端環境

    🔥 2132026-09-03深入閱讀arXiv 原文Terminal-Universe: Turning Agent Trajectories into Scalable Terminal Environments
  6. 6

    LLaDA-Image：以完全開放的訓練配方構建強大圖像生成器

    🔥 1962026-09-03深入閱讀arXiv 原文LLaDA-Image: Building Strong Image Generators with Fully Open Training Recipes
  7. 7

    Aspire：模型能否從模糊目標中自我進化？

    🔥 1732026-08-31深入閱讀arXiv 原文Aspire: Can Models Self-Evolve from Vague Goals?
  8. 8

    判斷何時不宜重用：自主大型語言模型後訓練中的條件式經驗遷移

    🔥 1392026-08-27深入閱讀arXiv 原文Knowing When Not to Reuse: Conditional Experience Transfer in Autonomous LLM Post-Training
  9. 9

    SolarWM：面向長時域視頻世界模型的開放數據與可擴展訓練

    🔥 1332026-09-02深入閱讀arXiv 原文SolarWM: Open Data and Scalable Training for Long-Horizon Video World Models
  10. 10

    隨機注意力：重新思考KV快取淘汰以實現高效推理

    🔥 1232026-09-03深入閱讀arXiv 原文Random Attention: Rethinking KV Cache Eviction for Efficient Reasoning
  11. 11

    EarlyEval：透過早期結果預測實現更經濟的代理評測

    🔥 1102026-09-02深入閱讀arXiv 原文EarlyEval: Cheaper Agent Evaluation via Early Outcome Prediction
  12. 12

    LatentPress：超越文字與視覺的上下文壓縮

    🔥 1022026-09-01深入閱讀arXiv 原文LatentPress: Context Compression Beyond Text and Vision
  13. 13

    同策略蒸馏真的在蒸馏吗？从噪声教师到自我改进

    🔥 982026-08-31深入閱讀arXiv 原文Does On-Policy Distillation Really Distill? From Noisy Teacher to Self-Improvement
  14. 14

    LoopArena：將模型作為迴圈工程的運行時控制器進行基準測試

    🔥 982026-08-28深入閱讀arXiv 原文LoopArena: Benchmarking Models as Runtime Controllers for Loop Engineering
  15. 15

    DreamX-Creator：普及2K分辨率的原生音视频生成

    🔥 902026-08-31深入閱讀arXiv 原文DreamX-Creator: Democratizing Native Audio-Video Generation at 2K Resolution
  16. 16

    DART-SD：多輪工具調用智能體自蒸餾的鑽石拓撲感知檢索與調優

    🔥 882026-08-19深入閱讀arXiv 原文DART-SD: Diamond-topology Aware Retrieval and Tuning for Self-Distillation of Multi-Turn Tool-Calling Agents
  17. 17

    超越資料擴展：以表徵為中心的視覺-語言-動作模型持續預訓練

    🔥 842026-08-27深入閱讀arXiv 原文Beyond Data Scaling: Representation-Centric Continued Pre-training for Vision-Language-Action Models
  18. 18

    SMELT：計算匹配MoE循環Transformer的擴展定律

    🔥 752026-09-01深入閱讀arXiv 原文SMELT: Scaling Laws for Compute-Matched MoE Looped Transformers
  19. 19

    Lucida：用於可組合真實到模擬場景建模的解析、生成與放置

    🔥 722026-08-31深入閱讀arXiv 原文Lucida: Parse, Generate, and Place for Composable Real-to-Sim Scene Modeling
  20. 20

    為何門控DeltaNet能經受4位元量化：混合式270億參數大型語言模型中遞迴部分的NVFP4 W4A4

    🔥 662026-09-03深入閱讀arXiv 原文Why Gated DeltaNet Survives 4-Bit Quantization: NVFP4 W4A4 for the Recurrent Half of a Hybrid 27B LLM

每天 5 分鐘，跟上 AI 前沿

每日精選 5 篇熱門論文，講成白話，配一道小謎題。

去看今日刊 →