[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-anthropic-ai-building-ai-recursive-self-improvement-zh":3,"article-related-anthropic-ai-building-ai-recursive-self-improvement-zh":31,"series-research-8a06c20a-c2d6-4cb0-a35c-69eab7f7f89a":82},{"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":23,"views":27,"created_at":28,"published_at":29,"topic_cluster_id":30},"8a06c20a-c2d6-4cb0-a35c-69eab7f7f89a","anthropic-ai-building-ai-recursive-self-improvement-zh","Anthropic 自己的數據已經證明：AI 正在幫 AI 進步","\u003Cp data-speakable=\"summary\">\u003Ca href=\"\u002Ftag\u002Fanthropic\">Anthropic\u003C\u002Fa> 的數據顯示，AI 已經在加速 AI \u003Ca href=\"\u002Fnews\u002Fwhat-vibe-coding-means-for-developers-zh\">開發\u003C\u002Fa>本身。\u003C\u002Fp>\u003Cp>我站在明確的一邊：AI 正在幫 AI 進步，而且這件事已經不是科幻，而是工程流程的現實。Anthropic 自己公布的資料很直接，到了 2026 年 5 月，超過 80% 的合併程式碼由 \u003Ca href=\"\u002Ftag\u002Fclaude\">Claude\u003C\u002Fa> 撰寫，工程師產出也比 2024 年高出約 8 倍。這不是單純的效率優化，而是研發結構改變，因為寫程式、審程式、測程式，已經有相當一部分交給模型接手。\u003C\u002Fp>\u003Ch2>第一個論點：AI 已經在實驗室內形成複利\u003C\u002Fh2>\u003Cp>Anthropic 最有力的證據，不是某個榜單分數，而是內部勞動分工的轉移。\u003Ca href=\"\u002Ftag\u002Fclaude-code\">Claude Code\u003C\u002Fa> 在 2025 年初推出前，Claude 參與撰寫的程式碼還只在個位數百分比；到 2026 年 5 月，這個比例已經突破 80%。這代表公司不再只是把 AI 當作建議層，而是把它放進生產層，由工程師負責定義問題、審查結果，而不是逐行手寫。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781257685705-1m6f.png\" alt=\"Anthropic 自己的數據已經證明：AI 正在幫 AI 進步\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>這件事重要，因為複利從\u003Ca href=\"\u002Fnews\u002Fvibe-coding-lets-you-ship-a-tiny-app-fast-zh\">工具\u003C\u002Fa>變成機器的一部分時才真正開始。Anthropic 表示，工程師現在每季交付的程式碼量，已經比 2021 到 2025 年間高出約 8 倍，而且當模型開始能更長時間自主工作後，曲線還進一步變陡。即使程式碼行數不能完全等同生產力，方向也很清楚：瓶頸已從「寫出程式」轉成「監督程式生成」。\u003C\u002Fp>\u003Ch2>第二個論點：外部基準也在同一方向上加速\u003C\u002Fh2>\u003Cp>Anthropic 的外部數據和內部數據互相印證。模型能穩定獨立完成的任務，正以大約每四個月翻倍的速度成長，快於先前約七個月的節奏。例子很具體：Claude Opus 3 在 2024 年 3 月能處理人類約 4 分鐘的工作；一年後，Sonnet 3.7 能處理約 90 分鐘的工作；再一年後，Opus 4.6 已經能處理 12 小時的任務。若這條曲線延續，日級任務今年就會進入可處理範圍，週級任務則可能在 2027 年出現。\u003C\u002Fp>\u003Cp>軟體與研究基準測試也在說同一件事。\u003Ca href=\"\u002Ftag\u002Fswe-bench\">SWE-bench\u003C\u002Fa> 在兩年內從低個位數進到飽和，CORE-Bench 這類測試模型能否重現已發表研究的基準，則從 2024 年約 20% 成功率，在 15 個月內接近飽和。這些不是花俏指標，它們測的是模型能不能真的執行會餵養下一代模型的工作。當模型能穩定重現、除錯、優化這條管線，距離「協助」和「自我改進」就會急速縮短。\u003C\u002Fp>\u003Ch2>反方可能怎麼說\u003C\u002Fh2>\u003Cp>嚴肅的反對意見也成立：Anthropic 距離一個能自己決定目標、判斷研究方向、端到端重設自身的模型，還差得很遠。公司自己也承認，Claude 已經能執行定義清楚的工作，甚至在某些實驗上接近或超過熟練人類，但它在判斷、目標選擇、以及長鏈條優先順序上仍然落後。這個落差是真實的，也正是「強力編碼\u003Ca href=\"\u002Fnews\u002Fmistral-vibe-cli-agent-still-matters-zh\">代理\u003C\u002Fa>」和「真正閉環自我改進系統」之間的分界。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781257681613-mf0c.png\" alt=\"Anthropic 自己的數據已經證明：AI 正在幫 AI 進步\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>另一個質疑是衡量方式本身有問題。程式碼行數只是粗糙代理，基準測試飽和也不等於通用智能。能在 SWE-bench 上大勝的模型，仍可能在組織判斷、安全權衡、或長期策略上失手。前沿 AI 最重要的工作不只是實作，而是決定該做什麼、該測什麼、以及該相信什麼。\u003C\u002Fp>\u003Cp>但這些反駁無法推翻核心結論。遞迴式自我改進不需要第一天就完全自治，它只需要足夠的能力，把研發管線一步一步從人轉到機器，直到機器承擔越來越多「改善下一代機器」的工作。Anthropic 自己的數據已經顯示這個轉移正在發生。等到完美閉環才重視，屬於分類錯誤。\u003C\u002Fp>\u003Ch2>你能做什麼\u003C\u002Fh2>\u003Cp>如果你是工程師，別再把 AI 當成更快的自動補全，而要把它當成會改變 review、測試、incident response 的勞動乘數；如果你是 PM，要把任務拆解、規格撰寫、評估設計視為每季都更重要的核心能力；如果你是創辦人，路線圖要建立在「軟體交付成本下降」與「判斷該交付什麼的成本上升」這兩件事同時發生的前提上。真正會贏的團隊，不是叫 AI 多寫幾行 code，而是把整條工作流重新設計成機器產出加上人類監督。","Anthropic 的內部與外部數據都顯示，AI 已經開始加速 AI 開發；這不是未來想像，而是今天就該警惕的工程現實。","www.anthropic.com","https:\u002F\u002Fwww.anthropic.com\u002Finstitute\u002Frecursive-self-improvement",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781257685705-1m6f.png","research","zh","d389cb06-cef8-48a6-abfc-0c5f5bcb6a26",[17,18,19,20,21,22],"Anthropic","AI 自我改進","遞迴式自我改進","AI 研發","程式碼生成","工程生產力",[24,25,26],"Anthropic 的內部數據顯示，AI 已經不是輔助層，而是進入生產層。","外部基準和研究重現能力也在快速提升，代表 AI 正在接近能改善 AI 開發流程的門檻。","最該調整的不是單一工具使用方式，而是整個研發、評估與審查流程。",0,"2026-06-12T09:47:24.801004+00:00","2026-06-12T09:47:24.79+00:00","0c35a120-52fc-41fc-afa3-d404eb934158",{"tags":32,"relatedLang":41,"relatedPosts":45},[33,34,36,38,40],{"name":19,"slug":19},{"name":18,"slug":35},"ai-自我改進",{"name":17,"slug":37},"anthropic",{"name":20,"slug":39},"ai-研發",{"name":21,"slug":21},{"id":15,"slug":42,"title":43,"language":44},"anthropic-ai-building-ai-recursive-self-improvement-en","Anthropic’s own data says AI is already building AI","en",[46,52,58,64,70,76],{"id":47,"slug":48,"title":49,"cover_image":50,"image_url":50,"created_at":51,"category":13},"59cf2061-712e-4a92-b3a7-5bdd8644c5a6","art-fine-tunes-multimodal-llms-via-pixels-zh","用像素微調多模態 LLM","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781266684477-t1np.png","2026-06-12T12:17:31.662347+00:00",{"id":53,"slug":54,"title":55,"cover_image":56,"image_url":56,"created_at":57,"category":13},"e08b8946-29a0-486a-b2c1-b23faf16b441","taxonomy-rwa-tokenization-blockchain-infrastructure-zh","RWA 代幣化的 23 維分類法","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781259482592-9fiv.png","2026-06-12T10:17:30.417901+00:00",{"id":59,"slug":60,"title":61,"cover_image":62,"image_url":62,"created_at":63,"category":13},"34681ebb-0d9d-4988-822a-45b6e5ad46d6","2026-llm-paper-lists-better-than-feeds-zh","2026 年的 LLM 論文清單，比資訊流更適合做研究","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781258570660-0l2n.png","2026-06-12T10:02:16.438561+00:00",{"id":65,"slug":66,"title":67,"cover_image":68,"image_url":68,"created_at":69,"category":13},"29143a1b-a610-4674-96a5-e3b1695350bd","project-glasswing-mythos-bug-chaining-zh","Project Glasswing 揭露 Mythos 會串漏洞","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781254982476-voas.png","2026-06-12T09:02:32.008908+00:00",{"id":71,"slug":72,"title":73,"cover_image":74,"image_url":74,"created_at":75,"category":13},"ba442703-edfa-4353-b256-db502d94a99e","mana-articulated-tool-manipulation-animation-zh","Mana把工具操作改寫成動畫","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781246882933-bvjm.png","2026-06-12T06:47:29.612828+00:00",{"id":77,"slug":78,"title":79,"cover_image":80,"image_url":80,"created_at":81,"category":13},"6911e614-4894-4f1f-a0ad-816e323793ef","retrieval-augmented-reinforcement-fine-tuning-analogy-zh","RA-RFT 讓模型學會類比推理","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781245983485-miul.png","2026-06-12T06:32:27.726554+00:00",[83,88,93,98,103,108,113,118,123,128],{"id":84,"slug":85,"title":86,"created_at":87},"f18dbadb-8c59-4723-84a4-6ad22746c77a","deepmind-bets-on-continuous-learning-ai-2026-zh","DeepMind 押注 2026 連續學習 AI","2026-03-26T08:16:02.367355+00:00",{"id":89,"slug":90,"title":91,"created_at":92},"f4a106cb-02a6-4508-8f39-9720a0a93cee","ml-papers-of-the-week-github-research-desk-zh","每週 ML 論文清單，為何紅到 GitHub","2026-03-27T01:11:39.284175+00:00",{"id":94,"slug":95,"title":96,"created_at":97},"c4f807ca-4e5f-47f1-a48c-961cf3fc44dc","ai-ml-conferences-to-watch-in-2026-zh","2026 AI 研討會投稿時程整理","2026-03-27T01:51:53.874432+00:00",{"id":99,"slug":100,"title":101,"created_at":102},"cf046742-efb2-4753-aef9-caed5da5e32e","adaptive-block-scaled-data-types-zh","IF4：神經網路量化的聰明選擇","2026-03-31T06:00:36.990273+00:00",{"id":104,"slug":105,"title":106,"created_at":107},"53a0dc54-0371-4e40-8d5e-74e94a73840c","geometry-aware-similarity-metrics-for-neural-representations-zh","超越距離測量：用微分幾何重新理解神經網路","2026-03-31T06:01:01.241968+00:00",{"id":109,"slug":110,"title":111,"created_at":112},"fee7d472-a775-4b1d-bbc2-1e8bca1bbf8b","on-the-fly-repulsion-in-the-contextual-space-for-rich-divers-zh","讓AI繪圖更有創意：用排斥力提升生成多樣性","2026-03-31T06:01:25.439673+00:00",{"id":114,"slug":115,"title":116,"created_at":117},"a9901203-d69b-447b-8854-15d14eab32b4","vision-aided-beam-prediction-cnn-eca-zh","影像輔助波束預測升級 CNN","2026-04-01T10:00:25.8073+00:00",{"id":119,"slug":120,"title":121,"created_at":122},"b55e7dd4-0a24-4b3d-804d-b0309a03f498","triple-band-fss-mimo-antenna-sub-6-ghz-zh","三頻 FSS MIMO 天線瞄準 sub-6 GHz","2026-04-01T13:18:36.857305+00:00",{"id":124,"slug":125,"title":126,"created_at":127},"f68290bd-e7f3-4b30-ba22-dcd4e0130a66","openclaw-1299-repos-eight-weeks-analysis-zh","OpenClaw 1299 個 Repo 的資料解讀","2026-04-02T05:03:45.208411+00:00",{"id":129,"slug":130,"title":131,"created_at":132},"ed9f80eb-eb02-4d35-8ad4-0ddf428751dd","beam-coherence-aware-combining-mmwave-mimo-zh","毫米波 MIMO 的雙階合併法","2026-04-02T05:27:26.897188+00:00"]