[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-longmemeval-v2-agent-memory-web-workflows-zh":3,"tags-longmemeval-v2-agent-memory-web-workflows-zh":37,"related-lang-longmemeval-v2-agent-memory-web-workflows-zh":48,"related-posts-longmemeval-v2-agent-memory-web-workflows-zh":52,"series-research-cec2d028-df49-4444-a0e2-e857109414bf":89},{"id":4,"title":5,"content":6,"summary":7,"source":8,"source_url":9,"author":10,"image_url":11,"keywords":12,"language":18,"translated_content":10,"views":19,"is_premium":20,"created_at":21,"updated_at":21,"cover_image":11,"published_at":22,"rewrite_status":23,"rewrite_error":10,"rewritten_from_id":24,"slug":25,"category":26,"related_article_id":27,"status":28,"google_indexed_at":29,"x_posted_at":10,"tweet_text":10,"title_rewritten_at":10,"title_original":10,"key_takeaways":30,"topic_cluster_id":34,"embedding":35,"is_canonical_seed":36},"cec2d028-df49-4444-a0e2-e857109414bf","LongMemEval-V2：測 agent 長期記憶","\u003Cp data-speakable=\"summary\">LongMemEval-V2 用 451 題測試 \u003Ca href=\"\u002Ftag\u002Fagent\">agent\u003C\u002Fa> 能否記住 Web 環境經驗，而不只是使用者歷史。\u003C\u002Fp>\u003Cp>對很多 agent 來說，記憶一直是個卡點。問題不只在於要不要記住使用者說過\u003Ca href=\"\u002Fnews\u002Fwhy-anthropic-200b-google-cloud-pledge-changes-ai-race-zh\">什麼\u003C\u002Fa>，而是能不能記住某個 Web 環境的規則、介面變化、常見失誤，還有哪些工作流程真的有用。這篇 \u003Ca href=\"https:\u002F\u002Farxiv.org\u002Fabs\u002F2605.12493\">LongMemEval-V2: Evaluating Long-Term Agent Memory Toward Experienced Colleagues\u003C\u002Fa>，就是要把這件事拿來直接測。\u003C\u002Fp>\u003Cp>它想回答的不是「agent 有沒有記憶功能」，而是更實際的問題：當 agent 在同一個環境裡反覆工作後，記憶系統能不能把那些經驗累積起來，讓它表現得像一個熟悉現場的同事，\u003Ca href=\"\u002Fnews\u002Fwhy-ai-coding-assistants-need-tighter-governance-zh\">而不是\u003C\u002Fa>每次都從零開始。\u003C\u002Fp>\u003Ch2>這篇在解什麼痛點\u003C\u002Fh2>\u003Cp>現有很多 agent 記憶評測，重點都放在使用者歷史、短對話軌跡，或是任務有沒有做完。這種測法有它的價值，但它會漏掉一大塊真實場景裡很重要的東西：環境本身的經驗。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778653249376-2wo2.png\" alt=\"LongMemEval-V2：測 agent 長期記憶\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>在 Web 工作流程裡，很多成功與失敗不是來自單次對話，而是來自你有沒有記住某個系統的靜態狀態、它會怎麼變、哪種流程比較穩、哪些坑很容易重踩。若 \u003Ca href=\"\u002Ftag\u002Fbenchmark\">benchmark\u003C\u002Fa> 沒有把這些納進來，就很容易高估記憶系統的實用性。\u003C\u002Fp>\u003Cp>LongMemEval-V2 的核心想法很直接：記憶不該只是存資料，而是要把反覆接觸環境後累積的經驗內化進去。也就是說，它評的不是背誦能力，而是經驗沉澱能力。\u003C\u002Fp>\u003Ch2>方法怎麼做\u003C\u002Fh2>\u003Cp>LongMemEval-V2 一共收了 451 題手工整理的問題。這些問題被整理成五種 Web agent 的記憶能力：static state recall、dynamic state tracking、workflow knowledge、environment gotc\u003Ca href=\"\u002Fnews\u002Falphagrpo-self-reflective-multimodal-generation-zh\">ha\u003C\u002Fa>s、premise awareness。這五類本身就很能反映作者想測的不是單純問答，而是 agent 對環境的整體理解。\u003C\u002Fp>\u003Cp>每一題都會配上歷史 trajectories，而且這些歷史可以多到 500 條 trajectories、總計 115M tokens。這代表它不是小型回憶題，而是很明顯的\u003Ca href=\"\u002Ftag\u002F長上下文\">長上下文\u003C\u002Fa>記憶測試。對記憶系統來說，重點不是能不能看懂一段歷史，而是能不能在一大堆歷史裡找出有用的那一段。\u003C\u002Fp>\u003Cp>作者把這個流程定義成 context gathering formulation。白話一點，就是記憶系統先讀歷史軌跡，再吐出可以用來回答問題的精簡證據。這樣的設計，測的不只是存不存得住，也測取回來的內容是不是對的、是不是剛好夠用。\u003C\u002Fp>\u003Cp>論文在這個框架下比較了兩種記憶方法：\u003C\u002Fp>\u003Cul>\u003Cli>\u003Cstrong>AgentRunbook-R\u003C\u002Fstrong>：一種效率導向的 RAG 記憶法，分成 raw state observations、events、strategy notes 這幾個知識池\u003C\u002Fli>\u003Cli>\u003Cstrong>AgentRunbook-C\u003C\u002Fstrong>：把 trajectories 存成檔案，再用 coding agent 在增強 sandbox 裡蒐集證據\u003C\u002Fli>\u003C\u002Ful>\u003Cp>這個對比很有意思。它不是只在比兩個模型誰比較會答題，而是在比兩種完全不同的記憶工作流：一種偏向檢索，一種偏向主動蒐證。這對實作端很重要，因為很多團隊現在都在想，單靠 \u003Ca href=\"\u002Ftag\u002Frag\">RAG\u003C\u002Fa> 式記憶到底夠不夠，還是要讓 agent 自己去整理、推敲、抓證據。\u003C\u002Fp>\u003Ch2>結果證明了什麼\u003C\u002Fh2>\u003Cp>摘要裡最明確的數字，是 AgentRunbook-C 的平均準確率達到 72.5%。這個成績高於最強的 RAG baseline，後者是 48.5%，也高於 off-the-shelf coding agent baseline 的 69.3%。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778653264341-yzk7.png\" alt=\"LongMemEval-V2：測 agent 長期記憶\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>這組結果透露出一個很清楚的訊號：單純把資訊檢索出來，和用 coding agent 去歷史軌跡裡蒐證，能力差距不小。換句話說，對長期 agent 記憶來說，「能找到資料」不等於「能把經驗用對」。\u003C\u002Fp>\u003Cp>論文也提到，AgentRunbook-C 在 accuracy-latency Pareto frontier 上有進步。這句話的意思很務實：它在準確率上更好，但不是沒有代價。作者同時明講，基於 coding agent 的方法有很高的延遲成本。也就是說，效果更好，不代表部署起來就一定輕鬆。\u003C\u002Fp>\u003Cp>摘要沒有公開更細的 benchmark breakdown，因此看不到五種記憶能力各自的表現差異。就目前公開資訊來看，我們只能確認整體平均準確率與方法間的落差，還不能進一步判斷哪一類記憶最難、哪一類最容易。\u003C\u002Fp>\u003Cp>不過，從作者的描述可以看出，LME-V2 被設計成一個有挑戰性的測試。就算最強方法拿到目前的最佳成績，也仍然留有不少改善空間。這代表長期 agent 記憶還不是一個已經被解完的題目。\u003C\u002Fp>\u003Ch2>對開發者有什麼影響\u003C\u002Fh2>\u003Cp>如果你在做的是專門跑 Web 工作流程的 agent，這篇論文很值得放進你的評估清單。它提醒我們，記憶不是單純的資料庫問題，而是「經驗管理」問題。agent 要記的不只是事實，還包括環境狀態、流程習慣、常見陷阱，以及哪些前提在這個系統裡成立。\u003C\u002Fp>\u003Cp>這對很多架構都會有影響。像是依賴 retrieval 的 agent、用 runbook 管理操作步驟的系統，或是要長期維持操作知識的工作流，都會碰到同一個問題：如果只記住使用者，卻沒記住環境，錯誤還是會一再重演。\u003C\u002Fp>\u003Cp>LongMemEval-V2 提供了一個比較像真實世界的壓力測試。它逼你回答一個很工程化的問題：你的記憶堆疊，到底只是存檔，還是真的把環境經驗吸收進去。對台灣開發團隊來說，這種測法特別適合拿來檢查內部 agent 是否真的能在固定系統裡越做越順，而不是每次都像第一次上線。\u003C\u002Fp>\u003Cp>但這篇也有明顯限制。摘要沒有說這些方法能不能泛化到 benchmark 以外的環境，也沒有交代不同環境設定下的敏感度。它也沒有告訴我們，五種記憶能力裡哪一種最難，或是 coding-agent 式蒐證在 production 的嚴格延遲要求下是否仍然可行。\u003C\u002Fp>\u003Cp>所以，這篇論文更像是一個方向明確的壓力測試，而不是一個已經定案的最佳解。它把討論重心從「agent 會不會記得」往前推了一步，變成「agent 能不能真的學會這個環境，並把學到的東西用在下一次任務裡」。\u003C\u002Fp>\u003Cp>對現在正在做 agent 的團隊來說，這個問題幾乎就是實戰核心。因為真正有價值的記憶，不是把歷史堆起來，而是讓系統在下一次遇到同樣情境時，少犯一次同樣的錯。\u003C\u002Fp>","LongMemEval-V2 用 451 題測試 agent 能否記住 Web 環境經驗，而不只是使用者歷史；結果顯示以 coding agent 蒐證的記憶法準確率最高，但延遲也更高。","arxiv.org","https:\u002F\u002Farxiv.org\u002Fabs\u002F2605.12493",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778653249376-2wo2.png",[13,14,15,16,17],"agent memory","web workflows","RAG","coding agent","long-context evaluation","zh",2,false,"2026-05-13T06:20:29.320872+00:00","2026-05-13T06:20:29.298+00:00","done","8343003c-562a-4a51-b1d5-7902c7450d63","longmemeval-v2-agent-memory-web-workflows-zh","research","442f0ac0-6fd2-460b-83ab-694f0627d98f","published","2026-05-13T09:00:10.253+00:00",[31,32,33],"LongMemEval-V2 不是測使用者記憶，而是測 agent 能否記住 Web 環境經驗。","451 題、最高 500 條 trajectories、115M tokens，讓它成為明顯的長上下文記憶測試。","AgentRunbook-C 以 72.5% 平均準確率勝過 RAG baseline 的 48.5%，但延遲成本也更高。","0c35a120-52fc-41fc-afa3-d404eb934158","[-0.031465575,-0.0061267028,0.022560608,-0.08633812,-0.013334848,-0.02703589,-0.000883767,0.007478521,0.010367735,0.019586608,-0.014242423,-0.010044369,0.015276049,-0.0048055714,0.13626471,0.026063168,-0.00528372,0.017888416,-0.00062759314,-0.038838957,0.014785183,0.010959735,0.0073959003,-0.014521127,-0.010143107,0.010581779,0.017097628,-0.0029981446,0.07640145,0.020384932,-0.011786511,0.012334363,0.0013607568,0.013243314,0.014545275,0.020844342,0.024818758,0.013292475,-0.00090883014,0.006289564,0.0032492536,-0.037197705,-0.008933114,-0.0006316249,-0.017832316,0.024818188,-0.019693978,-0.020613642,-0.020959685,0.026569445,-0.012639221,0.035377968,0.01782444,-0.1570906,0.010505206,-0.0107782865,-0.010612242,0.01423928,-0.0067397333,-0.018114762,-0.01211818,0.0014296325,-0.05332527,-0.034904998,0.0120431315,-0.006828899,0.028816031,0.005782249,-0.023326056,-0.013604545,-0.035407193,-0.010605209,-0.011512695,-0.03843777,0.0015237632,-0.04271829,-0.008288296,0.004134036,0.0030224163,0.022609014,-0.0029060203,-0.02372918,0.012990275,0.01634672,0.002381179,0.01066687,-0.004024056,-0.013565879,0.0074615157,0.008691165,0.00041379008,0.020106036,0.0036632745,-0.0058681197,-0.0092664445,-0.017529273,-0.016063277,0.019710561,-0.012986298,0.013047803,-0.002593929,-0.026059102,-0.004410138,0.016698696,0.008983584,-0.022074772,-0.0056915176,0.02155722,0.0019469867,0.016085902,0.0010206325,-0.013136605,0.005837983,-0.008235594,-0.03566446,-0.13089123,-0.0018162286,-0.012774196,-0.0025844304,-0.0026735188,-0.03103068,0.0350918,-0.02484315,0.024906129,-0.00488395,-0.0080345115,0.022647696,-0.023493102,0.011907885,0.028018437,-0.02406126,0.009573918,-0.0074753123,0.0039450815,-0.023254035,-0.0050571547,0.009167955,-0.005004272,-0.0050623016,-0.019412527,0.0012319465,0.035053562,0.019421699,-0.011205055,-0.013975806,-0.027175944,-0.042901482,0.0010838975,0.009134351,0.005572876,0.0136717195,-0.007886149,-0.0013278938,0.0054147933,0.026857318,-0.02331282,0.023086552,0.0215246,0.006258548,0.029937172,0.013246746,0.0063516987,0.015558477,-0.0011438841,0.010992059,-0.0024470584,-0.0029538847,-0.029592002,-0.012662517,0.02320105,-0.006821256,-0.012169806,0.0018505403,-0.010732112,0.021930035,-0.011243748,0.00475562,0.0011212375,0.007970871,-0.0037390545,0.0009922424,-0.0055521526,-0.025691524,0.010969386,-0.019321058,-0.0009979381,-0.009224187,0.036843184,0.031807132,0.017283564,-0.033368435,-0.0086015165,0.018310705,-0.010223501,-0.008519859,0.0026364145,0.002128196,0.020705843,0.0011045593,0.0024034083,0.016665382,-0.003415797,0.023777261,-0.016810847,0.0025813256,0.00887825,0.0013491348,-0.004331939,0.02003381,0.0024194005,-0.023874486,0.021885503,0.008884074,0.0038312774,-0.017610578,-0.0107626105,-0.018249637,-0.007466217,0.0067845043,-0.008028159,0.00745132,0.0011526776,0.026583022,0.00019292734,0.0094709685,-0.00037587667,-0.025777977,-0.018181337,-0.0063947644,0.022745632,-0.0014690852,0.014527899,-0.029699005,0.027473366,0.021404516,-0.0093941605,-0.01656962,0.009859722,0.027400278,0.011907074,-0.0017897018,0.016757943,-0.01984762,7.605836e-05,0.026034657,-0.01774974,0.0013103568,0.0016827798,-0.01779218,0.014650391,-0.010229924,0.0017536378,-0.005618147,-0.022238981,0.004481166,-0.008772359,-0.007383882,0.018631522,-0.013266445,-0.010499022,0.0076834233,0.019381925,-0.011954769,-0.020501086,-0.0029060917,0.0064227334,0.024816819,0.016908333,-0.042943265,-0.0070665446,-0.019366613,0.012043489,0.019861113,0.005952513,0.026544582,-0.009009261,-0.045764126,-0.004320317,0.007830094,0.013152168,0.012390471,-0.0021035015,0.0007408681,0.0014061993,-0.02472937,0.009671373,-0.006518544,-0.02413042,-0.017720092,-0.018768134,0.0021325024,0.005999809,0.0077532986,0.0011183163,-0.0403847,-0.012817717,0.0045184847,-0.0019638431,0.016743094,-0.002060328,0.021157758,0.010975613,0.005046151,0.032032106,-0.015973708,0.0019003489,0.0068506934,0.026897091,-0.021003159,-0.015480235,-0.0086861625,0.0030378127,-0.005722694,-0.01906275,0.0064650956,-0.023586769,-0.015241393,-0.00953688,-0.0165141,-0.014703925,0.019766727,-0.018767228,-0.005031043,-0.00088651775,-0.014061706,-0.0015150058,0.00065704016,0.016766774,0.027295657,-0.012929601,-0.0007910688,0.020863354,0.05024064,-0.015151324,0.021520535,0.009580513,0.0046202,-0.023830874,-0.003286139,0.015870351,-0.019291228,-0.008618437,-0.017857589,0.01938403,-0.003405475,0.06259415,0.01850357,0.019218015,-0.029370917,-0.017850386,0.009489077,-0.0028123953,0.006339617,-0.016300632,-0.018903006,0.008107954,-0.030133016,-0.004855244,0.025891287,0.033509716,-0.020860445,-0.0011883215,-0.0029298523,0.017292678,0.011173024,-0.02152392,0.019400334,0.003420113,-0.02151671,0.0118892025,-0.0058114077,0.0018476088,0.0087199155,0.000685539,0.018101897,0.006574502,-0.0048188474,0.014786757,0.0067526866,-0.008607417,-0.01122243,0.022908987,-0.026870897,-0.012040793,-0.024690187,0.032355197,0.027806805,0.035980523,0.012870228,0.0146066705,0.01490799,0.028218022,-0.03380896,0.01011941,0.018545147,-0.016341181,0.021240152,0.023471573,0.004568289,-0.013048099,0.0053351517,0.014854468,-0.020280275,-0.003239542,0.010405223,0.027656518,0.015665093,0.0049848254,0.005701249,-0.0035586227,0.010429213,0.020512834,0.022136085,-0.015947897,0.009822633,0.021910852,0.004486949,-0.00797787,0.0016547797,0.0033079414,-0.024735749,0.033388857,-0.0030504947,-0.003965077,-0.02654112,0.0044517983,-0.017698435,-0.010912422,-0.0071170893,-0.022181736,0.013816914,0.015641324,-0.03375633,-0.01572862,-0.019911822,-0.025846988,-0.009233381,6.127371e-05,-0.027917752,-0.014227336,-0.0020350225,-0.0080357455,-0.0031623496,0.0134341605,0.009019851,0.008464212,0.0011077257,-0.018024432,-0.015894769,0.00880083,0.01542105,0.015889676,-0.0019843257,0.011556698,-0.009511508,-0.01872608,-0.013397075,0.00051557034,-0.019294517,-0.008163661,0.017370002,-0.0014700314,-0.029035574,0.023977052,-0.0012751492,-0.0038384644,-0.0007869243,0.0025267424,-0.0061421646,0.012592226,-0.020838784,0.022554168,0.009228641,0.018214023,-0.011060673,-0.00020439259,-0.026084213,0.014549089,0.018165939,-0.00070714304,-0.007167614,0.0029131481,0.013094008,-0.03480772,-0.00690262,0.029219244,-0.0102338465,0.020351518,0.007767512,0.014551164,0.023095183,0.029106371,0.009349783,0.003259808,-0.01847508,-0.03113237,-0.014319237,-0.0125246225,0.010656731,-0.0008630011,-0.00045946584,-0.011898574,0.005011106,0.006043243,-0.0100050885,0.014577955,0.011222515,0.012845374,0.0071183713,-0.0007599743,-0.016630862,0.01586573,0.027060816,0.024868485,-0.0028126547,-0.032673597,0.012464436,0.011743483,0.0023603232,-0.0053972336,-0.0055899774,0.0077394187,-0.01722325,0.023399169,-0.02501325,-0.016370554,0.017413115,0.010309472,0.023181135,0.051580742,0.00187723,0.022830551,0.022682559,-0.039708707,0.003692815,-0.009423229,0.0419353,0.0014497258,0.00073096954,0.042849917,-0.037128925,0.014488853,-0.029400425,-0.013822371,0.018134668,-0.09682665,-0.010383814,0.0022763435,-0.005297374,-0.027485223,0.00035811355,0.002677249,-0.025807392,-0.0017515478,0.014096549,-0.0048957216,-0.030488094,0.0014096741,0.0075746826,-0.037423935,-0.01453523,-0.011054705,0.008444233,0.024566356,-0.015091696,0.038090453,-0.0047832895,0.011098061,0.014894123,-0.004582673,-0.018359674,0.011715178,0.006604258,0.020160781,-0.019372538,-0.018128222,0.0017692221,-0.009327394,0.025503976,-0.0009841492,-0.0045835413,0.045128666,-0.01171281,0.018869009,-0.0033797391,-0.0012378092,0.011987313,-0.01762148,-0.027938567,-0.017513383,0.017408585,0.003060892,0.007507472,3.2722895e-05,0.011450349,-0.04259885,-0.0041991933,0.011458951,-0.0014674986,-0.009364695,-0.012703702,-0.0245741,-0.011167374,-0.012802485,-0.003470794,-0.0034864817,-0.0032279731,-0.0016945256,0.005237695,0.007751642,0.020302156,0.018632034,0.02657093,-0.006020584,-0.000933812,-0.018589653,-0.0059321052,0.0065528625,0.024954285,-0.010903096,0.0001353022,0.009044297,0.02014788,0.015319147,0.01681896,-0.023226274,-0.020130532,-0.08174908,-0.015024801,0.024368463,-0.020381263,-0.0018021457,-0.017869702,0.0130918035,-0.011981513,-0.0013605098,-0.013631157,0.0061434517,0.00792158,-0.013168382,-0.019530399,0.016355611,0.0019778474,0.0034532293,0.013410474,0.008317028,-0.04536029,-0.0015628149,0.0021468136,0.023426993,-0.043362077,-0.0061619338,0.0008160051,-0.017079156,0.014682798,0.013733139,-0.0036631562,-0.009620693,-0.12882468,-0.0027200554,-0.0025480823,0.0021731555,-0.004435033,-0.016783308,0.0165907,0.019006984,0.019164529,0.008659919,-0.007906972,-0.04242604,-0.009865712,-0.008665839,-0.022479799,0.11288297,-0.00032734885,0.010519484,-0.028171131,-0.033800464,0.009597006,-0.05005508,-0.0075933044,-0.001749772,-0.0012406241,0.0023904457,0.0066303797,-0.012903183,-0.006967429,0.020486431,0.031378474,0.025394665,-0.00063137425,-0.015278425,0.00843253,0.009535538,0.0024616248,-0.027405206,0.0072462545,-0.009024964,0.017343726,0.01894076,0.0063366704,-0.02838586,-0.012311517,-0.004372163,-0.012103585,-0.028736046,0.017513007,0.0037701884,-0.027023401,-0.057218798,-0.014310859,-0.0091510955,-0.003151394,-0.00293368,0.02739359,0.0064936085,0.002223923,0.00626942,0.01899661,-0.013586412,-0.0018253205,0.02393989,-0.016615458,-0.0049063773,0.0035621913,0.024833173,0.0021673285,0.015000314,-0.0041040448,0.007972228,0.0009827523,-0.021913514,0.007950766,0.0026109186,0.005891244,0.022704856,0.021188831,-0.011688644,0.02010527,0.007847205,0.021180699,0.0019231561,0.02089034,-0.011395623,-0.012393319,-0.017579308,0.012160618,-0.015799504,-0.018060502,0.020411227,0.0030398834,0.0107274335,0.020018471,0.054454364,-0.00827667,-0.014793593,0.01846786,-0.003637095,-0.027339144,-0.021241097,0.008237945,-0.018402837,0.005622118,0.036847476,0.022224158,-0.0040623522,0.021617632,-0.010637659]",true,[38,40,42,44,46],{"name":15,"slug":39},"rag",{"name":13,"slug":41},"agent-memory",{"name":16,"slug":43},"coding-agent",{"name":14,"slug":45},"web-workflows",{"name":17,"slug":47},"long-context-evaluation",{"id":27,"slug":49,"title":50,"language":51},"longmemeval-v2-agent-memory-web-workflows-en","LongMemEval-V2 tests agent memory in web workflows","en",[53,59,65,71,77,83],{"id":54,"slug":55,"title":56,"cover_image":57,"image_url":57,"created_at":58,"category":26},"667b72b6-e821-4d68-80a1-e03340bc85f1","turboquant-seo-shift-small-sites-zh","TurboQuant 與小站 SEO 變化","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778840440690-kcw9.png","2026-05-15T10:20:27.319472+00:00",{"id":60,"slug":61,"title":62,"cover_image":63,"image_url":63,"created_at":64,"category":26},"381fb6c6-6da7-4444-831f-8c5eed8d685c","turboquant-vllm-comparison-fp8-kv-cache-zh","TurboQuant 與 FP8 實測結果","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778839867551-4v9g.png","2026-05-15T10:10:36.034569+00:00",{"id":66,"slug":67,"title":68,"cover_image":69,"image_url":69,"created_at":70,"category":26},"c15f45ee-a548-4dbf-8152-91de159c1a11","llmbda-calculus-agent-safety-rules-zh","LLMbda 演算替 AI 代理人立安全規則","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778825503412-mlbf.png","2026-05-15T06:10:34.832664+00:00",{"id":72,"slug":73,"title":74,"cover_image":75,"image_url":75,"created_at":76,"category":26},"0c02225c-d6ff-44f8-bc92-884c8921c4a3","low-complexity-beamspace-denoiser-mmwave-mimo-zh","更簡單的毫米波波束域去噪器","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778814650361-xtc2.png","2026-05-15T03:10:30.06639+00:00",{"id":78,"slug":79,"title":80,"cover_image":81,"image_url":81,"created_at":82,"category":26},"9d27f967-62cc-433f-8cdb-9300937ade13","ai-benchmark-wins-cyber-scare-defenders-zh","為什麼 AI 基準賽在資安領域的勝利，應該讓防守方警醒","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778807450006-nofx.png","2026-05-15T01:10:29.379041+00:00",{"id":84,"slug":85,"title":86,"cover_image":87,"image_url":87,"created_at":88,"category":26},"bc402dc6-5da6-46fc-9d66-d09cb215f72b","why-linux-security-needs-patch-wave-mindset-zh","為什麼 Linux 安全需要「補丁浪潮」思維","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778741449813-s2wn.png","2026-05-14T06:50:24.052583+00:00",[90,95,100,105,110,115,120,125,130,135],{"id":91,"slug":92,"title":93,"created_at":94},"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":96,"slug":97,"title":98,"created_at":99},"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":101,"slug":102,"title":103,"created_at":104},"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":106,"slug":107,"title":108,"created_at":109},"9f50561b-aebd-46ba-94a8-363198aa7091","openclaw-agents-manipulated-self-sabotage-zh","OpenClaw Agent 會自己搞砸自己","2026-03-28T03:03:18.786425+00:00",{"id":111,"slug":112,"title":113,"created_at":114},"11f22e92-7066-4978-a544-31f5f2156ec6","vega-learning-to-drive-with-natural-language-instructions-zh","Vega：使用自然語言指示進行自駕車控制","2026-03-28T14:54:04.847912+00:00",{"id":116,"slug":117,"title":118,"created_at":119},"a4c7cfec-8d0e-4fec-93cf-1b9699a530b8","drive-my-way-en-zh","Drive My Way：個性化自駕車風格的實現","2026-03-28T14:54:26.207495+00:00",{"id":121,"slug":122,"title":123,"created_at":124},"dec02f89-fd39-41ba-8e4d-11ede93a536d","training-knowledge-bases-with-writeback-rag-zh","用 WriteBack-RAG 強化知識庫提升檢索效能","2026-03-28T14:54:45.775606+00:00",{"id":126,"slug":127,"title":128,"created_at":129},"3886be5c-a137-40cc-b9e2-0bf18430c002","packforcing-efficient-long-video-generation-method-zh","PackForcing：短影片訓練也能生成長影片","2026-03-28T14:55:02.688141+00:00",{"id":131,"slug":132,"title":133,"created_at":134},"72b90667-d930-4cc9-8ced-aaa0f8968d44","pixelsmile-toward-fine-grained-facial-expression-editing-zh","PixelSmile：提升精細臉部表情編輯的新方法","2026-03-28T14:55:20.678181+00:00",{"id":136,"slug":137,"title":138,"created_at":139},"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"]