我只是想問一個相關的問題。 I was just going to ask kind of a related question.
他也說我們有 10 年的晶片製造實施經驗。 He was saying also that we have a 10-year chip manufacturing implementation.
我對此感到驚訝。 I was surprised about that.
是的，我認為對我這個勞動經濟學家來說有趣的是，這真的是我在文獻和新聞中看到的綠燈，好吧，如果我們把所有這些晶片製造都移回國內，這難道不會在藍領工作上創造某種復興嗎？ Yeah, and I think what was interesting to me as a labor economist is that it was really like a green flag I've seen in literature and news that, okay, if we're onshoring all of this chip manufacturing, isn't that going to create some sort of resurgence in blue collar jobs?
我想知道您對智慧機器人模型或人類勞動有什麼看法？ And I wondered if you had any thoughts about intelligent robotic models or human labor?
嗯，我不認為這會有多大影響，我的意思是，我甚至沒放你們參觀晶圓廠的影片給你們看。
Well, I don't think it's gonna be much of a, I mean, I didn't even play you guys a visit at a chip fab.
有人去過嗎？ Anybody?
你們，我猜有幾個人去過。 You guys, I'm a few of you have.
那個晶圓廠有多少工人？ How many workers were in that fab?
那是什麼？
What was it?
她有一些履歷，像你一樣，所以我不知道。 She has some CV, like you, so I don't know.
是的，我的意思是。
Yeah, I mean.
好吧，答案是零。（觀眾笑） Well, okay, so the answer's zero. (audience laughing)
原因是，他們不讓任何人進去，因為我們人類太笨拙、太髒了，我們做不到，所以都是機器人。 Like, the reason, they don't let anyone go in because we humans are too clumsy and dirty and we can't, so it's all robotic.
裡面是密封的。 It's sealed inside.
所以有工作是把東西帶給它們等等。 So there is work to bring stuff to them, et cetera.
如果機器人摔倒了或出了什麼問題，他們得穿上，你們可能看過這些，看起來像太空衣。 And if a robot falls over or something goes wrong, they have to put on, you've probably seen these, they look like spacesuits.
他們得進去，然後他們可能會調整一些東西，然後再出來，希望他們沒有弄壞任何東西。 They have to go in and then they kind of maybe adjust something, and then they go back out and hope they didn't break anything.
所以基本上是關燈生產。 So it's basically lights out.
是的，我不認為——有一些更複雜的勞動力是需要的。 Yeah, I don't think it's-- there is some more sophisticated labor required.
我不認為這是藍領階級的研究。 I don't think it's a blue-collar research.
事實上，蘋果將 MacBook 生產遷回德州的原因之一，並不是因為德州的勞動力便宜或什麼的，而是因為他們實際上不再需要那麼多勞動力了。 In fact, one of the reasons that Apple reshored MacBook production to Texas is not because labor is so cheap in Texas or anything. is that they don't actually require a whole lot of labor anymore.
所以我認為這是一個相當勞力密集的產業。 So it's a pretty labor, I think.
美國製造業在產出方面正在激增。 US manufacturing is surging in terms of output.
但在就業方面，並沒有真正增長多少。 But in terms of employment, it's not really growing all that much.
我們到這邊來。 Let's go over here.
您認為明年 AI 代理人或測試行動模型會出現轉折點嗎？ Do you see an inflection point coming for AI agents or test action models in the next year?
哦，是的。 Oh, yeah.
不，不。 No, no.
他說了艾瑞克——我聽到了類似的事情。 He said what Eric-- I'm hearing similar things.
其實，他把那三個趨勢說得很好。 Actually, he had a really nice way of putting those three trends.
我分別聽說過它們，但我認為把它們都放在一起很好。 I've heard about them all separately, but I think it was good to bring them all together.
今天早些時候，我跟吳恩達（Andrew Ng）聊過，他一直在敲打代理人這個鼓，特別是把它當作 2024 年的浪潮，吳恩達有一個很好的描述方式，就像你們知道的，如果你有一個大型語言模型，我不知道，寫一篇文章之類的，它一個字一個字地寫，一次性地寫完這篇文章。 Earlier today I was talking to Andrew Eng and he's been beating this drum about agents in particular as being sort of the wave of 2024, where Andrew had a nice way of describing it that as you guys know, if you have an LLM, I don't know, write an essay or something like that, it writes it one word at a time and it just goes through in one pass and writes the essay.
這很不錯。 And it's pretty good.
但想像一下，如果你必須那麼做，不能用退格鍵，沒有機會讓你，你知道，你不會先做大綱，你就是直接寫。 But imagine if you had to do that, No backspace, no chance to let you know, you don't make an outline first, you just kind of go through.
代理人現在會說，好的，首先做個大綱，那是你寫文章時的第一步，然後填寫每個段落，然後回去看看流程是否正確。 The agents now will say, okay, first make an outline, that's the first step you do when you write an essay, and then fill in each paragraph, then go back and see if the flow is right.
現在回去檢查語氣，這個層級適合我們的讀者嗎？ Now go back and check the voice, is this the right level for our audience?
現在，透過這樣的迭代，你可以寫出好得多的文章或完成任何類型的任務。這是一場真正的革命。 Now, and by iterating like that, you can write a much, much better essay or any kind of a task. is a real revolution.
有很多事情如果你這麼做，就可以做得更好。 There's all sorts of things you could just do much better if you do that.
然後關於 context window 的事情也很重要。 Then the thing about the context windows is also really important.
所以我只是引用我認識的聰明人的話。 So I'm just going to quote smart people that I know.
艾瑞克·霍維茨（Eric Horvitz），我在 GSB 的一個座談會上和他一起，上週你們有些人可能也在場。 Eric Horvitz, I was on a panel with him at the GSB. Some of you may have been there last week.
他有一個很好的分類法。 And he had this nice taxonomy.
人們在問關於微調的問題。 People were asking about fine tuning.
我想是蘇珊（Susan）在問關於微調的問題。 I think Susan was asking about fine tuning.
他說，嗯，其實有三種方法可以讓你採用一個模型並讓它更客製化。 And he said, well, there's really three ways that you can take a model and have it more customized.
一種是你可以微調它，基本上就像再訓練它一樣。 One is is you can fine tune it, which basically like train it some more.
另一個是使用越來越大的 context window，第三個是使用 RAG 或類似的技術，也就是檢索增強生成，它會去存取外部資料。 Another is with larger and larger context windows, and the third is with RAG or techniques like that that are retrieval augmented generation where it goes and accesses external data.
但這些 context window 現在似乎非常有效。 But these context windows seem to be remarkably effective now.
我想正如艾瑞克所說，我們認為這很難，也許彼得可以解釋。 I guess as Eric was saying, we thought it was hard, maybe Peter can explain.
但不知何故，我們能夠製造出大得多的，現在你可以載入像一整本書或一整套書。 But for some reason, we're able to make much, much bigger ones and now you can load like a whole book or a whole set of books.
你可以在裡面載入各種資訊，那可以給你所有的 context。 You can load all sorts of information in there and that can give you all of the context around.
所以那是一場相當大的革命。 So that's a pretty big revolution.
它開啟了一堆我們以前沒有的功能，包括讓事情更即時，正如艾瑞克所說。 It opens up a bunch of capabilities that we just didn't have before, including having things much more current as Eric was saying.
你想接著說嗎？ Did you want to follow up on that?
我的意思是，當然有更多的資本投入，但這有點迴避了問題和評論，為什麼所有這些資本都流向那裡而不是其他地方？ I mean, there's certainly a lot more capital going, But that kind of begs the questions and comments, why is all this capital going there as opposed to somewhere else?
我認為，如果你看看歷史的軌跡，有時候它看起來很平滑。 And I think if you look at the arc of history, sometimes it looks kind of smooth.
但如果你更仔細地看，會有很多跳躍。 But if you look more closely, there's a lot of jumps.
有一些大的發明和小的發明。 There are certain big inventions and smaller inventions.
安德魯·卡帕斯（Andrew Karpathy）說他在玩物理。 And Andrew Karpathy was saying that he was playing around with physics.
要真正在物理學上取得進展，成為頂尖的物理學家，你必須非常聰明，學習很多東西。 And to really make progress in physics, to be a top physicist, you have to be incredibly smart, study a whole lot.
如果你幸運的話，也許可以做出一些小的、漸進式的貢獻，有些人做到了。 And maybe if you're lucky, you could make some small, incremental contribution, and some people do.
但他說，現在在 AI 機器學習領域，我們似乎處於一個有很多唾手可得的果實的時代，已經有一些突破。 But he says that right now in AI machine learning, we seem to be in an era where there's just a lot of low-hanging fruit, that there have been some breakthroughs.
它不是耗盡了空間，像是把樹上的果實都摘光了，更像是組合學。 And instead of exhausting the space, like picking all the food off of a tree, it's more like combinatorics.
在第二台機器裡，他們談到積木。 In the second machine, they talk about building blocks.
當你把兩個積木或樂高積木放在一起，你可以做出越來越多的東西。 When you put two building blocks together, or LEGO blocks, you can make more and more.
現在，我們似乎處於一個充滿機會的時代，人們也意識到了這一點。 Right now, we seem to be in an era where there's just a lot of opportunity, and people are recognizing that.
一個發現孕育了另一個發現，孕育了另一個機會。 And one discovery begets another discovery, begets another opportunity.
正因為如此，它吸引了投資，更多的人參與進來。 And because of that, it attracts the investment, and more people are involved.
在經濟學中，有時候投入更多資源，你會得到遞減的回報，像是農業或礦業。有些地方會有遞增的回報，更多的工程師來到矽谷，讓現有的工程師更有價值，而不是更沒價值。 And in economics, sometimes when more resources go in, you get diminishing returns, like in agriculture or in mining. places there's a increasing returns and more engineers coming to Silicon Valley makes the existing engineers more valuable not less valuable.
我們似乎正處於一個那樣的時代，然後是額外投資、額外訓練資金的飛輪，所有這些都讓它們更強大。 So we seem to be in an era where that's happening and then the flywheel of of the additional investment, the additional dollars for training, all of that makes them more more powerful.
我不知道這會持續多久，但我不知道你……這似乎只是有些技術已經達到了這個非常肥沃的時期，並且有正向回饋最終有所幫助。 I don't know how long this will continue but I don't I don't you It just seems that there are some technologies that have hit this really fertile period and there's positive feedback that ends up helping.
我們現在似乎就在其中之一。 We seem to be in one of those right now.
所以受過訓練進入這個領域的人，所做的貢獻往往比在其他領域可能更快地變得相當顯著。 So people who are trained in getting in the field are making contributions that are often quite significant in a faster time than they might have in some other fields.
鼓勵大家，我認為你們現在做的都是對的。 Encouraging all of you guys, I think you're doing the right thing right now.
是的。 Yeah.
我們再問幾個問題，然後，是的。 Let's take a couple more questions. and then, yeah.
好的，這邊怎麼樣？ OK, how about over here?
並非每個人都能坐在一個房間裡，圍繞 AI 進行所有這些討論和辯論。 So not everyone can sit in a room and have all these discussions and debates around AI.
所以我很想聽聽您對非技術利益相關者的 AI 素養的看法，無論他們是必須做出某種程度上知情判斷的決策者，還是一般大眾，像是，你知道的，[聽不清楚] 的技術。 And so I'd like to get your thoughts on AI literacy for non-technical stakeholders, whether they're policymakers that have to make it in somewhat informed judgment, or the general public, like, you know, [INAUDIBLE] tech.
您如何看待解釋技術基礎知識與討論不一定有正確答案的抽象影響？ How do you think about explaining technical basics versus discussing abstract implications that don't necessarily have a right answer?
嗯，那是個難題。 Well, that's a hard one.
我會說最近在國會和其他地方的人們對這個主題的關注程度發生了巨大變化。 I would say there's been a sea change recently in terms of how much people in Congress and elsewhere are paying more attention to this topic.
以前這不是他們感興趣的話題。 It used to be not something that they were interested in.
現在每個人都試圖更了解它一點。 Now everyone's trying to understand it a little bit better.
我認為人們可以在很多方面做出貢獻。 And I think that there are a lot of margins where people can make contributions.
他們可以在技術方面做出貢獻。 They can make contributions in the technical side.
但如果說有什麼的話，我的賭注是，商業和經濟方面才是現在更大的瓶頸。 But if anything, I mean, my bet is that the business and economic side is where the bigger bottleneck is right now.
但即使你在技術方面做出了巨大的貢獻，將其轉化為能改變政策的東西仍然存在差距。 But even if you made enormous contributions to the technology side, there's still a gap converting that into something that will change policy.
所以，如果你對政治學感興趣或是個政治家，理解對民主、假訊息、權力和集中的影響是什麼。 So understand if you're into political science or a politician, understanding what are the implications for democracy and for misinformation and power and concentration.
這些都是還沒有被很好理解的事情。 Those are things that are not well understood at all.
我不知道電腦科學家是否一定是嘗試理解這個問題的合適人選。 I don't know that a computer scientist is necessarily the right person to try to understand that.
但是，對技術有足夠的了解，讓你曉得什麼是可能的，然後思考動態是什麼，就像亨利·季辛吉（Henry Kissinger）和艾瑞克·施密特（Eric Schmidt）在他的書中所做的那樣。 But understanding enough about the technologies so you know what might be possible and then and thinking through what are the dynamics, like Henry Kissinger was doing with Eric Schmidt in his book.
如果你是經濟學家，思考對勞動市場的影響、對集中的影響、對不平等的影響、工作、對生產力的影響以及驅動生產力的因素。 If you're an economist, thinking through the labor market implications, the implications for concentration, implications for inequality, jobs, the implications for productivity and what drives productivity.
這些都是現在非常成熟的議題。 Those are things that are very ripe right now.
你可以遍歷很多不同的領域，在那裡，充分了解技術可能的能力，然後思考其影響。 And you could go through lots of different fields where there's understanding well enough what the technology might be capable of, but then thinking through the implications.
我認為那是一些最大回報所在的地方。 That's, I think, where some of the biggest payoffs are.
讓我給你一個更具體的例子。 Let me give you a little bit more of a concrete example.
這是我上週要講的東西。 And this is something I was going to talk about last week.
電力也是一種通用技術。 Electricity was also a general purpose technology.
通用技術有這樣一個特點，它們本身可能就很有用。 And general purpose technologies have this characteristic that they're probably in and of themselves.
但通用技術（GPT）真正的威力之一，正如我所說的，是它們能帶來互補——它們能點燃互補的創新。 But one of the real powers of general purpose technologies, GPTs, as I was saying, is that they give complementary-- they ignite complementary innovations.
所以電力、燈泡、電腦、電動馬達，而電動馬達給你壓縮機、冰箱和空調。 So electricity, light bulbs, and computers, and electric motors, and electric motors give you compressors and refrigerators and air conditioning.
你可以從這一個創新中得到一整套、一連串的額外創新。 You can just have a whole set, a cascade of additional innovations from this one innovation.
大部分的價值來自於這些互補的創新。 Most of the value comes from these complementary innovations.
有一件事人們不夠了解，那就是一些最重要的互補創新是組織和人力資本的互補性。 One thing people don't appreciate enough is that some of the most important complementary innovations are organizational and human capital complementarities.
有了電，當他們第一次把電引進工廠時，史丹佛的保羅·大衛（Paul David）研究了那些工廠發生了什麼。 So with electricity, when they first introduced electricity into factories, Paul Davids here at Stanford studied what happened to those factories.
令人驚訝的是，沒什麼變化。 And surprisingly, not much.
當工廠開始電氣化時，它們的生產力並不比以前由蒸汽機驅動的工廠高多少。 The factories, when they started electrifying, they were not significantly more productive than the previous factories that were powered by steam engines.
他說，嗯，這有點奇怪，因為這似乎是一項相當重要的技術。 He's like, well, that's kind of weird, because this seems like a pretty important technology.
這只是一時的風潮嗎？ Is it just a fad?
顯然不是。 Obviously not.
電氣化前的工廠由蒸汽機驅動。 The factories before electricity were powered by steam engines.
他們通常在中間有一台大蒸汽機，然後用曲軸和滑輪來驅動所有設備。 They typically had a big steam engine in the middle, and then crankshafts and pulleys that powered all the equipment.
所有東西都是分散的，但你會試著讓它盡可能靠近蒸汽機，因為如果你把曲軸做得太長，扭力會把它弄斷。 And it was all distributed, but you tried to have it as close to the steam engine as possible, because if you make the crankshaft too long, it would break the torsion.
當他們引進電力時，他發現在一個又一個的工廠裡，他們會把蒸汽機拔掉，然後買他們能找到的最大的電動馬達，把它放在蒸汽機原來的位置，然後啟動它。 When they introduced electricity, he found that in factory after factory, they would pull out the steam engine, and they would get the biggest electric motor they could find and put it where the steam engine used to be and fire it up.
但這對生產並沒有太大改變。 But it didn't really change production a whole lot.
你可以看到那不是什麼大事。 You can see that that's not a big deal.
然後他們開始在一個新的地點從頭開始建造全新的工廠。 So then it started building entirely new factories from scratch in a new location.
那些看起來像什麼？ What did those look like?
就像舊的一樣。 Just like the old ones.
他們會採用同樣的模式。 It would take the same model.
某個工程師會畫一張藍圖，也許拿過來，在蒸汽機的位置畫一個大叉，不，不，在這裡放一個電動馬達，然後他們就去蓋一座新工廠。 Some engineer would make a blueprint, maybe take it, make a big X where the steam engines, no, no, put electric motor here, and they'd go and build a fresh factory.
同樣，生產力沒有太大提升。 Again, not a big improvement in productivity.
大概過了 30 年，你才開始看到一種完全不同的工廠，那裡的動力源不是集中在中間的一個大動力源，而是分散的動力，因為電動馬達，如你們所知，你可以做大的，可以做中等的，也可以做非常非常小的。 It took about 30 years before you started seeing a fundamentally different kind of factory, where instead of having the central power source, a big one in the middle, you had distributed power because electric motors, as you guys know, you can make them big, you can make a medium, you can make them really, really small.
你可以用不同的方式把他們都連接起來。 You can have them all connected in different ways.
於是，他們開始讓每台設備都有一個獨立的馬達，而不是一個大的。 So, they started having each piece of equipment have a separate motor instead of one big one.
他們稱之為「單元驅動」（unit drive），而不是「群組驅動」（group drive）。 They called it unit drive instead of group drive.
我去哈佛商學院的貝克圖書館讀了 1914 年左右的書，裡面全都是關於單元驅動與群組驅動的辯論。 I went and read the books in Baker Library at Harvard Business School from like 1914 and it was like this whole debate about unit drive versus group drive.
嗯，當他們開始這麼做時，他們就有了一種新的工廠佈局，通常是在單一層樓，機器的擺放不是根據它需要多少動力，而是根據其他東西，物料的流動。 Well, when they started doing that, then they had a new layout of factories where it was typically on a single story where the machinery was not based on how much power it needed, but based on something else, the flow of materials.
你開始有了這些流水線系統。 And you started having these assembly line systems.
那帶來了生產力的巨大提升，像是生產力翻倍，在某些情況下甚至三倍。 That led to a huge improvement in productivity, like a doubling of productivity, or tripling in some cases.
所以教訓不是說電力是一時的風潮或失敗品，被過度吹捧了。 So the lesson is not that electricity was a fad or a dud and was overhyped.
電力是一項從根本上有價值的技術，但直到他們有了那個流程創新，那個重新思考如何進行生產的組織創新，你才得到了巨大的回報。 Electricity was a fundamentally valuable technology, but it wasn't until they had that process innovation, that organizational innovation of rethinking how to do production, that you got the big payoff.
有很多像這樣的故事。 There's a lot of stories like that.
我只告訴了你其中一個。 I only told you one of them.
我們時間不多，沒辦法告訴你其他的，但在我的一些書和文章中，如果你看看蒸汽機和其他的，你會發現類似的代際延遲，幾十年後人們才意識到這項技術可以讓你做一些和你以前做的事情完全不同的事。 We don't have that much time. tell you the other ones, but in some of my books and articles, if you look at the steam engine and others, you had similar generational lags decades before people realized that this technology could allow you to do something completely different than you used to do.
我認為 AI 在某些方面有點像那樣，將會有很多組織創新，會有新的商業模式，新的組織經濟的方式，是我們以前沒想過的。 I think AI is a bit like that in some ways, that there's going to be a lot of organizational innovations, going to be new business models, new ways of organizing an economy that we hadn't thought of before.
現在人們大多只是在改造。 Right now people are mostly just retrofitting.
我可以再講一整套互補的技能變化。 I could go through a whole nother set of skill changes that are complimentary.
我不知道它們都是什麼。 I don't know what they all are.
你必須有創造力去思考它們，但那正是差距所在。 You have to be creative to think about them, but that's what the gap is.
在早期電腦的例子中，如果你看看投資規模與硬體和軟體的對比，組織資本和人力資本的投資實際上是十倍之多。 In the case of early computers, it was literally, it's literally like 10 times more investment in organizational capital and human capital if you look at the size of the investments to the hardware and software.
所以那非常大。 So that's very big.
話雖如此，我對這些想法持開放態度，願意調整，因為 ChatGPT 和其他一些工具，它們被採用的速度非常快，而且它們能夠更快地改變事情，部分原因是你不需要在同樣的程度上學習 Python。 That said, I'm open to adjusting my thoughts on this a bit because ChatGPT and some of the other tools, they have been adopted very quickly and they have much more quickly been able to change things in part because you don't need to learn Python to the same degree.
你只需用英文就可以做很多事情，或者你只需將它們放在現有組織之上就可以獲得很多價值。 You can do a lot of things just in English or you can get a lot of value just by putting them on top of the existing organization.
所以有些事情發生得更快。 So some of it's happening faster.
在一些你可能為這裡的閱讀材料讀過的論文中，我們很快就看到了 15%、20%、30% 的生產力提升。 And in some of the papers that you may have read for the readings here, we had like 15, 20, 30% productivity gains pretty quickly.
但我懷疑，一旦人們想出這些互補的創新，影響會更大。 But my suspicion is that it'll be even bigger once people figure out these complementary innovations.
所以這是回答你問題的長篇大論，不只是技術技能，而是弄清楚所有其他的事情，所有重新思考事情的方式。 And so that's a long way of answering your question about, it's not just the technical skills, it's figuring out all the other stuff, all the ways of rethinking things.
所以你們這些在商學院和經濟學系的人，在你們被賦予這套驚人的技術之後，有很多機會重新思考你們的領域。 So those of you who are at the business school and economics, there's a lot of opportunity there to rethink your areas now that you've been given this amazing set of technologies.
是的，問題。 Yeah, question.
似乎您對轉變速度的看法比之前更為謹慎。 It seems like you're expressing more caution than there was with regard to the speed of transformation.
我感覺您自己也這樣覺得，對嗎？ Am I correct in feeling yourself?
嗯，所以我想區分兩件事。 Well, so I would make a distinction between two things.
我會聽從他和別人在技術方面的意見。 I'll defer to him and others on the technologies.
我們將會聽到其他幾位的看法。 We're going to hear from several other folks.
有些人和他一樣樂觀，甚至在技術方面比他更樂觀。 There are people who are equally optimistic as him, or even more optimistic on the technology side.
也有人比較不樂觀。 There's also people who are less optimistic.
但光有技術不足以創造生產力。 But technology alone is not enough to create productivity.
所以你可以有一個驚人的技術，然後出於各種原因，A，也許人們就是想不出有效的使用方法。 So you can have an amazing technology and then for various reasons, A, maybe people just don't figure out an effective way to use it.
另一個可能是監管問題。我的一些電腦科學同事引進並開發了更好的放射學系統來判讀醫學影像。 Another is it may be regulatory things. and some of my computer science colleagues introduced and developed better radiology systems for reading medical images.
它們沒有被採用，因為文化因素，人們就是不想要。 They weren't adopted because of cultural, people just didn't want them.
他們不想要，而且有安全上的原因。 They didn't want, and there are safety reasons.
當我分析哪些任務 AI 最能幫助，哪些職業受影響最大時，我很驚訝航空公司飛行員竟然排在很前面，但我認為很多人如果沒有飛行員跟你一起下去，會感到不舒服。 When I did an analysis of which tasks AI could help the most and which professions were most affected, I was surprised that airline pilots was kind of near the top but I think that a lot of people would not feel comfortable not having the pilot go down with you.
所以他們有點——你希望有人類在裡面。 So they sort of-- you want to have the human in there.
所以有很多不同的事情可能會顯著地減緩它。 So there are a lot of different things that might slow it down significantly.
我認為那是我們需要意識到的。 And I think that's something we need to be conscious of.
如果我們能解決那些瓶頸，那對生產力的貢獻可能比只專注於技術本身更大。 And if we could address those bottlenecks, that would probably do more for productivity than just working on the technology alone.
是的，問題。 Yeah, question.
艾瑞克對大學裡的資料中心有有趣的評論。 So Eric had an interesting comment on data centers in universities.
我認為這是一個更大的問題，像是—— And I think this is a larger point of like--
我本來想問他為什麼不自己開張支票？
And I was gonna ask him why doesn't he write a checkbook?
(觀眾笑) 很多人都在問他這個問題。 (audience laughing) People are asking him that question.
有點像是，大學的生態圈該扮演什麼角色？
Sort of like what is the role of the university ecosystem?
顯然，這裡有一個更大的... 我相信在座的所有資工系教授—— Obviously there is this larger, I'm sure all of the CS professors here--
我認為，我的意思是，如果能有更多資金，那當然很棒。
So I'll take, I mean, I think it'd be great if there were more funding.
我的意思是，聯邦政府有個叫做「國家AI資源」的計畫，有提供一些幫助，但金額大概是幾百萬、幾千萬美元，而不是幾十億，更不用說幾千億美元了。 I mean, the federal government has something called the National AI Resource that is helping a little bit, but it's in like the millions of dollars, tens of millions of dollars, not billions of dollars, let alone hundreds of billions of dollars.
不過，Eric 課前確實有跟我提到，他們正在醞釀一個規模可能大很多的計畫。 Although Eric did mention to me before class that they're working on something that could be much, much bigger.
他正在力推一個規模大很多的計畫。 He's pushing for something much, much bigger.
我不知道會不會成功。 I don't know if it'll happen.
那是用來訓練這些超大型模型的。 That's for training these really large models.
我曾經和 Jeff Hinton 有過一段非常有趣的對話。 I had a really interesting conversation with Jeff Hinton once.
如你所知，Jeff Hinton 可說是深度學習的教父之一。 Jeff Hinton, as you know, is sort of like one of the godfathers of deep learning.
我問他，他覺得做研究時，哪種硬體最有用。 And I asked him what kind of hardware he found most useful for doing his work.
當時他正坐在他的筆電前，就只是輕輕地敲了敲他的 MacBook。 And he was sitting at his laptop, and he kind of just tapped his MacBook.
這就提醒了我，還有另一整塊研究領域，也許大學才具有比較優勢，那就是創新演算法，而非訓練耗資千億的模型，像是任何在 Transformer 之後出現的新東西。 And it just reminded me there's a whole other set of research that maybe universities have a competitive advantage in, which is not training $100 billion models, but is innovating new algorithms, like whatever comes after Transformers.
人們可以透過很多其他方式做出貢獻。 And there's a lot of other ways that people can make contributions.
所以，或許這裡有點像是分工的概念。 So maybe there's a little bit of a divisional waiver.
我完全贊成也支持我的同事們去爭取更多買 GPU 的預算。 I'm all for and support my colleagues asking for more budgets for GPUs.
但學術界能做出最大貢獻的地方，不見得總是在這裡。 But that's not always where academics can make the biggest contributions.
有些貢獻是來自於想法、思考事物的新穎不同視角，以及新的方法。 Some of it comes from ideas and new ways of different perspective about thinking about things, new approaches.
而那很可能就是我們的優勢所在。 And that's likely where we have an advantage.
我上週和 Sendhil Mullainathan 吃晚餐。 I had dinner with Sendhil Malanathan last week.
他剛從芝加哥大學搬到麻省理工學院。 He just moved from Chicago to MIT.
他是一位研究員。 And he was a researcher.
我們當時在聊大學的比較優勢是什麼。 We were talking about what is the comparative advantage
他提出一個論點，你知道的，耐心就是其中之一。 And he made the case, you know, patience is one of them.
大學裡有些人正在進行非常長期的計畫。 That there are people in universities who are working on very long-term projects.
你知道的，有人在研究核融合。 You know, there's people working on fusion.
他們研究核融合已經很長一段時間了。 They've been working on fusion for a long time.
不是因為他們今年、或可能十年後、甚至二十年後，能靠蓋一座核融合發電廠賺大錢。 Not because they're going to get, you know, a lot of money this year or 10 years from now probably from building a fusion plant or even 20 years.
我不知道核融合還需要多久。 I don't know how long it is for fusion.
但你知道，這就是即使時程遙遠，人們也願意投入心力的事情。 But you know, it's just something that people are willing to work on even if the timelines are a little further.
企業比較難以負擔這麼長的研發時程。 It's harder for companies to afford to have those kinds of timelines.
所以，就大學可能做到的事情而言，這裡存在著比較優勢或是一種分工。 So there's a comparative advantage or divisional labor in terms of what universities might be able to do.
我們只剩下幾分鐘了。 We have just a couple of minutes left.
這還蠻有趣的。 This is kind of fun.
所以我們再問一兩個問題就好。 So we'll just do one or two more questions.
然後我想花點時間談一下專案。 And then I want to talk a little bit about the projects.
好的，請說。 Yeah. Go ahead.
是的，我想發表一個評論。 Yeah. I'm a comment.
我在想我們之前討論到的 AI 的湧現能力。 I was wondering about the emerging capabilities of AI that we discussed.
是的。 Yeah.
Eric 似乎比較傾向於架構上的差異和設計更好的模型，而我們上堂課反而比較多在談小模型。 It seemed that Eric was leaning more towards the architectural differences and design better models versus last class, we talked about more small instead.
所以我想知道您如何... >> 嗯，他說了全部三種。 So I wonder how you sort of >> Well, he said all three.
你們還記得規模化法則 (scaling laws) 嗎？ So you guys remember the scaling laws?
它大概有三個部分。 It had like three parts to it.
我想我之前有放上 Dario 和他團隊提出的那個規模化法則。 I think I put the scaling law that like Dario and team.
就是更多的運算、更多的資料，以及演算法的改進，包含更多的參數。我印象中聽到 Eric 說這三者都很重要，但不要忽略了最後一項，像是新的架構。 So there's more compute, more data, and algorithmic improvements, including more parameters, and all three of them, I think I heard Eric say all three of them were important, but not to be dismissed this last one like new architectures.
我想這三者都很重要。 All three of them I think are being important.
我想那裡面其實還包含了另一個問題。 So I think there was another question in there though also.
就是有了這些大型語言模型後，我們離通用人工智慧 (AGI) 系統近了多少？ So how much closer are we to like an AGI type system with these larger language models?
Eric 認為我們離通用人工智慧系統並沒有那麼近，不過我也不認為那有個很明確的定義。 So Eric doesn't think we're like that close to AGI type systems, although I don't think it's like a sharp definition.
事實上，那也是我本來要問他的問題之一，但我們時間不夠了。 In fact, that was one of the I was going to ask him that question, but we ran out of time.
如果能聽他描述一下會很棒。 It would have been good to hear him describe it.
但當我跟他聊的時候，就覺得那真的不是個定義很明確的東西。 But when I was talking to him, It's just not that sharply a defined thing.
你知道，在某些方面，通用人工智慧已經出現了。 You know, in some ways AGI is already here.
Peter Norvig 寫過一篇文章，標題是《通用人工智慧已然降臨》(AGI is Already Here)。 Peter Norvig wrote an article called AGI is Already Here.
我不確定有沒有在閱讀教材裡。 I don't know if it's in the reading packet.
我想如果沒有的話，我會把它放進去。 I think if it's not, I'll put it in there.
那是篇和 Blaise Iarca、Gary Iarca 合寫的有趣短文。 It's a fun little article with Blaise Iarca, Gary Iarca.
很多 20 年前人們會說「這就是通用人工智慧」的事情，現在的大型語言模型 (LLMs) 差不多都能做到。 And a lot of the things that 20 years ago people would have said, this is what AGI is, that's kind of what LLMs are doing.
也許做得沒那麼好，但它確實是在用更通用的方式解決問題。 Not as well, maybe, but it's sort of solving problems in a more general way.
另一方面，顯然它們目前在很多事情上做得比人類差得多。 On the other hand, there's obviously many things they do much worse than humans currently.
諷刺的是，體力活是人類目前仍保有比較優勢的項目之一。 Ironically, physical tasks are one of the ones that humans have a comparative advantage in right now.
你們可能知道莫拉維克悖論 (Moravec's paradox)。 And you guys may know of Moravec's paradox.
漢斯·莫拉維克 (Hans Moravec) 指出，那些三、四歲小孩能做到的事情，像是扣襯衫釦子或走上樓梯，通常很難讓機器做到。 Hans Moravec pointed out that often the kinds of things that a three-year-old or a four-year-old can do, like buttoning a shirt or walking upstairs are very hard to get a machine to be able to do.
然而，許多博士都覺得很困難的事情，像是解決凸優化問題 (convex optimization problems)，機器卻常常很擅長。 Whereas a lot of things that a lot of PhDs have trouble doing, like solving convex optimization problems, are things that machines are often quite good at.
所以這不完全是... 對人類簡單、對電腦困難的事，和對人類困難、對電腦簡單的事，這兩者不是在同一個尺度上的。 So it's not quite a-- things that are easy for humans and hard for computers and other things that are hard for humans and easy for computers, they're not the same scale.
下週，我們的來賓是 OpenAI 的技術長 Mira Murati，她也曾短暫擔任過 OpenAI 的執行長。 And next week, we have Mira Moradi, Chief Technology Officer of OpenAI, briefly the CEO of OpenAI.
所以帶著你們的問題來問她吧。 And so come with your questions for her.
我們到時見。 We'll see you.
