{ "text": "嗨各位\n今天我們來聊聊最近 AI 圈的\n當紅炸子雞，Jev\n它也是一個 AI 模型，但特別的是\n它不跟你聊天\n不幫你寫文章，也不會寫程式\n你丟問題給它，它甚至連一個字都不會回你\n因為它只專注做一件事，就是做決定\nJev 在九月十五號發布\n創辦人 Diogo Almeida\n是當年一起做出 ChatGPT 的人之一\n他發的那則 X 貼文\n到現在已經接近三千萬次觀看\n當天在工程師論壇 Hacker News 上拿到 1679 分\n底下四百多則留言\n隔天 Vercel 也直接把 Jev 上架到自己的平台\n讓用戶可以使用\n根據官方公布的數據\nJev 跟現在主流的大語言模型相比\n速度快了 20 到 200 倍\n價格更是便宜了 40 到 400 倍\n最扯的是它的輸出完全免費\n意思就是\n不管你叫它吐出一百個字還是一百萬個字\n它通通一毛錢都不會跟你收\n我認為這款模型\n將會對 AI Automation 帶來很大的影響\n原因很簡單\n現在大家用 AI 搭出來的自動化流程裡\n有一大堆步驟其實只是在做判斷\n比方說，這封信要不要回？\n這張客服單該轉給哪個部門？\n這則留言是不是詐騙廣告？\n這些問題的答案通常只是一個選項\n但我們現在卻都習慣\n請又貴又慢的大模型來處理\n這其實非常浪費時間和金錢\n而 Jev 想要搶佔的領地就是這一塊\n它對標的不是 Fable、Astra 這類幾乎全能的大模型\nAstra 這類幾乎全能的大模型\n你也不會拿它來幫你開發產品\n它最適合的位子\n是放在你的工作流裡面當一個決策節點\n主打的優勢就是快，便宜，以及極低的幻覺\n今天這支影片\n我會先跟你分享 Jev 到底是什麼\n它跟傳統大模型應該怎麼分工，接著\n我們會來看幾個社群上非常有趣的 Demo\n然後，我會用一個最近剛接手的實際案例\n帶你看看這個模型\n可以怎麼被完美整合進工作流裡面\n那我們就直接開始吧！\n先來聊聊它的來歷\nJev 是由一家叫做 TypeSafe AI 的公司開發的\n這家公司過去一直非常低調，沒有對外曝光\n而 Jev 就是他們公開亮相的第一款模型\n前面有提到\n創辦人 Diogo Almeida\n是當年參與打造 ChatGPT 的團隊成員\n講得更精準一點\nChatGPT 之所以聽得懂人話\n會跟你一來一往的聊天\n背後靠的是一套叫做 RLHF 的訓練方法\n而他正是這套方法的共同發明人\n他自己也分享過，在做完 ChatGPT 之後\n他花了兩年的時間\n默默研究了一套全新的訓練方法\n最後生出來的成果就是 Jev\n一個完全不會跟你聊天的模型\n有趣的是\nJev 這個名字取自經濟學中的傑文斯悖論\n英文是 Jevons Paradox\n意思就是當一樣東西變得越便宜\n效率越高時，大家反而會用得越多\n它的運作方式\n跟我們熟悉的大模型完全不一樣\n當你問 ChatGPT 或 Claude 問題時\n它是一個字一個字把答案生出來的\n想得越多，算得越久，你就得等越久\n但 Jev 不會自己寫答案\n你只需要給它三樣東西：一份資料、一個問題，以及你事先列好的幾個選項\n一個問題，以及你事先列好的幾個選項\n它就會直接回傳給你每個選項的機率\n舉個簡單的例子，你丟一張發票給它\n問它這張發票有沒有問題？\n並給它三個選項正常\n詐騙，需要人工檢查\n它就會秒回你這張發票正常的機率是 88%\n剩下的機率則分配給另外兩個選項\n而這整個判斷過程\n通常花不到 0.3 秒就能完成\n如果你聽到這邊還是一頭霧水\n那我用個生活化一點的比喻\n一般的大模型就像是在寫申論題\n而 Jev 則像是在寫選擇題\n申論題得一句一句慢慢寫，不但寫得慢\n還有可能不小心離題\n但選擇題只要圈一個答案就好\n所以速度極快，而且答案絕對不會發散\n一定只會是你給的選項其中之一\n這也是為什麼 Jev 能夠做到低幻覺\n因為它沒有機會\n去瞎掰一個選項以外的答案出來\n目前 Jev 能夠回答的題型，主要分為三種：是非題、單選題，以及打分題\n主要分為三種是非題\n單選題，以及打分題\n是非題很好理解，你問它一句話是不是真的\n它會回傳一個 0 到 1 之間的機率\n越接近 1 就代表它越肯定是\n比方說，你可以丟一張發票給 Jev\n問它這張發票是不是詐騙？\nJev 可能會回你 90%\n你甚至可以把幾個常見的詐騙特徵\n寫進條件裡再問一次\n機率或許就會變得更高\n因為 Jev 有了更多的參考依據\n來增加它的信心水準\n拿到這個數字之後\n最簡單暴力的應用方式就是\n在自動化退款的程式裡設定一個門檻\n比方說只要詐騙機率超過九成\n就直接自動拒絕\n其餘的退款請求\n再自動寄信轉交給真人客服處理\n這樣一來\n你就幫客服省下了大量處理送分題的時間\n第二種是單選題，你可以先列出好幾個選項\nJev 會從裡面挑出一個最合適的\n同時告訴你每個選項的機率各是多少\n選項的內容完全由你定義\n最多可以列到 255 個\n而且每個選項你還可以搭配一句簡單的說明\n讓它更清楚這個選項的定義\n比方說，你可以拿它來自動整理信箱\n讓 Jev 幫你精準判斷\n每一封信應該歸類到哪個資料夾\n第三種是打分題\n有些事情不是非黑即白地選一個類別\n而是要看程度\n這時候你可以先定好一把尺\n或者說畫一個光譜\n從低到高寫出每一個級距代表什麼意思\n最多可以分到 11 級\nJev 就會告訴你\n現在的狀況落在那把尺的哪個位置\n比方說你想幫業務篩選潛在客戶\n你可以設定 0 分是沒有預算的學生\n1 分是小團隊\n2 分是有明確需求跟時程的公司\n3 分則是預算充足的大企業\n這時候如果有一間大企業\n透過網站表單來詢問\nJev 可能就會幫它打出 2.9 分的高分\n拿到這個分數後\n你就可以把它串接到後面的自動化規則\n比方說設定 2.5 分以上的客戶\n要觸發緊急通知給業務主管\n確保團隊當天就能立刻回信搶下訂單\n所以 Jev 的能力跟定位其實非常明確\n它能回答的就是是不是\n哪一個，到什麼程度這三種問題\n而如果你仔細拆解\n現在各種自動化工作流裡的判斷邏輯\n其實大部分也離不開這三種\n整個模型用起來的感覺就是資料丟進去\n它不跟你廢話\n瞬間給你結論\n像是一個小而美的自動化開關\n在價格方面，Jev 也非常有競爭力\n就像前面提到的，它的輸出是完全免費的\n計費方式只看讀進去的文字量\n每一百萬個 Token 只收 0.042 美元\n你可能對一百萬個 Token 沒什麼概念\n最簡單的理解方式就是\n這大約等於好幾本厚厚的小說的文字量\n從這裡就可以看出\nJev 真的是便宜到不可思議\n不過話說回來，不管它再怎麼便宜或快速\n它都還是無法取代\nFable 跟 Astra 這種全能型模型\n或者說它打從一開始就沒打算要取代它們\n因為這兩者的定位根本完全不同\nTypeSafe 幫 Jev 創造了一個新的分類\n叫做 System 1 模型\n這個詞出自於經典名著快思慢想\n書裡提到\n人的大腦其實有兩套系統系統一靠的是直覺\n反應極快，系統二靠的是推理\n速度慢但是想得深\n就像你剛學開車的時候，看後照鏡\n打方向燈，換檔，每一個動作都要用力去想\n這就是系統二，但開了幾年車之後\n這些動作全部變成了下意識的反射動作\n這就是系統一\nFable 跟 Astra 就是那個會深思熟慮的系統二\n而 Jev 就是那個靠直覺秒回的系統一\n所以，這兩者的分工非常清楚\n如果你今天需要 AI 幫你構思\n幫你寫作\n那就交給大模型\n但如果你今天只是需要一個快速\n準確的判斷，那就交給 Jev\n對 Jev 有了基本的認知後\n我們來看看社群上的創作者們\n都拿它來做什麼吧\n國外 YouTuber Matthew Berman\n用它做了一個虛擬小鎮\n裡面住了 50 個 AI 居民\n當他對全鎮廣播一句麵包店大拍賣！\n時，短短 0.6 秒之內\n這 50 個居民就各自決定好自己要怎麼反應了\n其中 39 個不為所動繼續做自己的事\n6 個跑去湊熱鬧看看\n4 個直接加入搶購行列\n還有 1 個熱心地跑去通知別人\n我覺得這個實驗非常有趣\n未來我們絕對會在遊戲領域\n看到大量類似的應用\n比方說遊戲裡的 NPC 行為\n或許就能透過這種方式變得更加擬真\n多樣化，而且運算成本還極低\n第二個例子\n是 AI ML API 這個團隊\n在 X 上發布的一場對弈實驗\n他們讓 Jev 分別跟 Fable 和 Astra 下西洋棋\n規則是每方只有五分鐘時間的快棋\n每下一步棋，就等於呼叫一次模型\n有趣的是，在對上 Fable 的那一局\n如果單論棋力\nFable 絕對是完全碾壓\n下到第 29 步時已經佔盡優勢\n甚至還多吃了一隻皇后，但致命傷是\nFable 每想一步棋都要花上 6 到 15 秒\n而 Jev 每一步只花 2.6 秒\n最後的結果是\nFable 把五分鐘的時間全耗光了\n因為超時而被判輸\n可以說，最聰明的模型，最後卻輸在想太久\n不過，當 Jev 對上 Astra 時情況就不太一樣了\nAstra 只用了短短 18 步就直接把 Jev 將軍死棋\n是真的有點瘋狂\n除此之外\n社群上也有很多人\n開始把 Jev 拿來做大量分類的任務\n比方說，有人拿了 1000 封 Email\n來進行壓力測試\n讓模型去判斷每一封信是不是發票\n是不是合作邀約等等\n結果 Jev 跑完這 1000 封信只花了短短 70 秒\n成本更是只要驚人的 9 美分\n不到台幣三塊錢\n但如果把同一批信交給 GPT-4o 來跑\n不僅要花上 5 分鐘\n成本還高達 62 美分，要台幣二十幾塊錢\nOpenCode 創始團隊的 Ryan Vogel\n也做了一個類似的專案\n他一口氣把自己信箱裡的 1700 封信\n全丟給 Jev ，讓它幫忙分門別類\n排定處理優先順序，打上垃圾信件的分數\n最後再判斷這封信需不需要親自回覆\n全部 1700 封信跑完，總共竟然只花了 18 美分\n另外， Jev 也非常適合拿來做路由\n也就是幫你決定這件事應該交給誰來做\nYouTuber Riley Brown 就實作了一個模型路由器\n當你的問題丟進去後\nJev 自己不負責回答\n它只負責快速判斷\n這道題該派給哪個模型處理\n如果是改錯字這種小事\n就丟給最便宜的模型\n如果是要寫一整套複雜的程式\n再轉發給最貴，最聰明的模型\nTypeSafe 官方也釋出了一個有趣的數據\n如果讓 Agent 自己挑要觸發哪個 skill\n拿錯的機率高達 17%\n但如果先讓 Jev 幫忙快速過濾挑選\n出錯的機率瞬間降到了只有 7.3%\n還有一位叫做 kitze 的開發者\n用它做了一款\n名為 Unclutter 的免費瀏覽器外掛\n當你網頁一打開，它就會在背景呼叫 Jev\n即時判斷網頁上的哪些區塊是廣告\n哪些是沒有營養的垃圾內容\n然後瞬間幫你清理得乾乾淨淨\n還給你一個清爽的閱讀版面\n我認為 Jev 對大多數人來說\n最實用的場景絕對是分類和路由\n因為只要你想做自動化流程\n這兩個環節幾乎是必經之路\n我在之前的影片中就不斷強調過\n在 AI 時代做自動化\n你要學會把大模型當成工作流裡的一個零件\n而不是把它當成一個只會聊天的機器人\n而 Jev ，可以說\n就是把這個理念貫徹到極致的產物\n它直接拔掉了聊天的能力\n生下來就是為了當一個極致高效的零件\n最近剛好有一位來找我諮詢的客戶\n他們公司想做一個客服 AI Agent\n用來自動處理每天雪片般飛來的客服單\n剛好碰上 Jev 推出\n我就跟他建議\n把整條自動化流程的第一站直接換成 Jev\n客服單一進來，二話不說先交給 Jev 做分類\n假設今天收到一張客訴單\n上面寫著我被重複扣款兩次\n今天一定要給我處理好！\n同時後台資料顯示\n這位客人目前使用的是高階的付費方案\n而且帳號已經使用了四百多天\n當 Jev 拿到這張單以及客人的背景資料後\n我們可以同時向它拋出五個問題\n第一，這張單的分類是什麼？\n是帳務問題，系統故障，還是其他？\n第二，這張單的嚴重程度\n如果從 0 到 10 分來打分\n是幾分？\n第三，客人有沒有在信件中要求退款？\n第四，客人有沒有具體寫出\n問題發生的操作步驟？\n第五，客人現在的情緒狀態是平靜\n不耐煩，還是憤怒？\n這五個關鍵問題的答案\n不到一秒鐘就能全部生成出來\n接著，只要把這些答案\n串接到我們預先寫好的分類邏輯裡\n系統就能自動把這張緊急的客訴單\n精準地派送到負責處理帳務的對應部門\n甚至直接亮起紅燈提醒主管處理\n不過，就如同前面所說的， Jev 不會寫字\n所以最後回覆給客人的那封安撫信\n還是得交給聰明的大模型來代筆\n但信寫完之後\n我們發送前可以再次呼叫 Jev\n讓它當最後的把關者\n檢查這封回覆有沒有真的回答到客人的痛點？\n語氣看起來像是在誠懇道歉\n還是在推卸責任？\n有沒有不小心把公司內部的\n機密規定給洩漏出去？\n加上 Jev 這道檢查關卡\n不僅速度快，成本幾乎可以忽略不計\n還能大幅降低 AI 產生幻覺\n亂回覆客人的公關危機風險\n講到這裡，我們已經知道 Jev 是什麼\n看了社群上的各種玩法\n也看了它放進客服流程裡的樣子\n最後我就簡單做個總結吧\n如果你本身就在做 AI 自動化\n或者手邊每天都有海量的信件，社群留言\n表單或客服單需要處理\n那這些痛點簡直可以說就是為了 Jev 而生的\n看完這支影片，你今天就可以馬上開始構思\n該怎麼把 Jev 整合進你的工作流裡\n如果你是寫程式的開發者\n它非常適合被放在檢查的關卡\n幫你掃描註解，審查程式碼邏輯\n因為它實在太便宜了，便宜到你每一次存檔\n每一次改 code 都可以毫無壓力地讓它跑一遍\n至於開發者具體的實作方式\n我已經寫成了一篇詳細的文章\n就不在這支影片裡佔用大家時間了\n同時，我也特別製作了\n一組好用的提示詞工具\n可以幫你快速盤點\n你現有的專案或自動化工作流中\n到底有哪些環節最適合引入 Jev 來優化\n對這些資源感興趣的朋友\n歡迎到我們的 Skool 社群\n查看完整文章並領取提示詞工具\n相關連結我都會放在下方的資訊欄\n而如果你平常只是習慣把 ChatGPT 或 Claude\n當作聊天機器人來問問題\n那你高機率不會直接用到 Jev\n但我相信\n你每天在用的各種軟體和產品背後\n很快就會開始大量出現\n這種專門做判斷的模型應用\n比方說，哪天 Claude\n把要不要觸發某個工具或技能\n這件事直接外包給 Jev 來做決策\n也不是不可能的事\n畢竟這真的是又快又便宜的解法\n我之前在頻道裡就一直強調過一個觀念\n不要問現在哪個模型最強\n你要問的是\n哪個模型最適合你當下的工作流\n這個道理放在任何領域都通用\n就像好久以前大家就知道用高階\n聰明的模型來做 Planning\n再用低階，快速的模型去執行苦力實作\n而現在，我們的武器庫裡\n又多了一個專職的決策模型\n它可以幫助我們用更低的成本\n更快的速度，達到比以前更好的自動化效果\n這對於所有喜歡\n自己動手 Build 東西的創作者\n或者是第一線的工程師來說\n絕對是一大福音\n如果你也想親自試試看 Jev 的威力\n我會把他們的官方網站連結放在下方資訊欄\n不過目前他們還處於早期預約階段\n需要先登記你的 Email 排隊\n才有機會獲得使用權限\n有興趣的朋友強烈建議先去卡位預約起來\n才能在第一時間親自體驗\n那以上就是今天影片的全部內容啦\n希望這集關於 Jev 的介紹對各位有所幫助\n如果你喜歡今天的影片\n別忘了幫我按讚，訂閱\n並分享給身邊對 AI 自動化有興趣的朋友\n你們的支持就是我持續創作下去最大的動力！\n那我們就下支影片見囉，掰掰！", "language": "zh", "duration": 820.0, "segments": [ { "id": 0, "start": 0.0, "end": 0.56, "text": "嗨各位", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 1, "start": 0.92, "end": 2.359, "text": "今天我們來聊聊最近 AI 圈的", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 2, "start": 2.359, "end": 3.289, "text": "當紅炸子雞，Jev", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 3, "start": 3.75, "end": 5.84, "text": "它也是一個 AI 模型，但特別的是", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 4, "start": 5.96, "end": 6.82, "text": "它不跟你聊天", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 5, "start": 6.82, "end": 8.859, "text": "不幫你寫文章，也不會寫程式", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 6, "start": 9.159, "end": 12.46, "text": "你丟問題給它，它甚至連一個字都不會回你", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 7, "start": 12.46, "end": 14.88, "text": "因為它只專注做一件事，就是做決定", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 8, "start": 15.6, "end": 17.299, "text": "Jev 在九月十五號發布", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 9, "start": 17.299, "end": 18.5, "text": "創辦人 Diogo Almeida", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 10, "start": 18.5, "end": 20.7, "text": "是當年一起做出 ChatGPT 的人之一", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 11, "start": 20.7, "end": 21.879, "text": "他發的那則 X 貼文", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 12, "start": 21.879, "end": 23.68, "text": "到現在已經接近三千萬次觀看", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 13, "start": 23.7, "end": 27.299, "text": "當天在工程師論壇 Hacker News 上拿到 1679 分", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 14, "start": 27.299, "end": 28.6, "text": "底下四百多則留言", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 15, "start": 28.6, "end": 31.299, "text": "隔天 Vercel 也直接把 Jev 上架到自己的平台", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 16, "start": 31.459, "end": 32.74, "text": "讓用戶可以使用", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 17, "start": 32.939, "end": 34.7, "text": "根據官方公布的數據", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 18, "start": 34.7, "end": 37.0, "text": "Jev 跟現在主流的大語言模型相比", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 19, "start": 37.0, "end": 38.4, "text": "速度快了 20 到 200 倍", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 20, "start": 38.759, "end": 40.52, "text": "價格更是便宜了 40 到 400 倍", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 21, "start": 40.639, "end": 42.659, "text": "最扯的是它的輸出完全免費", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 22, "start": 43.0, "end": 43.72, "text": "意思就是", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 23, "start": 43.72, "end": 46.4, "text": "不管你叫它吐出一百個字還是一百萬個字", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 24, "start": 46.599, "end": 48.04, "text": "它通通一毛錢都不會跟你收", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 25, "start": 48.759, "end": 49.74, "text": "我認為這款模型", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 26, "start": 49.74, "end": 51.979, "text": "將會對 AI Automation 帶來很大的影響", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 27, "start": 52.18, "end": 53.04, "text": "原因很簡單", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 28, "start": 53.099, "end": 55.34, "text": "現在大家用 AI 搭出來的自動化流程裡", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 29, "start": 55.34, "end": 57.799, "text": "有一大堆步驟其實只是在做判斷", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 30, "start": 57.799, "end": 59.799, "text": "比方說，這封信要不要回？", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 31, "start": 59.86, "end": 61.619, "text": "這張客服單該轉給哪個部門？", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 32, "start": 61.68, "end": 63.52, "text": "這則留言是不是詐騙廣告？", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 33, "start": 63.52, "end": 65.36, "text": "這些問題的答案通常只是一個選項", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 34, "start": 65.36, "end": 66.699, "text": "但我們現在卻都習慣", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 35, "start": 66.699, "end": 68.279, "text": "請又貴又慢的大模型來處理", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 36, "start": 68.279, "end": 70.379, "text": "這其實非常浪費時間和金錢", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 37, "start": 70.839, "end": 73.459, "text": "而 Jev 想要搶佔的領地就是這一塊", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 38, "start": 73.549, "end": 74.5, "text": "它對標的不是 Fable、Astra 這類幾乎全能的大模型", "source": "video_source_subtitle", "authority": "video_source_subtitle", "text_original": "它對標的不是 Fable", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "id": 39, "start": 74.919, "end": 76.36, "text": "Astra 這類幾乎全能的大模型", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 40, "start": 76.36, "end": 78.12, "text": "你也不會拿它來幫你開發產品", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 41, "start": 78.29, "end": 79.099, "text": "它最適合的位子", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 42, "start": 79.099, "end": 81.4, "text": "是放在你的工作流裡面當一個決策節點", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 43, "start": 81.459, "end": 84.739, "text": "主打的優勢就是快，便宜，以及極低的幻覺", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 44, "start": 85.44, "end": 86.26, "text": "今天這支影片", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 45, "start": 86.26, "end": 88.18, "text": "我會先跟你分享 Jev 到底是什麼", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 46, "start": 88.349, "end": 90.36, "text": "它跟傳統大模型應該怎麼分工，接著", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 47, "start": 90.36, "end": 92.76, "text": "我們會來看幾個社群上非常有趣的 Demo", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 48, "start": 92.94, "end": 95.4, "text": "然後，我會用一個最近剛接手的實際案例", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 49, "start": 95.58, "end": 96.639, "text": "帶你看看這個模型", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 50, "start": 96.639, "end": 98.68, "text": "可以怎麼被完美整合進工作流裡面", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 51, "start": 98.68, "end": 100.36, "text": "那我們就直接開始吧！", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 52, "start": 100.36, "end": 101.559, "text": "先來聊聊它的來歷", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 53, "start": 101.76, "end": 104.779, "text": "Jev 是由一家叫做 TypeSafe AI 的公司開發的", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 54, "start": 104.779, "end": 107.239, "text": "這家公司過去一直非常低調，沒有對外曝光", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 55, "start": 107.5, "end": 109.639, "text": "而 Jev 就是他們公開亮相的第一款模型", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 56, "start": 109.94, "end": 110.959, "text": "前面有提到", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 57, "start": 110.959, "end": 111.94, "text": "創辦人 Diogo Almeida", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 58, "start": 111.94, "end": 114.279, "text": "是當年參與打造 ChatGPT 的團隊成員", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 59, "start": 114.279, "end": 115.739, "text": "講得更精準一點", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 60, "start": 115.739, "end": 117.26, "text": "ChatGPT 之所以聽得懂人話", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 61, "start": 117.26, "end": 118.779, "text": "會跟你一來一往的聊天", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 62, "start": 118.779, "end": 121.4, "text": "背後靠的是一套叫做 RLHF 的訓練方法", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 63, "start": 121.4, "end": 123.54, "text": "而他正是這套方法的共同發明人", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 64, "start": 123.559, "end": 126.019, "text": "他自己也分享過，在做完 ChatGPT 之後", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 65, "start": 126.139, "end": 127.54, "text": "他花了兩年的時間", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 66, "start": 127.54, "end": 129.74, "text": "默默研究了一套全新的訓練方法", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 67, "start": 129.74, "end": 131.113, "text": "最後生出來的成果就是 Jev", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 68, "start": 131.339, "end": 133.419, "text": "一個完全不會跟你聊天的模型", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 69, "start": 133.46, "end": 134.199, "text": "有趣的是", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 70, "start": 134.419, "end": 137.179, "text": "Jev 這個名字取自經濟學中的傑文斯悖論", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 71, "start": 137.32, "end": 138.839, "text": "英文是 Jevons Paradox", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 72, "start": 139.139, "end": 141.44, "text": "意思就是當一樣東西變得越便宜", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 73, "start": 141.74, "end": 143.5, "text": "效率越高時，大家反而會用得越多", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 74, "start": 144.389, "end": 145.1, "text": "它的運作方式", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 75, "start": 145.1, "end": 146.759, "text": "跟我們熟悉的大模型完全不一樣", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 76, "start": 147.059, "end": 149.139, "text": "當你問 ChatGPT 或 Claude 問題時", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 77, "start": 149.419, "end": 151.44, "text": "它是一個字一個字把答案生出來的", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 78, "start": 151.44, "end": 153.919, "text": "想得越多，算得越久，你就得等越久", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 79, "start": 153.979, "end": 155.539, "text": "但 Jev 不會自己寫答案", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 80, "start": 155.539, "end": 157.419, "text": "你只需要給它三樣東西：一份資料、一個問題，以及你事先列好的幾個選項", "source": "video_source_subtitle", "authority": "video_source_subtitle", "text_original": "你只需要給它三樣東西一份資料", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "id": 81, "start": 157.6, "end": 159.339, "text": "一個問題，以及你事先列好的幾個選項", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 82, "start": 159.479, "end": 161.339, "text": "它就會直接回傳給你每個選項的機率", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 83, "start": 162.059, "end": 164.159, "text": "舉個簡單的例子，你丟一張發票給它", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 84, "start": 164.259, "end": 165.72, "text": "問它這張發票有沒有問題？", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 85, "start": 165.82, "end": 167.539, "text": "並給它三個選項正常", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 86, "start": 167.94, "end": 169.419, "text": "詐騙，需要人工檢查", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 87, "start": 169.66, "end": 172.889, "text": "它就會秒回你這張發票正常的機率是 88%", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 88, "start": 173.119, "end": 175.46, "text": "剩下的機率則分配給另外兩個選項", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 89, "start": 175.539, "end": 176.86, "text": "而這整個判斷過程", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 90, "start": 176.88, "end": 179.16, "text": "通常花不到 0.3 秒就能完成", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 91, "start": 179.46, "end": 180.979, "text": "如果你聽到這邊還是一頭霧水", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 92, "start": 180.979, "end": 182.58, "text": "那我用個生活化一點的比喻", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 93, "start": 182.72, "end": 184.8, "text": "一般的大模型就像是在寫申論題", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 94, "start": 185.08, "end": 186.679, "text": "而 Jev 則像是在寫選擇題", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 95, "start": 187.339, "end": 189.339, "text": "申論題得一句一句慢慢寫，不但寫得慢", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 96, "start": 189.339, "end": 190.539, "text": "還有可能不小心離題", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 97, "start": 190.539, "end": 192.419, "text": "但選擇題只要圈一個答案就好", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 98, "start": 192.44, "end": 195.059, "text": "所以速度極快，而且答案絕對不會發散", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 99, "start": 195.059, "end": 197.08, "text": "一定只會是你給的選項其中之一", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 100, "start": 197.16, "end": 199.139, "text": "這也是為什麼 Jev 能夠做到低幻覺", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 101, "start": 199.24, "end": 200.039, "text": "因為它沒有機會", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 102, "start": 200.039, "end": 201.8, "text": "去瞎掰一個選項以外的答案出來", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 103, "start": 202.32, "end": 203.779, "text": "目前 Jev 能夠回答的題型，主要分為三種：是非題、單選題，以及打分題", "source": "video_source_subtitle", "authority": "video_source_subtitle", "text_original": "目前 Jev 能夠回答的題型", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "id": 104, "start": 203.779, "end": 205.199, "text": "主要分為三種是非題", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 105, "start": 205.52, "end": 207.58, "text": "單選題，以及打分題", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 106, "start": 207.58, "end": 209.58, "text": "是非題很好理解，你問它一句話是不是真的", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 107, "start": 209.72, "end": 211.86, "text": "它會回傳一個 0 到 1 之間的機率", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 108, "start": 211.86, "end": 213.94, "text": "越接近 1 就代表它越肯定是", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 109, "start": 213.94, "end": 216.093, "text": "比方說，你可以丟一張發票給 Jev", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 110, "start": 216.32, "end": 217.839, "text": "問它這張發票是不是詐騙？", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 111, "start": 218.059, "end": 219.9, "text": "Jev 可能會回你 90%", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 112, "start": 220.16, "end": 222.039, "text": "你甚至可以把幾個常見的詐騙特徵", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 113, "start": 222.039, "end": 223.32, "text": "寫進條件裡再問一次", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 114, "start": 223.46, "end": 224.82, "text": "機率或許就會變得更高", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 115, "start": 224.839, "end": 226.399, "text": "因為 Jev 有了更多的參考依據", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 116, "start": 226.399, "end": 227.58, "text": "來增加它的信心水準", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 117, "start": 227.94, "end": 229.1, "text": "拿到這個數字之後", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 118, "start": 229.1, "end": 230.74, "text": "最簡單暴力的應用方式就是", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 119, "start": 230.74, "end": 232.899, "text": "在自動化退款的程式裡設定一個門檻", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 120, "start": 233.059, "end": 235.08, "text": "比方說只要詐騙機率超過九成", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 121, "start": 235.16, "end": 236.36, "text": "就直接自動拒絕", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 122, "start": 236.399, "end": 237.479, "text": "其餘的退款請求", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 123, "start": 237.619, "end": 240.059, "text": "再自動寄信轉交給真人客服處理", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 124, "start": 240.059, "end": 240.699, "text": "這樣一來", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 125, "start": 240.72, "end": 243.539, "text": "你就幫客服省下了大量處理送分題的時間", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 126, "start": 243.96, "end": 246.72, "text": "第二種是單選題，你可以先列出好幾個選項", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 127, "start": 246.86, "end": 249.119, "text": "Jev 會從裡面挑出一個最合適的", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 128, "start": 249.119, "end": 251.22, "text": "同時告訴你每個選項的機率各是多少", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 129, "start": 251.639, "end": 253.16, "text": "選項的內容完全由你定義", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 130, "start": 253.16, "end": 254.699, "text": "最多可以列到 255 個", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 131, "start": 254.759, "end": 257.16, "text": "而且每個選項你還可以搭配一句簡單的說明", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 132, "start": 257.16, "end": 259.239, "text": "讓它更清楚這個選項的定義", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 133, "start": 259.239, "end": 261.459, "text": "比方說，你可以拿它來自動整理信箱", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 134, "start": 261.559, "end": 262.98, "text": "讓 Jev 幫你精準判斷", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 135, "start": 262.98, "end": 265.16, "text": "每一封信應該歸類到哪個資料夾", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 136, "start": 265.7, "end": 267.2, "text": "第三種是打分題", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 137, "start": 267.2, "end": 269.92, "text": "有些事情不是非黑即白地選一個類別", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 138, "start": 269.92, "end": 270.839, "text": "而是要看程度", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 139, "start": 271.16, "end": 272.779, "text": "這時候你可以先定好一把尺", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 140, "start": 272.88, "end": 274.019, "text": "或者說畫一個光譜", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 141, "start": 274.14, "end": 276.54, "text": "從低到高寫出每一個級距代表什麼意思", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 142, "start": 276.559, "end": 277.799, "text": "最多可以分到 11 級", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 143, "start": 277.959, "end": 278.959, "text": "Jev 就會告訴你", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 144, "start": 278.959, "end": 281.22, "text": "現在的狀況落在那把尺的哪個位置", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 145, "start": 281.22, "end": 283.42, "text": "比方說你想幫業務篩選潛在客戶", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 146, "start": 283.559, "end": 285.679, "text": "你可以設定 0 分是沒有預算的學生", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 147, "start": 285.76, "end": 286.76, "text": "1 分是小團隊", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 148, "start": 286.859, "end": 289.0, "text": "2 分是有明確需求跟時程的公司", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 149, "start": 289.0, "end": 291.38, "text": "3 分則是預算充足的大企業", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 150, "start": 291.38, "end": 292.66, "text": "這時候如果有一間大企業", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 151, "start": 292.66, "end": 294.16, "text": "透過網站表單來詢問", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 152, "start": 294.16, "end": 296.88, "text": "Jev 可能就會幫它打出 2.9 分的高分", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 153, "start": 296.88, "end": 297.739, "text": "拿到這個分數後", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 154, "start": 297.739, "end": 299.899, "text": "你就可以把它串接到後面的自動化規則", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 155, "start": 299.92, "end": 301.899, "text": "比方說設定 2.5 分以上的客戶", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 156, "start": 301.899, "end": 303.7, "text": "要觸發緊急通知給業務主管", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 157, "start": 303.7, "end": 306.519, "text": "確保團隊當天就能立刻回信搶下訂單", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 158, "start": 306.519, "end": 308.519, "text": "所以 Jev 的能力跟定位其實非常明確", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 159, "start": 308.76, "end": 309.94, "text": "它能回答的就是是不是", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 160, "start": 310.2, "end": 312.26, "text": "哪一個，到什麼程度這三種問題", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 161, "start": 312.64, "end": 313.7, "text": "而如果你仔細拆解", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 162, "start": 313.7, "end": 315.839, "text": "現在各種自動化工作流裡的判斷邏輯", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 163, "start": 316.0, "end": 317.42, "text": "其實大部分也離不開這三種", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 164, "start": 317.42, "end": 320.399, "text": "整個模型用起來的感覺就是資料丟進去", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 165, "start": 320.489, "end": 321.2, "text": "它不跟你廢話", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 166, "start": 321.2, "end": 322.079, "text": "瞬間給你結論", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 167, "start": 322.16, "end": 324.04, "text": "像是一個小而美的自動化開關", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 168, "start": 324.619, "end": 326.48, "text": "在價格方面，Jev 也非常有競爭力", "source": "video_source_subtitle", "authority": "video_source_subtitle", "text_original": "在價格方面 Jev 也非常有競爭力", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "id": 169, "start": 326.88, "end": 329.72, "text": "就像前面提到的，它的輸出是完全免費的", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 170, "start": 329.72, "end": 331.519, "text": "計費方式只看讀進去的文字量", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 171, "start": 331.519, "end": 333.839, "text": "每一百萬個 Token 只收 0.042 美元", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 172, "start": 334.299, "end": 336.339, "text": "你可能對一百萬個 Token 沒什麼概念", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 173, "start": 336.559, "end": 337.94, "text": "最簡單的理解方式就是", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 174, "start": 338.0, "end": 340.72, "text": "這大約等於好幾本厚厚的小說的文字量", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 175, "start": 340.72, "end": 341.64, "text": "從這裡就可以看出", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 176, "start": 341.64, "end": 343.019, "text": "Jev 真的是便宜到不可思議", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 177, "start": 343.619, "end": 346.119, "text": "不過話說回來，不管它再怎麼便宜或快速", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 178, "start": 346.289, "end": 347.32, "text": "它都還是無法取代", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 179, "start": 347.32, "end": 349.239, "text": "Fable 跟 Astra 這種全能型模型", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 180, "start": 349.579, "end": 352.279, "text": "或者說它打從一開始就沒打算要取代它們", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 181, "start": 352.279, "end": 354.16, "text": "因為這兩者的定位根本完全不同", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 182, "start": 354.299, "end": 356.299, "text": "TypeSafe 幫 Jev 創造了一個新的分類", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 183, "start": 356.299, "end": 357.6, "text": "叫做 System 1 模型", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 184, "start": 358.019, "end": 360.42, "text": "這個詞出自於經典名著快思慢想", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 185, "start": 360.779, "end": 361.54, "text": "書裡提到", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 186, "start": 361.54, "end": 364.26, "text": "人的大腦其實有兩套系統系統一靠的是直覺", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 187, "start": 364.26, "end": 366.2, "text": "反應極快，系統二靠的是推理", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 188, "start": 366.239, "end": 367.6, "text": "速度慢但是想得深", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 189, "start": 367.94, "end": 370.0, "text": "就像你剛學開車的時候，看後照鏡", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 190, "start": 370.0, "end": 372.82, "text": "打方向燈，換檔，每一個動作都要用力去想", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 191, "start": 372.82, "end": 375.04, "text": "這就是系統二，但開了幾年車之後", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 192, "start": 375.04, "end": 377.72, "text": "這些動作全部變成了下意識的反射動作", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 193, "start": 377.72, "end": 378.679, "text": "這就是系統一", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 194, "start": 378.679, "end": 381.66, "text": "Fable 跟 Astra 就是那個會深思熟慮的系統二", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 195, "start": 381.739, "end": 384.359, "text": "而 Jev 就是那個靠直覺秒回的系統一", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 196, "start": 384.359, "end": 386.26, "text": "所以，這兩者的分工非常清楚", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 197, "start": 386.26, "end": 388.16, "text": "如果你今天需要 AI 幫你構思", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 198, "start": 388.16, "end": 388.88, "text": "幫你寫作", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 199, "start": 388.88, "end": 389.98, "text": "那就交給大模型", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 200, "start": 390.059, "end": 392.0, "text": "但如果你今天只是需要一個快速", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 201, "start": 392.0, "end": 393.66, "text": "準確的判斷，那就交給 Jev", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 202, "start": 394.14, "end": 395.859, "text": "對 Jev 有了基本的認知後", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 203, "start": 395.859, "end": 397.559, "text": "我們來看看社群上的創作者們", "source": "video_source_subtitle", "authority": "video_source_subtitle", "text_original": "我們來看看社群上的兄弟們", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "id": 204, "start": 397.559, "end": 399.399, "text": "都拿它來做什麼吧", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 205, "start": 399.399, "end": 400.579, "text": "國外 YouTuber Matthew Berman", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 206, "start": 400.579, "end": 401.92, "text": "用它做了一個虛擬小鎮", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 207, "start": 402.0, "end": 403.679, "text": "裡面住了 50 個 AI 居民", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 208, "start": 403.679, "end": 406.399, "text": "當他對全鎮廣播一句麵包店大拍賣！", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 209, "start": 406.399, "end": 408.059, "text": "時，短短 0.6 秒之內", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 210, "start": 408.059, "end": 411.44, "text": "這 50 個居民就各自決定好自己要怎麼反應了", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 211, "start": 411.44, "end": 413.339, "text": "其中 39 個不為所動繼續做自己的事", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 212, "start": 413.339, "end": 414.679, "text": "6 個跑去湊熱鬧看看", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 213, "start": 414.7, "end": 416.279, "text": "4 個直接加入搶購行列", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 214, "start": 416.279, "end": 418.42, "text": "還有 1 個熱心地跑去通知別人", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 215, "start": 418.42, "end": 419.779, "text": "我覺得這個實驗非常有趣", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 216, "start": 419.92, "end": 421.459, "text": "未來我們絕對會在遊戲領域", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 217, "start": 421.459, "end": 422.679, "text": "看到大量類似的應用", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 218, "start": 422.7, "end": 424.7, "text": "比方說遊戲裡的 NPC 行為", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 219, "start": 424.7, "end": 426.779, "text": "或許就能透過這種方式變得更加擬真", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 220, "start": 427.059, "end": 428.839, "text": "多樣化，而且運算成本還極低", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 221, "start": 429.5, "end": 430.26, "text": "第二個例子", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 222, "start": 430.26, "end": 432.26, "text": "是 AI ML API 這個團隊", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 223, "start": 432.26, "end": 434.42, "text": "在 X 上發布的一場對弈實驗", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 224, "start": 434.839, "end": 437.7, "text": "他們讓 Jev 分別跟 Fable 和 Astra 下西洋棋", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 225, "start": 437.899, "end": 440.26, "text": "規則是每方只有五分鐘時間的快棋", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 226, "start": 440.38, "end": 443.079, "text": "每下一步棋，就等於呼叫一次模型", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 227, "start": 443.16, "end": 445.1, "text": "有趣的是，在對上 Fable 的那一局", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 228, "start": 445.1, "end": 445.92, "text": "如果單論棋力", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 229, "start": 445.92, "end": 447.32, "text": "Fable 絕對是完全碾壓", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 230, "start": 447.339, "end": 449.54, "text": "下到第 29 步時已經佔盡優勢", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 231, "start": 449.6, "end": 452.2, "text": "甚至還多吃了一隻皇后，但致命傷是", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 232, "start": 452.519, "end": 455.059, "text": "Fable 每想一步棋都要花上 6 到 15 秒", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 233, "start": 455.079, "end": 457.799, "text": "而 Jev 每一步只花 2.6 秒", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 234, "start": 457.799, "end": 458.679, "text": "最後的結果是", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 235, "start": 458.679, "end": 460.32, "text": "Fable 把五分鐘的時間全耗光了", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 236, "start": 460.32, "end": 461.54, "text": "因為超時而被判輸", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 237, "start": 462.2, "end": 465.899, "text": "可以說，最聰明的模型，最後卻輸在想太久", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 238, "start": 465.92, "end": 469.519, "text": "不過，當 Jev 對上 Astra 時情況就不太一樣了", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 239, "start": 469.519, "end": 472.16, "text": "Astra 只用了短短 18 步就直接把 Jev 將軍死棋", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 240, "start": 472.2, "end": 473.339, "text": "是真的有點瘋狂", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 241, "start": 473.799, "end": 474.619, "text": "除此之外", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 242, "start": 474.619, "end": 475.839, "text": "社群上也有很多人", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 243, "start": 475.839, "end": 478.839, "text": "開始把 Jev 拿來做大量分類的任務", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 244, "start": 478.839, "end": 480.82, "text": "比方說，有人拿了 1000 封 Email", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 245, "start": 480.82, "end": 481.94, "text": "來進行壓力測試", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 246, "start": 481.94, "end": 484.14, "text": "讓模型去判斷每一封信是不是發票", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 247, "start": 484.14, "end": 485.72, "text": "是不是合作邀約等等", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 248, "start": 485.799, "end": 488.519, "text": "結果 Jev 跑完這 1000 封信只花了短短 70 秒", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 249, "start": 488.519, "end": 490.38, "text": "成本更是只要驚人的 9 美分", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 250, "start": 490.519, "end": 491.839, "text": "不到台幣三塊錢", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 251, "start": 491.839, "end": 494.48, "text": "但如果把同一批信交給 GPT-4o 來跑", "source": "video_source_subtitle", "authority": "video_source_subtitle", "text_original": "但如果把同一批信交給 GPT-5.6 Luna 來跑", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "id": 252, "start": 494.619, "end": 495.82, "text": "不僅要花上 5 分鐘", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 253, "start": 496.059, "end": 499.739, "text": "成本還高達 62 美分，要台幣二十幾塊錢", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 254, "start": 499.739, "end": 501.48, "text": "OpenCode 創始團隊的 Ryan Vogel", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 255, "start": 501.48, "end": 502.76, "text": "也做了一個類似的專案", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 256, "start": 502.98, "end": 505.119, "text": "他一口氣把自己信箱裡的 1700 封信", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 257, "start": 505.119, "end": 506.899, "text": "全丟給 Jev ，讓它幫忙分門別類", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 258, "start": 507.04, "end": 509.6, "text": "排定處理優先順序，打上垃圾信件的分數", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 259, "start": 509.779, "end": 512.119, "text": "最後再判斷這封信需不需要親自回覆", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 260, "start": 512.58, "end": 516.26, "text": "全部 1700 封信跑完，總共竟然只花了 18 美分", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 261, "start": 516.78, "end": 518.739, "text": "另外， Jev 也非常適合拿來做路由", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 262, "start": 518.799, "end": 521.28, "text": "也就是幫你決定這件事應該交給誰來做", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 263, "start": 521.82, "end": 524.219, "text": "YouTuber Riley Brown 就實作了一個模型路由器", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 264, "start": 524.26, "end": 525.7, "text": "當你的問題丟進去後", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 265, "start": 525.7, "end": 527.0, "text": "Jev 自己不負責回答", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 266, "start": 527.16, "end": 528.219, "text": "它只負責快速判斷", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 267, "start": 528.219, "end": 530.239, "text": "這道題該派給哪個模型處理", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 268, "start": 530.5, "end": 532.02, "text": "如果是改錯字這種小事", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 269, "start": 532.02, "end": 533.38, "text": "就丟給最便宜的模型", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 270, "start": 533.679, "end": 535.46, "text": "如果是要寫一整套複雜的程式", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 271, "start": 535.46, "end": 537.32, "text": "再轉發給最貴，最聰明的模型", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 272, "start": 537.88, "end": 539.919, "text": "TypeSafe 官方也釋出了一個有趣的數據", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 273, "start": 540.0, "end": 542.28, "text": "如果讓 Agent 自己挑要觸發哪個 skill", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 274, "start": 542.32, "end": 544.07, "text": "拿錯的機率高達 17%", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 275, "start": 544.239, "end": 546.58, "text": "但如果先讓 Jev 幫忙快速過濾挑選", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 276, "start": 546.58, "end": 549.5, "text": "出錯的機率瞬間降到了只有 7.3%", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 277, "start": 549.919, "end": 551.4, "text": "還有一位叫做 kitze 的開發者", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 278, "start": 551.46, "end": 552.34, "text": "用它做了一款", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 279, "start": 552.34, "end": 554.419, "text": "名為 Unclutter 的免費瀏覽器外掛", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 280, "start": 554.539, "end": 557.139, "text": "當你網頁一打開，它就會在背景呼叫 Jev", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 281, "start": 557.44, "end": 559.58, "text": "即時判斷網頁上的哪些區塊是廣告", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 282, "start": 559.64, "end": 561.32, "text": "哪些是沒有營養的垃圾內容", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 283, "start": 561.58, "end": 563.739, "text": "然後瞬間幫你清理得乾乾淨淨", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 284, "start": 563.739, "end": 565.52, "text": "還給你一個清爽的閱讀版面", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 285, "start": 565.719, "end": 567.46, "text": "我認為 Jev 對大多數人來說", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 286, "start": 567.46, "end": 569.359, "text": "最實用的場景絕對是分類和路由", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 287, "start": 569.479, "end": 571.059, "text": "因為只要你想做自動化流程", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 288, "start": 571.059, "end": 572.64, "text": "這兩個環節幾乎是必經之路", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 289, "start": 573.34, "end": 575.359, "text": "我在之前的影片中就不斷強調過", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 290, "start": 575.359, "end": 576.5, "text": "在 AI 時代做自動化", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 291, "start": 576.599, "end": 579.28, "text": "你要學會把大模型當成工作流裡的一個零件", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 292, "start": 579.299, "end": 582.099, "text": "而不是把它當成一個只會聊天的機器人", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 293, "start": 582.2, "end": 583.0, "text": "而 Jev ，可以說", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 294, "start": 583.0, "end": 584.82, "text": "就是把這個理念貫徹到極致的產物", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 295, "start": 585.059, "end": 586.4, "text": "它直接拔掉了聊天的能力", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 296, "start": 586.579, "end": 588.84, "text": "生下來就是為了當一個極致高效的零件", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 297, "start": 589.299, "end": 591.34, "text": "最近剛好有一位來找我諮詢的客戶", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 298, "start": 591.4, "end": 593.479, "text": "他們公司想做一個客服 AI Agent", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 299, "start": 593.479, "end": 596.119, "text": "用來自動處理每天雪片般飛來的客服單", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 300, "start": 596.46, "end": 597.88, "text": "剛好碰上 Jev 推出", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 301, "start": 598.059, "end": 598.9, "text": "我就跟他建議", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 302, "start": 598.9, "end": 601.46, "text": "把整條自動化流程的第一站直接換成 Jev", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 303, "start": 601.82, "end": 604.44, "text": "客服單一進來，二話不說先交給 Jev 做分類", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 304, "start": 604.919, "end": 606.52, "text": "假設今天收到一張客訴單", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 305, "start": 606.58, "end": 608.679, "text": "上面寫著我被重複扣款兩次", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 306, "start": 608.679, "end": 610.26, "text": "今天一定要給我處理好！", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 307, "start": 610.26, "end": 611.44, "text": "同時後台資料顯示", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 308, "start": 611.44, "end": 613.76, "text": "這位客人目前使用的是高階的付費方案", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 309, "start": 613.82, "end": 615.78, "text": "而且帳號已經使用了四百多天", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 310, "start": 615.82, "end": 618.719, "text": "當 Jev 拿到這張單以及客人的背景資料後", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 311, "start": 618.719, "end": 620.64, "text": "我們可以同時向它拋出五個問題", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 312, "start": 621.219, "end": 622.82, "text": "第一，這張單的分類是什麼？", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 313, "start": 622.94, "end": 625.679, "text": "是帳務問題，系統故障，還是其他？", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 314, "start": 625.799, "end": 627.4, "text": "第二，這張單的嚴重程度", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 315, "start": 627.4, "end": 629.099, "text": "如果從 0 到 10 分來打分", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 316, "start": 629.099, "end": 629.739, "text": "是幾分？", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 317, "start": 629.739, "end": 632.619, "text": "第三，客人有沒有在信件中要求退款？", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 318, "start": 632.64, "end": 634.479, "text": "第四，客人有沒有具體寫出", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 319, "start": 634.479, "end": 636.039, "text": "問題發生的操作步驟？", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 320, "start": 636.179, "end": 638.559, "text": "第五，客人現在的情緒狀態是平靜", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 321, "start": 638.559, "end": 640.28, "text": "不耐煩，還是憤怒？", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 322, "start": 640.28, "end": 641.659, "text": "這五個關鍵問題的答案", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 323, "start": 641.659, "end": 644.0, "text": "不到一秒鐘就能全部生成出來", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 324, "start": 644.0, "end": 645.4, "text": "接著，只要把這些答案", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 325, "start": 645.4, "end": 647.539, "text": "串接到我們預先寫好的分類邏輯裡", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 326, "start": 647.539, "end": 649.419, "text": "系統就能自動把這張緊急的客訴單", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 327, "start": 649.52, "end": 651.82, "text": "精準地派送到負責處理帳務的對應部門", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 328, "start": 651.82, "end": 654.119, "text": "甚至直接亮起紅燈提醒主管處理", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 329, "start": 654.659, "end": 656.84, "text": "不過，就如同前面所說的， Jev 不會寫字", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 330, "start": 656.979, "end": 658.799, "text": "所以最後回覆給客人的那封安撫信", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 331, "start": 658.799, "end": 660.52, "text": "還是得交給聰明的大模型來代筆", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 332, "start": 661.039, "end": 662.219, "text": "但信寫完之後", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 333, "start": 662.219, "end": 663.926, "text": "我們發送前可以再次呼叫 Jev", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 334, "start": 664.28, "end": 665.96, "text": "讓它當最後的把關者", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 335, "start": 666.02, "end": 669.26, "text": "檢查這封回覆有沒有真的回答到客人的痛點？", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 336, "start": 669.26, "end": 671.119, "text": "語氣看起來像是在誠懇道歉", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 337, "start": 671.299, "end": 672.76, "text": "還是在推卸責任？", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 338, "start": 672.76, "end": 674.599, "text": "有沒有不小心把公司內部的", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 339, "start": 674.599, "end": 676.26, "text": "機密規定給洩漏出去？", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 340, "start": 676.26, "end": 677.739, "text": "加上 Jev 這道檢查關卡", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 341, "start": 677.78, "end": 680.299, "text": "不僅速度快，成本幾乎可以忽略不計", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 342, "start": 680.38, "end": 682.34, "text": "還能大幅降低 AI 產生幻覺", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 343, "start": 682.679, "end": 684.179, "text": "亂回覆客人的公關危機風險", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 344, "start": 684.7, "end": 686.78, "text": "講到這裡，我們已經知道 Jev 是什麼", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 345, "start": 686.9, "end": 688.26, "text": "看了社群上的各種玩法", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 346, "start": 688.26, "end": 690.52, "text": "也看了它放進客服流程裡的樣子", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 347, "start": 690.52, "end": 692.859, "text": "最後我就簡單做個總結吧", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 348, "start": 692.859, "end": 694.32, "text": "如果你本身就在做 AI 自動化", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 349, "start": 694.34, "end": 696.919, "text": "或者手邊每天都有海量的信件，社群留言", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 350, "start": 697.059, "end": 698.5, "text": "表單或客服單需要處理", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 351, "start": 698.64, "end": 701.88, "text": "那這些痛點簡直可以說就是為了 Jev 而生的", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 352, "start": 701.88, "end": 704.38, "text": "看完這支影片，你今天就可以馬上開始構思", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 353, "start": 704.479, "end": 707.159, "text": "該怎麼把 Jev 整合進你的工作流裡", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 354, "start": 707.159, "end": 708.46, "text": "如果你是寫程式的開發者", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 355, "start": 708.66, "end": 710.479, "text": "它非常適合被放在檢查的關卡", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 356, "start": 710.5, "end": 712.559, "text": "幫你掃描註解，審查程式碼邏輯", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 357, "start": 712.659, "end": 715.539, "text": "因為它實在太便宜了，便宜到你每一次存檔", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 358, "start": 715.539, "end": 718.219, "text": "每一次改 code 都可以毫無壓力地讓它跑一遍", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 359, "start": 718.219, "end": 720.02, "text": "至於開發者具體的實作方式", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 360, "start": 720.159, "end": 721.919, "text": "我已經寫成了一篇詳細的文章", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 361, "start": 722.099, "end": 724.679, "text": "就不在這支影片裡佔用大家時間了", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 362, "start": 724.679, "end": 725.88, "text": "同時，我也特別製作了", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 363, "start": 725.88, "end": 727.02, "text": "一組好用的提示詞工具", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 364, "start": 727.159, "end": 728.179, "text": "可以幫你快速盤點", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 365, "start": 728.179, "end": 730.26, "text": "你現有的專案或自動化工作流中", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 366, "start": 730.26, "end": 732.88, "text": "到底有哪些環節最適合引入 Jev 來優化", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 367, "start": 732.88, "end": 734.159, "text": "對這些資源感興趣的朋友", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 368, "start": 734.159, "end": 735.2, "text": "歡迎到我們的 Skool 社群", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 369, "start": 735.2, "end": 737.2, "text": "查看完整文章並領取提示詞工具", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 370, "start": 737.239, "end": 739.32, "text": "相關連結我都會放在下方的資訊欄", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 371, "start": 739.46, "end": 740.919, "text": "而如果你平常只是習慣把 ChatGPT 或 Claude", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 372, "start": 740.919, "end": 743.08, "text": "當作聊天機器人來問問題", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 373, "start": 743.14, "end": 744.919, "text": "那你高機率不會直接用到 Jev", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 374, "start": 744.979, "end": 745.679, "text": "但我相信", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 375, "start": 745.719, "end": 748.059, "text": "你每天在用的各種軟體和產品背後", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 376, "start": 748.059, "end": 749.219, "text": "很快就會開始大量出現", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 377, "start": 749.219, "end": 751.32, "text": "這種專門做判斷的模型應用", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 378, "start": 751.32, "end": 752.46, "text": "比方說，哪天 Claude", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 379, "start": 752.659, "end": 754.299, "text": "把要不要觸發某個工具或技能", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 380, "start": 754.299, "end": 756.099, "text": "這件事直接外包給 Jev 來做決策", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 381, "start": 756.28, "end": 757.7, "text": "也不是不可能的事", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 382, "start": 757.7, "end": 759.719, "text": "畢竟這真的是又快又便宜的解法", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 383, "start": 760.119, "end": 762.559, "text": "我之前在頻道裡就一直強調過一個觀念", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 384, "start": 762.599, "end": 764.179, "text": "不要問現在哪個模型最強", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 385, "start": 764.2, "end": 765.059, "text": "你要問的是", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 386, "start": 765.059, "end": 767.239, "text": "哪個模型最適合你當下的工作流", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 387, "start": 767.44, "end": 769.32, "text": "這個道理放在任何領域都通用", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 388, "start": 769.32, "end": 771.82, "text": "就像好久以前大家就知道用高階", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 389, "start": 772.26, "end": 773.239, "text": "聰明的模型來做 Planning", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 390, "start": 773.44, "end": 776.7, "text": "再用低階，快速的模型去執行苦力實作", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 391, "start": 776.7, "end": 777.82, "text": "而現在，我們的武器庫裡", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 392, "start": 777.82, "end": 779.7, "text": "又多了一個專職的決策模型", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 393, "start": 779.799, "end": 781.44, "text": "它可以幫助我們用更低的成本", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 394, "start": 781.559, "end": 784.2, "text": "更快的速度，達到比以前更好的自動化效果", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 395, "start": 784.299, "end": 785.2, "text": "這對於所有喜歡", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 396, "start": 785.2, "end": 786.78, "text": "自己動手 Build 東西的創作者", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 397, "start": 786.78, "end": 788.299, "text": "或者是第一線的工程師來說", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 398, "start": 788.479, "end": 789.5, "text": "絕對是一大福音", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 399, "start": 790.26, "end": 792.08, "text": "如果你也想親自試試看 Jev 的威力", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 400, "start": 792.099, "end": 794.9, "text": "我會把他們的官方網站連結放在下方資訊欄", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 401, "start": 795.28, "end": 797.239, "text": "不過目前他們還處於早期預約階段", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 402, "start": 797.239, "end": 798.78, "text": "需要先登記你的 Email 排隊", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 403, "start": 798.78, "end": 800.419, "text": "才有機會獲得使用權限", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 404, "start": 800.46, "end": 803.159, "text": "有興趣的朋友強烈建議先去卡位預約起來", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 405, "start": 803.159, "end": 805.419, "text": "才能在第一時間親自體驗", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 406, "start": 805.419, "end": 807.659, "text": "那以上就是今天影片的全部內容啦", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 407, "start": 807.659, "end": 810.0, "text": "希望這集關於 Jev 的介紹對各位有所幫助", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 408, "start": 810.219, "end": 811.4, "text": "如果你喜歡今天的影片", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 409, "start": 811.4, "end": 812.94, "text": "別忘了幫我按讚，訂閱", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 410, "start": 812.94, "end": 815.2, "text": "並分享給身邊對 AI 自動化有興趣的朋友", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 411, "start": 815.32, "end": 818.02, "text": "你們的支持就是我持續創作下去最大的動力！", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 412, "start": 818.02, "end": 819.559, "text": "那我們就下支影片見囉，掰掰！", "source": "video_source_subtitle", "authority": "video_source_subtitle" } ], "transcript_selection": { "schema": "videonote.transcript_selection.v1", "status": "SELECTED", "requested_language": "zh", "duration": 820.0, "reference_path": "/Users/sagawa/AI_WORK/video_notes/url/20260921_220558__爆紅新模型_Jev_解析_不聊天只做決定_快_200_倍又便宜_400_倍/reference_subtitle.vtt", "reference_kind": "manual", "reference_language": "zh-TW", "selected_source": "video_source_subtitle", "source_caption_kind": "manual", "source_caption_processing": "verbatim_direct", "youtube_overlap_clamps": [], "general_high_confidence_repairs": [], "decision_reason": "complete_source_subtitle_fast_path_no_whisper", "whisper_rejected": false, "source_profile": { "segment_count": 413, "duration_coverage": 0.941194 } }, "presegmentation_correction": { "applied": true, "correction_count": 6, "deterministic_fix_count": 0 } }