[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-why-fine-tuning-still-beats-prompt-only-ai-zh":3,"article-related-why-fine-tuning-still-beats-prompt-only-ai-zh":31,"series-research-85a3dfd4-9f87-4028-8ec1-44636ec804b8":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},"85a3dfd4-9f87-4028-8ec1-44636ec804b8","why-fine-tuning-still-beats-prompt-only-ai-zh","為什麼微調仍然勝過只靠提示詞的 AI","\u003Cp data-speakable=\"summary\">微調仍是把基礎模型做成可靠專用工具的最佳方法，因為它改變模型本身，而不只是包裝在外的提示詞。\u003C\u002Fp>\u003Cp>我站在微調這一邊：只靠提示詞，AI 很難在生產環境裡穩定表現；要把模型做成能重複交付的工具，還是得訓練。原因很直接，提示詞只能影響一次輸出，微調卻會把任務偏好寫進模型參數。當同一類輸入要反覆得到一致答案時，後者才是真正可控的做法。\u003C\u002Fp>\u003Ch2>第一個論點\u003C\u002Fh2>\u003Cp>微調之所以贏，不是因為它聽起來更「深」，而是因為它真的改了模型。大型語言模型在下游資料集上微調，通常能比基礎模型有更好的任務表現，這在 NLP 裡已是常規做法。GPT 類模型被廣泛拿去做下游微調，目的就是把通用語言能力轉成特定任務能力，這不是修飾輸出，而是重塑分布。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780120970410-mtab.png\" alt=\"為什麼微調仍然勝過只靠提示詞的 AI\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>最具體的例子就是 \u003Ca href=\"\u002Ftag\u002Fchatgpt\">ChatGPT\u003C\u002Fa> 本身。它不是單靠提示詞堆出來的產品，而是經過後訓練與對齊的 GPT 變體。Sparrow 這類對齊系統也依賴 post-training。這些產品之所以不只依賴 prompt，是因為 prompt 很脆弱，換個寫法、換個上下文，輸出就可能漂移；微調則能把風格、規則與領域知識固定下來。\u003C\u002Fp>\u003Ch2>第二個論點\u003C\u002Fh2>\u003Cp>效率技術的流行，反而證明大家要的是訓練，不是提示詞。LoRA 之所以受歡迎，是因為它能用幾百萬個可訓練參數去適配一個擁有數十億參數的模型，並且已經被整合進 Hugging Face 工具鏈與 Stable Diffusion 生態。這代表\u003Ca href=\"\u002Fnews\u002F5-takeaways-replit-enterprise-push-zh\">企業\u003C\u002Fa>與研究\u003Ca href=\"\u002Fnews\u002F5-openai-product-updates-for-teams-zh\">團隊\u003C\u002Fa>真正想要的是「可訓練但便宜」的適配方式，而不是永遠停留在 \u003Ca href=\"\u002Ftag\u002Fprompt-engineering\">prompt engineering\u003C\u002Fa>。\u003C\u002Fp>\u003Cp>更進一步的 ReFT 也說明同一件事。Stanford 的研究把它描述為只修改不到 1% 的表示層，仍然是在 frozen base model 上做任務特定干預。這不是對微調的否定，而是把微調做得更精準。市場一路往更窄、更省的訓練方式走，恰恰因為訓練比靜態提示更能解決真實任務。\u003C\u002Fp>\u003Ch2>反方可能怎麼說\u003C\u002Fh2>\u003Cp>反方最強的論點不是「微調沒用」，而是「微調會傷到泛化」。研究確實指出，模型一旦朝某個資料集適配，可能在分布外表現變差，甚至扭曲預訓練特徵。對通用助理來說，這是實打實的風險：模型在目標任務上更準，卻可能在其他場景變鈍。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780120967279-96zq.png\" alt=\"為什麼微調仍然勝過只靠提示詞的 AI\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>這個擔憂是真的，所以不該被輕描淡寫帶過。若你的產品目標是高度通用、\u003Ca href=\"\u002Fnews\u002F5-reasons-cursor-is-growing-so-fast-zh\">快速\u003C\u002Fa>試驗、需求還在變，prompt-only 的確更便宜，也更容易回收。微調不是萬能藥，前期資料品質差、評估不完整，反而會把錯誤寫進模型裡。\u003C\u002Fp>\u003Cp>但這不構成放棄微調的理由。正確做法是承認專用化會縮窄能力，再用工程手段補回去。已有研究提出把微調後權重與原始模型做 interpolation，可在保留任務收益的同時改善分布外行為。也就是說，問題不是「能不能微調」，而是「有沒有把微調納入可回滾、可評估、可修正的流程」。\u003C\u002Fp>\u003Ch2>你能做什麼\u003C\u002Fh2>\u003Cp>如果你是工程師、PM 或創辦人，別把 prompt 當預設解法。當任務穩定、可量測、而且會大量重複時，先做微調，再決定要用全量微調、LoRA，還是更受限的 representation 方法。先建立 eval set，先測一致性與分布外表現，再把 base model 留著做回退或插值。真正該追求的不是「最會說話的模型」，而是「在你的場景裡最可靠的模型」。\u003C\u002Fp>","微調仍是把基礎模型做成可靠專用工具的最佳方法，因為它改變模型本身，而不只是包裝在外的提示詞。","en.wikipedia.org","https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FFine-tuning_(deep_learning)",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780120970410-mtab.png","research","zh","ec8ce723-21d9-45ee-b847-ae02eeb9b6fa",[17,18,19,20,21,22],"fine-tuning","prompt engineering","LoRA","LLM","robustness","task adaptation",[24,25,26],"提示詞只能影響輸出，微調才會改變模型本身的任務偏好。","LoRA 與 ReFT 的流行，證明產業要的是更省的訓練，不是放棄訓練。","微調會帶來泛化風險，但可以用評估、插值與回退機制管理。",4,"2026-05-30T06:02:21.657693+00:00","2026-05-30T06:02:21.649+00:00","0c35a120-52fc-41fc-afa3-d404eb934158",{"tags":32,"relatedLang":41,"relatedPosts":45},[33,34,36,38,39],{"name":21,"slug":21},{"name":19,"slug":35},"lora",{"name":18,"slug":37},"prompt-engineering",{"name":17,"slug":17},{"name":20,"slug":40},"llm",{"id":15,"slug":42,"title":43,"language":44},"why-fine-tuning-still-beats-prompt-only-ai-en","Why fine-tuning still beats prompt-only 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},"f374155a-c29e-478c-b7a5-679cad1c51e4","crdts-keep-replicas-in-sync-without-locks-zh","CRDT 讓副本不用鎖也能同步","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781011086259-4p4k.png","2026-06-09T13:17:34.493426+00:00",{"id":53,"slug":54,"title":55,"cover_image":56,"image_url":56,"created_at":57,"category":13},"4b3b5a50-45b7-4238-a38b-160f82e323ff","post-deterministic-systems-autonomous-infra-zh","後決定性分散系：自治基礎設施新框架","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781010194792-5ogb.png","2026-06-09T13:02:32.717551+00:00",{"id":59,"slug":60,"title":61,"cover_image":62,"image_url":62,"created_at":63,"category":13},"04e45398-9814-4907-b416-fcb5b8d69508","causal-learnability-formal-language-tasks-zh","用因果法量化任務可學性","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780987696075-l4g0.png","2026-06-09T06:47:34.438642+00:00",{"id":65,"slug":66,"title":67,"cover_image":68,"image_url":68,"created_at":69,"category":13},"75bcc569-5e89-45c8-b809-6f169e929f4b","rl-training-hands-off-control-gradually-zh","RL 先接管再放手","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780986786312-03yo.png","2026-06-09T06:32:32.849589+00:00",{"id":71,"slug":72,"title":73,"cover_image":74,"image_url":74,"created_at":75,"category":13},"e3ecab4b-7cc7-4246-baf6-e1c170d86ca5","omnigamearena-vlm-game-agent-benchmark-zh","OmniGameArena 讓 VLM 遊戲代理更好比","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780985893022-70pl.png","2026-06-09T06:17:32.189729+00:00",{"id":77,"slug":78,"title":79,"cover_image":80,"image_url":80,"created_at":81,"category":13},"6f25a29c-cbb8-4f53-9af7-1656b394333a","turboquant-cuts-kv-cache-memory-6x-google-tests-zh","TurboQuant 在 Google 測試中省下 6x KV 快取","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780906682236-sqe2.png","2026-06-08T08:17:21.878314+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"]