
這裡有很多問題。 There are a lot of questions here.
是的,女士。 Yes, ma'am.
嗨。 Hi.
在最一開始——在最一開始,您提到 context window 的擴展、代理人和文本到行動的組合將產生難以想像的影響。 So at the very beginning-- at the very beginning, you mentioned that there's the combination of the context window expansion, the agents, and the text-to-action is going to have unimaginable impacts.
首先,為什麼這個組合很重要? First of all, why is the combination important?
其次,我知道您不像水晶球,不一定能預知未來,但您為什麼認為這超乎我們的想像? And second of all, I know that you're not like a crystal ball and you can't necessarily tell the future, but why do you think it's beyond anything that we could imagine?
我認為主要是因為 context window 讓你能解決即時性的問題。 I think largely because the context window allows you to solve the problem of recency.
目前的模型訓練大約需要一年——有 18 個月——六個月的準備、六個月的訓練、六個月的微調。 The current models take a year to train, roughly six-- there's 18 months-- six months of preparation, six months of training, six months of fine-tuning.
所以它們總是過時的。 So they're always out of date.
Context window,你可以輸入發生的事情。 Context window, you can feed what happened.
比如,你可以問它關於哈瑪斯-以色列戰爭的問題,對吧,在一個 context 裡。 Like, you can ask it questions about the Hamas-Israel war, right, in a context.
這非常強大。 That's very powerful.
它變得像 Google 一樣即時。 It becomes current like Google.
在代理人的例子中,我舉個例子。 In the case of agents, I'll give you an example.
我成立了一個基金會,資助一個非營利組織。 I set up a foundation, which is funding a non-profit.
它始於,有一個,我不知道房間裡有沒有化學家,我不太懂化學。 It starts with, there's a, I don't know if there's chemists in the room, I don't really understand chemistry.
有一個叫做 ChemCrow 的工具,C-R-O-W,是一個基於大型語言模型的系統,它學習化學。 There's a tool called ChemCrow, C-R-O-W, which was an LLM-based system that learned chemistry.
他們做的是運行它來產生關於蛋白質的化學假設,然後他們有一個實驗室連夜進行測試,然後它就學習了。 And what they do is they run it to generate chemistry hypotheses about proteins, and then they have a lab which runs the tests overnight, and then it learns.
這在化學、材料科學等領域是一個巨大的加速器。 That's a huge acceleration, accelerant in chemistry, material science, and so forth.
所以那是一個代理模型。 So that's an agent model.
我認為「文本到行動」可以理解為擁有大量便宜的程式設計師。 And I think the text to action can be understood by just having a lot of cheap programmers.
我認為我們不了解當每個人都有自己的程式設計師時會發生什麼——這又是你的專業領域——當每個人都有自己的程式設計師時會發生什麼。 And I don't think we understand what happens-- and this is, again, your area of expertise-- what happens when everyone has their own programmer.
我不是說開關燈。 And I'm not talking about turning on and off the lights.
我想像——另一個例子——出於某種原因,你不喜歡 Google。 I imagine-- another example-- for some reason, you don't like Google.
所以你說,給我做一個 Google 的競爭對手。 So you say, build me a Google competitor.
是的,你個人——給我做一個 Google 的競爭對手,搜尋網路,建立一個使用者介面,做一個好的複製品,以有趣的方式加入生成式 AI,在 30 秒內完成,看看它是否有效。 Yeah, you personally-- build me a Google competitor, search the web, build a UI, make a good copy, add generative AI in an interesting way, do it in 30 seconds, and see if it works.
所以很多人相信,包括 Google 在內的現有業者,都可能受到這種攻擊。 So a lot of people believe that the incumbents, including Google, are vulnerable to this kind of an attack.
現在,我們拭目以待。 Now, we'll see.
Slido 上寄來了一堆問題。 There were a bunch of questions that were sent over by Slido.
其中一些被頂了上來。 Some of them were uploaded.
這裡有一個,我們去年談過一點。 Here's one, we talked a little bit about this last year.
我們如何阻止 AI 影響公眾輿論、散播假訊息,尤其是在即將到來的選舉期間? How can we stop AI from influencing public opinion, misinformation, especially during the upcoming election?
短期和長期的解決方案是什麼? What are the short and long-term solutions from?
在即將到來的選舉和全球範圍內,大部分的假訊息將會出現在社群媒體上。
Most of the misinformation in this upcoming election and globally will be on social media.
而社群媒體公司並沒有足夠好的組織來監管它。 And the social media companies are not organized well enough to police it.
如果你看看 TikTok,舉個例子,有很多指控說 TikTok 偏袒某一種假訊息。 If you look at TikTok, for example, there are lots of accusations that TikTok is favoring one kind of misinformation over another,
有很多人聲稱,雖然我沒有看到證據,說中國人強迫他們這麼做。 And there are many people who claim without proof that I'm aware of that the Chinese are forcing them to do it.
我認為我們這裡一團亂。 I think we have a mess here.
這個國家將必須學習批判性思考。 And the country is going to have to learn critical thinking.
對美國來說,這可能是一個不可能的挑戰。 That may be an impossible challenge for the US.
但有人告訴你某件事,不代表那是真的。 But the fact that somebody told you something does not mean that it's true.
會不會走向另一個極端,有些事情真的是真的,卻沒有人再相信了? Could it go too far the other way, that there's things that really are true and nobody believes anymore?
有些人稱之為「知識論危機」(epistemological crisis),現在,你知道,伊隆(Elon)說:「不,我從沒做過那件事,證明給我看。」 You get some people call it a epistemological crisis, that now, you know, Elon says, "No, I never did that, prove it."
好吧,我們用唐納·川普(Donald Trump)做例子。
Well, let's use Donald Trump.
好的。
Okay.
聽著,我認為我們的社會存在信任問題。
Look, I think we have a trust problem in our society.
民主制度可能會失敗。 Democracies can fail.
我認為對民主最大的威脅是假消息,因為我們將會變得非常擅長製造假消息。 And I think that the greatest threat to democracy is misinformation, because we're gonna get really good at it.
當我管理 YouTube 時,我們在 YouTube 上最大的問題是,人們會上傳假的影片,結果導致有人死亡。 When I ran, managed YouTube, the biggest problems we had on YouTube were that people would upload false videos and people would die as a result.
我們有「禁止死亡」的政策,很驚人吧。 And we had a no-death policy, shocking.
我們只是,試圖解決這個問題真是太可怕了。 And we just went, it was just horrendous to try to address this.
而這是在生成式 AI 出現之前。 And this is before generative AI.
嗯,所以——
Well, so--
我沒有好的答案。
I don't have a good answer.
一個技術性的,這不是答案,但有一件事似乎可以緩解這個問題,我不明白為什麼沒有更廣泛地使用,那就是公鑰認證。
One technical, it's not an answer, but one thing that seems like it could mitigate that I don't understand why it's more widely used is public key authentication.
當喬·拜登(Joe Biden)講話時,為什麼不像 SSL 那樣進行數位簽章? That when Joe Biden speaks, why isn't it digitally signed like SSL is?
或者當,你知道,名人或公眾人物或其他人,他們不能有一個公鑰嗎? Or when, you know, that celebrities or public figures or others, couldn't they have a public key?
是的,這是一種公鑰的形式,然後是某種形式的確定性,知道系統是如何進來的。
Yeah, it's a form of public key and then some form of certainty of knowing how the system came in.
是的,當我把我的信用卡送到亞馬遜時,我知道那是亞馬遜。
Yeah, when I send my credit card to Amazon, I know it's Amazon.
我寫了一篇論文並與 Jonathan Haidt 一起發表了,他就是那個研究焦慮世代的人。
I wrote a paper and published it with Jonathan Haid, who's the one working on the anxiety generation stuff.
它的影響完全是零。 It had exactly zero impact.
他是一個非常好的溝通者,我可能不是。 And he's a very good communicator, I probably am not.
所以我得出的結論是,這個系統並沒有像你說的那樣組織起來。 So my conclusion was that the system is not organized what you said.
你有一篇論文主張我們剛才說的? You had a paper advocating what we just said?
主張你的提議。 Advocating your proposal.
好的,我的提議。 Okay, my proposal.
不,你說的。 No, what you said.
是的,對的。 Yeah, right.
我的結論是,執行長們通常都在最大化營收。 And my conclusion is the CEOs in general are maximizing revenue.
為了最大化營收,你最大化參與度。 To maximize revenue, you maximize engagement.
為了最大化參與度,你最大化憤怒。 To maximize engagement, you maximize outrage.
演算法選擇憤怒,因為那能產生更多營收,對吧? The algorithms choose outrage because that generates more revenue, right?
因此,有一種偏向於支持瘋狂事物的偏見。而且是各個方面的。 Therefore, there's a bias to favor crazy stuff. and on all sides.
我不是在做黨派聲明。 I'm not making a partisan statement here.
那是個問題。 That's a problem.
這在民主國家必須得到解決。 That's got to get addressed in a democracy.
而我對 TikTok 的解決方案,我們稍早私下談過,是我小時候,有一個叫做「平等時間原則」(equal time rule)的東西,因為 TikTok 其實不是社群媒體,它其實是電視,對吧? And my solution to TikTok, we talked about this earlier privately, is there was when I was a boy, there was something called the equal time rule, because TikTok is really not social media, it's really television, right?
有一個程式設計師在幫你做數字,順便說一句,數字是每天 90 分鐘,每個 TikTok 用戶 200 個 TikTok 影片。 There's a programmer making you the numbers, by the way, are 90 minutes a day, 200 TikTok videos per TikTok user in the And the government is not going to do the equal time rule, but it's the right thing to do.
某種形式的平衡是必要的。 Some form of balance that is required.
好的,讓我們再問一些問題。
All right, let's take some more questions.
兩個簡短的問題。
Two quick questions.
第一,大型語言模型的經濟影響,像是勞動市場的影響,比你最初預期的要慢,這是 Chegg 和一些服務業的人,然後第二,你認為學術界值得,或應該得到 AI 補助嗎,還是你認為他們應該直接和業界的大公司合作? One, economic impact of LM's, slower like labor market impacts, slower than you originally anticipated, and this is Chag and a couple of service people, and then two, do you think academia deserves, or should get AI subsidies, Or do you think they should just partner with big players out there?
他們不能——我非常、非常努力地推動為大學建立資料中心。 And they could not-- I pushed really, really hard on getting data centers for universities.
如果我是這裡電腦科學系的教員,我會非常不爽,因為我無法和我的研究生一起建立能做那種博士研究的演算法。 If I were a faculty member in the computer science department here, I would be beyond upset that I can't build the algorithms with my graduate students that will do the kind of PhD research.
我被迫和這些公司合作。 And I'm forced to work with these.
在我看來,公司在這方面還不夠慷慨。 And the companies have not, in my view, been generous enough with respect to that.
我交談過的教職員,其中很多你都認識,他們花了很多時間等待 Google Cloud 的點數,他們拼湊著用,那太糟糕了。 The faculty members that I talk with, many of whom you know, spend lots of time waiting for their credits from Google Cloud, and they piece, that's terrible.
這是一場爆炸性的發展,我們希望美國贏,我們希望美國的大學、美國,有很多理由認為正確的做法是把資源給他們。 This is an explosion, we want America to win, we want American universities, America, there's lots of reasons to think that the right thing to do is to get it to them.
所以我正在為此努力。 So I'm working hard on that.
你第一個問題是勞動市場影響? And your first question was labor market impact?
我把這個問題留給真正的專家。 I'll defer to the real expert here.
作為你的業餘經濟學家,由艾瑞克(Eric)教導,我從根本上相信,那種大學教育、高技能的工作會沒事,因為人們會和這些系統一起工作。 As your amateur economist, taught by Eric, I fundamentally believe that the sort of college education, high skills task will be fine because people will work with these systems.
我認為這些系統和任何其他科技浪潮沒有不同。 I think the systems is no different from any other technology wave.
危險的工作和需要很少人類判斷的工作將被取代。 The dangerous jobs and the jobs which require very little human judgment will get replaced.
我們大概還有五分鐘。
We got about five minutes left.
所以我們快點問些問題吧。 So let's go really quick with some questions.
我讓你來選,艾瑞克。 I'll let you pick them, Eric.
是的,女士。
Yes, ma'am.
我對「文本到行動」對電腦科學教育的影響非常好奇。
I'm really curious about the text to action impact on, for example, computer science education.
我想知道您對於電腦科學教育應如何轉變以迎接這個時代有什麼想法。 I'm wondering what you have thoughts on how CS education should transform to meet the age.
嗯,我假設大學部的電腦科學家作為一個群體,身邊總會有一個程式設計師夥伴。
Well, I'm assuming that computer scientists as a group in undergraduate school will always have a programmer buddy with them.
所以當你學習第一個 for 迴圈等等時,你會有一個工具成為你的天然夥伴。 So when you learn your first for loop and so forth and so on, you'll have a tool that will be your natural partner.
教學就是這樣進行的。教授,他或她,會講授概念,但你會用那種方式參與其中。 And that's how the teaching will go on. that the professor, he or she, will talk about the concepts, but you'll engage with it that way.
這就是我的猜測。 And that's my guess.
是的,女士,在你後面。 Yes, ma'am, behind you.
是的,您能多談談您感到興奮的非 Transformer 架構嗎? Yeah, can you talk more about the non-transformer architectures that you're excited about?
我認為被談論過的一個是狀態模型,但現在隨著所有 context 的增加,所以我想知道您在這個領域看到了什麼。 I think one that's been talked about is state models, but then now with all the context More, so I was curious what you're seeing in space.
我對數學的理解不夠深入。 I don't understand the math well enough.
我很高興我們為數學家創造了工作機會,因為這裡的數學太複雜了。 I'm really pleased that we have produced jobs for mathematicians, because the math here is so complicated.
但基本上,它們是做梯度下降、矩陣乘法不同方式,更快更好。 But basically, they are different ways of doing gradient descent, matrix multiply, faster and better.
而 Transformer,如你所知,是一種系統化的同時進行乘法的方式。 And transformers, as you know, is a sort of systematic way of multiplying at the same time.
這是我思考的方式。 That's the way I think about it.
它與此類似,但數學不同。 And it's similar to that, but different math.
我們看看,這邊。 Let's see, over here.
是的,先生。 Yes, sir.
請說。 Go ahead.
是的。 Yeah.
您在關於國家安全的論文中提到,您有中國和美國以及其他擁有千億級能力的國家,今天的下一個十個集群要麼是美國的盟友,要麼是很好地準備成為美國的盟友。 You mentioned in your paper on national security as you have China and the US and the other quadrillion capabilities today, the next 10 from that next cluster down are all either US allies or teed up nicely to the US allies.
我很好奇您對那 10 個處於中間、尚未正式成為盟友的國家的看法。 I'm curious what your take is on those 10 that are sort of in the middle, that aren't formally allies.
他們有多大可能加入確保我們優勢的僵局,又有哪些因素會阻礙他們加入。 What is stuff, how likely are they to get on board with securing our superiority deadlock and what would hold them back and wanting to get on board.
最有趣的國家是印度,因為頂尖的 AI 人才從印度來到美國。 The most interesting country is India, because the top AI people come from India to the US.
我們應該讓印度保留一些頂尖人才。 And we should let India keep some of its top talent.
不是全部,但一部分。 Not all of them, but some of them.
他們沒有我們這裡如此豐富的訓練設施和計畫。 And they don't have the kind of training facilities and programs that we so richly have here.
對我來說,印度在這方面是個重要的搖擺國家。 To me, India is the big swing state in that regard.
中國已經離開。 China's lost.
它不會回來了。 It's not going to come back.
他們不會改變政權,儘管人們希望他們這麼做。 They're not going to change the regime as much as people wish them to do.
日本和韓國顯然在我們的陣營。 Japan and Korea are clearly in our camp.
台灣是一個很棒的國家,但軟體很糟糕,所以那行不通。 Taiwan is a fantastic country whose software is terrible, so that's not gonna work.
硬體驚人。 Amazing hardware.
世界其他地方,沒有太多其他好的大國選擇。 And in the rest of the world, there are not a lot of other good choices that are big.
歐洲因為布魯塞爾而搞砸了。 Europe has screwed up because of Brussels.
這不是什麼新鮮事。 It's not a new fact.
我花了 10 年和他們鬥爭。 I spent 10 years fighting them.
我非常努力地讓他們修改歐盟法案,但他們仍然有所有那些使得在歐洲進行我們這種研究非常困難的限制。 And I worked really hard to get them to fix the EU Act, and they still have all the restrictions that make it very difficult to do our kind of research in Europe.
我的法國朋友們把所有時間都花在和布魯塞爾的鬥爭上。 My French friends have spent all their time battling Brussels.
馬克宏(Macron),我的私交好友,正在為此努力奮鬥。 And Macron, who's a personal friend, is fighting hard for this.
所以我認為法國有機會。 And so France, I think, has a chance.
我看不到德國會跟上,剩下的都不夠大。 I don't see Germany coming, and the rest is not big enough.
是的,女士? Yes, ma'am?
我知道您是工程師出身,而且您有編譯器。 So I know you're an engineer by training, and you have a compiler.
鑑於您預見這些模型將擁有的能力,我們還應該花時間學習寫程式嗎? Given the capabilities that you envision these models having, should we still spend time learning to code?
是的,因為最終,這是老問題,為什麼你能說英語還要學英語? Yeah, because ultimately, it's the old thing of, why do you study English if you can speak English?
你會變得更好。 You get better at it.
你真的需要了解這些系統是如何運作的,我非常強烈地感覺到——是的,先生? You really do need to understand how these systems work, and I feel very strongly-- yes, sir?
是的,我好奇您是否探索過分散式設定,我這麼問是因為,當然,建立一個大型叢集很困難,但 MacBook 很強大。 Yeah, I'm curious if you've explored the distributed setting, and I'm asking because, sure, making a large cluster is difficult, but MacBooks are powerful.
全世界有很多小型機器。 There's a lot of small machines across the world.
所以您認為 Folding@home 或類似的想法對訓練這些系統有效嗎? So do you think folding at home or a similar idea works for training these systems?
是的,我們非常仔細地研究過這個。 Yeah, we've looked very hard at this.
演算法的運作方式是,你有一個非常大的矩陣,基本上你有一個乘法函數。 So the way the algorithms work is you have a very large matrix and you have essentially a multiplication function.
把它想成是來來回回、來來回回。 So think of it as going back and forth and back and forth.
這些系統完全受限於記憶體到 CPU 或 GPU 的速度。 And these systems are completely limited by the speed of memory to CPU or GPU.
事實上,下一代 NVIDIA 晶片已將所有這些功能整合到一個晶片中。 And in fact, the next iteration of NVIDIA chips has combined all those functions into one chip.
晶片現在變得如此之大,以至於它們把所有東西都黏在一起。 The chips are now so big that they glue them all together.
事實上,封裝非常敏感,封裝和晶片本身都是在無塵室中組裝的。 And in fact, the package is so sensitive, the package is put together in a clean room as well as the chip itself.
所以答案看起來是超級電腦和光速,特別是記憶體互連,真的主導了一切。 So the answer looks like supercomputers and speed of light, especially memory interconnect, really dominated.
所以我認為暫時不太可能。 So I think unlikely for a while.
有沒有辦法分割大型語言模型?
Is there a way to segment the LLM?
傑夫·迪恩(Jeff Dean)去年在這裡演講時,談到將它分成不同的部分,你可以分開訓練,然後再進行某種類似聯邦式的整合。 So Jeff Dean, last year when he spoke here, talked about having these different parts of it that you would train separately and then kind of federate.
為了做到那樣,你必須有一千萬個這樣的東西,然後你問問題的方式會太慢。
In order to do that, you'd have to have 10 million such things then the way you ask the questions would be too slow.
他說的是八個、十個或十二個。 He's talking about eight or 10 or 12.
是的,是的,是的,是的,是的。
Yeah, yeah, yeah, yeah, yeah.
所以不是降到——的層級。 So not down to the level of--
不是在他的層級。
Not at his level.
是的,好的。
Yeah, all right.
後面,是的,最後面。
See in the back, yes, way back.
我知道,像是 FTP2 發布後,《紐約時報》控告 Open Data 使用他們的作品進行訓練。
I know, like after FTP2 was released in the New York Times sued Open Data for using their works for training.
您認為這會走向何方,對資料隱私又意味著什麼? Where do you think that's gonna go and what that means for data privacy?
我以前做過很多音樂授權方面的工作。
I used to do a lot of work on the music licensing stuff.
我學到的是,在 60 年代,有一系列訴訟導致了一項協議,每當你的歌曲被播放時,你就會得到一筆規定的版稅。 What I learned was that in the '60s, there was a series of lawsuits that resulted in an agreement where you get a stipulated royalty whenever your song is played.
他們甚至不知道你是誰。 They don't even know who you are.
錢就直接付進銀行了。 It's just paid into a bank.
我猜結果會一樣。 And my guess is it'll be the same thing.
會有很多訴訟,然後會有某種規定的協議,就只是說,你必須支付你收入的 x% 才能使用 ASCAP EMI。 There'll be lots of lawsuits, and there'll be some kind of stipulated agreement, which will just say, you have to pay x percent of whatever revenue you have in order to use the ASCAP EMI.
ASCAP EMI。 ASCAP EMI.
查一下吧。 Look them up.
這很長——對你們來說會顯得很古老,但我認為最終會是這樣——是的,先生? It's a long-- it will seem very old to you, but I think that's how it will ultimately-- yes, sir?
似乎有少數幾家公司在主導 AI,而且他們會繼續主導。
It seems like there's a few players that are dominating AI and they'll continue to dominate.
他們似乎與所有反壟斷法規關注的大公司重疊。 And they seem to overlap with the large companies that all the antitrust regulation is kind of focused on.
您如何看待這兩種趨勢,是的,您認為監管機構會拆分這些公司嗎?那又會如何影響,是的。 How do you see those two trends kind of, yeah, do you see regulators breaking up these companies and how will that affect the, yeah.
在我的職業生涯中,我幫助微軟被拆分,但它沒有被拆分。
So in my career, I helped Microsoft get broken up and it wasn't broken up.
我為 Google 不被拆分而奮鬥,而它也沒有被拆分。 And I fought for Google to not be broken up and it's not been broken up.
所以對我來說,趨勢似乎是不會被拆分。 So it sure looks to me like the trend is not to be broken up.
只要公司避免成為老約翰·戴維森·洛克菲勒(John D. Rockefeller Sr.),我研究過這個,查一下,反托拉斯法就是這麼來的,我認為政府不會採取行動。 As long as the companies avoid being John D. Rockefeller the senior, and I studied this, look it up, it's how antitrust law came, I don't think the governments will act.
你看到這些大公司主導的原因是,誰有資本來建造這些資料中心? The reason you're seeing these large companies dominate is who has the capital to build these data centers?
所以我的朋友里德(Reed)和我的朋友穆斯塔法(Mustafa)—— So my friend Reed and my friend Mustafa--
他下週,兩週後會來。
He's coming next week, two weeks from now.
讓里德跟你談談他們決定把 Inflection 基本上拆分到微軟的決定。 have Reed talk to you about the decision that they made to take inflection and essentially piece part it into Microsoft.
基本上,他們決定他們無法籌集到數百億美元。 Basically, they decided they couldn't raise the tens of billions of dollars.
你剛才提到的那個數字是公開的嗎?
Is that number public that you mentioned earlier?
不是。
No.
讓里德告訴你那個數字。 Have Reed give you that one.
好的,也許里德可以說。
Okay, maybe Reed can say it.
我知道你得走了,我不想耽誤你,但我想留給你—— I know you gotta go, I don't wanna hold you, but I wanna leave you with--
我們該問這位先生嗎?
Should we do this gentleman?
我也有個問題想問你。
I also have a question for you.
再一個。
One more.
是的,請說。
Yeah, go ahead.
非常感謝。(觀眾笑)非常感謝,我會快點問。
Thank you so much. (audience laughing) Thank you so much, I'll make it quick.
我想知道這一切將會把那些不參與開發志願者模型和無法獲得運算資源的國家帶向何方,例如? I was wondering where all of this is going to lead countries who are non-participants in development volunteer models and access to compute, for example?
富者愈富,貧者盡力而為。
The rich get richer, and the poor do the best they can.
他們將不得不,事實是,這是富國的遊戲,對吧? They'll have to, the fact of the matter is, this is a rich country's game, right?
巨大的資本,大量技術強大的人才,強大的政府支持,對吧? Huge capital, lots of technically strong people, strong government support, right?
有兩個例子。 There are two examples.
還有很多其他國家有各種各樣的問題,他們沒有那些資源,他們必須找到合作夥伴,他們必須與其他人聯合,類似這樣的事情。 There are lots of other countries that have all sorts of problems, they don't have those resources, they'll have to find a partner, they'll have to join with somebody else, something like that.
我想留下這個問題,因為我想上次你在 AGI House 的一個黑客松,我知道你花了很多時間幫助年輕人,因為他們創造了很多財富,而且你非常熱情地談到想做這件事。
I wanna leave it, 'cause I think the last time You were at a hackathon at AGI House, and I know you spent a lot of time helping young people as they create a lot of wealth, and you spoke very passionately about wanting to do that.
在他們職業生涯的這個階段,當他們為這門課寫商業計劃書、政策提案或研究提案時,你有什麼建議嗎? Do you have any advice for folks here as they're writing their business plans for this class, or policy proposals, or research proposals, at this stage of their careers going forward?
嗯,我在商學院教一門關於這個的課,所以你應該來上我的課。
Well, I teach a class in the business school on this, so you should come to my class.
我對你們能多快地建立新想法的展示感到震驚。 I am struck by the speed with which you can build demonstrations of new ideas.
在我做的一個黑客松中,獲勝的團隊,指令是,讓無人機在兩座塔之間飛行,它得到了一個虛擬的無人機空間。 So in one of the hackathons I did, the winning team, the command was, fly the drone between two towers, and it was given a virtual drone space.
它想出了如何駕駛無人機,什麼是「之間」這個詞的意思,用 Python 生成了程式碼,然後在模擬器中讓無人機飛過塔樓。 And it figured out how to fly the drone, what the word between meant, generated the code in Python, and flew the drone in the simulator through the tower.
我只是,讓優秀的專業程式設計師來做,可能需要一兩個星期。 I just, it would have taken a week or two from good professional programmers to do that.
我告訴你們,快速原型製作的能力,成為企業家的一部分問題是一切都發生得更快。 I'm telling you that the ability to prototype quickly, part of the problem with being an entrepreneur is everything happens faster.
現在,如果你不能在一天內用這些各種工具建立你的原型,你需要思考一下。 Well now, if you can't get your prototype built in a day using these various tools, you need to think about that.
因為你的競爭對手就是這麼做的。 Because that's who your competitor is doing.
所以我最大的建議是,當你開始考慮一家公司時,寫一份商業計劃書是好的。 So I guess my biggest advice is when you start thinking about a company, it's fine to write a business plan.
事實上,你應該讓電腦為你寫商業計劃書,只要是合法的。 In fact, you should ask the computer to write your business plan for you, as long as it's legal.
不,不,是的,我應該在你離開後談談這個。(觀眾笑)
No, no, yeah, I should talk about that after you leave this. (audience laughing)
但我認為,盡快使用這些工具將你的想法原型化非常重要,因為你可以肯定有另一個人在另一家公司、另一所大學、一個你從未去過的地方做著完全相同的事情。
But I think it's very important to prototype your idea using these tools as quickly as you can, because you can be sure there's another person doing exactly that same thing in another company, in another university, in a place that you've never been.
好的,非常感謝你,艾倫。
All right, well, thanks very much, Aaron.
謝謝大家。
Thank you all.
我得趕緊走了。
I'm gonna rush off.
謝謝。 Thank you.
[掌聲] 謝謝。 [APPLAUSE] Thank you.
[掌聲] 其實,讓我接著最後一點說,因為我想我第一堂課沒談到使用大型語言模型(LLM),這門課歡迎使用 LLM 做作業。 [APPLAUSE] So actually, let me pick up on that very last point, because I don't think I talked about in the first class about using LLMs, which is a welcome in this class for the assignments.
但必須完全公開。 But it has to be full disclosure.
所以當你使用它們時,無論是每週的作業還是期末專案或其他任何東西,就像你問你友善的叔叔或同學或其他人給你建議一樣,你應該這麼做,或者如果你有筆記,你把它們包含進去。 So when you use them, whether it's for the weekly assignments or for the final project or whatever, just like you would if you asked your friendly uncle or classmate or anybody else and give you advice you should do that or if you have notes that you include in there.
所以我想做的是,我想談談 AI,它是 GPT 嗎?以及這在商業和影響方面意味著什麼。 So what I thought I'd do is I want to talk a little bit about AIs, is it GPT and what that means in terms of business and implications.
但在我們開始之前,我只是想看看你們有沒有什麼問題想接著艾瑞克(Eric)提到的事情問,我會試著傳達他的一些想法,我們談談那些出現的事情,然後我們可以繼續。 But before we do that, I just want to see if there are any questions you want to pick up on things that Eric brought up that I'll try and channel some of his thoughts and we about the things that came up and then we can move on.
是的,請說。 Yeah, go ahead.
我想問的一個問題是關於監管,如果目標是維持霸權,你要如何創造正確的誘因,讓每個人,無論是盟友還是非盟友,都有動力去遵守?
One of the questions I want to ask is in relation to regulation, if the goal is to maintain supremacy, how do you create the right incentives so that everyone, allies and non-allies, are motivated to follow it?
你是說在相互競爭的公司之間嗎?
You mean among companies that are competing with each other?
公司或國家。
Companies or in countries.
國家。
Countries.
在美國和歐盟,它不只是變成一種妨礙或阻礙那些選擇遵守法規的國家的發展。
In the US and in the EU, and it doesn't just become sort of a ham or obstruct kind of development for the ones that choose to follow the regulations.
這非常棘手。
It's super tricky.
有一本書,《競合策略》(Co-opetition),是貝利·奈勒波夫(Barry Nalebuff)寫的,因為在某些地方,法規確實可以幫助公司和幫助一個行業生存。 There's a book, Co-opetition, that Mary Nailboff wrote about this, because there are definitely places where regulation can help companies and help an industry survive.
所以法規不一定會減緩事情。 So regulation doesn't necessarily slow things.
我的意思是,標準就是個好例子。 I mean, standards are a good example.
澄清了這一點可以讓公司更容易競爭。 And having that clarified can make it easier for companies to compete.
所以我跟很多這些公司的高層談過,有些地方他們希望有一些共同的標準。 So I've talked to a lot of the executives of these companies, and there are places where they wish there were some common standards.
有時候在一些危險的事情上也會有一點逐底競爭。 And sometimes there's a bit of a race to the bottom as well in some of the dangerous things.
Google 的人說他們之所以沒有那麼快行動,另一個原因是他們覺得這些大型語言模型可能會被濫用或有危險,但他們的情況有點是被迫的。 One of the other reasons that the folks at Google say that they didn't move as fast is they felt like these LMs could be misused or dangerous, but their hand was sort of forced.
我跟另一家大公司的一些人聊過,他們說:「我們本來不打算發布這個功能,但現在競爭對手都在做,所以我們也必須發布了。」 And I was talking to some folks at one of the other big companies, and they said, "We weren't going to release this feature, but now competitors are doing it, so we're going to have to release it as well."
所以這就是監管——可能在協調監管方面會有一些利益。 So this is where regulation-- there might be some interest in coordinating on regulation.
但顯然,更明顯的是,它被用來阻礙競爭。 But it's also, obviously, the more obvious thing is that it is used to hinder competition.
例如,很多人認為一些大公司之所以非常反對某些開源和讓事情更廣泛開源,是因為他們想減緩競爭對手的速度。 And a lot of people, for instance, think that the reasons that some of the big companies are very opposed to some of the open source and making things more widely open source is they want to slow down competitors.
所以這兩件事都在發生。 So there's both of those things going on.
是的,那邊有個簡短的問題。 Yeah, quick question over there.
是的。 Yeah.
我只是想接著問—— I just want to follow up on--
我剛才在說,我們還應該學寫程式嗎? I was talking about, should we still learn to code?
我們還應該學英文嗎? Should we still study English?
那些還有用嗎? Are those gonna be useful?
艾瑞克(Eric)的回答是,是的,受過大學教育、高技能的工作或任務仍然會是安全的,但其他所有在停車場或開車的工作可能就不會了。 And Eric's reply, yes, college educated, high skilled jobs or tasks are still gonna be safe, but everything else that's in a parking or driving might not be.
這有點像故意的。 That's kind of like intentional.
我想我們可以談談,也許幾分鐘後我們會再談談,但思考一下 AI 系統在哪裡只是取代了人們正在做的事情,又在哪裡補充了他們,這很有趣。
I think we could talk, maybe we'll talk some more about that in a few minutes, but it is interesting to think about where the AI systems sort of just replace what people are doing versus they complement them.
在程式設計領域,目前看來,它們對真正頂尖的程式設計師並沒有那麼有幫助。 And in coding right now, it appears that they're not actually that helpful for the really best coders.
它們對中等程度的程式設計師非常有幫助。 They're very helpful for moderately good coders.
但如果你對程式設計一無所知,它們也沒什麼幫助。 But if you don't know anything at all about coding, they're not helpful either.
所以這有點像一個倒 U 型。 So it's kind of an inverted U.
你可以明白為什麼會是這樣。 And you can see why that would be the case.
如果你連它們現在產生的程式碼都看不懂,那些程式碼常常有 bug 或不完全正確。 That if you don't even understand the code that they generate right now is often buggy or isn't exactly right.
所以如果你連解釋或理解發生了什麼都做不到,你就無法非常有效地使用它。 So if you can't even interpret or understand what's going on, You can't use it very effectively.
目前看來,對於最頂尖的程式設計師來說,產生的程式碼還沒有達到那個水平。 And for now, the very best coders, it appears that the code that is generated just isn't at that level.
所以你得到了那個 U 型曲線。 So you get that U shape.
但那意味著如果你不懂任何程式碼,你確實需要懂一些才能讓它有用。 But that means if you don't know any code, you do need to have some in order for it to be useful.
我認為這對現在很多應用程式來說都是如此,你必須有一些基本的理解才能充分利用它。 And I think that's true for a lot of applications right now, that you have to have some basic understanding in order to get the most of it.
我認為這是否會永遠如此,是一個有趣的開放性問題。 I think it's an interesting open question if that's always going to be the case.
我在上堂課很簡短地放了一張幻燈片,上面有 0 到 5 級的自動駕駛汽車。 I put up at the last class very briefly this slide that had level 0 through 5 autonomous cars.
其中一件事——其實,我們現在可以談談——我正在試圖釐清的是,如果你把那個範式應用到經濟中的所有任務上會怎麼樣? And one of the things that-- actually, we can talk about now-- I'm trying to sort through is, what if you took that paradigm and you applied it to all tasks in the economy?
像是,它們會經歷多少? Like, how many would they go through?
對於自動駕駛汽車,我們還沒有真正達到第 5 級,雖然我不知道你們有多少人搭過……我的意思是,Waymo 的其中一輛車。 So with autonomous cars, we aren't really at level 5 very much, although I don't know how many you guys have written I mean, one of the Waymo cars.
所以,那輛車看起來很不錯,雖然我和 Sebastian Thrun 一起搭乘時,他說現在的成本非常高。 So, that one seems pretty good, although Sebastian Thrun, who I rode in it with, says it's just incredibly expensive right now.
他們可能虧損 50 到 100 美元。 They probably lose $50 to $100.
他不知道,他不在那裡了。 He doesn't know he's not there.
他創辦了這個專案,但他已經不在了。 He started the program, but he's not there anymore.
但光是營運的所有成本,就不實際。 But just all of the costs of running it, it's not practical.
也許它會沿著曲線下降,光達(LIDAR)會變得更便宜等等。 Maybe it'll get down the curve, LIDAR will get cheaper, et cetera.
但我們有很多在第 2、3 級,甚至可以說是第 4 級的自動駕駛汽車,人類仍然參與其中。 But we have a lot of autonomous cars at level 2, 3, even 4, arguably, where humans are still involved.
你也看到很多其他的任務,像是寫程式——我剛才談過。 And you see a lot of other tasks, like coding-- I just talked about that.
另一方面,西洋棋——那張幻燈片,或者前一張幻燈片,我談到了有時被稱為「進階西洋棋」或「自由式西洋棋」。 On the other hand, chess-- that slide, or the slide before it, I talked about what's sometimes called advanced chess or freestyle chess.
當蓋瑞·卡斯帕洛夫(Gary Kasparov)在 1998 年輸給「深藍」(Deep Blue)後,他開始了一系列的比賽,讓人和機器可以一起工作。 When Gary Kasparov, after he lost to Deep Blue in 1998, He started this set of competitions where humans and machines could work together.
很長一段時間,當我在 2012、2013 年做 TED 演講時,當時的情況是,一個人與一台機器合作可以擊敗「深藍」或任何西洋棋電腦。 And for a long time, when I gave my TED talk, it was true, my TED talk in 2012, 2013, it was true at that time that a human working with a machine could beat Deep Blue or any chess computer.
所以當時最厲害的下棋實體就是這些組合。 And so the very best chess-playing entities were these combinations.
現在已經不是這樣了。 That's not true anymore.
AlphaZero 和其他類似的程式,它們從人類的貢獻中得不到任何東西。 AlphaZero and other programs like that, they would get nothing from a human contributing.
對西洋棋機器來說,那只會是一種煩擾。 It would just be like kind of an annoyance to the chess machine.
所以它經歷了從零級,機器什麼都做不了,到它們一起工作的時期,再到完全自主的時期,跨度大約是,我不知道,20 年左右。 So that went through level zero, machines not being able to do anything, through a period where they work together to a period where it's fully autonomous in a span of, I don't 20 years or so.
如果有人想做一個研究專案,或者如果你們現在有任何想法,什麼樣的標準可以判斷經濟中哪些種類的任務會處於那個中間地帶,這會很有趣。 It would be interesting, if anybody wants to work on a research project, or if any of you guys have thoughts right now, what are the criteria for which kinds of tasks in the economy will be in that middle zone?
因為那個中間地帶對我們人類來說是個不錯的地方,機器在幫助我們,但人類在創造價值方面仍然不可或缺。 Because that middle zone is kind of a nice one for us humans, where the machines are helping us, but humans are still indispensable to creating value.
那是一個可以有更高生產力、更多財富和表現的區域,但也更有可能實現共同繁榮,因為勞動力本質上是分散的,而技術和資本,正如艾瑞克剛才所說,可能會非常集中。 And that's a zone where you can have higher productivity, more wealth and performance, but also more likely to have shared prosperity because labor is sort of inherently distributed, whereas technology and capital, as Eric was just saying, potentially could be very concentrated.
你對此有想法嗎? Do you have a thought on that?