{"text": "每周拆解一個頂級開源項目,這周是DeepTutor,港大開源的AI私人導師工作台。各位觀眾老爺們,肯定都有類似經歷,收藏夾裡躺著很多高質量學習資料,技術教程書,學習資料,考試資料等等。收藏時很熱血,真到自學就卡住了,沒人按你的知識儲備和學習目標排路線,沒人抽查重點,看懂了也很容易以為自己真會了,學過的歷史也串不成圖譜,錯題和薄弱點很快散掉。所以DeepTutor真正要做的,是先根據你的基礎和目標定制學習路線,學的過程中,他用閃卡和測驗不斷抽查重點,學完之後再把錯題和薄弱點沉淀到金屬記憶裡,這三件事合起來才像一個真的私人導師。今天就拿兩個高新AI Agent學習資料實測,Learn Cloud Code,7萬多星,偏工程實戰,適合跟著手挫Agent,AI Agent Book,3萬多星,偏系統原理,一個練手,一個補底層,正好一起學。接下來把兩份資料導進DeepTutor,新建知識庫,填好名字,再把兩個項目的Markdown一起選進去。等索引完成,同一個知識庫裡,兩邊文件都會變成ready,後面的路線,出題和學習,就都從這批真實資料裡來。資料導入之後,接著就要自定義學習路線,關鍵在於先告訴他我是誰,基礎怎樣,這次到底想學成什麼。一類目標是實戰手挫開發Agent Harness,另一類是從原理層面系統理解Agent的底層架構,目標不同,路線也就不同。手挫Agent的開發學習路線的大綱,從最小Agent的loop循環開始,先跑出第一版Agent,再逐步補齊工程細節。Agent的原理學習路線這邊,會先從認識Agent的架構開始,再順著上下文,工具,記憶和評估,把底層框架串起來。拿到教材之後就可以進入學習了,每一張都會被拆成可讀的概念,例子和練習,比如第一張先看目錄,再進入最小Agent的loop和工具調用機制。學習過程中,內容不會只攤開給你看,Flashcards會逼你停下來回憶核心概念,避免看懂了就以為自己會了。請接著QuickCheck會圍繞Stop Reason,工具調用和Agent loop直接出題,讓你先判斷,再揭示答案。如果還想考,自己測驗快會繼續追問實現細節,Stop Reason怎麼分支,工具結果怎麼回寫,循環什麼時候退出,這樣學習才從看過,變成真的會用。最後,學習結果會進入金屬路徑,最小loop,權限,hooks,subagin,上下文壓縮,記憶,mcp和生產化都會留下進度,後面可以按薄弱點繼續推進。最後再看原理,Deep Tutor能把這些串起來,是因為同一個學習閉環把資料,目標,教材,測驗和記憶接在一起,先檢索兩份資料,再按目標上。在目標生成路線,作題結果和學習進度,又回到金屬路徑裡。所以,它接住的是完整學習動作,兩份資料進入知識庫,目標決定教材路線,學習時不斷抽查,錯題和薄弱點再沉澱進金屬路徑。所以,這期我最想說的,不是Deep Tutor又多了幾個AI功能,而是它把收藏夾裡的硬資料,變成今天就能開始的路線,問題和反饋。資料不缺,缺的是一套能逼你往前走的學習機制,別只收藏,拿它開鏈。每週拆解一個頂級開源項目,下期繼續。", "segments": [{"id": 0, "seek": 0, "start": 0.0, "end": 5.6000000000000005, "text": "每周拆解一個頂級開源項目,這周是DeepTutor,港大開源的AI私人導師工作台。", "tokens": [50365, 23664, 37390, 6852, 228, 17278, 8990, 8313, 224, 35315, 8949, 47402, 8313, 227, 11386, 11, 2664, 37390, 1541, 11089, 595, 51, 22163, 11, 22307, 3582, 8949, 47402, 1546, 48698, 20083, 4035, 25647, 16276, 41315, 15433, 1543, 50645], "temperature": 0, "avg_logprob": -0.12093292819997652, "compression_ratio": 1.2567567567567568, "no_speech_prob": 9.146044165075917e-12}, {"id": 1, "seek": 0, "start": 5.6000000000000005, "end": 13.4, "text": "各位觀眾老爺們,肯定都有類似經歷,收藏夾裡躺著很多高質量學習資料,技術教程書,學習資料,考試資料等等。", "tokens": [50645, 22073, 39368, 10439, 8164, 118, 4623, 11, 14356, 107, 12088, 48121, 40092, 7384, 120, 15145, 4287, 115, 11, 18681, 21782, 237, 1787, 122, 11066, 17514, 118, 19382, 20778, 12979, 43450, 26748, 21372, 34025, 35380, 33404, 11, 32502, 9890, 241, 21936, 29649, 29801, 11, 21372, 34025, 35380, 33404, 11, 26504, 22099, 35380, 33404, 36000, 1543, 51035], "temperature": 0, "avg_logprob": -0.12093292819997652, "compression_ratio": 1.2567567567567568, "no_speech_prob": 9.146044165075917e-12}, {"id": 2, "seek": 0, "start": 13.4, "end": 25.0, "text": "收藏時很熱血,真到自學就卡住了,沒人按你的知識儲備和學習目標排路線,沒人抽查重點,看懂了也很容易以為自己真會了,學過的歷史也串不成圖譜,錯題和薄弱點很快散掉。", "tokens": [51035, 18681, 21782, 237, 6611, 4563, 32745, 44110, 11, 6303, 4511, 9722, 21372, 3111, 32681, 21632, 2289, 11, 4913, 4035, 26613, 18961, 6498, 43143, 12827, 110, 20854, 12565, 21372, 34025, 11386, 46299, 44647, 24658, 31699, 11, 4913, 4035, 46022, 42623, 12624, 8216, 11, 4200, 29624, 2289, 6404, 4563, 49212, 3588, 6344, 17645, 6303, 6236, 2289, 11, 21372, 8816, 1546, 4287, 115, 45399, 6404, 940, 110, 1960, 11336, 2523, 244, 11823, 250, 11, 13133, 14475, 12565, 27746, 226, 5702, 109, 8216, 4563, 10251, 7017, 96, 29327, 1543, 51615], "temperature": 0, "avg_logprob": -0.12093292819997652, "compression_ratio": 1.2567567567567568, "no_speech_prob": 9.146044165075917e-12}, {"id": 3, "seek": 2500, "start": 25.0, "end": 38.0, "text": "所以DeepTutor真正要做的,是先根據你的基礎和目標定制學習路線,學的過程中,他用閃卡和測驗不斷抽查重點,學完之後再把錯題和薄弱點沉淀到金屬記憶裡,這三件事合起來才像一個真的私人導師。", "tokens": [50365, 7239, 11089, 595, 51, 22163, 6303, 15789, 4275, 10907, 1546, 11, 1541, 10108, 31337, 36841, 18961, 26008, 14971, 236, 12565, 11386, 46299, 12088, 25491, 21372, 34025, 24658, 31699, 11, 21372, 1546, 8816, 29649, 5975, 11, 5000, 9254, 5196, 225, 32681, 12565, 9592, 105, 24023, 245, 1960, 4307, 115, 46022, 42623, 12624, 8216, 11, 21372, 14128, 26362, 8623, 16075, 13133, 14475, 12565, 27746, 226, 5702, 109, 8216, 3308, 231, 14176, 222, 4511, 19117, 9636, 105, 16958, 29713, 114, 11066, 11, 2664, 10960, 20485, 6973, 14245, 21670, 18888, 12760, 8990, 8034, 20083, 4035, 25647, 16276, 1543, 51015], "temperature": 0, "avg_logprob": -0.11260841109535911, "compression_ratio": 1.2036553524804177, "no_speech_prob": 1.806775562496199e-11}, {"id": 4, "seek": 2500, "start": 38.0, "end": 51.0, "text": "今天就拿兩個高新AI Agent學習資料實測,Learn Cloud Code,7萬多星,偏工程實戰,適合跟著手挫Agent,AI Agent Book,3萬多星,偏系統原理,一個練手,一個補底層,正好一起學。", "tokens": [51015, 12074, 3111, 24351, 34623, 12979, 12560, 48698, 27174, 21372, 34025, 35380, 33404, 10376, 9592, 105, 11, 11020, 1083, 8061, 15549, 11, 22, 14099, 6392, 20682, 11, 7437, 237, 23323, 29649, 10376, 23209, 11, 3330, 102, 14245, 9678, 19382, 11389, 8501, 104, 32, 6930, 11, 48698, 27174, 9476, 11, 18, 14099, 6392, 20682, 11, 7437, 237, 25368, 33725, 19683, 13876, 11, 8990, 47027, 11389, 11, 8990, 8083, 250, 22816, 9636, 97, 11, 15789, 2131, 29567, 21372, 1543, 51665], "temperature": 0, "avg_logprob": -0.11260841109535911, "compression_ratio": 1.2036553524804177, "no_speech_prob": 1.806775562496199e-11}, {"id": 5, "seek": 5100, "start": 51.0, "end": 58.0, "text": "接下來把兩份資料導進DeepTutor,新建知識庫,填好名字,再把兩個項目的Markdown一起選進去。", "tokens": [50365, 40012, 16075, 16313, 36266, 35380, 33404, 25647, 18214, 11089, 595, 51, 22163, 11, 12560, 34157, 6498, 43143, 6346, 104, 11, 13331, 104, 2131, 15940, 22381, 11, 8623, 16075, 34623, 8313, 227, 11386, 1546, 15168, 5093, 29567, 21625, 18214, 6734, 1543, 50715], "temperature": 0, "avg_logprob": -0.07719694453177692, "compression_ratio": 1.2088607594936709, "no_speech_prob": 1.9541358114993912e-11}, {"id": 6, "seek": 5100, "start": 58.0, "end": 65.0, "text": "等索引完成,同一個知識庫裡,兩邊文件都會變成ready,後面的路線,出題和學習,就都從這批真實資料裡來。", "tokens": [50715, 10187, 7732, 95, 41301, 41509, 11, 13089, 8990, 6498, 43143, 6346, 104, 11066, 11, 16313, 14901, 17174, 20485, 7182, 6236, 26327, 11336, 1201, 11, 5661, 8833, 1546, 24658, 31699, 11, 7781, 14475, 12565, 21372, 34025, 11, 3111, 7182, 21068, 2664, 3416, 117, 6303, 10376, 35380, 33404, 11066, 3763, 1543, 51065], "temperature": 0, "avg_logprob": -0.07719694453177692, "compression_ratio": 1.2088607594936709, "no_speech_prob": 1.9541358114993912e-11}, {"id": 7, "seek": 5100, "start": 65.0, "end": 72.0, "text": "資料導入之後,接著就要自定義學習路線,關鍵在於先告訴他我是誰,基礎怎樣,這次到底想學成什麼。", "tokens": [51065, 35380, 33404, 25647, 14028, 26362, 11, 14468, 19382, 3111, 4275, 9722, 12088, 37430, 21372, 34025, 24658, 31699, 11, 14899, 41668, 113, 3581, 19488, 10108, 31080, 5000, 15914, 20757, 11, 26008, 14971, 236, 42836, 11, 2664, 9487, 33883, 7093, 21372, 11336, 7598, 1543, 51415], "temperature": 0, "avg_logprob": -0.07719694453177692, "compression_ratio": 1.2088607594936709, "no_speech_prob": 1.9541358114993912e-11}, {"id": 8, "seek": 7200, "start": 72.0, "end": 79.0, "text": "一類目標是實戰手挫開發Agent Harness,另一類是從原理層面系統理解Agent的底層架構,目標不同,路線也就不同。", "tokens": [50365, 2257, 40092, 11386, 46299, 1541, 10376, 23209, 11389, 8501, 104, 8949, 14637, 32, 6930, 389, 16937, 11, 22762, 2257, 40092, 1541, 21068, 19683, 13876, 9636, 97, 8833, 25368, 33725, 13876, 17278, 32, 6930, 1546, 22816, 9636, 97, 7360, 114, 43362, 11, 11386, 46299, 47123, 11, 24658, 31699, 6404, 3111, 47123, 1543, 50715], "temperature": 0, "avg_logprob": -0.049756223505193535, "compression_ratio": 1.3498349834983498, "no_speech_prob": 2.2574360783456093e-11}, {"id": 9, "seek": 7200, "start": 79.0, "end": 87.0, "text": "手挫Agent的開發學習路線的大綱,從最小Agent的loop循環開始,先跑出第一版Agent,再逐步補齊工程細節。", "tokens": [50715, 11389, 8501, 104, 32, 6930, 1546, 8949, 14637, 21372, 34025, 24658, 31699, 1546, 3582, 9261, 109, 11, 21068, 8661, 7322, 32, 6930, 1546, 46623, 2172, 103, 40063, 108, 21017, 11, 10108, 32585, 7781, 18049, 42096, 32, 6930, 11, 8623, 2215, 238, 31429, 8083, 250, 44628, 232, 23323, 29649, 41588, 27694, 1543, 51115], "temperature": 0, "avg_logprob": -0.049756223505193535, "compression_ratio": 1.3498349834983498, "no_speech_prob": 2.2574360783456093e-11}, {"id": 10, "seek": 7200, "start": 87.0, "end": 95.0, "text": "Agent的原理學習路線這邊,會先從認識Agent的架構開始,再順著上下文,工具,記憶和評估,把底層框架串起來。", "tokens": [51115, 32, 6930, 1546, 19683, 13876, 21372, 34025, 24658, 31699, 22821, 11, 6236, 10108, 21068, 22041, 43143, 32, 6930, 1546, 7360, 114, 43362, 21017, 11, 8623, 8313, 228, 19382, 5708, 4438, 17174, 11, 23323, 39806, 11, 16958, 29713, 114, 12565, 39115, 7384, 108, 11, 16075, 22816, 9636, 97, 21416, 228, 7360, 114, 940, 110, 21670, 1543, 51515], "temperature": 0, "avg_logprob": -0.049756223505193535, "compression_ratio": 1.3498349834983498, "no_speech_prob": 2.2574360783456093e-11}, {"id": 11, "seek": 9500, "start": 95.0, "end": 105.0, "text": "拿到教材之後就可以進入學習了,每一張都會被拆成可讀的概念,例子和練習,比如第一張先看目錄,再進入最小Agent的loop和工具調用機制。", "tokens": [50365, 24351, 4511, 21936, 4422, 238, 26362, 36781, 18214, 14028, 21372, 34025, 2289, 11, 23664, 2257, 18889, 7182, 6236, 23238, 6852, 228, 11336, 4429, 7422, 222, 1546, 26181, 33679, 11, 17797, 7626, 12565, 47027, 34025, 11, 36757, 18049, 18889, 10108, 4200, 11386, 9567, 226, 11, 8623, 18214, 14028, 8661, 7322, 32, 6930, 1546, 46623, 12565, 23323, 39806, 32943, 9254, 17543, 25491, 1543, 50865], "temperature": 0, "avg_logprob": -0.08150101311599152, "compression_ratio": 1.205128205128205, "no_speech_prob": 1.778466783564081e-11}, {"id": 12, "seek": 9500, "start": 105.0, "end": 113.0, "text": "學習過程中,內容不會只攤開給你看,Flashcards會逼你停下來回憶核心概念,避免看懂了就以為自己會了。", "tokens": [50865, 21372, 34025, 8816, 29649, 5975, 11, 28472, 25750, 21121, 14003, 5917, 97, 8949, 17798, 16529, 11, 25680, 1299, 40604, 6236, 2215, 120, 2166, 36135, 27769, 8350, 29713, 114, 8987, 116, 7945, 26181, 33679, 11, 3330, 123, 2347, 235, 4200, 29624, 2289, 3111, 3588, 6344, 17645, 6236, 2289, 1543, 51265], "temperature": 0, "avg_logprob": -0.08150101311599152, "compression_ratio": 1.205128205128205, "no_speech_prob": 1.778466783564081e-11}, {"id": 13, "seek": 9500, "start": 113.0, "end": 120.0, "text": "請接著QuickCheck會圍繞Stop Reason,工具調用和Agent loop直接出題,讓你先判斷,再揭示答案。", "tokens": [51265, 16302, 14468, 19382, 8547, 618, 47769, 6236, 2523, 235, 23141, 252, 26163, 39693, 11, 23323, 39806, 32943, 9254, 12565, 32, 6930, 6367, 43297, 7781, 14475, 11, 21195, 2166, 10108, 2437, 97, 4307, 115, 11, 8623, 11673, 255, 25696, 28223, 28899, 1543, 51615], "temperature": 0, "avg_logprob": -0.08150101311599152, "compression_ratio": 1.205128205128205, "no_speech_prob": 1.778466783564081e-11}, {"id": 14, "seek": 12000, "start": 120.0, "end": 129.0, "text": "如果還想考,自己測驗快會繼續追問實現細節,Stop Reason怎麼分支,工具結果怎麼回寫,循環什麼時候退出,這樣學習才從看過,變成真的會用。", "tokens": [50365, 13119, 7824, 7093, 26504, 11, 17645, 9592, 105, 24023, 245, 10251, 6236, 38459, 39947, 11361, 10376, 9581, 41588, 27694, 11, 26163, 39693, 11614, 6627, 24400, 11, 23323, 39806, 35181, 11614, 8350, 4510, 104, 11, 2172, 103, 40063, 108, 7598, 14010, 46361, 7781, 11, 8377, 21372, 34025, 18888, 21068, 4200, 8816, 11, 26327, 11336, 8034, 6236, 9254, 1543, 50815], "temperature": 0, "avg_logprob": -0.11274456618419244, "compression_ratio": 1.2584541062801933, "no_speech_prob": 1.6258347099173953e-11}, {"id": 15, "seek": 12000, "start": 129.0, "end": 139.0, "text": "最後,學習結果會進入金屬路徑,最小loop,權限,hooks,subagin,上下文壓縮,記憶,mcp和生產化都會留下進度,後面可以按薄弱點繼續推進。", "tokens": [50815, 20578, 11, 21372, 34025, 35181, 6236, 18214, 14028, 19117, 9636, 105, 24658, 2172, 239, 11, 8661, 7322, 46623, 11, 40212, 43446, 11, 71, 1212, 82, 11, 30131, 559, 259, 11, 5708, 4438, 17174, 45964, 21030, 106, 11, 16958, 29713, 114, 11, 76, 66, 79, 12565, 8244, 33299, 23756, 7182, 6236, 24456, 4438, 18214, 13127, 11, 5661, 8833, 6723, 26613, 27746, 226, 5702, 109, 8216, 38459, 33597, 18214, 1543, 51315], "temperature": 0, "avg_logprob": -0.11274456618419244, "compression_ratio": 1.2584541062801933, "no_speech_prob": 1.6258347099173953e-11}, {"id": 16, "seek": 12000, "start": 139.0, "end": 149.0, "text": "最後再看原理,Deep Tutor能把這些串起來,是因為同一個學習閉環把資料,目標,教材,測驗和記憶接在一起,先檢索兩份資料,再按目標上。", "tokens": [51315, 20578, 8623, 4200, 19683, 13876, 11, 11089, 595, 18392, 284, 8225, 16075, 29869, 940, 110, 21670, 11, 1541, 11471, 13089, 8990, 21372, 34025, 5196, 231, 40063, 108, 16075, 35380, 33404, 11, 11386, 46299, 11, 21936, 4422, 238, 11, 9592, 105, 24023, 245, 12565, 16958, 29713, 114, 14468, 3581, 29567, 11, 10108, 40741, 95, 7732, 95, 16313, 36266, 35380, 33404, 11, 8623, 26613, 11386, 46299, 5708, 1543, 51815], "temperature": 0, "avg_logprob": -0.11274456618419244, "compression_ratio": 1.2584541062801933, "no_speech_prob": 1.6258347099173953e-11}, {"id": 17, "seek": 14900, "start": 149.0, "end": 162.0, "text": "在目標生成路線,作題結果和學習進度,又回到金屬路徑裡。所以,它接住的是完整學習動作,兩份資料進入知識庫,目標決定教材路線,學習時不斷抽查,錯題和薄弱點再沉澱進金屬路徑。", "tokens": [50365, 3581, 11386, 46299, 8244, 11336, 24658, 31699, 11, 11914, 14475, 35181, 12565, 21372, 34025, 18214, 13127, 11, 17047, 8350, 4511, 19117, 9636, 105, 24658, 2172, 239, 11066, 1543, 7239, 11, 11284, 14468, 21632, 24620, 14128, 27662, 21372, 34025, 12572, 11914, 11, 16313, 36266, 35380, 33404, 18214, 14028, 6498, 43143, 6346, 104, 11, 11386, 46299, 33540, 12088, 21936, 4422, 238, 24658, 31699, 11, 21372, 34025, 6611, 1960, 4307, 115, 46022, 42623, 11, 13133, 14475, 12565, 27746, 226, 5702, 109, 8216, 8623, 3308, 231, 25621, 109, 18214, 19117, 9636, 105, 24658, 2172, 239, 1543, 51015], "temperature": 0, "avg_logprob": -0.15011427799860635, "compression_ratio": 1.2081218274111676, "no_speech_prob": 2.0076744086661158e-11}, {"id": 18, "seek": 16200, "start": 162.0, "end": 172.0, "text": "所以,這期我最想說的,不是Deep Tutor又多了幾個AI功能,而是它把收藏夾裡的硬資料,變成今天就能開始的路線,問題和反饋。", "tokens": [50365, 7239, 11, 2664, 16786, 1654, 8661, 7093, 4622, 1546, 11, 7296, 11089, 595, 18392, 284, 17047, 6392, 2289, 23575, 3338, 48698, 32311, 8225, 11, 11070, 1541, 11284, 16075, 18681, 21782, 237, 1787, 122, 11066, 1546, 38114, 105, 35380, 33404, 11, 26327, 11336, 12074, 3111, 8225, 21017, 1546, 24658, 31699, 11, 17197, 12565, 22138, 41520, 233, 1543, 50865], "temperature": 0, "avg_logprob": -0.08670676549275716, "compression_ratio": 0.9575757575757575, "no_speech_prob": 1.9924081481881295e-11}, {"id": 19, "seek": 17200, "start": 172.0, "end": 180.0, "text": "資料不缺,缺的是一套能逼你往前走的學習機制,別只收藏,拿它開鏈。每週拆解一個頂級開源項目,下期繼續。", "tokens": [50365, 35380, 33404, 1960, 38109, 118, 11, 38109, 118, 24620, 2257, 1881, 245, 8225, 2215, 120, 2166, 29510, 8945, 9575, 1546, 21372, 34025, 17543, 25491, 11, 16158, 14003, 18681, 21782, 237, 11, 24351, 11284, 8949, 46241, 230, 1543, 23664, 38003, 6852, 228, 17278, 8990, 8313, 224, 35315, 8949, 47402, 8313, 227, 11386, 11, 4438, 16786, 38459, 1543, 50765], "temperature": 0, "avg_logprob": -0.08727747707043663, "compression_ratio": 1.0, "no_speech_prob": 2.266140400331018e-11}], "language": "zh"}