浙大吳飛“舌戰”阿里賈揚清：AI內卷與年薪百萬，哪個纔是真實？

{"type":"doc","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"十四五規劃提出，把科技自立自強作爲國家發展的戰略支撐，瞄準人工智能、量子信息、生物醫藥等前沿領域，實施一批具有前瞻性、戰略性的國家重大科技項目。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"人工智能將引領新一輪的技術變革與數智化浪潮，與此同時，外界對於人工智能的隱憂與質疑也從未中斷。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"哪一面，纔是人工智能的真實？"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"爲了推動AI技術的應用創新，促進人工智能領域的學術交流、人才培養，中國人工智能學會聯合杭州市餘杭區政府發起首屆全球人工智能技術創新大賽，阿里雲天池平臺爲本次大賽提供平臺和算力支撐。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"作爲大賽的預熱節目，"},{"type":"text","marks":[{"type":"strong"}],"text":"“AI青年說”近期邀請到浙江大學求是特聘教授、人工智能研究所所長吳飛與阿里雲計算平臺負責人賈揚清，交流人工智能技術與應用的發展趨勢，更重要的是，聊聊“AI內卷與年薪百萬，哪個纔是真實？”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"（以下爲對話節選）"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"如何與人工智能結緣？"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"吳飛："},{"type":"text","marks":[{"type":"strong"},{"type":"strong"}],"text":"我是2009年的時候開始對人工智能產生興趣，當時我對自己的研究方向產生了深深的焦慮，覺得自己當時研究所寫的東西"},{"type":"text","marks":[{"type":"strong"}],"text":"“面目可憎，言語無味”"},{"type":"text","text":"。這個時候，我得到去美國加州大學伯克利分校統計系鬱彬教授課題組做訪問學者的機會，深入學習很多有關機器學習模型理論和算法的知識。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"原來在冰冷的機器後面，有這麼多深奧的模型、算法和這麼新鮮的理論來支持。一年之後，我回到浙大，從此走上了人工智能的學術道路。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"賈揚清："},{"type":"text","text":" 我最開始是學習自動化的。以前，我們開玩笑說，自動化主要做兩件事情，一個是燒鍋爐的，一個是開電梯的。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"後來在研究生期間，我就對於人工智能和統計機器學習產生了更多的興趣，當時人工智能並沒有那麼熱，有一句流行的話，"},{"type":"text","marks":[{"type":"strong"}],"text":"“人工智能是在80%的時間裏，以80%的正確率，解決80%的問題。”"},{"type":"text","text":" 但是我們也不知道，那80%的時間解決了哪些80%的問題，大家一直處於相對低谷的狀態。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"2006年的時候， Geoffrey Hinton在《Science》發表了一篇講受限玻爾茲曼機（restricted Boltzmann machine）的文章，我開始對人工神經網絡等基礎理論以及大規模訓練產生興趣。我是2009年去了加州大學伯克利分校，也有幸跟吳老師在伯克利相處了一年多的時間。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"攻讀博士期間，我們發現神經網絡、深度學習的方法變得越來越重要，我們最開始從稀疏編碼（Sparse Coding）等方面入手，構建一系列的軟件棧以及相應的科學研究，來把基於深度學習的算法做得越來越好。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"2012年，AlexNet的出現讓全世界都突然意識到深度學習的重要，在此之前，2010年左右，在語音領域RNN等方法已經開始被應用起來。當時我發現大家都在紛紛湧向深度學習算法，相應的軟件工具平臺卻比較匱乏的，所以我們在伯克利就開始做Caffe以及後來一系列的深度學習框架。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"在接下來的幾年，我們見證了深度學習算法的大規模應用，無論是算法創新方面，還是方法的落地方面，都開始有了非常大的市場。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"無論是阿里的城市大腦算法，還是達摩院的很多新型算法模型，很榮幸我的研究能夠成爲一個底座，支持大家在AI技術及應用上有進一步的創新和發展。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"十四五規劃把人工智能列爲前沿科技領域的“最高優先級”。這對人工智能專業而言，意味着什麼？"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"吳飛："},{"type":"text","text":"2017年，國務院公佈了中國新一代人工智能發展規劃，規劃明確指出：人工智能是引領未來的戰略性技術，必將推動人類社會和生活模式以及學習方式的巨大改變。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"我想對年輕人而言，人工智能就意味着未來，因爲它本身是一個使能技術，不斷推動着社會的快速前進。規劃還指出，高校要設置人工智能本科專業。2018年，教育部批准了35所高校設置人工智能本科專業。到現在爲止，全國一共有345所高校設置了人工智能本科專業。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"圖靈在1949年談及圖靈機時表示，“（圖靈機模型的提出）這不過是將來之事的前奏，也是將來之事的影子。”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"因此，對年輕人而言，無論是基礎研究、關鍵核心技術的突破，還是賦能場景的實現，人工智能已經悄然地走進社會，像水和電一樣的不斷賦能社會的發展。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"未來AI有哪些值得關注的趨勢？"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"賈揚清："},{"type":"text","text":" 在2012年以前，要做一個計算機視覺的識別系統，基本上就要去讀一個博士，才能做出來，而且效果還不一定好。"},{"type":"text","marks":[{"type":"strong"}],"text":"今天我們發現，從0到1的積累差不多已經完成了，或者說已經比較成熟。算法的標準化和應用化會變成趨勢。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"今天如果我想做一個無人駕駛的demo，就並不需要去學計算機視覺的博士。因爲今天有非常多開源的模型，讓我們能夠非常迅速地把算法能力給補齊。"},{"type":"text","marks":[{"type":"strong"}],"text":"算法的標準化、工程化，以及怎樣迅速地把標準算法和特定的業務場景結合起來，是我們今天在計算機視覺、語音、自然語言理解等領域的大趨勢。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"當然科研還在繼續往前走，我們在尋找新的高精尖方向。但同時我認爲，怎樣把現有結果大規模應用到不同場景中去，是一個非常大的趨勢。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"另外，我們會發現，以前單純的垂直場景，比如像計算機視覺、語音、自然語言處理等已經開始逐漸融合，變得共通，這就需要用到大規模、多模態模型。如我們所見，谷歌、OpenAI、DeepMind等公司都在這方面做出非常多的探索。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"前段時間OpenAI推出GPT-3模型，這帶來的啓發是："},{"type":"text","marks":[{"type":"strong"}],"text":"我們需要有一個通用、多模態的模型，來統一理解各種形態的數據和各種形態的輸入。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"我覺得這代表了另外一個趨勢。"},{"type":"text","marks":[{"type":"strong"}],"text":"一方面是現有結果的工程化和規模化，另一方面是各子領域之間的相互貫通，以及做到更加深入、更加本質的理解。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"吳飛："},{"type":"text","text":" 我補充一下，有兩個方向的趨勢值得關注。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"第一個是從0到1，按照朱松純教授的說法，現在的機器智能是“大數據小任務”，比如GPT-3有1750億的參數，並且使用上千GB的訓練數據把它訓練出來。"},{"type":"text","text":" 但如何讓人工智能或機器智能具有人腦小樣本學習的能力、動物的直覺能力以及舉一反三的能力，這是我們面臨的巨大挑戰。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"第二個我覺得是從1到N，人工智能已經是一門使能技術。就像我們徐匡迪院士所言，“人工智能需要數學家參與進來。”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"人工智能取得突破性的領域方向，一定是從腦科學、材料學等領域得到了啓發，所以AI接下來的突破，一定是基於大數據、多學科交叉下的領域突破。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"賈揚清：我也補充一點。今天的人工智能系統，特別是在感知領域，無論是計算機視覺還是語音，都存在一種“一攬子買賣”的狀態。"},{"type":"text","text":"例如有一個輸入和一個輸出，標註是人或者車，這就是一個所謂的“one shot”過程，目標集有時是手工指定的。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"怎樣從單點的目標或者單點的預測（prediction）到更加完整的知識體系，即所謂的大知識。正如吳老師剛提到的“大數據小任務”，“大數據大知識”是我們今天需要打通的一件事情：在簡單標籤的基礎上再構建一個知識體系，無論是邏輯關係還是其他關係。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"傳統的專家系統，更多的是通過人工手寫的方式來做的，很難規模化。傳統的機器學習方法還是小任務單點預測的狀態。"},{"type":"text","marks":[{"type":"strong"}],"text":"怎樣能夠讓機器學習系統更加自動、規模化地生成結構化的知識和結構化的體系，我覺得，在接下來幾年，這方面可能產生最大的突破，或者至少是需求最大的方面。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"同時它也引出了另外一點，就是吳老師剛纔提到的：人工智能怎樣賦能行業，例如製藥業的數據是非常少的，對於邏輯推理和知識抽象的要求非常高。因爲它沒有辦法像圖像識別一樣能使用幾百萬張圖片作爲數據。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"簡而言之，"},{"type":"text","marks":[{"type":"strong"}],"text":"未來有兩個趨勢：一個是從大數據到大知識，另一個是怎樣通過抽象出來的知識體系，來賦能其他領域，以更好地使用AI技術。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"人工智能目前存在的障礙與問題"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"吳飛："},{"type":"text","text":" 發展人工智能一定要有豐沃的土壤，也就是人工智能發展的生態。"},{"type":"text","marks":[{"type":"strong"}],"text":"今天要支撐起中國人工智能的發展，一定要依靠底層的算力、底層的軟硬件以及中間算法和上層的應用軟件。從底到高支撐起中國人工智能的發展，就是它的技術生態。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"可能是之前我們忙於搞模型、搞算法、搞應用。而現在已經到了一個關口，需要阿里巴巴等企業的投入，需要浙江大學、清華大學、北京大學等高校的投入，需要政府的關切以及各行各業人才的投入，同頻共振、相向而行，建設一個良好的生態。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"賈揚清："},{"type":"text","text":" 今天談論得非常不足的一點是，"},{"type":"text","marks":[{"type":"strong"}],"text":"怎樣讓所有人都能用到人工智能的技術。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"此前我們更關注的是怎樣發展高精尖的技術。縱觀歷史，我們說人工智能是通用化的技術，或者我們相信未來它將是技術發展的核心點，這就需要讓所有人都能更容易地接觸到人工智能。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"幾十年前，英語專業有一個專門的職業叫翻譯，而在國際化的影響下，如今很多人自身就能掌握中英雙語。我們以前說：“計算機要從娃娃抓起。”而今天我們幾乎每個人都有使用計算機或操作計算機的能力，無論是電腦還是手機。只有達到這種普惠程度的時候，我們才能夠把一個技術稱爲通用的、普惠的技術。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"今天人們要觸達AI技術還是相對困難的。無論是在硬件上（搭一個帶有GPU的機器），還是在軟件上（安裝一套人工智能框架以及底層數據的軟件等），都還比較困難。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"設想一下，如果任何一個懂電腦的人都可以在5秒鐘之內開始嘗試寫AI算法的代碼，並且很快能識別一張圖像的內容。他也許並不懂底層的技術是怎樣實現的，但創造性的火花很容易就此迸發。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"高精尖的技術往往是比較抓眼球，容易被大家接受。而從長遠角度看，這種潛移默化的普惠化、潤物細無聲的過程會收穫更好的效果。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"讀博對於人工智能意味着什麼？"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"賈揚清："},{"type":"text","text":" 學習人工智能與是否讀博的關係並不是很大，每一個行業都會對博士有特別定義。在AI領域有非常多優秀的本科生和碩士生，他們並不比博士的能力差。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"從個人經歷來看，我的博士生涯教會我的最重要一點不是“怎樣做一件事情”，而是讓我更多地思考“我們應該解決什麼樣的問題”。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"博士更多的是要定義問題，需要從特別複雜的場景中抽象出要解決的問題的定義。然後第二位的纔是「怎樣解決這個問題」，這是博士生涯對我幫助最大的一點。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"從培養的角度講，如果大家希望培養一下自己定義問題的能力，我覺得博士還是很值得讀的。5年的時間不長也不短，但能力的提升是終身獲益的事。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"吳飛："},{"type":"text","text":" 我完全同意揚清的觀點。**讀博的問題要辯證看待。**如果同學們渴望充電，有渴望學習的飢餓感，通過職業培訓或者通過工作環境中的積累無法解決，那就去讀博士。並且讀博期間一定要注意，學習目的不只是獲得文憑，也不只是發表文章，而是爲了培養能力。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"剛纔揚清也講了，這是一個終身的事情，人們常說讀完博士之後，他能不能發展好要再看5年，5年之後如果繼續往前發展，就說明他已經走上了人生不斷向前發展的軌道。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"排除疫情的因素，出國深造還有沒有必要？"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"賈揚清："},{"type":"text","text":" 這個問題沒有非黑即白的答案。首先，現在國際交流越來越多，線上會議只有時差問題，就像看世界盃一樣，出國可能沒有10年前或20年前那麼必要了。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"另一方面，我一直相信不同文化、不同思想的融會貫通是創造新的思想和方法的豐沃土壤。如果有機會的話，我也建議大家出國去看一看，無論是參加一個學術會議，還是訪學計劃，或者投資更長的時間去讀博。"},{"type":"text","marks":[{"type":"strong"}],"text":"更加重要的是，從自己出發，打開心扉，去了解不同的文化與技術氛圍"},{"type":"text","text":"。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"吳飛："},{"type":"text","text":" 我是2009年10月份從北京首都機場出發，飛往美國舊金山。當時我已經36歲了，那是我第一次出國。在伯克利的時光，我基本是兩點一線，從宿舍到伯克利的實驗室。一年到頭，我只有春節期間休息了幾天，記得除夕約了揚清吃了頓餃子，那也是我沒去實驗室的僅有的一兩天。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"很多人都有出國的經歷，但並不是所有出國的人都抓住了機會，也並不是每個人都能完成自己的志向。在美國高水平的大學裏，如果能夠把自己全部的精力都用於學習，是非常好的人生體驗。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"從伯克利回來之後，我的大腦裏面會經常浮想起美國西海岸蔚藍的天空以及宿舍夜裏火車經過的鳴笛聲。這不是在回憶美國的生活，而是在回憶當時的奮鬥。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"AI內卷與年薪百萬，哪個纔是真實？"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"賈揚清："},{"type":"text","text":" 內卷是一個很有意思的概念，包括大家逛知乎常看到一個詞語，叫調參俠。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"在AI領域，有很多創新的事情可以做。內卷與否，是我們自己內心的表現。如果覺得自己做不出新的東西，只能調調參數，做一些增量工作，這肯定是內卷的。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"容易的地方就會內卷的，難的領域往往會有突破。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"應該給自己更多的信心與勇氣。AI已經是個日新月異的領域了，廣泛應用到互聯網、交通、電力、航空等領域，每年都有新的突破，何不去做創新呢。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"吳飛"},{"type":"text","text":"：揚清有意避開了“年薪百萬不是夢”，但對高校老師而言“年薪百萬就是夢”。教師的工資明顯沒有學生高，這是人工智能領域的實際情況。我們培養了很多的學生，現在已是年薪百萬。這說明人工智能專業的職業發展前景非常好。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"未來的發展趨勢不會改變，整個國家把人工智能作爲一種戰略性技術進行發展的大方針不變，各行各業都會對人工智能有極大熱情。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"爲什麼會內卷，是因爲大家都跑到一條賽道上去了，比如都擠到“大數據小任務”的賽道，變成了所謂的“調參俠”。於是沒人做農業、司法、環境等更具開創性的場景應用，就發生了內卷。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"想象一下，你讀完人工智能的博士或者研究生，如果進入一個稀缺人工智能的行業，那麼你就是這個行業的AI之王，哪裏會有內卷。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"賈揚清"},{"type":"text","text":"：我也順便談下“年薪百萬”這個事。像我今天早餐是自己做的，午飯是在食堂解決的，平時真的花不了多少。其實真讓自己開心的事並不是年薪百萬，而是我們做的東西有人用，大家都喜歡。我特別崇敬吳老師這些學校裏面的老師。第一，他們在探索前沿研究；第二，他們在培養AI以及各個技術領域的人才。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"全球人工智能技術創新大賽，對學界與業界的意義"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"吳飛："},{"type":"text","text":" 中國人工智能學會從爲國家培養人才的角度來舉辦大賽，賽題也來自於產業界的需求，這非常好。人工智能最貴重的一點就是貴其能用，賦能社會，每個賽題都來自產業的需求。同學們用在高校學習得到的知識和技術，來解決產業的問題，就打破了學科的壁壘，達到產教融合的效果。同學藉此瞭解到業界解決問題的一些方法與思路，相信以後回到高校，可以更高屋建瓴地看待學習的內容。所以全球人工智能技術創新大賽是高等學校和產業界攜手，面向人工智能應用進行人才培養的好方式。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"賈揚清："},{"type":"text","text":" 競賽一直是推動創新的好機制，以前在讀書的時候有挑戰杯這樣的比賽，這對我們在課程之外嘗試一些新的東西是非常有用的。比如Facebook等公司也有黑客馬拉松的機制，某程度上可以讓大家跳出本職工作，嘗試新的點子，產出新的技術或者產品。藉着吳老師剛纔說的“隨心所欲，隨遇而安”，我覺得這個比賽大家可以“生死看淡，不服就幹”。相信大家都能夠從比賽中體會到快樂，收穫到知識。"}]}]}