[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-wengli-returns-openai-rsi-team-focus-zh":3,"article-related-wengli-returns-openai-rsi-team-focus-zh":32,"series-industry-5e4adf65-51c8-42cc-8144-b657ff7619e1":77},{"id":4,"slug":5,"title":6,"content":7,"summary":8,"source":9,"source_url":10,"author":11,"image_url":12,"cover_image":12,"category":13,"language":14,"translated_content":11,"related_article_id":15,"keywords":16,"key_takeaways":25,"views":29,"created_at":30,"published_at":31,"topic_cluster_id":11},"5e4adf65-51c8-42cc-8144-b657ff7619e1","wengli-returns-openai-rsi-team-focus-zh","翁荔回OpenAI：5個關鍵看點","\u003Cp>翁荔為什麼又回 \u003Ca href=\"\u002Ftag\u002Fopenai\">OpenAI\u003C\u002Fa>？\u003C\u002Fp>\u003Cp data-speakable=\"summary\">翁荔離開 Thinking Machines Lab 後回到 OpenAI，接手自進化 RSI 團隊。\u003C\u002Fp>\u003Ctable>\u003Cthead>\u003Ctr>\u003Cth>項目\u003C\u002Fth>\u003Cth>重點\u003C\u002Fth>\u003Cth>可比資訊\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd>OpenAI RSI\u003C\u002Ftd>\u003Ctd>自進化研究\u003C\u002Ftd>\u003Ctd>內部高優先級\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Thinking Machines Lab\u003C\u002Ftd>\u003Ctd>前東家\u003C\u002Ftd>\u003Ctd>6 位聯創中 3 位回流 OpenAI\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>翁荔\u003C\u002Ftd>\u003Ctd>研究負責人\u003C\u002Ftd>\u003Ctd>曾任 VP of Research and Safety\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Lil’Log\u003C\u002Ftd>\u003Ctd>個人技術博客\u003C\u002Ftd>\u003Ctd>長期寫 Transformer、RL、Agent\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003Ch2>1. RSI 團隊才是這次回歸的核心\u003C\u002Fh2>\u003Cp>翁荔回到 \u003Ca href=\"https:\u002F\u002Fopenai.com\u002F\">OpenAI\u003C\u002Fa>，不是去一般研究組，而是接手 RSI 團隊。RSI 指的是自進化方向，目標是讓\u003Ca href=\"\u002Fnews\u002Fbuild-small-language-model-deepmind-zh\">模型\u003C\u002Fa>訓練並改進自己的後繼模型。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1785567775278-k78k.png\" alt=\"翁荔回OpenAI：5個關鍵看點\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>這個位置之所以受矚目，是因為它被描述為 OpenAI 內部優先級很高的研究線。放一位資深研究負責人進去，代表公司把這條線看得很重。\u003C\u002Fp>\u003Cul>\u003Cli>方向：模型改進自己的後繼模型\u003C\u002Fli>\u003Cli>定位：高優先級研究\u003C\u002Fli>\u003Cli>性質：研究與訓練方法的前沿嘗試\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>2. 她離開 TML 的原因是健康，不是轉行\u003C\u002Fh2>\u003Cp>翁荔本週一才宣布離開 Thinking Machines Lab，公開原因是健康問題。她表示自己已經無法用初創公司需要的節奏工作，因此想轉向更可預測、邊界更清楚的環境。\u003C\u002Fp>\u003Cp>這也說明她並沒有離開 AI 研究，而是在調整工作強度。她在離職聲明裡還說，自己仍會持續閱讀前沿 AI 論文。\u003C\u002Fp>\u003Cul>\u003Cli>離開原因：健康\u003C\u002Fli>\u003Cli>需求改變：更穩定的工作節奏\u003C\u002Fli>\u003Cli>去向：仍在 AI 研究一線\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>3. 她在 OpenAI 的資歷本來就很深\u003C\u002Fh2>\u003Cp>翁荔不是第一次加入 OpenAI。她本科畢業於北京大學，後來出國讀博，早期研究重心是強化學習。2018 年加入 OpenAI 後，她待了六年，做過強化學習、機器人、應用 AI 和基礎模型相關工作。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1785567770013-9bry.png\" alt=\"翁荔回OpenAI：5個關鍵看點\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>她後來升任 VP of Research and Safety，負責 Safety Systems 團隊，研究模型如何 jailbreak、如何產生有害內容，以及如何透過後訓練替 \u003Ca href=\"\u002Ftag\u002Fchatgpt\">ChatGPT\u003C\u002Fa> 加上護欄。\u003C\u002Fp>\u003Cpre>\u003Ccode>早期：強化學習、機器人研究\n中期：應用 AI、基礎模型\n後期：安全系統、後訓練、護欄設計\u003C\u002Fcode>\u003C\u002Fpre>\u003Ch2>4. TML 與 OpenAI 的人才回流已經很明顯\u003C\u002Fh2>\u003Cp>這次回流不只是個人選擇，也反映出 Thinking Machines Lab 的聯創陣容正在縮小。六位聯創裡，已有三人回到 OpenAI：Barret Zoph、Luke Metz 和翁荔。\u003C\u002Fp>\u003Cp>Andrew Tulloch 則去了 \u003Ca href=\"\u002Ftag\u002Fmeta\">Meta\u003C\u002Fa>，剩下 Mira Murati 和 John Schulman 還在。對一家成立不久、又拿到巨額融資的實驗室來說，這種流動會直接影響研究連續性。\u003C\u002Fp>\u003Cul>\u003Cli>六位聯創中，三位已回 OpenAI\u003C\u002Fli>\u003Cli>Andrew Tulloch 去了 Meta\u003C\u002Fli>\u003Cli>仍在 TML 的核心人物：Mira Murati、John Schulman\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>5. 她的影響力不只來自職位\u003C\u002Fh2>\u003Cp>翁荔在 AI 圈還有另一個身份：高產技術寫作者。她長期經營 \u003Ca href=\"https:\u002F\u002Flilianweng.github.io\u002F\">Lil’Log\u003C\u002Fa>，寫過 Transformer、強化學習、\u003Ca href=\"\u002Ftag\u002Fagent\">Agent\u003C\u002Fa> 等主題，很多文章都成了入門材料。\u003C\u002Fp>\u003Cp>這也能解釋她為何適合 RSI 這種跨研究、\u003Ca href=\"\u002Fnews\u002Fti-shi-gong-cheng-vs-hui-quan-gong-cheng-vs-tu-pu-gong-cheng-zh\">工程\u003C\u002Fa>和方法論的團隊。能把複雜研究拆成可理解框架的人，往往也更擅長推進前沿項目。\u003C\u002Fp>\u003Cul>\u003Cli>博客：Lil’Log\u003C\u002Fli>\u003Cli>常寫主題：Transformer、RL、Agent\u003C\u002Fli>\u003Cli>影響：常被當作技術入門資料\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>哪種讀者最該看這條新聞\u003C\u002Fh2>\u003Cp>如果你關心技術路線，重點看 RSI，因為它代表模型自我改進的研究野心。如果你關心產業格局，重點看 OpenAI 和 TML 之間的人才回流，這說明前沿 AI 實驗室的競爭已經不只是搶人，還包括研究節奏與組織穩定性。\u003C\u002Fp>\u003Cp>如果你是研究員或創業者，這條\u003Ca href=\"\u002Fnews\u002Fopenai-newsroom-announcements-digest-zh\">新聞\u003C\u002Fa>最值得記住的不是誰去了哪裡，而是健康、節奏和研究方向，已經和薪資、頭銜一樣，成為頂尖人才做決定時的核心變數。\u003C\u002Fp>","5 個看點看懂翁荔回 OpenAI、接手 RSI 團隊，以及 OpenAI 與 TML 之間的人才回流。","zhuanlan.zhihu.com","https:\u002F\u002Fzhuanlan.zhihu.com\u002Fp\u002F2066077900957393039",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1785567775278-k78k.png","industry","zh","4f3a42d9-39c7-4299-a6c5-7ad1164da664",[17,18,19,20,21,22,23,24],"翁荔","OpenAI","Thinking Machines Lab","RSI","自進化","人才流動","Lil’Log","AI 研究",[26,27,28],"翁荔回到 OpenAI，接手的是高優先級的 RSI 自進化團隊。","她離開 TML 的主因是健康，而不是退出 AI 研究。","OpenAI 與 TML 之間的人才回流，反映前沿實驗室競爭正在升級。",0,"2026-08-01T07:02:31.085891+00:00","2026-08-01T07:02:31.057+00:00",{"tags":33,"relatedLang":36,"relatedPosts":40},[34],{"name":18,"slug":35},"openai",{"id":15,"slug":37,"title":38,"language":39},"lilian-weng-returns-openai-rsi-team-en","Lilian Weng returns to OpenAI to lead RSI","en",[41,47,53,59,65,71],{"id":42,"slug":43,"title":44,"cover_image":45,"image_url":45,"created_at":46,"category":13},"9ae7f70e-d9ef-4fc0-a328-dc64154a6333","anthropic-texas-buildout-drawing-15b-debt-zh","Anthropic 德州擴建正在撬動 150 億美元債務","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1785565959574-9t35.png","2026-08-01T06:32:17.374264+00:00",{"id":48,"slug":49,"title":50,"cover_image":51,"image_url":51,"created_at":52,"category":13},"ee0f2de5-3337-4167-a623-14a3b352a1bc","cognizant-claude-partnership-pilots-to-production-zh","Cognizant 把 Claude 變成上線流程","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1785564189980-cu2v.png","2026-08-01T06:02:48.042108+00:00",{"id":54,"slug":55,"title":56,"cover_image":57,"image_url":57,"created_at":58,"category":13},"7df8569c-4934-4733-9a2a-445420b0c7a4","ti-shi-gong-cheng-vs-hui-quan-gong-cheng-vs-tu-pu-gong-cheng-zh","提示工程 vs 迴圈工程 vs 圖譜工程","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1785547966803-qov5.png","2026-08-01T01:32:25.912529+00:00",{"id":60,"slug":61,"title":62,"cover_image":63,"image_url":63,"created_at":64,"category":13},"c869bb61-6e6b-4cfb-b9a9-8142daaf0d1a","pwcs-ai-blunder-verification-beats-prompt-engineering-zh","PwC 的 AI 失誤證明：驗證比提示工程更重要","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1785546161684-tn7d.png","2026-08-01T01:02:18.564567+00:00",{"id":66,"slug":67,"title":68,"cover_image":69,"image_url":69,"created_at":70,"category":13},"32e46e55-e017-43f1-8ada-bd33c57cbf32","alphafold-breakup-turns-science-into-gemini-work-zh","AlphaFold 拆組成 Gemini 工程","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1785524596007-62sw.png","2026-07-31T19:02:46.126821+00:00",{"id":72,"slug":73,"title":74,"cover_image":75,"image_url":75,"created_at":76,"category":13},"0c487fac-7e5e-4ae0-ad18-59ca7c37cd01","rust-to-zig-rewrite-progress-update-zh","Rust 轉 Zig：重寫已過最難關","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1785501166060-u7ke.png","2026-07-31T12:32:19.644785+00:00",[78,83,88,93,98,103,108,113,118,123],{"id":79,"slug":80,"title":81,"created_at":82},"ee073da7-28b3-4752-a319-5a501459fb87","ai-in-2026-what-actually-matters-now-zh","2026 AI 真正重要的事","2026-03-26T07:09:12.008134+00:00",{"id":84,"slug":85,"title":86,"created_at":87},"83bd1795-8548-44c9-9a7e-de50a0923f71","trump-ai-framework-power-speech-state-preemption-zh","川普 AI 框架瞄準電力、言論與州權","2026-03-26T07:12:18.695466+00:00",{"id":89,"slug":90,"title":91,"created_at":92},"ea6be18b-c903-4e54-97b7-5f7447a612e0","nvidia-gtc-2026-big-ai-announcements-zh","NVIDIA GTC 2026 重點拆解","2026-03-26T07:14:26.62638+00:00",{"id":94,"slug":95,"title":96,"created_at":97},"4bcec76f-4c36-4daa-909f-54cd702f7c93","claude-users-spreading-out-and-getting-better-zh","Claude 用戶更分散，也更會用","2026-03-26T07:22:52.325888+00:00",{"id":99,"slug":100,"title":101,"created_at":102},"bd903b15-2473-4178-9789-b7557816e535","openclaw-raises-hard-question-for-ai-models-zh","OpenClaw 逼問 AI 模型價值","2026-03-26T07:24:54.707486+00:00",{"id":104,"slug":105,"title":106,"created_at":107},"eeac6b9e-ad9d-4831-8eec-8bba3f9bca6a","gap-google-gemini-checkout-fashion-search-zh","Gap 把結帳搬進 Gemini","2026-03-26T07:28:23.937768+00:00",{"id":109,"slug":110,"title":111,"created_at":112},"0740e53f-605d-4d57-8601-c10beb126f3c","google-pushes-gemini-transition-to-march-2026-zh","Google 把 Gemini 轉換延到 2026 年 3…","2026-03-26T07:30:12.825269+00:00",{"id":114,"slug":115,"title":116,"created_at":117},"e660d801-2421-4529-8fa9-86b82b066990","metas-llama-4-benchmark-scandal-gets-worse-zh","Meta Llama 4 分數風波又擴大","2026-03-26T07:34:21.156421+00:00",{"id":119,"slug":120,"title":121,"created_at":122},"183f9e7c-e143-40bb-a6d5-67ba84a3a8bc","accenture-mistral-ai-sovereign-enterprise-deal-zh","Accenture 攜手 Mistral AI 賣主權 AI","2026-03-26T07:38:14.818906+00:00",{"id":124,"slug":125,"title":126,"created_at":127},"191d9b1b-768a-478c-978c-dd7431a38149","mistral-ai-faces-its-hardest-year-yet-zh","Mistral AI 迎來最硬的一年","2026-03-26T07:40:23.716374+00:00"]