{ "text": "本集节目由VOAI绝好身疮技术支援\nVOAI提供富有情绪的高拟增台湾口音AI声音\n并提供上传图片与音档一键生成虚拟人的功能\n是最具台湾味的AI平台\n使用优惠码AILRB26\n吉祥VOAI网站以及API方案95折优惠\n大家听到的这集Podcast是透过VOAI API生成的哦\n更多资讯请下资讯栏\n每天都有一堆新的AI工具冒出来\n是不是常常不知道该从哪里开始\n别担心\n这里是AI懒人报\n每天帮你精选5个最多人讨论的AI工具影片\n每天10分钟\n让AI走进你的生活中\n让你不知不觉成为效率大师\n今天这个工具最有趣的地方是\n它可能让你再也不用半夜起来\n修坏掉的Jason\n这对我们这些每天跟AI流程\n打交道的人来说简直是福音\n好,那我们赶快开始吧\n先从一个很多人都有过的痛苦经验说起\n你花了好几个小时设计一个自动化流程\n测试的时候跑得很完美\n你觉得这次终于搞定了\n然后你把它排成\n让他每天早上自动跑\n第一天没问题\n第二天没问题\n第三天早上你打开电脑\n一堆错误\n为什么\n因为AI今天心情好\n把一个应该是数字的栏位输出成字串了\n或者他觉得自己很聪明\n在Jason里多加了一个他觉得有用的栏位\n结果你的后端完全吃不下去\n所以你不是去改流程\n你是去写验证逻辑\n写从事机制\n写各种备缘处理\n花了整个周末\n结果整个系统变得比手动作还复杂\n说真的\n我觉得这是用AI做自动化\n最让人崩溃的地方\n你明明是为了省时间\n结果花更多时间在管理AI的输出\n这个问题的根源在哪\n在于现在主流的大型语言模型\n包括Chattgbt Cloud\n还有最近很红的Astra跟Fable\n他們的設計邏輯都是預測下一個Token\n就算你在Prompt里寫得再清楚\n請只輸出JSON\n他還是有一定機率會偷偷加一個Markdown的反引號\n或者換行多一個空格\n或者把Tru寫成加引號的Tru\n這並不是模型本\n這其實是他的架構\n本質上就是機率性的\n沒辦法百分之百保證格式\n就像你請一個很會寫文章的朋友幫你填表格\n他可能會寫得很精彩\n但格式就不一定每次都對了\nJerby就是冲着这个问题来的\nJerby是Type Safe AI这家公司开发的\n他们把它归类为System 1 Model\n跟我们熟悉的大型原模型是完全不同的东西\n它不是用来聊天的\n也不是用来写文章或解释概念的\n它只做一件事\n看你给它的资料\n对照你定义的schema\n输出一个结构完全正确的决策结果\n你可以把它想象成一个非常严谨的表格填写员\n你给它什么格式\n它就保证给你什么格式\n这里最关键的技术概念叫做\nConstraint Decoding\n也就是在推理层面就做好限制\n一般的模型\n在生成每一个Token的时候\n理论上可以选任何字\n但Jav的做法是\n你传给他一个Schema\n你可以先把它想成资料格式规则\n他在生成每个Token的时候\n会直接把所有经费违反这个规则的选项\n从词汇表里遮掉\n如果这个栏位的形别是布林子\n那他在那個位置能選的詞就只有2跟false\n其他字根本不在選項裡\n這並不是說他很努力不出錯\n這是他在架構上就不可能出錯\n這兩件事的差別非常大\n前面那種你還是需要驗證\n後面這種你可以直接信任輸出\n就像你相信一個計算機算出來的結果一樣\nTypeSafe AI把這個特性叫做schema safe\n意思就是輸出永遠符合你定義的型別結構\n即對開發者來說\n这意味着你可以把原本那些验证成\n从事逻辑\n而是修正的程式码\n大部分都砍掉\n你定义好行别\n交给Jeff\n就会拿到一个结构完全正确的结果\n每次都是\n这就像你请了一个永远不会犯错的秘书\n他交给你的文件永远都是格式正确的\n你完全不用再检查一遍\nJeff提供三个主要的功能界面\n第一个叫Choice\n你给他一个输入\n再给他一个选项清单\n他直接回传最符合的那个选项\n最多可以有255个选项\n这是像一个超级精准的选择题机器\n你问他问题\n他马上给你正确答案\n第二个叫score\n他会对输入打一个分数\n同时附上confidence\n也就是信心分数\n让你知道他对这个分数有多确定\n这就像一个很专业的评论员\n不只给分数\n还会告诉你\n他打这个分数的把握有多大\n第三个叫no-o\n这个比较特别\n他会输出一个0到1之间的几率值\n代表某件事发生的可能性\n而且這個機率是經過校準的\n不是隨便說說的\n意思是如果NOLE說退款機率是0.8\n就代表10封類似的信裡\n大概有8封真的會是退款需求\n這個數字是可靠的\n這三個功能合在一起\n你就可以做很多事了\n比如說你有一個客服系統\n每天收到幾千封信\n你想要自動分類\n判斷優先順序\n偵測有沒有退款需求\n用傳統的大型語言模型\n你要寫很長的prompt\n還要解析它輸出的文字\n然後還要擔心它今天心情不好\n输出格式跑掉\n用Jeff\n你就把信的内容传进去\n定义好Schema\n他直接给你一个Jason\n类别是什么\n优先顺序\n几分\n是不是退款请求\n信心分数多少\n全部结构完整\n直接可以用\n这就像你把一大堆文件\n丢给一个超级有效率的助理\n他马上就能把所有资讯整理好\n分门别类\n让你一眼就能看懂\n社群里有人做了一个测试\n一次处理1000封信\n平均回应时间大约是200毫秒\n整批跑完费用是多少\n非常少\n因为JF的定价非常低\n输入是每百万Token四分美金\n输出是免费的\n这个价格跟主流的大型原模型\n比起来差距非常大\n简直是佛星价\n说到速度\nTypeSafe AI官方的说法是\nJF可以比传统Transformer模型\n快20到200倍\n便宜40到400倍\n我知道这种数字看起来像行销化数\n所以要说清楚\n这个速度优势\n是在特定的结构化决策任务上\n而不是在所有任务上\n你不能拿Jeff来写一篇文章\n他根本不会\n但如果你的任务是\n看这段资料\n告诉我他属于哪个类别\n那这个速度差距是真实存在的\n这就像你请一个\n短跑选手去跑马拉松\n他可能跑不完\n但如果你让他去跑100公尺\n他绝对是世界级的\n为什么Jeff可以这么快\n因为他根本不需要\n做一般语言模型做的那一事\n他不需要生成自然语言\n不需要维护复杂的上下文\n也不需要做什么深度推理\n他的工作就是看schema\n看輸入輸出\n一個符合schema的決策\n整個推理路徑短很多\n所以快很多\n這就像一個專注於單一任務的機器人\n他只做他被設計來做的事\n所以效率極高\n這個速度差距在某些使用情境上\n會讓整件事從可以用變成真正好用\n最明顯的例子是瀏覽器自動化\n現在有很多工具\n可以讓AI控制你的瀏覽器\n你給他一個目標\n他自己去點頁面\n填資料跳到下一頁\n但如果你真的試過用Astra或Fable做這件事\n你會發現他慢的很痛苦\nAgent 結一張網頁的截圖傳給模型\n然後你等2秒、3秒、有時候5秒\n模型才決定要點哪裡\n這段時間裡網頁可能已經更新了\n或者跳出一個彈窗\n然後Agent的動作就失效了\n這就像你請一個動作很慢的人\n幫你搶限量商品\n等到他反應過來\n商品早就被搶光了\n這就是為什麼瀏覽器自動化到現在\n還是很難真正穩定部署的原因之一\n延遲太高\n反應跟不上動態網頁的節奏\n但如果你把決策層換成Jeff\n延遲從秒級降到毫秒級\nAgent看到DOM結構\n馬上對照Schema\n馬上輸出下一個動作\n他不需要思考\n他就是執行\n這樣的速度才能讓瀏覽器自動化\n從一個很酷的示範\n變成一個可以真正跑在生產環境的東西\n這就像你給機器人裝上超快速反應晶片\n他能瞬間判斷並執行任務\n讓整個流程順暢無比\n好,現在來到我覺得最有意思的部分\n也就是第二個重點\nJeff的定位很特別\n他並不是要取代Astra或Fable\n他其實是要跟他們一起用\n你可以把它想像成一個團隊裡的兩個角色\n一個負責策略,一個負責執行\n我把這個架構叫做雙腦架構\n概念很直觀\n你用大型語言模型做需要深度推理的事\n用Jav做需要快速執行的事\n舉個具體的例子\n假設你在建一個自動化的軟體開發agent\n他可以看需求、寫程式、跑測試、修bug\n當一個新的功能需求進來的時候\n你先把它交給Fable或Astra\n這些模型有深度的推理能力\n他們可以分析整個專案程式碼\n理解依賴關係\n想清楚要怎麼實作\n然后产出一份详细的执行计划\n就是慢的贵的但值得\n因为这个阶段真的需要深度思考\n这就像一个公司的策略长\n负责制定整体方向和计划\n计划出来之后\nFable退场\nJeff接手\nJeff比较适合负责\n判断下一步要做什么\n输出固定格式的指令\n如果测试失败了\nJeff不需要重新思考整个架构\n他只需要看错误讯息\n对照行别定义\n修掉那個行別不符的地方\n再跑一次幾毫秒就完成\n真正寫大段程式碼還是交給Fable或Astra\n這就像公司的執行長\n負責把策略長的計劃\n分解成一個個具體任務\n並快速高效的執行\n你把昂貴的推理集中在規劃階段\n把執行交給一個高速低成本\n格式保證正確的引擎\n這樣的組合才能讓多A-系統真正規模化\n以前這種架構的問題是成本和延遲\n如果一個複雜的任務需要50個為決策\n每個決策要2秒花3分美金\n整個流程就要花1分多鐘\n1塊多美金單次執行\n這樣的單位經濟學根本沒辦法做自動化\n因為跑幾次就虧了\n但Jeff把每個為決策的成本和延遲\n都壓到幾乎可以忽略的程度\n這個架構才真正變得可行\n這就像以前你需要請好幾個顧問來做決策\n現在你只需要請一個策略顧問\n然後把執行交給一個超便宜又超快的機器人\n那實際上怎麼把JAB接近你現有的流程\n有幾個很具體的切入點\n最直接的方式\n是找出你現有AI流程裡最容易壞掉的節點\n通常就是那些讓LLM從非結構化的資料裡\n抽取資訊\n然後格式化成某個固定格式的步驟\n比如說從一封信裡抓出客戶名稱\n訂單編號\n問題類別\n然後寫進CRM\n這種步驟犯成JAB是最直接的\n你定義好schema讓介位做抽取\n格式問題直接消失\n這就像你把一個經常出錯的環節\n換成一個永遠不會出錯的自動化機器人\n瞬間解決了所有問題\n另一個很適合的使用情境是vibe coding\n就是你用AI輔助快速開發的那種模式\n現在的vibe coding流程通常是\n讓AI生成程式碼\n你去跑,跑出錯誤\n把錯誤貼回去給AI\n等它修,再跑\n這個回圈很慢,而且很煩\n因為Jeff速度過快\n对程式语法的理解是原生的\n你可以在IDE里设定一个背景监听器\n每次有编译错误或测试失败\n自动把错误丢给Jeff\n让他在背景直接修掉\n你甚至不需要手动介入\n他已经修好了\n这就像你身边有一个超级程式码修正助手\n在你写程式的时候\n他随时帮你检查并修复错误\n让你专心在创作上\n这个部分说完\n我要讲一个很重要\n但很多人在兴奋的时候会忘掉的是\n速度越快\n出错的代价越高\n当你有一个模型可以在毫秒内生成\n并执行程式码\n你根本来不及在他出错之前阻止他\n如果Jeff在自动化流程里\n有权限存取你的档案系统\n执行终端机指令\n更新资料库\n而他以200倍的速度在跑\n那出了问题的时候\n你看到的已经是结果\n不是过程\n这就像你开一台超级跑车\n速度很快\n但如果失控造成的损害\n也会非常巨大\n所以安全边界必须跟着调整\n几个原则要记住\n第一\n就算是高准确定的模型\nAI生成的程式码\n预设要当成不信任的来处理\n不要因为他很少出错\n就给他最高权限\n第二\n如果你要在本机跑Jab做自动化\nDocker容器\n把执行环境隔离起来\n给他最小必要的权限\n不要给root\n第三\n如果Jab在做浏览器自动化\n不要用你平常已经登录的浏览器设定档\n用独立的\n一次性的工作阶段\n這樣就算他誤判了某個DOM元素\n也不會動到你的生產環境\n這就像你使用任何強大的工具一樣\n必須先做好安全防護\n才能避免不必要的風險\nJeff幫你省下了寫程式碼的時間\n但這些省下來的時間\n你要拿去建更完善的自動化測試和安全驗證\n你的角色從寫程式的人\n變成設計系統架構\n和做安全審計的人\n這就像以前你需要自己造車\n現在你只需要設計好車子的功能\n和安全系統\n然後讓機器人去製造\n最後來說說我對Jeff這件事的整體看法\n過去一年多\nAI產業的主旋律\n是把模型做得越來越大\n越來越聰明\n越來越貴\n每個人都在追AGI\n都在比誰的模型推理能力更強\nExtra跟Fable在這條路上做得很好\n他們真的很厲害\n但真正在用AI做產品的人\n工程師\n獨立開發者\n小團隊\n我們不是每天都在解歷史難題\n我們大部分時間需要的是\n這個資料是不是不領值\n這個請求應該要分派給哪個服務\n這封信的優先順序\n是高還是低\n這些問題不需要深度推理\n需要的是快、準、格式正確\n這就像我們日常生活\n大部分時候我們需要的不是哲學家\n而是一個能快速精準回答問題的工具人\nJeff代表的是AI執行層的工業化\n他承認一件事\n未來的自動化不會是一個超級大腦\n慢慢把所有事情都做完\n而是一堆高度專門化的引擎分工合作\n各自在自己最擅長的部分做到最好\n這就像一個現代化的工廠\n每個環節都有專門的機器人負責\n協同合作達到最高的效率\n對那些花了無數個夜晚\n在除錯Jason格式問題的人來說\n這個東西出現的時機真的很剛好\n如果你覺得今天的內容讓你有點收穫\n那就幫我到Apple Podcast按讚追蹤\n有個五星好評吧\nFB IG Flats 收AI懶人報就找到我了\n你的支持是我持續優化內容的最大動力\n最近的AI發展依然快的讓我害怕\n不想出過明天再繼續收聽AI懶人報囉\n我是湯懶懶\n我們明天見\n拜拜", "language": "multilingual", "duration": 918.000938, "segments": [ { "start": 0.0, "end": 2.7, "text": "本集节目由VOAI绝好身疮技术支援", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 2.7, "end": 6.44, "text": "VOAI提供富有情绪的高拟增台湾口音AI声音", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 6.44, "end": 10.04, "text": "并提供上传图片与音档一键生成虚拟人的功能", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 10.04, "end": 12.04, "text": "是最具台湾味的AI平台", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 12.04, "end": 14.8, "text": "使用优惠码AILRB26", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 14.8, "end": 19.02, "text": "吉祥VOAI网站以及API方案95折优惠", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 19.02, "end": 23.52, "text": "大家听到的这集Podcast是透过VOAI API生成的哦", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 23.52, "end": 25.52, "text": "更多资讯请下资讯栏", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 26.92, "end": 29.97, "text": "每天都有一堆新的AI工具冒出来", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 29.97, "end": 32.59, "text": "是不是常常不知道该从哪里开始", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 32.59, "end": 33.57, "text": "别担心", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 33.57, "end": 35.31, "text": "这里是AI懒人报", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 35.31, "end": 39.23, "text": "每天帮你精选5个最多人讨论的AI工具影片", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 39.23, "end": 40.51, "text": "每天10分钟", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 40.51, "end": 42.67, "text": "让AI走进你的生活中", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 42.67, "end": 45.13, "text": "让你不知不觉成为效率大师", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 45.13, "end": 47.29, "text": "今天这个工具最有趣的地方是", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 47.29, "end": 50.25, "text": "它可能让你再也不用半夜起来", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 50.25, "end": 51.29, "text": "修坏掉的Jason", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 51.29, "end": 53.61, "text": "这对我们这些每天跟AI流程", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 53.61, "end": 55.47, "text": "打交道的人来说简直是福音", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 55.47, "end": 56.83, "text": "好,那我们赶快开始吧", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 56.83, "end": 60.19, "text": "先从一个很多人都有过的痛苦经验说起", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 60.19, "end": 63.15, "text": "你花了好几个小时设计一个自动化流程", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 63.15, "end": 64.71, "text": "测试的时候跑得很完美", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 64.71, "end": 66.43, "text": "你觉得这次终于搞定了", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 66.43, "end": 67.43, "text": "然后你把它排成", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 67.43, "end": 69.27, "text": "让他每天早上自动跑", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 69.27, "end": 70.63, "text": "第一天没问题", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 70.63, "end": 71.39, "text": "第二天没问题", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 71.39, "end": 72.91, "text": "第三天早上你打开电脑", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 72.91, "end": 73.91, "text": "一堆错误", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 73.91, "end": 74.39, "text": "为什么", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 74.39, "end": 75.79, "text": "因为AI今天心情好", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 75.79, "end": 79.31, "text": "把一个应该是数字的栏位输出成字串了", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 79.31, "end": 81.11, "text": "或者他觉得自己很聪明", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 81.11, "end": 83.73, "text": "在Jason里多加了一个他觉得有用的栏位", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 83.73, "end": 86.07, "text": "结果你的后端完全吃不下去", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 86.07, "end": 87.97, "text": "所以你不是去改流程", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 87.97, "end": 89.35, "text": "你是去写验证逻辑", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 89.35, "end": 90.37, "text": "写从事机制", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 90.37, "end": 91.61, "text": "写各种备缘处理", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 91.61, "end": 92.81, "text": "花了整个周末", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 92.81, "end": 95.55, "text": "结果整个系统变得比手动作还复杂", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 95.55, "end": 96.29, "text": "说真的", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 96.29, "end": 98.11, "text": "我觉得这是用AI做自动化", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 98.11, "end": 99.59, "text": "最让人崩溃的地方", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 99.59, "end": 100.97, "text": "你明明是为了省时间", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 100.97, "end": 103.51, "text": "结果花更多时间在管理AI的输出", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 103.51, "end": 105.29, "text": "这个问题的根源在哪", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 105.29, "end": 108.23, "text": "在于现在主流的大型语言模型", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 108.23, "end": 110.07, "text": "包括Chattgbt Cloud", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 110.07, "end": 112.93, "text": "还有最近很红的Astra跟Fable", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 112.93, "end": 116.51, "text": "他們的設計邏輯都是預測下一個Token", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 116.51, "end": 118.89, "text": "就算你在Prompt里寫得再清楚", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 118.89, "end": 120.41, "text": "請只輸出JSON", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 120.41, "end": 124.05, "text": "他還是有一定機率會偷偷加一個Markdown的反引號", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 124.05, "end": 126.09, "text": "或者換行多一個空格", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 126.09, "end": 128.51, "text": "或者把Tru寫成加引號的Tru", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 128.51, "end": 130.35, "text": "這並不是模型本", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 130.35, "end": 131.75, "text": "這其實是他的架構", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 131.75, "end": 133.35, "text": "本質上就是機率性的", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 133.35, "end": 135.19, "text": "沒辦法百分之百保證格式", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 135.19, "end": 138.65, "text": "就像你請一個很會寫文章的朋友幫你填表格", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 138.65, "end": 140.19, "text": "他可能會寫得很精彩", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 140.19, "end": 142.31, "text": "但格式就不一定每次都對了", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 142.31, "end": 144.61, "text": "Jerby就是冲着这个问题来的", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 144.61, "end": 147.73, "text": "Jerby是Type Safe AI这家公司开发的", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 147.73, "end": 150.13, "text": "他们把它归类为System 1 Model", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 150.13, "end": 153.37, "text": "跟我们熟悉的大型原模型是完全不同的东西", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 153.37, "end": 155.19, "text": "它不是用来聊天的", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 155.19, "end": 158.17, "text": "也不是用来写文章或解释概念的", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 158.17, "end": 159.15, "text": "它只做一件事", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 159.15, "end": 160.33, "text": "看你给它的资料", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 160.33, "end": 162.07, "text": "对照你定义的schema", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 162.07, "end": 165.29, "text": "输出一个结构完全正确的决策结果", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 165.29, "end": 168.85, "text": "你可以把它想象成一个非常严谨的表格填写员", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 168.85, "end": 170.13, "text": "你给它什么格式", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 170.13, "end": 171.71, "text": "它就保证给你什么格式", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 171.71, "end": 174.29, "text": "这里最关键的技术概念叫做", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 174.29, "end": 175.45, "text": "Constraint Decoding", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 175.45, "end": 177.67, "text": "也就是在推理层面就做好限制", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 177.67, "end": 179.21, "text": "一般的模型", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 179.21, "end": 181.21, "text": "在生成每一个Token的时候", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 181.21, "end": 182.95, "text": "理论上可以选任何字", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 182.95, "end": 184.95, "text": "但Jav的做法是", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 184.95, "end": 187.25, "text": "你传给他一个Schema", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 187.25, "end": 190.03, "text": "你可以先把它想成资料格式规则", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 190.03, "end": 192.09, "text": "他在生成每个Token的时候", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 192.09, "end": 196.03, "text": "会直接把所有经费违反这个规则的选项", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 196.03, "end": 197.59, "text": "从词汇表里遮掉", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 197.59, "end": 201.21, "text": "如果这个栏位的形别是布林子", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 201.21, "end": 204.87, "text": "那他在那個位置能選的詞就只有2跟false", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 204.87, "end": 206.83, "text": "其他字根本不在選項裡", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 206.83, "end": 208.89, "text": "這並不是說他很努力不出錯", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 208.89, "end": 211.25, "text": "這是他在架構上就不可能出錯", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 211.25, "end": 213.45, "text": "這兩件事的差別非常大", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 213.45, "end": 215.49, "text": "前面那種你還是需要驗證", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 215.49, "end": 217.79, "text": "後面這種你可以直接信任輸出", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 217.79, "end": 220.37, "text": "就像你相信一個計算機算出來的結果一樣", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 220.37, "end": 223.21, "text": "TypeSafe AI把這個特性叫做schema safe", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 223.21, "end": 226.77, "text": "意思就是輸出永遠符合你定義的型別結構", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 226.77, "end": 228.99, "text": "即對開發者來說", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 228.99, "end": 231.63, "text": "这意味着你可以把原本那些验证成", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 231.63, "end": 232.67, "text": "从事逻辑", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 232.67, "end": 234.57, "text": "而是修正的程式码", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 234.57, "end": 235.77, "text": "大部分都砍掉", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 235.77, "end": 236.97, "text": "你定义好行别", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 236.97, "end": 237.85, "text": "交给Jeff", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 237.85, "end": 240.59, "text": "就会拿到一个结构完全正确的结果", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 240.59, "end": 241.21, "text": "每次都是", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 241.21, "end": 244.25, "text": "这就像你请了一个永远不会犯错的秘书", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 244.25, "end": 246.95, "text": "他交给你的文件永远都是格式正确的", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 246.95, "end": 248.61, "text": "你完全不用再检查一遍", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 248.61, "end": 250.75, "text": "Jeff提供三个主要的功能界面", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 250.75, "end": 252.25, "text": "第一个叫Choice", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 252.25, "end": 253.51, "text": "你给他一个输入", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 253.51, "end": 255.21, "text": "再给他一个选项清单", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 255.21, "end": 257.51, "text": "他直接回传最符合的那个选项", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 257.51, "end": 259.25, "text": "最多可以有255个选项", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 259.25, "end": 261.47, "text": "这是像一个超级精准的选择题机器", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 261.47, "end": 262.35, "text": "你问他问题", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 262.35, "end": 264.43, "text": "他马上给你正确答案", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 264.43, "end": 265.49, "text": "第二个叫score", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 265.49, "end": 267.59, "text": "他会对输入打一个分数", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 267.59, "end": 269.17, "text": "同时附上confidence", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 269.17, "end": 270.41, "text": "也就是信心分数", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 270.41, "end": 272.81, "text": "让你知道他对这个分数有多确定", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 272.81, "end": 274.59, "text": "这就像一个很专业的评论员", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 274.59, "end": 275.49, "text": "不只给分数", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 275.49, "end": 276.29, "text": "还会告诉你", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 276.29, "end": 278.51, "text": "他打这个分数的把握有多大", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 278.51, "end": 280.35, "text": "第三个叫no-o", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 280.35, "end": 281.17, "text": "这个比较特别", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 281.17, "end": 284.05, "text": "他会输出一个0到1之间的几率值", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 284.05, "end": 286.21, "text": "代表某件事发生的可能性", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 286.21, "end": 288.45, "text": "而且這個機率是經過校準的", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 288.45, "end": 289.97, "text": "不是隨便說說的", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 290.21, "end": 293.15, "text": "意思是如果NOLE說退款機率是0.8", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 293.15, "end": 294.85, "text": "就代表10封類似的信裡", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 294.85, "end": 297.25, "text": "大概有8封真的會是退款需求", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 297.25, "end": 298.51, "text": "這個數字是可靠的", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 298.51, "end": 299.91, "text": "這三個功能合在一起", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 299.91, "end": 301.25, "text": "你就可以做很多事了", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 301.57, "end": 303.31, "text": "比如說你有一個客服系統", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 303.31, "end": 304.77, "text": "每天收到幾千封信", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 304.77, "end": 305.97, "text": "你想要自動分類", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 305.97, "end": 307.15, "text": "判斷優先順序", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 307.15, "end": 308.67, "text": "偵測有沒有退款需求", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 308.67, "end": 310.51, "text": "用傳統的大型語言模型", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 310.51, "end": 311.81, "text": "你要寫很長的prompt", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 311.81, "end": 313.67, "text": "還要解析它輸出的文字", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 313.67, "end": 315.47, "text": "然後還要擔心它今天心情不好", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 315.47, "end": 316.53, "text": "输出格式跑掉", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 316.53, "end": 317.41, "text": "用Jeff", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 317.41, "end": 319.19, "text": "你就把信的内容传进去", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 319.19, "end": 320.19, "text": "定义好Schema", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 320.19, "end": 321.51, "text": "他直接给你一个Jason", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 321.51, "end": 322.61, "text": "类别是什么", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 322.61, "end": 323.73, "text": "优先顺序", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 323.73, "end": 324.37, "text": "几分", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 324.37, "end": 325.85, "text": "是不是退款请求", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 325.85, "end": 327.11, "text": "信心分数多少", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 327.11, "end": 328.35, "text": "全部结构完整", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 328.35, "end": 329.15, "text": "直接可以用", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 329.15, "end": 331.29, "text": "这就像你把一大堆文件", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 331.29, "end": 333.07, "text": "丢给一个超级有效率的助理", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 333.07, "end": 335.25, "text": "他马上就能把所有资讯整理好", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 335.25, "end": 335.89, "text": "分门别类", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 335.89, "end": 337.01, "text": "让你一眼就能看懂", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 337.01, "end": 338.65, "text": "社群里有人做了一个测试", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 338.65, "end": 339.99, "text": "一次处理1000封信", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 339.99, "end": 342.43, "text": "平均回应时间大约是200毫秒", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 342.43, "end": 344.19, "text": "整批跑完费用是多少", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 344.19, "end": 345.45, "text": "非常少", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 345.45, "end": 347.15, "text": "因为JF的定价非常低", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 347.15, "end": 349.59, "text": "输入是每百万Token四分美金", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 349.59, "end": 350.75, "text": "输出是免费的", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 350.75, "end": 353.23, "text": "这个价格跟主流的大型原模型", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 353.23, "end": 354.47, "text": "比起来差距非常大", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 354.47, "end": 355.79, "text": "简直是佛星价", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 355.79, "end": 357.39, "text": "说到速度", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 357.39, "end": 359.47, "text": "TypeSafe AI官方的说法是", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 359.47, "end": 362.49, "text": "JF可以比传统Transformer模型", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 362.49, "end": 364.21, "text": "快20到200倍", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 364.21, "end": 366.29, "text": "便宜40到400倍", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 366.29, "end": 368.99, "text": "我知道这种数字看起来像行销化数", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 368.99, "end": 370.01, "text": "所以要说清楚", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 370.01, "end": 370.97, "text": "这个速度优势", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 370.97, "end": 373.45, "text": "是在特定的结构化决策任务上", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 373.45, "end": 374.95, "text": "而不是在所有任务上", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 374.95, "end": 376.91, "text": "你不能拿Jeff来写一篇文章", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 376.91, "end": 377.61, "text": "他根本不会", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 377.61, "end": 379.47, "text": "但如果你的任务是", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 379.47, "end": 380.55, "text": "看这段资料", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 380.55, "end": 382.49, "text": "告诉我他属于哪个类别", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 382.49, "end": 385.19, "text": "那这个速度差距是真实存在的", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 385.19, "end": 386.35, "text": "这就像你请一个", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 386.35, "end": 388.09, "text": "短跑选手去跑马拉松", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 388.09, "end": 389.33, "text": "他可能跑不完", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 389.33, "end": 392.43, "text": "但如果你让他去跑100公尺", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 392.43, "end": 393.79, "text": "他绝对是世界级的", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 393.79, "end": 395.25, "text": "为什么Jeff可以这么快", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 395.25, "end": 396.63, "text": "因为他根本不需要", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 396.63, "end": 398.33, "text": "做一般语言模型做的那一事", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 398.33, "end": 400.05, "text": "他不需要生成自然语言", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 400.05, "end": 402.39, "text": "不需要维护复杂的上下文", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 402.39, "end": 404.23, "text": "也不需要做什么深度推理", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 404.23, "end": 405.93, "text": "他的工作就是看schema", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 405.93, "end": 407.77, "text": "看輸入輸出", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 407.77, "end": 409.93, "text": "一個符合schema的決策", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 409.93, "end": 411.43, "text": "整個推理路徑短很多", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 411.43, "end": 412.57, "text": "所以快很多", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 412.57, "end": 415.17, "text": "這就像一個專注於單一任務的機器人", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 415.17, "end": 417.63, "text": "他只做他被設計來做的事", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 417.63, "end": 419.03, "text": "所以效率極高", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 419.03, "end": 422.03, "text": "這個速度差距在某些使用情境上", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 422.03, "end": 425.43, "text": "會讓整件事從可以用變成真正好用", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 425.43, "end": 427.99, "text": "最明顯的例子是瀏覽器自動化", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 427.99, "end": 429.23, "text": "現在有很多工具", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 429.23, "end": 431.17, "text": "可以讓AI控制你的瀏覽器", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 431.17, "end": 432.13, "text": "你給他一個目標", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 432.13, "end": 433.77, "text": "他自己去點頁面", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 433.77, "end": 435.77, "text": "填資料跳到下一頁", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 435.77, "end": 438.61, "text": "但如果你真的試過用Astra或Fable做這件事", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 438.61, "end": 440.13, "text": "你會發現他慢的很痛苦", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 440.13, "end": 443.37, "text": "Agent 結一張網頁的截圖傳給模型", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 443.37, "end": 445.97, "text": "然後你等2秒、3秒、有時候5秒", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 445.97, "end": 447.93, "text": "模型才決定要點哪裡", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 447.93, "end": 450.29, "text": "這段時間裡網頁可能已經更新了", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 450.29, "end": 451.77, "text": "或者跳出一個彈窗", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 451.77, "end": 453.89, "text": "然後Agent的動作就失效了", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 453.89, "end": 455.97, "text": "這就像你請一個動作很慢的人", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 455.97, "end": 457.49, "text": "幫你搶限量商品", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 457.49, "end": 458.61, "text": "等到他反應過來", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 458.61, "end": 460.69, "text": "商品早就被搶光了", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 460.69, "end": 462.97, "text": "這就是為什麼瀏覽器自動化到現在", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 462.97, "end": 465.67, "text": "還是很難真正穩定部署的原因之一", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 465.67, "end": 466.73, "text": "延遲太高", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 466.73, "end": 469.31, "text": "反應跟不上動態網頁的節奏", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 469.31, "end": 471.91, "text": "但如果你把決策層換成Jeff", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 471.91, "end": 474.57, "text": "延遲從秒級降到毫秒級", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 474.57, "end": 476.27, "text": "Agent看到DOM結構", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 476.27, "end": 477.87, "text": "馬上對照Schema", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 477.87, "end": 479.81, "text": "馬上輸出下一個動作", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 479.81, "end": 480.73, "text": "他不需要思考", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 480.73, "end": 481.67, "text": "他就是執行", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 481.67, "end": 484.07, "text": "這樣的速度才能讓瀏覽器自動化", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 484.07, "end": 485.83, "text": "從一個很酷的示範", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 485.83, "end": 489.57, "text": "變成一個可以真正跑在生產環境的東西", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 489.57, "end": 492.63, "text": "這就像你給機器人裝上超快速反應晶片", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 492.63, "end": 495.49, "text": "他能瞬間判斷並執行任務", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 495.49, "end": 497.47, "text": "讓整個流程順暢無比", "language": "zh", "chunk": 0, "local_language_run": 1, "language_run": 0 }, { "start": 497.954, "end": 499.954, "text": "好,現在來到我覺得最有意思的部分", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 499.954, "end": 501.254, "text": "也就是第二個重點", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 501.254, "end": 502.754, "text": "Jeff的定位很特別", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 502.754, "end": 505.354, "text": "他並不是要取代Astra或Fable", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 505.354, "end": 506.954, "text": "他其實是要跟他們一起用", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 507.554, "end": 509.854, "text": "你可以把它想像成一個團隊裡的兩個角色", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 509.854, "end": 511.954, "text": "一個負責策略,一個負責執行", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 511.954, "end": 514.114, "text": "我把這個架構叫做雙腦架構", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 514.114, "end": 515.314, "text": "概念很直觀", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 515.314, "end": 518.194, "text": "你用大型語言模型做需要深度推理的事", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 518.194, "end": 520.634, "text": "用Jav做需要快速執行的事", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 520.634, "end": 521.954, "text": "舉個具體的例子", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 521.954, "end": 524.914, "text": "假設你在建一個自動化的軟體開發agent", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 524.914, "end": 528.834, "text": "他可以看需求、寫程式、跑測試、修bug", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 528.834, "end": 531.074, "text": "當一個新的功能需求進來的時候", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 531.074, "end": 533.394, "text": "你先把它交給Fable或Astra", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 533.714, "end": 535.794, "text": "這些模型有深度的推理能力", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 535.794, "end": 538.594, "text": "他們可以分析整個專案程式碼", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 538.594, "end": 540.114, "text": "理解依賴關係", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 540.114, "end": 541.634, "text": "想清楚要怎麼實作", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 541.634, "end": 544.534, "text": "然后产出一份详细的执行计划", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 544.534, "end": 546.474, "text": "就是慢的贵的但值得", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 546.474, "end": 549.374, "text": "因为这个阶段真的需要深度思考", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 549.934, "end": 552.674, "text": "这就像一个公司的策略长", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 552.674, "end": 555.174, "text": "负责制定整体方向和计划", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 555.374, "end": 556.374, "text": "计划出来之后", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 556.374, "end": 557.194, "text": "Fable退场", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 557.194, "end": 558.174, "text": "Jeff接手", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 558.394, "end": 560.234, "text": "Jeff比较适合负责", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 560.234, "end": 561.994, "text": "判断下一步要做什么", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 562.094, "end": 563.934, "text": "输出固定格式的指令", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 564.034, "end": 565.234, "text": "如果测试失败了", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 565.234, "end": 568.234, "text": "Jeff不需要重新思考整个架构", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 568.334, "end": 570.234, "text": "他只需要看错误讯息", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 570.334, "end": 571.634, "text": "对照行别定义", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 571.634, "end": 573.934, "text": "修掉那個行別不符的地方", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 573.934, "end": 576.494, "text": "再跑一次幾毫秒就完成", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 577.134, "end": 580.594, "text": "真正寫大段程式碼還是交給Fable或Astra", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 580.594, "end": 582.274, "text": "這就像公司的執行長", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 582.274, "end": 583.994, "text": "負責把策略長的計劃", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 583.994, "end": 586.234, "text": "分解成一個個具體任務", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 586.234, "end": 588.134, "text": "並快速高效的執行", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 588.734, "end": 591.034, "text": "你把昂貴的推理集中在規劃階段", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 591.034, "end": 593.434, "text": "把執行交給一個高速低成本", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 593.434, "end": 595.334, "text": "格式保證正確的引擎", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 595.574, "end": 599.634, "text": "這樣的組合才能讓多A-系統真正規模化", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 599.634, "end": 603.034, "text": "以前這種架構的問題是成本和延遲", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 603.034, "end": 606.274, "text": "如果一個複雜的任務需要50個為決策", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 606.274, "end": 608.394, "text": "每個決策要2秒花3分美金", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 608.394, "end": 610.094, "text": "整個流程就要花1分多鐘", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 610.094, "end": 612.034, "text": "1塊多美金單次執行", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 612.034, "end": 615.174, "text": "這樣的單位經濟學根本沒辦法做自動化", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 615.174, "end": 616.874, "text": "因為跑幾次就虧了", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 616.874, "end": 620.034, "text": "但Jeff把每個為決策的成本和延遲", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 620.034, "end": 621.974, "text": "都壓到幾乎可以忽略的程度", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 621.974, "end": 623.974, "text": "這個架構才真正變得可行", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 623.974, "end": 627.294, "text": "這就像以前你需要請好幾個顧問來做決策", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 627.294, "end": 629.374, "text": "現在你只需要請一個策略顧問", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 629.374, "end": 632.874, "text": "然後把執行交給一個超便宜又超快的機器人", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 633.274, "end": 635.974, "text": "那實際上怎麼把JAB接近你現有的流程", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 635.974, "end": 637.774, "text": "有幾個很具體的切入點", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 637.774, "end": 638.874, "text": "最直接的方式", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 638.874, "end": 642.574, "text": "是找出你現有AI流程裡最容易壞掉的節點", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 642.774, "end": 646.274, "text": "通常就是那些讓LLM從非結構化的資料裡", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 646.274, "end": 647.274, "text": "抽取資訊", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 647.274, "end": 649.874, "text": "然後格式化成某個固定格式的步驟", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 649.874, "end": 652.574, "text": "比如說從一封信裡抓出客戶名稱", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 652.574, "end": 653.574, "text": "訂單編號", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 653.574, "end": 654.674, "text": "問題類別", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 654.674, "end": 656.374, "text": "然後寫進CRM", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 656.874, "end": 659.274, "text": "這種步驟犯成JAB是最直接的", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 659.274, "end": 661.614, "text": "你定義好schema讓介位做抽取", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 661.614, "end": 663.174, "text": "格式問題直接消失", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 663.174, "end": 665.974, "text": "這就像你把一個經常出錯的環節", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 665.974, "end": 668.874, "text": "換成一個永遠不會出錯的自動化機器人", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 668.874, "end": 670.514, "text": "瞬間解決了所有問題", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 670.514, "end": 673.514, "text": "另一個很適合的使用情境是vibe coding", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 673.514, "end": 676.634, "text": "就是你用AI輔助快速開發的那種模式", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 676.634, "end": 678.734, "text": "現在的vibe coding流程通常是", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 678.734, "end": 680.734, "text": "讓AI生成程式碼", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 680.734, "end": 682.274, "text": "你去跑,跑出錯誤", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 682.274, "end": 683.874, "text": "把錯誤貼回去給AI", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 683.874, "end": 685.234, "text": "等它修,再跑", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 685.234, "end": 687.134, "text": "這個回圈很慢,而且很煩", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 687.134, "end": 689.134, "text": "因為Jeff速度過快", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 689.274, "end": 691.574, "text": "对程式语法的理解是原生的", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 691.574, "end": 694.714, "text": "你可以在IDE里设定一个背景监听器", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 694.714, "end": 697.174, "text": "每次有编译错误或测试失败", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 697.174, "end": 698.934, "text": "自动把错误丢给Jeff", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 698.934, "end": 701.074, "text": "让他在背景直接修掉", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 701.074, "end": 702.434, "text": "你甚至不需要手动介入", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 702.434, "end": 703.414, "text": "他已经修好了", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 703.414, "end": 706.514, "text": "这就像你身边有一个超级程式码修正助手", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 706.514, "end": 707.574, "text": "在你写程式的时候", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 707.574, "end": 709.774, "text": "他随时帮你检查并修复错误", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 709.774, "end": 712.534, "text": "让你专心在创作上", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 712.534, "end": 713.574, "text": "这个部分说完", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 713.574, "end": 714.574, "text": "我要讲一个很重要", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 714.574, "end": 717.134, "text": "但很多人在兴奋的时候会忘掉的是", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 717.134, "end": 718.174, "text": "速度越快", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 718.174, "end": 719.914, "text": "出错的代价越高", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 719.914, "end": 722.514, "text": "当你有一个模型可以在毫秒内生成", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 722.514, "end": 723.874, "text": "并执行程式码", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 723.874, "end": 727.114, "text": "你根本来不及在他出错之前阻止他", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 727.114, "end": 729.114, "text": "如果Jeff在自动化流程里", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 729.114, "end": 731.114, "text": "有权限存取你的档案系统", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 731.114, "end": 732.454, "text": "执行终端机指令", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 732.454, "end": 733.514, "text": "更新资料库", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 733.514, "end": 735.634, "text": "而他以200倍的速度在跑", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 735.634, "end": 736.734, "text": "那出了问题的时候", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 736.734, "end": 738.234, "text": "你看到的已经是结果", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 738.234, "end": 739.434, "text": "不是过程", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 739.434, "end": 741.334, "text": "这就像你开一台超级跑车", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 741.334, "end": 742.434, "text": "速度很快", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 742.434, "end": 744.374, "text": "但如果失控造成的损害", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 744.374, "end": 745.634, "text": "也会非常巨大", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 745.634, "end": 748.034, "text": "所以安全边界必须跟着调整", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 748.034, "end": 749.334, "text": "几个原则要记住", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 749.334, "end": 749.934, "text": "第一", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 749.934, "end": 751.594, "text": "就算是高准确定的模型", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 751.594, "end": 753.034, "text": "AI生成的程式码", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 753.034, "end": 755.734, "text": "预设要当成不信任的来处理", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 755.734, "end": 757.334, "text": "不要因为他很少出错", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 757.334, "end": 758.534, "text": "就给他最高权限", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 759.174, "end": 759.874, "text": "第二", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 759.874, "end": 762.174, "text": "如果你要在本机跑Jab做自动化", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 762.674, "end": 763.574, "text": "Docker容器", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 763.574, "end": 765.494, "text": "把执行环境隔离起来", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 765.494, "end": 767.294, "text": "给他最小必要的权限", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 767.294, "end": 768.294, "text": "不要给root", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 768.794, "end": 769.394, "text": "第三", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 769.394, "end": 771.834, "text": "如果Jab在做浏览器自动化", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 771.834, "end": 774.894, "text": "不要用你平常已经登录的浏览器设定档", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 774.894, "end": 776.134, "text": "用独立的", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 776.134, "end": 777.794, "text": "一次性的工作阶段", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 777.794, "end": 780.434, "text": "這樣就算他誤判了某個DOM元素", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 780.434, "end": 782.694, "text": "也不會動到你的生產環境", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 782.694, "end": 785.294, "text": "這就像你使用任何強大的工具一樣", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 785.294, "end": 786.834, "text": "必須先做好安全防護", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 786.834, "end": 788.694, "text": "才能避免不必要的風險", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 788.694, "end": 791.254, "text": "Jeff幫你省下了寫程式碼的時間", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 791.254, "end": 792.634, "text": "但這些省下來的時間", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 792.634, "end": 796.234, "text": "你要拿去建更完善的自動化測試和安全驗證", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 796.234, "end": 798.194, "text": "你的角色從寫程式的人", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 798.194, "end": 799.794, "text": "變成設計系統架構", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 799.794, "end": 801.654, "text": "和做安全審計的人", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 801.654, "end": 804.034, "text": "這就像以前你需要自己造車", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 804.034, "end": 806.494, "text": "現在你只需要設計好車子的功能", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 806.494, "end": 807.694, "text": "和安全系統", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 807.694, "end": 809.354, "text": "然後讓機器人去製造", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 809.934, "end": 812.834, "text": "最後來說說我對Jeff這件事的整體看法", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 813.234, "end": 814.234, "text": "過去一年多", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 814.234, "end": 815.934, "text": "AI產業的主旋律", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 816.034, "end": 817.934, "text": "是把模型做得越來越大", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 818.034, "end": 819.034, "text": "越來越聰明", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 819.034, "end": 820.034, "text": "越來越貴", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 820.334, "end": 821.934, "text": "每個人都在追AGI", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 821.934, "end": 824.954, "text": "都在比誰的模型推理能力更強", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 825.554, "end": 828.054, "text": "Extra跟Fable在這條路上做得很好", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 828.054, "end": 829.054, "text": "他們真的很厲害", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 829.254, "end": 831.994, "text": "但真正在用AI做產品的人", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 831.994, "end": 833.054, "text": "工程師", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 833.054, "end": 834.354, "text": "獨立開發者", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 834.354, "end": 835.394, "text": "小團隊", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 835.394, "end": 837.954, "text": "我們不是每天都在解歷史難題", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 838.334, "end": 840.394, "text": "我們大部分時間需要的是", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 840.394, "end": 842.534, "text": "這個資料是不是不領值", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 842.534, "end": 845.694, "text": "這個請求應該要分派給哪個服務", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 845.694, "end": 847.434, "text": "這封信的優先順序", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 847.434, "end": 848.694, "text": "是高還是低", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 848.694, "end": 851.034, "text": "這些問題不需要深度推理", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 851.034, "end": 854.694, "text": "需要的是快、準、格式正確", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 854.694, "end": 856.194, "text": "這就像我們日常生活", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 856.194, "end": 858.334, "text": "大部分時候我們需要的不是哲學家", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 858.334, "end": 861.294, "text": "而是一個能快速精準回答問題的工具人", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 861.294, "end": 864.054, "text": "Jeff代表的是AI執行層的工業化", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 864.054, "end": 865.454, "text": "他承認一件事", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 865.454, "end": 867.694, "text": "未來的自動化不會是一個超級大腦", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 867.694, "end": 868.994, "text": "慢慢把所有事情都做完", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 868.994, "end": 871.954, "text": "而是一堆高度專門化的引擎分工合作", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 871.954, "end": 874.794, "text": "各自在自己最擅長的部分做到最好", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 874.794, "end": 877.494, "text": "這就像一個現代化的工廠", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 877.494, "end": 879.694, "text": "每個環節都有專門的機器人負責", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 879.694, "end": 882.014, "text": "協同合作達到最高的效率", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 882.014, "end": 884.014, "text": "對那些花了無數個夜晚", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 884.014, "end": 886.514, "text": "在除錯Jason格式問題的人來說", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 886.514, "end": 888.554, "text": "這個東西出現的時機真的很剛好", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 888.554, "end": 890.714, "text": "如果你覺得今天的內容讓你有點收穫", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 890.714, "end": 892.914, "text": "那就幫我到Apple Podcast按讚追蹤", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 892.914, "end": 894.214, "text": "有個五星好評吧", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 894.754, "end": 898.714, "text": "FB IG Flats 收AI懶人報就找到我了", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 898.914, "end": 902.214, "text": "你的支持是我持續優化內容的最大動力", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 902.514, "end": 905.214, "text": "最近的AI發展依然快的讓我害怕", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 905.314, "end": 908.714, "text": "不想出過明天再繼續收聽AI懶人報囉", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 908.714, "end": 909.714, "text": "我是湯懶懶", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 909.714, "end": 910.714, "text": "我們明天見", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 910.714, "end": 911.129, "text": "拜拜", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 } ], "segmented_multilingual_transcription": { "status": "STABLE_SEGMENTED_MULTILINGUAL_OK", "created_at": "2026-09-18T16:24:17", "manifest": "/Users/sagawa/AI_WORK/video_notes/url/20260918_160431__EP404_-_Jev_橫空出世_前OpenAI研究員打造專注結構化決策的工具_速度_200x_徹底解決_LLM_格式不穩定的致命缺陷/segmented_work/chunk_manifest.json", "audio": "/Users/sagawa/AI_WORK/video_notes/url/20260918_160431__EP404_-_Jev_橫空出世_前OpenAI研究員打造專注結構化決策的工具_速度_200x_徹底解決_LLM_格式不穩定的致命缺陷/audio_16k_mono.wav", "audio_sha256": "76a6cf8b6fceb57f55143cc608fc3af2a4171d1e33ee1dff72435f685eece2cb", "audio_duration": 918.000938, "model": "/Users/sagawa/AI_WORK/cache/huggingface/hub/models--mlx-community--whisper-large-v3-turbo/snapshots/a4aaeec0636e6fef84abdcbe3544cb2bf7e9f6fb", "language_profile": "/Users/sagawa/AI_WORK/video_notes/url/20260918_160431__EP404_-_Jev_橫空出世_前OpenAI研究員打造專注結構化決策的工具_速度_200x_徹底解決_LLM_格式不穩定的致命缺陷/multilingual_work/multilingual_language_profile.json", "dominant_languages": [ "zh" ], "suppressed_languages": [ "en" ], "chunk_count": 2, "completed_chunks": 2, "resumed_chunks": 0, "merged_segment_count": 450, "detected_languages": [ "zh" ], "transcript_languages": [ "zh" ], "missing_languages": [], "unexpected_languages": [], "language_sequence": [ "zh" ], "speech_region_seconds": 909.791, "speech_covered_seconds": 898.2250000000004, "speech_coverage_ratio": 0.9872871901348774, "tail_repeat_run": 1, "tail_repeat_text": "本集节目由VOAI绝好身疮技术支援", "chunk_needs_review": [], "qc_ok": true, "qc_errors": [], "complete_review_artifact": true } }