{
 "generated": "2026-11-05",
 "sha": "707df050bf85",
 "node": {
  "id": "ep:14",
  "type": "Episode",
  "name": "EP.14",
  "slug": "14",
  "ep": 14,
  "date": "2024-08-16",
  "page": "/buzzwords/14/",
  "n_models": 0,
  "title": "有人把蒙特卡洛树搬进了 agent 的规划",
  "oneliner": "MCTS 被搬进 agent 规划与 DPO 数据生成；SWE-bench 这条路上各家开始自建数据集；Sakana 的 AI Scientist 自动写论文",
  "minutes": 43,
  "template": "archive",
  "status": "published",
  "kind": "archive",
  "kind_zh": "档案重建",
  "entry": "agent",
  "by": "episode-import",
  "at": "2026-11-04T16:15:12",
  "n_stories": 6,
  "n_topics": 6,
  "n_items": 30,
  "n_sections": 1,
  "n_material": 30,
  "degree": 0,
  "n_entities": 10,
  "scale": "6 个故事线 / 专题 · 30 个条目 · 讲到 10 个实体"
 },
 "page": "https://xiaopingfeng.com/graph/e/ep/14/",
 "out": [],
 "in": [
  {
   "type": "part_of",
   "sentence": "想学 RAG，这个系列基本不可或缺属于EP.14",
   "direction": "in",
   "id": "story:14-t1",
   "name": "想学 RAG，这个系列基本不可或缺",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "事件驱动的多 agent 框架属于EP.14",
   "direction": "in",
   "id": "story:14-t2",
   "name": "事件驱动的多 agent 框架",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "SWE-bench Verified，以及各建各的数据集属于EP.14",
   "direction": "in",
   "id": "story:14-t3",
   "name": "SWE-bench Verified，以及各建各的数据集",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "Genie：号称 30 分，而榜首才 20 分属于EP.14",
   "direction": "in",
   "id": "story:14-t4",
   "name": "Genie：号称 30 分，而榜首才 20 分",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "把蒙特卡洛树搬进 agent 的规划属于EP.14",
   "direction": "in",
   "id": "story:14-t5",
   "name": "把蒙特卡洛树搬进 agent 的规划",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "Sakana 的 AI Scientist：自己写论文，自己发表属于EP.14",
   "direction": "in",
   "id": "story:14-t6",
   "name": "Sakana 的 AI Scientist：自己写论文，自己发表",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "The Ultimate Prompt Monitoring Pipeline属于EP.14",
   "direction": "in",
   "id": "item:14-m1-6800a3",
   "name": "The Ultimate Prompt Monitoring Pipeline",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "@jerryjliu0：This is one of the cleanest implementations of a complex multi-agent system that I've seen. Props to @seld…属于EP.14",
   "direction": "in",
   "id": "item:14-m1-6676b6",
   "name": "@jerryjliu0：This is one of the cleanest implementations of a complex multi-agent system that I've seen. Props to @seld…",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "GitHub - run-llama/multi-agent-concierge: An example of multi-agent orchestration with llama-index属于EP.14",
   "direction": "in",
   "id": "item:14-m1-726be5",
   "name": "GitHub - run-llama/multi-agent-concierge: An example of multi-agent orchestration with llama-index",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "@dotey：OpenAI 正在 ChatGPT 的 Mac 客户端测试新的提示词，这段提示词可以帮助用户直接通过对话触发共享屏幕的请求，弹出共享屏幕的界面，而不需要用户去点击共享屏幕按钮。属于EP.14",
   "direction": "in",
   "id": "item:14-m1-a7a576",
   "name": "@dotey：OpenAI 正在 ChatGPT 的 Mac 客户端测试新的提示词，这段提示词可以帮助用户直接通过对话触发共享屏幕的请求，弹出共享屏幕的界面，而不需要用户去点击共享屏幕按钮。",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "(Original)属于EP.14",
   "direction": "in",
   "id": "item:14-m1-aadfe7",
   "name": "(Original)",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "Introducing SWE-bench Verified属于EP.14",
   "direction": "in",
   "id": "item:14-m1-df22a4",
   "name": "Introducing SWE-bench Verified",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "SWE-bench Leaderboards属于EP.14",
   "direction": "in",
   "id": "item:14-m1-43b202",
   "name": "SWE-bench Leaderboards",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "@imxiaohu：兄弟们... 一个领先亚马逊和 Cognition，超越所有的现有AI模型 更像人类工程师的 AI ：Genie Cosine创建了一个基于真实开发者活动的数据集，专门用来训练，使其能够像人类一样思考和处理问题，而不…属于EP.14",
   "direction": "in",
   "id": "item:14-m1-35241e",
   "name": "@imxiaohu：兄弟们... 一个领先亚马逊和 Cognition，超越所有的现有AI模型 更像人类工程师的 AI ：Genie Cosine创建了一个基于真实开发者活动的数据集，专门用来训练，使其能够像人类一样思考和处理问题，而不…",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "Technical Report: Building Genie 1.0属于EP.14",
   "direction": "in",
   "id": "item:14-m1-1ac471",
   "name": "Technical Report: Building Genie 1.0",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "@JinaAI_：Based. Semantic chunking is overrated. Especially when you write a super regex that leverages all possible bo…属于EP.14",
   "direction": "in",
   "id": "item:14-m1-1ec413",
   "name": "@JinaAI_：Based. Semantic chunking is overrated. Especially when you write a super regex that leverages all possible bo…",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "gist.github.com属于EP.14",
   "direction": "in",
   "id": "item:14-m1-0d507e",
   "name": "gist.github.com",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "Segmenter API属于EP.14",
   "direction": "in",
   "id": "item:14-m1-bbd0bb",
   "name": "Segmenter API",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "@manateelazycat：《AI服务器， 开源开发者专属福利！》 🚀 重磅发布: #懒猫微服 - 你的专属AI私有云服务器！ 👨‍💻 源自国内顶尖操作系统团队 🛠️ 三年匠心打磨， 底层架构精心设计 🎉 流浪地球机甲风， 今天终于…属于EP.14",
   "direction": "in",
   "id": "item:14-m2-8f1a3a",
   "name": "@manateelazycat：《AI服务器， 开源开发者专属福利！》 🚀 重磅发布: #懒猫微服 - 你的专属AI私有云服务器！ 👨‍💻 源自国内顶尖操作系统团队 🛠️ 三年匠心打磨， 底层架构精心设计 🎉 流浪地球机甲风， 今天终于…",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "京东-欢迎登录属于EP.14",
   "direction": "in",
   "id": "item:14-m2-466a13",
   "name": "京东-欢迎登录",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "@aigcchu：用flux做游戏页面设计的想法太棒了。我在原工作流的基础上家了Gemma2的大模型做中文提示词的理解和提示词的优化。 后面计划出一个可以识别Axure实时生成页面效果的工作流。 本期工作流分享如下：属于EP.14",
   "direction": "in",
   "id": "item:14-m2-b1b2b1",
   "name": "@aigcchu：用flux做游戏页面设计的想法太棒了。我在原工作流的基础上家了Gemma2的大模型做中文提示词的理解和提示词的优化。 后面计划出一个可以识别Axure实时生成页面效果的工作流。 本期工作流分享如下：",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "ComfyUI Workflows - Developer Community属于EP.14",
   "direction": "in",
   "id": "item:14-m2-aecbeb",
   "name": "ComfyUI Workflows - Developer Community",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "Oops! The page was not found.属于EP.14",
   "direction": "in",
   "id": "item:14-m2-ac2faa",
   "name": "Oops! The page was not found.",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "multion.ai · introducing agent q research breakthrough for the next generation of a属于EP.14",
   "direction": "in",
   "id": "item:14-m2-57239a",
   "name": "multion.ai · introducing agent q research breakthrough for the next generation of a",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "Agent Q: Advanced Reasoning and Learning for Autonomous AI Agents属于EP.14",
   "direction": "in",
   "id": "item:14-m2-19311b",
   "name": "Agent Q: Advanced Reasoning and Learning for Autonomous AI Agents",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "Grok-2 Beta Release属于EP.14",
   "direction": "in",
   "id": "item:14-m2-f5f97a",
   "name": "Grok-2 Beta Release",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "@xai 的帖子属于EP.14",
   "direction": "in",
   "id": "item:14-m2-312778",
   "name": "@xai 的帖子",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "GitHub - ItzCrazyKns/Perplexica: Perplexica is an AI-powered answering engine.属于EP.14",
   "direction": "in",
   "id": "item:14-m2-5d2fc3",
   "name": "GitHub - ItzCrazyKns/Perplexica: Perplexica is an AI-powered answering engine.",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery属于EP.14",
   "direction": "in",
   "id": "item:14-m2-6c189e",
   "name": "The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "《人工智能科学家：迈向全自动开放式科学发现》属于EP.14",
   "direction": "in",
   "id": "item:14-m2-b3104c",
   "name": "《人工智能科学家：迈向全自动开放式科学发现》",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "Oops! The page was not found.属于EP.14",
   "direction": "in",
   "id": "item:14-m2-f76fcc",
   "name": "Oops! The page was not found.",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "GitHub - SakanaAI/AI-Scientist: The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery 🧑‍🔬属于EP.14",
   "direction": "in",
   "id": "item:14-m2-eee4b0",
   "name": "GitHub - SakanaAI/AI-Scientist: The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery 🧑‍🔬",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "blogs/msftgraphrag at master · tomasonjo/blogs · GitHub属于EP.14",
   "direction": "in",
   "id": "item:14-m2-e2930f",
   "name": "blogs/msftgraphrag at master · tomasonjo/blogs · GitHub",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "清华唐杰团队新作：一口气生成2万字，大模型开卷长输出-36氪属于EP.14",
   "direction": "in",
   "id": "item:14-m2-ed86c2",
   "name": "清华唐杰团队新作：一口气生成2万字，大模型开卷长输出-36氪",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "GitHub - THUDM/LongWriter: [ICLR 2025] LongWriter: Unleashing 10,000+ Word Generation from Long Context LLMs属于EP.14",
   "direction": "in",
   "id": "item:14-m2-9424e2",
   "name": "GitHub - THUDM/LongWriter: [ICLR 2025] LongWriter: Unleashing 10,000+ Word Generation from Long Context LLMs",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "Prompt caching - Claude API Docs属于EP.14",
   "direction": "in",
   "id": "item:14-m2-3e5833",
   "name": "Prompt caching - Claude API Docs",
   "status": "derived",
   "evidence": []
  }
 ]
}