[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-vibe-research-ai-tools-workflows-zh":3,"article-related-vibe-research-ai-tools-workflows-zh":41,"series-tools-a109dac1-43f3-4a6b-982c-13b59e8f61e9":93},{"id":4,"title":5,"content":6,"summary":7,"source":8,"source_url":9,"author":10,"image_url":11,"keywords":12,"language":23,"translated_content":10,"views":24,"is_premium":25,"created_at":26,"updated_at":26,"cover_image":11,"published_at":27,"rewrite_status":28,"rewrite_error":10,"rewritten_from_id":29,"slug":30,"category":31,"related_article_id":32,"status":33,"google_indexed_at":34,"x_posted_at":10,"tweet_text":10,"title_rewritten_at":10,"title_original":10,"key_takeaways":35,"topic_cluster_id":39,"embedding":40,"is_canonical_seed":25},"a109dac1-43f3-4a6b-982c-13b59e8f61e9","Vibe Research：用 AI 加速研究流程","\u003Cp data-speakable=\"summary\">Vibe research 是把 LLM、\u003Ca href=\"\u002Ftag\u002Fagent\">agent\u003C\u002Fa>、coding 工具和 review loop 串起來，讓研究流程更快變成可執行的工作。\u003C\u002Fp>\u003Cp>說真的，這東西不只是聊天而已。它是在把研究工作拆成可交給 AI 的步驟。從讀論文、改程式，到跑實驗、比結果，都能接進同一條流程。\u003C\u002Fp>\u003Cp>這件事很實際。因為研究最卡的地方，常常不是想法，而是中間那段雜事。模型可以先讀文件，也可以去看 repo、改檔案、跑測試，然後把結果丟回來給人判斷。\u003C\u002Fp>\u003Cp>講白了，重點不是模型會不會寫字。重點是它能不能真的進工作流。這也是 vibe research 跟一般聊天機器人差最多的地方。\u003C\u002Fp>\u003Ctable>\u003Cthead>\u003Ctr>\u003Cth>流程環節\u003C\u002Fth>\u003Cth>AI 做什麼\u003C\u002Fth>\u003Cth>為什麼有用\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd>文獻回顧\u003C\u002Ftd>\u003Ctd>摘要論文與抽出主張\u003C\u002Ftd>\u003Ctd>縮短第一輪閱讀時間\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>程式修改\u003C\u002Ftd>\u003Ctd>讀 repo 並編輯檔案\u003C\u002Ftd>\u003Ctd>讓實驗更容易反覆迭代\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>實驗迴圈\u003C\u002Ftd>\u003Ctd>跑測試並比較輸出\u003C\u002Ftd>\u003Ctd>加快重複性評估\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>審查系統\u003C\u002Ftd>\u003Ctd>對照規則或 rubric 檢查結果\u003C\u002Ftd>\u003Ctd>提早抓出薄弱結論\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003Ch2>Vibe research 到底是什麼\u003C\u002Fh2>\u003Cp>Vibe research 不是正式標準，也不是單一產品。它比較像一種工作法。把大型語言模型、\u003Ca href=\"\u002Ftag\u002Fagentic-coding\">agentic coding\u003C\u002Fa> 工具、實驗追蹤，還有人類審查放在同一個迴圈裡。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778904653705-zekc.png\" alt=\"Vibe Research：用 AI 加速研究流程\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>這種做法的目標很直接。就是讓研究更可執行。想法不再停在筆記裡，而是能一路走到程式、實驗和結果。\u003C\u002Fp>\u003Cp>你可以把它想成研究版的自動化管線。不是叫 AI 幫你想完所有答案，而是讓它幫你處理那些重複、瑣碎、但很花時間的步驟。\u003C\u002Fp>\u003Cp>現在很多工具都在往這方向走。\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fclaude-code\" target=\"_blank\" rel=\"noopener\">Claude Code\u003C\u002Fa> 可以從 terminal 看 repo、改檔案。\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fopenai\u002Fcodex\" target=\"_blank\" rel=\"noopener\">OpenAI Codex\u003C\u002Fa> 把 code-oriented 助手放進開發流程。\u003Ca href=\"https:\u002F\u002Fwww.cursor.com\" target=\"_blank\" rel=\"noopener\">Cursor\u003C\u002Fa> 則讓你在同一個介面裡問問題、改程式、保留上下文。\u003C\u002Fp>\u003Cul>\u003Cli>LLM 適合做論文摘要、假設草稿、結果整理。\u003C\u002Fli>\u003Cli>Agent 可以改程式、跑指令、照 checklist 做事。\u003C\u002Fli>\u003Cli>Coding 工具把研究脈絡綁在 repo 上。\u003C\u002Fli>\u003Cli>Review 系統能在結論前多看一眼。\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>為什麼流程比模型更重要\u003C\u002Fh2>\u003Cp>研究工作裡，模型本身很重要，但流程設計\u003Ca href=\"\u002Fnews\u002Faws-repository-wide-security-scanner-matters-zh\">更重要\u003C\u002Fa>。因為複雜任務一多，單靠一個強模型不夠。它可能講得很順，卻不一定對。\u003C\u002Fp>\u003Cp>如果有 tests、logs、rubric 和 review，情況就不同。就算模型沒那麼強，還是能省下不少時間。因為它至少知道下一步該做\u003Ca href=\"\u002Fnews\u002Fwhy-docker-microvm-sandboxes-ai-agents-zh\">什麼\u003C\u002Fa>。\u003C\u002Fp>\u003Cp>這也是很多團隊現在的做法。不是只丟一個 prompt，而是做一整個 loop。先規劃，再改 code，再跑實驗，再檢查結果，最後才做人類決策。\u003C\u002Fp>\u003Cp>這裡有一個很重要的觀念。失敗要能被看見。不要把錯誤包裝成漂亮文字。那種東西看起來很像答案，其實只是幻覺。\u003C\u002Fp>\u003Cblockquote>\u003Cp>“The future of software development is going to be less about writing code and more about orchestrating AI systems.” — Andrej Karpathy\u003C\u002Fp>\u003C\u002Fblockquote>\u003Cp>Karpathy 這句話放在 vibe research 上也很準。研究者正在從「自己一行一行寫」變成「管理一個會做事的系統」。\u003C\u002Fp>\u003Cp>這代表好習慣也要跟著變。任務定義要清楚。實驗要可重現。審查標準要寫明白。因為 agent 跑得快，亂起來也很快。\u003C\u002Fp>\u003Cp>我覺得這點很現實。你如果沒有把流程定好，AI 只會幫你更快地把混亂放大。\u003C\u002Fp>\u003Ch2>這些工具實際上差在哪\u003C\u002Fh2>\u003Cp>不同工具，吃的是不同環節。有人強在 code edit。有人強在\u003Ca href=\"\u002Ftag\u002F長上下文\">長上下文\u003C\u002Fa>閱讀。有人強在把專案維持得比較整齊。這差很多。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778904668491-899n.png\" alt=\"Vibe Research：用 AI 加速研究流程\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>如果你在做 model research，需求跟產品實驗不一樣。如果你是工程導向的分析，也不一樣。所以工具選擇不能只看名氣。\u003C\u002Fp>\u003Cp>下面這幾個例子很常見。\u003Ca href=\"https:\u002F\u002Fwww.cursor.com\" target=\"_blank\" rel=\"noopener\">Cursor\u003C\u002Fa> 適合 codebase 很重的工作。\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fclaude-code\" target=\"_blank\" rel=\"noopener\">Claude Code\u003C\u002Fa> 適合 terminal 工作流。\u003Ca href=\"https:\u002F\u002Fopenai.com\u002Findex\u002Fintroducing-codex\u002F\" target=\"_blank\" rel=\"noopener\">Codex\u003C\u002Fa> 適合想把 code help 接到更大模型堆疊的團隊。\u003Ca href=\"https:\u002F\u002Fwww.langchain.com\" target=\"_blank\" rel=\"noopener\">LangChain\u003C\u002Fa> 則常拿來串 agent、工具和 retrieval。\u003C\u002Fp>\u003Cp>真正要比的，不是功能列表。是它能吃掉多少研究流程。只幫一小段，團隊還是要一直切 context。能接住整個 loop，才真的省事。\u003C\u002Fp>\u003Cp>這裡也很容易看出 review 系統的價值。模型可以先寫摘要，但 review \u003Ca href=\"\u002Fnews\u002Fwhy-solanas-developer-surge-matters-more-than-ethereums-lead-zh\">la\u003C\u002Fa>yer 可以去對照 logs、\u003Ca href=\"\u002Ftag\u002Fbenchmark\">benchmark\u003C\u002Fa> 數字、原始假設。這能提早抓出那種「文筆很好，證據很弱」的內容。\u003C\u002Fp>\u003Cul>\u003Cli>\u003Ca href=\"https:\u002F\u002Fwww.cursor.com\" target=\"_blank\" rel=\"noopener\">Cursor\u003C\u002Fa>：適合直接在 codebase 裡做快速修改。\u003C\u002Fli>\u003Cli>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fclaude-code\" target=\"_blank\" rel=\"noopener\">Claude Code\u003C\u002Fa>：適合命令列重度工作。\u003C\u002Fli>\u003Cli>\u003Ca href=\"https:\u002F\u002Fopenai.com\u002Findex\u002Fintroducing-codex\u002F\" target=\"_blank\" rel=\"noopener\">Codex\u003C\u002Fa>：適合接在更完整的模型堆疊裡。\u003C\u002Fli>\u003Cli>\u003Ca href=\"https:\u002F\u002Fwww.langchain.com\" target=\"_blank\" rel=\"noopener\">LangChain\u003C\u002Fa>：適合把 agent 與工具流程化。\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>數字怎麼看，才不會被話術騙\u003C\u002Fh2>\u003Cp>如果原始素材裡有 3 個以上數字，就應該把數字拉出來看。因為研究流程很容易被形容詞帶歪。數字比較誠實。\u003C\u002Fp>\u003Cp>這篇素材裡至少有 4 個明確環節。文獻回顧、程式修改、實驗迴圈、審查系統。每一段都能對應到不同工具，也能對應到不同成本。\u003C\u002Fp>\u003Cp>你如果把這些步驟拆開，就會發現 AI 最有用的地方不是單點能力，而是減少切換。少一次人工搬資料，少一次手動改檔案，少一次重跑測試，時間就省下來了。\u003C\u002Fp>\u003Cp>下面這種比較方式比較務實。\u003C\u002Fp>\u003Cul>\u003Cli>文獻回顧：AI 先摘要，再由人確認論點。\u003C\u002Fli>\u003Cli>程式修改：AI 先提 patch，再由人看 diff。\u003C\u002Fli>\u003Cli>實驗迴圈：AI 幫忙跑測試，人看結果是否可信。\u003C\u002Fli>\u003Cli>審查系統：AI 先做初篩，人做最後裁決。\u003C\u002Fli>\u003C\u002Ful>\u003Cp>如果你只看模型分數，很容易看走眼。因為研究不是單次問答。研究是多輪迭代。每一輪都會產生資料，也會產生誤差。\u003C\u002Fp>\u003Cp>所以我會建議團隊看兩個數字。第一個是每次迭代少花多少分鐘。第二個是錯誤率有沒有下降。這兩個比「模型很會講」有用多了。\u003C\u002Fp>\u003Ch2>這種工作法的背景是什麼\u003C\u002Fh2>\u003Cp>vibe research 其實是更大趨勢的一部分。大家開始把 LLM 當成工作系統，而不是單純問答機器。這在\u003Ca href=\"\u002Ftag\u002F軟體開發\">軟體開發\u003C\u002Fa>圈尤其明顯。\u003C\u002Fp>\u003Cp>原因很簡單。很多工作本來就不是一次完成。它需要查資料、寫程式、跑實驗、看 log、再修正。AI 剛好可以切進這些環節。\u003C\u002Fp>\u003Cp>但這也代表風險變高。流程如果沒設好，錯誤會跑得比人快。尤其是研究場景，錯一個假設，後面全部都會歪掉。\u003C\u002Fp>\u003Cp>所以現在比較成熟的團隊，會把 agent 當成 junior assistant。權限有限。任務有限。輸出也要能被追溯。這樣才不會變成一團黑箱。\u003C\u002Fp>\u003Cp>另一個背景是工具鏈變完整了。現在有模型、IDE、terminal agent、評測框架、版本控制。這些東西串起來後，研究就比較像工程，而不是純手工藝。\u003C\u002Fp>\u003Ch2>接下來該怎麼做\u003C\u002Fh2>\u003Cp>如果你想試 vibe research，先從一個重複任務開始。不要一開始就想把整個研究部門都 AI 化。那通常只會把複雜度拉高。\u003C\u002Fp>\u003Cp>比較好的做法，是先挑一段最常重做的流程。像是文獻摘要、實驗跑批次、結果整理，或是把 code 改成可測試版本。先量時間，再看錯誤率。\u003C\u002Fp>\u003Cp>我的判斷很直接。真正有用的 vibe research，不是讓 AI 看起來很忙。是讓人更容易驗證結果。你如果能讓 agent 改程式、跑 test、再把結果說清楚，這套流程才算站得住腳。\u003C\u002Fp>\u003Cp>接下來最值得做的事，就是把 review 規則寫死。然後看它能不能真的幫你少掉一半的手動檢查。做得到，就留下。做不到，就別硬上。\u003C\u002Fp>","Vibe research 把 LLM、agent、coding 工具和 review loop 串成流程，讓研究從讀文獻到跑實驗都更可執行。","zhuanlan.zhihu.com","https:\u002F\u002Fzhuanlan.zhihu.com\u002Fp\u002F2036384819442137022",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778904653705-zekc.png",[13,14,15,16,17,18,19,20,21,22],"vibe research","AI research workflows","LLM","agent","coding tools","review loop","Claude Code","Cursor","Codex","LangChain","zh",2,false,"2026-05-16T04:10:33.15767+00:00","2026-05-16T04:10:32.919+00:00","done","45e55e46-b8c2-4dcc-ad24-8e9adf5d1e64","vibe-research-ai-tools-workflows-zh","tools","93bb7b5a-144a-4887-8dde-625a400a0432","published","2026-05-16T09:00:17.325+00:00",[36,37,38],"vibe research 是把 LLM、agent、coding 工具和 review loop 串成研究流程。","真正重要的是 workflow，不是單一模型分數。","最實用的切入點，是先從一個重複任務開始量化時間與錯誤率。","c3c88dd2-a940-438a-b359-0e5a24562273","[0.0200243,-0.000205386,0.002665761,-0.099747404,0.00230873,0.006176991,-0.038329024,0.0019338126,0.007092704,-0.0026340997,-0.01507428,-0.027260792,0.0028512308,0.0057126246,0.12948427,0.019153027,0.0047280616,-0.0065038647,0.010714232,-0.009632879,-0.0039317166,0.03076023,0.009673299,-0.015248627,-0.00929295,-0.009169899,0.014178722,0.0040479363,0.035792645,-0.005494363,0.01698718,0.023983533,-0.019318245,0.005620658,-0.0047360356,0.037083175,0.011241093,-0.017086744,0.036489744,0.014475043,-0.0154538145,-0.0099806925,-0.015677746,0.005006128,0.021087743,0.015053056,-0.00044335477,-0.040277116,-0.010251714,0.015232731,-0.028526515,0.03301873,-0.016088154,-0.16191064,0.018032396,0.00039782946,-0.013817344,0.021195782,0.03719547,0.00865268,-0.013551923,-0.01107893,-0.039785817,-0.009229782,-0.009299011,-0.019355614,0.023812601,0.0034541616,0.00082710147,-0.019853013,-0.035728686,0.0140882125,0.010938502,-0.017054852,-0.0046184584,-0.031799227,-0.019873966,0.009612886,0.010192716,0.040633734,0.023177397,-0.021788906,-0.01009822,-0.0030778686,0.0027656525,-0.0050536455,-0.00011040348,-0.0076817386,0.006489966,0.014611221,0.0031878215,0.023245163,0.013290263,-0.0054932022,0.008981347,-0.015565216,-0.0039730505,-0.006231313,-0.009060654,0.008272292,-0.02922062,-0.034618292,0.014849435,0.016827397,0.021335125,0.014512729,-0.0054013096,-0.01598688,0.011737037,0.012393788,-0.011835577,-0.008542566,-0.009243637,-0.0028208883,-0.03175719,-0.1273012,-0.005193066,0.008414047,0.001104045,-0.0038013293,-0.011709076,0.024157535,0.0060776384,0.014191601,0.019760014,-0.0056474293,0.009144622,0.0049280887,-0.019810596,0.010530743,-0.0311466,0.014587736,-0.025097035,0.002937837,-0.0008870536,0.01902562,-0.018059399,0.0099317,0.005391089,-0.026880402,0.001566376,0.031854123,0.004645945,-0.018718513,-0.035980273,-0.013884297,-0.04672793,0.01178929,0.018141296,-0.012311275,0.006499581,-0.008563156,0.0023965023,-0.023219263,0.015959859,-0.018778231,0.0015437279,0.0030757524,-0.00031314368,0.013113073,0.0011657793,0.0074516013,-0.017520303,0.030831305,0.019266235,0.013146383,0.0052536284,0.014187803,0.002235382,0.030277967,0.013752244,-0.007124069,0.0060930336,0.019665826,-0.000102355225,0.021097673,-0.012213427,-0.015941586,0.0015249427,-0.024689192,0.034905788,0.0082959775,-0.014357822,0.017809115,0.00876019,0.008559429,0.010672848,0.01717432,0.022734884,0.032153636,-0.018216526,0.009261924,0.03781598,-0.013122704,-0.0046009906,-0.01712278,0.013654234,-0.0153686,0.0015503992,0.036933426,0.007785645,-0.008089169,0.01710888,-0.02432087,0.0091246925,-0.032967113,-0.013145737,-0.0028073129,0.015568345,0.007223538,0.0040686578,0.01049122,0.028330503,-0.02284312,-0.001356445,-0.014887428,0.0075170742,-0.010058947,0.0055941837,0.0018714294,0.020752791,-0.019437509,0.010109176,0.0098604625,0.014976691,-0.016006473,-0.01722941,-0.0031093734,-0.013773678,-0.010769919,-0.012423749,0.007850574,0.000100864076,0.009085362,0.00076408725,0.012954059,-0.002978952,0.021712514,0.017008968,0.014018739,-0.018410986,-0.008078974,0.008434822,-0.01839252,0.016246933,0.0016600058,0.0025077243,-0.015376061,0.0006387419,-0.008953383,0.0011699426,-0.021636466,0.005651479,0.0070833615,0.004053187,0.0027300513,0.0077472967,0.0036839629,-0.006016044,0.004799095,-0.00891884,0.014353768,-0.02617483,-0.009037941,-0.010776189,0.009141983,0.02890638,0.0177774,0.0054197903,0.014221236,0.009997219,0.048023265,0.0019102149,-0.011812342,0.018439792,-0.011650564,-0.03736338,0.038621824,-0.00039824267,-0.0252952,-0.01586659,0.01913187,0.010636979,0.0155260805,0.00093306415,0.0047404207,-0.008528706,-0.025756981,-0.007846361,-8.4824605e-05,-0.008361436,0.013912518,-0.0016967434,-0.019459201,0.00027054324,-0.04118547,-0.0052500907,0.031445917,-0.003970678,0.027245266,0.00930252,-0.0035721145,0.005843878,0.04212482,0.012126514,0.0013019359,-0.0051427702,0.029135322,-0.010491863,-0.0039042863,-0.023878545,-0.01353078,0.03182979,-0.01122141,-0.0008379127,-0.021012688,0.0022881327,-0.01833291,0.0056790286,-0.022803662,0.023911178,-0.006076203,-0.009084784,-0.011155178,-0.021011598,0.003544395,-0.00090676843,-0.019628808,-0.008287899,-0.0077568395,0.0040823417,0.0060794237,0.020450596,-0.027703375,0.021494705,0.0064013614,-0.032070123,0.013372762,-0.006060997,0.011478754,-0.0055358303,-0.0189167,-0.002385938,0.02477745,-0.0077927317,0.023488361,-0.0076334435,0.010601575,-0.006015453,-0.040732037,0.011971468,0.0068695224,0.0014835773,-0.021799086,-0.04885765,0.023765683,-0.009909166,-0.0046526263,0.0434347,-0.0019108702,0.025257006,-0.016915673,-0.007286465,0.0026517662,0.015789863,-0.033244897,0.0066152634,0.040268406,0.0018758321,0.011463793,0.0064894566,-0.02144374,0.005547467,-0.013852177,-0.009562344,-0.0037073323,-0.012249776,-0.0074396413,-0.023717217,0.0010050263,-0.030170398,0.034439452,-0.037064098,-0.014370678,-0.008724911,0.0070634424,-0.010932513,-0.0005953992,-0.00083281606,-0.0015168847,0.031553935,-0.015636558,-0.031971965,-0.00659761,0.011117652,-0.0148712555,-0.029181536,0.004215452,-0.006677912,0.016992286,0.0062107015,-0.01835547,0.00013848122,0.013047199,0.00050113985,0.008932531,-0.006498793,0.020556694,-0.008878353,-0.029428585,-0.02878944,-0.0115016885,0.010885213,-0.0052120984,0.009490127,0.014812765,0.0023092637,-0.016599791,0.014079215,0.005191338,-0.015251032,-0.007586037,-0.0068396325,0.00038549854,-0.01846957,0.0071573006,-0.005563532,-0.014963301,0.012284866,0.0008145402,0.0032579587,-0.0010361986,-0.013857988,-0.01950963,-0.032573182,0.0051334845,-0.033528138,0.001927109,-0.0029155125,-0.012174101,-0.007251311,-0.01628384,0.01323089,0.01958219,0.021408752,-0.026160745,0.011107893,-0.024201592,-0.03021218,-0.0051429397,0.027605634,0.006834565,0.010485676,-0.019528802,0.008070835,-0.011220238,0.0005803064,0.01963622,-0.016192082,-0.001513125,-0.00040218973,0.0002546502,0.0028792927,-0.00095665257,0.01641257,-0.0069426117,-0.00973927,-0.029862626,-0.007827673,0.017184796,-0.012203675,-0.010281015,0.02037209,0.0043568825,0.0024769597,-0.024463128,-0.005883109,0.00023719665,-0.004117416,0.010199624,-0.005131861,0.005690041,-0.013706375,-0.009822554,0.008995444,0.012180503,-0.020423954,0.012605773,0.010840844,0.006644379,0.009686975,0.02524388,0.0073624104,-0.015154555,-0.01581115,-0.005665198,-0.019953968,-0.0453424,-0.018100206,-0.020933019,0.005975381,-0.029996084,-0.0135411825,-0.020790262,0.014814632,-0.036727704,0.004259921,0.0126865255,-0.009654425,-0.03228228,-0.00052106497,-0.010699363,-0.003203459,0.0037085023,-0.004592926,-0.02848223,0.013983015,-0.0040236264,-0.0010431308,0.033333328,-0.009229732,-0.023822598,0.0035174193,0.014960644,-0.010980805,-0.017416704,0.020098552,-0.005724349,0.037735835,0.0288386,-0.00314144,0.041827906,0.0024578932,-0.0070531946,-0.0034590017,-0.019291297,0.022768157,0.03797745,0.0058497265,0.009107052,-0.0059269047,-0.012787101,-0.01319886,-0.019517766,0.04934943,-0.08946229,0.0093277,0.010175427,-0.013598297,0.020450907,-0.017110419,-0.008795545,-0.016537115,0.009329989,0.008951506,0.010068416,-0.0033962897,0.03814019,0.018893866,-0.0057163443,0.0032893647,-0.011066734,-0.020616297,-0.019595409,-0.021718485,0.005343574,0.021779928,0.026160784,-0.009110108,0.023188068,-0.011870623,0.00531974,-0.0081853345,0.01673885,0.016478758,-0.012434654,-0.010736124,0.015440539,-0.009538756,0.024202745,0.012251584,0.023834666,-0.0040392266,0.020075742,0.0016478291,-0.008657475,0.007852358,-0.03622476,-0.04388183,-0.023142485,0.010509152,-0.009038806,-0.023385085,0.0008245588,0.011805197,-0.044954855,-0.018354759,-0.024391642,-0.034760058,0.010205342,0.031124238,-0.04042769,0.0010039341,-0.032809716,0.015752228,0.002858437,0.00019953113,-0.024392586,0.049523514,-0.014887605,0.019107422,-0.0077175805,0.008594007,-0.0069511253,-0.008481341,0.008040778,0.014833165,0.009391726,-0.01143764,-0.021816555,0.026174903,-0.0061372733,0.017769404,-0.0044421116,0.019608658,-0.031672396,-0.0019272725,-0.06851261,-0.026789183,-0.0066218833,-0.0055003366,0.022573385,-0.0024064423,0.024943575,-0.019637128,0.009161142,0.008034782,0.009975675,0.0026340508,-0.009141723,-0.024052655,0.011677625,0.009784562,-0.038442373,0.0018631966,0.017396677,-0.013622544,-0.035129584,-0.0020127788,0.026087293,-0.040854502,-0.0038634688,0.0057878993,-0.02367159,-0.003975115,-0.010768769,-0.0041874074,-0.0097737415,-0.13482298,0.0045999954,-0.0019746814,-0.008863485,0.019205932,0.012898731,-0.020509915,0.0042117075,-0.002603693,-0.015656443,-0.016262053,-0.028183008,-0.018792093,-0.0050761304,-0.015153113,0.11965435,-0.015549845,-0.00843083,-0.030108636,-0.0051656445,-0.035588566,0.0072981743,-0.019179836,0.016789937,-0.0044070785,-0.0020388502,0.04342953,-0.0056932415,-0.00798259,0.036219317,0.009062297,0.020163612,-0.022267638,-0.0057755946,0.0155153945,-0.03179921,0.008735643,-0.022375517,0.04001666,0.013573905,0.018881189,0.015944075,0.015327635,-0.008871082,0.014635236,0.0074556596,-0.030384857,-0.0069293533,-0.015135059,-0.02717229,-0.00930313,-0.0836733,0.010655367,-0.02837168,-0.0025807086,0.015651962,0.0028488562,-0.002744731,0.05626533,0.012567041,0.009637789,-0.027330771,-0.023557806,0.019447492,-0.020593297,-0.009952083,0.020056168,0.032893874,0.011872991,0.03305689,0.018413046,-0.006998767,0.007084388,0.0020014378,-0.00919509,-0.010541771,-0.0061748745,0.00122656,0.015209134,-0.02276864,-0.018959817,0.020882662,0.023388665,-0.032646228,-0.01595504,0.013372909,0.011875115,-0.011259598,-0.009682281,-0.0075835874,-0.0066545205,0.012654482,0.0043771146,0.016168794,0.013933042,0.021553125,-0.0041109617,0.014962941,-0.0033682224,0.009919156,0.0088425325,-0.03009738,-0.0030202493,-0.011350764,0.006682298,0.0037409693,0.00033876786,0.02030953,0.038131636,0.010462357]",{"tags":42,"relatedLang":52,"relatedPosts":56},[43,44,46,48,50],{"name":16,"slug":16},{"name":17,"slug":45},"coding-tools",{"name":15,"slug":47},"llm",{"name":14,"slug":49},"ai-research-workflows",{"name":13,"slug":51},"vibe-research",{"id":32,"slug":53,"title":54,"language":55},"vibe-research-ai-tools-workflows-en","Vibe Research: AI Tools for Faster Research","en",[57,63,69,75,81,87],{"id":58,"slug":59,"title":60,"cover_image":61,"image_url":61,"created_at":62,"category":31},"cb68bb90-3638-4334-87c7-02580f59877a","aws-repository-wide-security-scanner-matters-zh","為什麼 AWS 的全倉庫安全掃描比更快的 SAST 更重要","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778901047875-n4l9.png","2026-05-16T03:10:24.757504+00:00",{"id":64,"slug":65,"title":66,"cover_image":67,"image_url":67,"created_at":68,"category":31},"7c966206-36f7-4d6b-b2e5-088a4732ede4","why-docker-microvm-sandboxes-ai-agents-zh","為什麼 Docker 的 microVM 沙盒才是 AI 代理的正解","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778893837158-gpxf.png","2026-05-16T01:10:20.668094+00:00",{"id":70,"slug":71,"title":72,"cover_image":73,"image_url":73,"created_at":74,"category":31},"d058a76f-6548-4135-8970-f3a97f255446","why-gemini-api-pricing-is-cheaper-than-it-looks-zh","為什麼 Gemini API 定價其實比看起來更便宜","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778869845081-j4m7.png","2026-05-15T18:30:25.797639+00:00",{"id":76,"slug":77,"title":78,"cover_image":79,"image_url":79,"created_at":80,"category":31},"68e4be16-dc38-4524-a6ea-5ebe22a6c4fb","why-vidhub-huiyuan-hutong-bushi-quan-shebei-tongyong-zh","為什麼 VidHub 會員互通不是「買一次全設備通用」","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778789450987-advz.png","2026-05-14T20:10:24.048988+00:00",{"id":82,"slug":83,"title":84,"cover_image":85,"image_url":85,"created_at":86,"category":31},"7a1e174f-746b-4e82-a0e3-b2475ab39747","why-buns-zig-to-rust-experiment-is-right-zh","為什麼 Bun 的 Zig-to-Rust 實驗是對的","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778767879127-5dna.png","2026-05-14T14:10:26.886397+00:00",{"id":88,"slug":89,"title":90,"cover_image":91,"image_url":91,"created_at":92,"category":31},"e742fc73-5a65-4db3-ad17-88c99262ceb7","why-openai-api-pricing-is-product-strategy-zh","為什麼 OpenAI API 定價是產品策略，不是註腳","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778749859485-chvz.png","2026-05-14T09:10:26.003818+00:00",[94,99,104,109,114,119,124,129,134,139],{"id":95,"slug":96,"title":97,"created_at":98},"de769291-4574-4c46-a76d-772bd99e6ec9","googles-biggest-gemini-launches-in-2026-zh","Google 2026 最大 Gemini 盤點","2026-03-26T07:26:39.21072+00:00",{"id":100,"slug":101,"title":102,"created_at":103},"855cd52f-6fab-46cc-a7c1-42195e8a0de4","surepath-real-time-mcp-policy-controls-zh","SurePath 推出即時 MCP 政策控管","2026-03-26T07:57:40.77233+00:00",{"id":105,"slug":106,"title":107,"created_at":108},"9b19ab54-edef-4dbd-9ce4-a51e4bae4ebb","mcp-in-2026-the-ai-tool-layer-teams-use-zh","2026 年 MCP：團隊真的在用的 AI 工具層","2026-03-26T08:01:46.589694+00:00",{"id":110,"slug":111,"title":112,"created_at":113},"af9c46c3-7a28-410b-9f04-32b3de30a68c","prompting-in-2026-what-actually-works-zh","2026 提示工程，真正有用的是什麼","2026-03-26T08:08:12.453028+00:00",{"id":115,"slug":116,"title":117,"created_at":118},"05553086-6ed0-4758-81fd-6cab24b575e0","garry-tan-open-sources-claude-code-toolkit-zh","Garry Tan 開源 Claude Code 工具包","2026-03-26T08:26:20.068737+00:00",{"id":120,"slug":121,"title":122,"created_at":123},"042a73a2-18a2-433d-9e8f-9802b9559aac","github-ai-projects-to-watch-in-2026-zh","2026 必看 20 個 GitHub AI 專案","2026-03-26T08:28:09.619964+00:00",{"id":125,"slug":126,"title":127,"created_at":128},"a5f94120-ac0d-4483-9a8b-63590071ac6a","claude-code-vs-cursor-2026-zh","Claude Code 與 Cursor 深度對比：202…","2026-03-26T13:27:14.279193+00:00",{"id":130,"slug":131,"title":132,"created_at":133},"0975afa1-e0c7-4130-a20d-d890eaed995e","practical-github-guide-learning-ml-2026-zh","2026 機器學習入門 GitHub 實用指南","2026-03-27T01:16:49.712576+00:00",{"id":135,"slug":136,"title":137,"created_at":138},"bfdb467a-290f-4a80-b3a9-6f081afb6dff","aiml-2026-student-ai-ml-lab-repo-review-zh","AIML-2026：像課綱的學生實驗 Repo","2026-03-27T01:21:51.467798+00:00",{"id":140,"slug":141,"title":142,"created_at":143},"80cabc3e-09fc-4ff5-8f07-b8d68f5ae545","ai-trending-github-repos-and-research-feeds-zh","AI Trending：把 AI 資源收成一張表","2026-03-27T01:31:35.262183+00:00"]