本集简介
双语字幕
仅展示文本字幕，不包含中文音频；想边听边看，请使用 Bayt 播客 App。
大家好，听众朋友们。
Hello, listeners.
本周，我们要分享一个特别的节目——No Priors播客。
This week, we're sharing something special, the No Priors podcast.
No Priors 是你了解人工智能革命的指南。
No Priors is your guide to the AI revolution.
在技术变革的关键时刻，联合主持人埃拉德·吉尔和萨拉·郭向全球顶尖的AI工程师、研究人员和创始人提出最重要的问题。
At this moment of inflection in technology, cohosts Elad Gil and Sarah Guo ask the world's leading AI engineers, researchers, and founders the biggest questions.
比如RunwayML的创始人兼首席执行官克里斯托巴尔·瓦伦苏埃拉，或微软首席技术官凯文·斯科特。
People like Cristobal Valenzuela, founder and CEO of RunwayML, or Kevin Scott, CTO at Microsoft.
他们会问：通用人工智能（AGI）还有多远？
They ask questions like how far away is AGI, artificial general intelligence?
哪些市场面临被颠覆的风险？
What markets are at risk for disruption?
商业、文化和社会将如何改变？
How will commerce, culture, and society change?
前沿研究现在进展如何？
What's happening in the state of the art research?
在这一集中，英伟达的传奇创始人兼首席执行官黄仁勋谈论了英伟达如何推动人工智能模型的发展、他们最新的芯片、他是如何运营英伟达的，以及他最看好的人工智能应用。
In this episode, Jensen Huang, legendary founder and CEO of Nvidia, talks about how Nvidia is powering AI models, their latest chips, how he runs Nvidia, and the AI applications he's most excited about.
你可以在任何你收听播客的地方找到《No Priors》。
You can find No Priors wherever you get your podcasts.
再次感谢收听。
And thanks again for listening.
计算机编程现在已经完全被颠覆了。
Computer programming has now been completely disrupted.
是的。
Mhmm.
这是计算历史上第一次，编程计算机的语言变成了人类语言。
That for the very first time in the history of computing, the language of programming a computer is human.
是的。
Mhmm.
任何人类语言。
Any human language.
甚至不需要语法正确。
It doesn't even have to be grammatically correct.
任何人都能编程计算机，这真是相当了不起。
And and it's it's fairly incredible that anyone can program a computer now.
对。
Mhmm.
这意义重大。
That's a big deal.
这是《No Priors》播客。
This is the No Priors podcast.
我是莎拉·郭。
I'm Sarah Guo.
我是埃尔德·吉尔。
I'm Elad Gil.
我们投资、提供咨询并帮助创立科技公司。
We invest in, advise, and help start technology companies.
在这个播客中，我们与人工智能领域的顶尖创始人和研究人员探讨最重要的问题。
In this podcast, we're talking with the leading founders and researchers in AI about the biggest questions.
如今人工智能领域的许多讨论都集中在新模型和新应用上。
So much of the conversation in AI today is about new models and new applications.
但这一切都是建立在英伟达，特别是英伟达GPU的基础之上的。
But it's all built on the back of Nvidia, and specifically Nvidia GPUs.
英伟达A100被称为当今人工智能生态系统的核心动力。
The Nvidia a 100 has been called the workhorse of today's AI eco system.
本周在《No Priors》节目中，我们非常荣幸邀请到英伟达的创始人、总裁兼首席执行官黄仁勋。
And this week on No Priors, we're so excited to have Jensen Huang, the founder, president, and CEO of Nvidia.
在过去三十年里，他创建了一家公司，最初的目标是加速图形处理，如今已彻底改变了计算方式，成长为一家市值高达6740亿美元的巨头——据我今天早上了解，但这位首席执行官却称这是一家他始终在努力拯救的公司。
So over the last thirty years, he's created a company that began with a goal to accelerate graphics and has transformed the way we compute to become a $674,000,000,000 behemoth as far as I could tell this morning, but that the CEO describes as a company that he's always trying to save.
感谢你和我们一起做这次对话，黄仁勋。
Thank you for doing this with us, Jensen.
很高兴能参与，莎拉。
Delighted to do it, Sarah.
我们从头开始好吗？
Why don't we start at the beginning?
你在创办公司之前曾在LSI和AMD工作过。
You worked at LSI and AMD before starting a company.
这是怎么发生的？
How did that happen?
他们给了我一份工作。
They gave me a job.
让我想想。
Let's see.
我当时在俄勒冈州立大学，有一天是校园招聘日，我面试了很多公司，其中有两家公司让我特别有共鸣。
I was at Oregon State University and it was a campus company day and I interviewed at a lot of companies and two companies really really connected with me.
我非常喜欢设计芯片和计算机。
I I love designing chips and designing computers.
当时，在我们的计算机科学实验室里，有一张AMD的29032位CPU的海报。
And at the time in our lab, in the computer science lab, there was a poster of a 29,032 bit CPU from AMD.
你知道，我一直觉得能制造出这样的芯片会很酷。
And, you know, I always thought it'd be kinda cool to build that.
另一方面，当时还有一家初创公司，由硅谷传奇人物Gulf Corrigan创办，他们致力于用软件设计芯片。
On the other hand, there was an another company that was a startup at the time built by one of the legends of Silicon Valley, Gulf Corrigan, and they started a company to design chips using software.
不是手工设计芯片，而是通过可编程逻辑来设计。
To design chips not by hand, but by using programmable logic.
你可以用语言描述它，系统会自动综合生成芯片。
And and you would describe it in language and it would synthesize it to chips.
当然，我选择了去AMD。
And of course, I chose to go to AMD.
结果我去那里是设计微处理器的，而我的办公室同事——不是实验室伙伴——去了LSI。
And it turned out that I went there to design microprocessors and my lab partner not my lab, my office mate ended up going to LSI.
她在我去之后、她去之前一直劝我也去，而LSI团队说：‘我们当时在招募一位来自俄勒冈州立大学的学生，非常希望他来LSI Logic工作，结果发现他就是她的同事。’
And she insisted I go there after I was there and after she went there and the LSI team said, hey, we were recruiting this kid from Oregon State and we really wanted him to come work at LSI Logic and turned out to have been her office mate.
于是他们都联系了我，我决定去那里，因为那时正是EDA行业的起步阶段，也是首次用计算机设计芯片的开端，这可能是我人生中最棒的决定之一。
And so they all reached out to me and I decided to go there because it was at the beginning of the EDA industry, it was at the beginning of designing chips using computers, and it was probably one of the best things that ever happened to me.
那时正是每家公司都能自行设计芯片的开端，也正是因此，我结识了一些非常出色的计算机架构师。
And it was in the beginning of the ability for every company to build their own chips, And it's the reason why I met some really great computer architects.
安迪·贝赫托尔海姆是太阳公司的创始人。
Andy Bechtelsheim was the founder of Sun.
我在硅图公司有机会与一群杰出的架构师共事。
I got to work with a bunch of great architects at Silicon Graphics.
还有约翰·鲁宾斯坦，他曾任职于一家名为达纳计算机的公司，后来成为苹果公司的工程副总裁。
And John Rubenstein who was at a company called Dana Computer who became the vice president of engineering for Apple.
当然，还有英伟达的两位创始人，克里斯·马尔科夫斯基和柯蒂斯·斯普林，以及我自己。
And then, of course, the two founders of Nvidia, Chris Malkowski and Curtis Spring, myself.
所以我有幸与一些极其出色的计算机架构师共事，学到了很多关于用芯片构建计算机的知识，这就是我的早期经历。
And and so I got a chance to work with some really amazing computer architects and I learned a lot about building computers with chips and so that's my early days.
你在LSI与联合创始人一起表现得非常出色。
And you were a star at LSI with your cofounders.
你是在什么时候意识到自己必须创办一家公司的？
At what point did you know I have to start a company?
这并不是我的主意。
It wasn't my idea.
是他们的主意。
It was theirs.
克里斯和柯蒂斯想离开太阳公司。
Chris and Curtis wanted to leave Sun.
他们有自己的理由，而我在LSI Logic工作得非常好，也很喜欢我的工作。
They had their own reasons, and I was doing really well at LSI Logic, and and I enjoyed my job.
我们有两个孩子，劳里和我，就像你一样，他们一直不停地劝我。
And we have two kids, Laurie and I and just like you, they wouldn't stop hounding me.
他们说：嘿，我们想创办一家公司，真的需要你加入，但我告诉他们我确实需要一份工作，他们得先想清楚该怎么做。
And they said, hey, we wanna start this company and we really need you to come along and and I told them that I really needed to have a job and they need to figure out what to do.
当时，计算机设计的方式在通用计算和使用加速器之间有着明显的分歧。
At the time, the value was the way of designing computers was rather split between general purpose computing versus using accelerators.
当时，99%的价值被认为存在于通用计算中，只有1%的人相信加速计算。
And about 99% of the value was believed in general purpose computing and about 1% believed in acceleration.
二十五年来，99%的判断都是对的。
For twenty five years, ninety nine percent was right.
我们决定创办一家专注于加速计算的公司。
We decided to start a company on accelerated computing.
当时，加速计算真正能做的，就是找到那些通用计算几乎无法解决或根本无法解决的应用或问题。
And at the time, the only thing you could really do with accelerated computing is find applications or find problems that were barely solvable or unsolvable by general purpose computing.
因此，我们公司致力于解决普通计算机无法处理的问题。
So that's kind of what we dedicated our company to do, to solve problems that normal computers can't.
如果你将这一使命贯彻到底，它引领我们走向了自动驾驶汽车、机器人技术、气候科学问题、数字生物学，当然还有最著名的人工智能。
And if you follow that mission to its limit, it led us to self driving cars, it led us to robotics, it led us to climate science problems, digital biology, and of course, one of the most famous ones is artificial intelligence.
在当前这波人工智能浪潮之前，你们就已经在研究这一大类应用了。
You were working on this huge set of applications before the current wave of artificial intelligence.
Nvidia在人工智能方面的原始技术优势是什么？你们是什么时候开始意识到这对你们来说会成为一个重要的应用场景的？
What was the original technical advantage of Nvidia in artificial intelligence, and when did you begin to realize that this was gonna be an important use case for you guys?
因此，我们提升了加速器的灵活性，使其更接近通用计算，并发明了一种名为CUDA的新计算模型。
So we had expanded the flexibility of our accelerators to to be more general purpose, And we invented a new computing model called CUDA.
我们现在录这个播客，大概是下午四点左右。
And we're doing this podcast like at 04:00 or something like that in the afternoon.
那时候正是精力最低落的时候。
It was like at the lowest point of energy.
对吧？
Isn't that right?
是的。
Yeah.
所以我们需要一些
So this we need some
这就是为什么我们需要
That's why we need
我们需要一些极客。
We need some nerds.
极客。
Nerds.
这就是我们需要一些极客的原因，
That's why we need some nerds,
家伙们。
guys.
你，极客。
You, nerds.
现在还有软糖集群了。
Now with gummy clusters as well.
所以这是非常令人兴奋的新技术。
So that's very exciting new technology.
我们需要一些能量。
We need some energy.
我们现在需要一些加速计算。
We need some accelerated computing right now.
所以我们希望让我们的图形处理器变得越来越通用。
So we wanted to make our graphics processors more and more general.
最初这样做的原因是，我们需要实现的一些效果涉及通用图像处理。
And the reason for that in the beginning was because some of the effects that we had to do related to general purpose image processing.
后期效果，你渲染完一张图像后，还要进行图像后期处理。
Post effects, you render an image and you do post image effects.
当然，其他应用方面，我们希望让场景活起来，因此需要进行物理计算。
Other applications of course, we wanted to bring the scene to life and so we had to do physics processing.
你必须进行物理计算，比如粒子物理、流体动力学等等。
And you have to do physics, have to do particle physics, fluid dynamics, so on and so forth.
因此，我们不断扩大加速计算平台的适用范围，使其越来越通用。
And so we expanded the aperture of our accelerated computing platform to be more and more and more general purpose.
通用性的问题是，你越通用，在特定领域获得的加速效果就越少。
The problem with general purposeness is that the more general purpose you are, the less acceleration you get in a particular domain.
所以你必须非常非常谨慎地找到这个平衡点。
And so you got to find that line really really carefully.
这正是我们公司的优势之一：在每一代产品中，都能在特定应用上实现远超CPU的惊人加速。
And that's one of the gifts of our company, to find that line between on the one hand, every single generation bringing enormous amounts of acceleration well beyond what CPU could do to the application.
如果你变得太过通用，那就和CPU没什么区别了。
And so if you become too general purpose, you're like just like a CPU.
你如何用CPU来加速CPU？
How can you accelerate a CPU with a CPU?
因此，你必须找到一种方法，在这条边界上走好平衡。
And so you have to find a way to walk that line.
另一方面，如果你不扩大所服务的应用范围，你所能获得的研发投入就不足以跟上CPU的步伐，而CPU拥有全球任何芯片中最大的研发预算。
On the other hand, if you don't expand the aperture of the applications that you serve, the r and d dollars that you're able to generate wouldn't be enough to stay ahead of the CPU, which had the largest r and d budget of any chip on the planet.
所以，如果你仔细思考这个问题，它实际上几乎不可能解决：当时你所服务的应用市场，或许只有十亿美元规模。
So if you think about this problem, it's actually really nearly impossible because you have a small application, let's call it, you know, a billion dollar market at the time.
而在这十亿美元的市场中，你每年投入一亿五千万美元。
And out of that billion dollar market, you're investing a $150,000,000 a year.
在每年一亿五千万美元的投入下，你如何跟上一个规模达数千亿美元的行业？
Out of the $150,000,000 a year, how do you keep up with a few $100,000,000,000 industry?
这根本说不通。
It's not even sensible.
因此，你必须非常非常谨慎地找到这样一个利基：1.5亿美元的投资能异常且疯狂地加速这一特定应用。
And so you have to find that niche very, very carefully where a $150,000,000 would accelerate this particular application abnormally and insanely.
随着时间推移，你可以扩展应用领域，使其从10亿美元增长到50亿美元、100亿美元，以此类推，而不会跌下悬崖。
And then over time, you could expand your application space so that it goes from $1,000,000,000 to $5,000,000,000 to $10,000,000,000, so on and so forth, without falling off that cliff.
这就是我们所走的那条微妙的界限。
That is the fine line that we walked.
因此，我们不断扩展这里的通用性，这引导我们走向分子动力学模拟，而这张图看起来正是如此。
And so we kept expanding the general purpose in this and it led us to molecular dynamic simulation which is what this image seems to look like.
地震处理是另一个行业，我们就这样一点一点地扩大了应用范围。
And seismic processing was another industry and just slowly by surely, we expanded our aperture.
但我们做得很出色的一点是，无论用户是将我们的平台用于通用计算还是加速计算，我们都始终维持架构兼容性。
But one of the things that we did well was to make sure that irrespective of whether somebody used our platform for general purpose computing, accelerated computing, We always maintain the architecture compatibility.
这样做的原因是，我们希望打造一个能吸引开发者的平台。
And the reason for that is because we wanted a platform that would attract developers.
如果世界上所有的NVIDIA芯片都不兼容，那么开发者即使发现CUDA对他们来说极其出色，又该如何选择呢？
If every single Nvidia chip in the world was incompatible, then how would a developer be able to pick one up even if they learned that CUDA was gonna be incredible for them?
他们怎么可能会选择它，并说我要开发一个能在上面运行的应用程序？
How would they pick that up and say, I'm gonna develop an application that's gonna run on that?
他们得去弄清楚到底该用哪款芯片？
Which chip would they have to go figure out?
但没人能搞清楚这一点。
And nobody could figure that out.
所以我们说，如果我们相信某种架构，并希望它成为一个新的计算平台，那就必须确保我们的每一块芯片都表现得完全一致。
And so we said, if we believe in an architecture and want this to be a new computing platform, then let's make sure that every one of our chips perform exactly the same way.
就像x86、ARM，或者任何其他计算平台一样。
Just like an x 86, just like ARM, just like any computing platform.
所以在前五到十年里，你知道，CUDA的客户非常少，但我们让每一块芯片都兼容CUDA。
And so for the first five, ten years, you know, we had very few customers for CUDA, but we made every chip CUDA compatible.
你可以回溯历史，看看我们的毛利率。
And you can go back in history and looked at our gross margins.
一开始很糟糕，后来变得更糟了。
It started out poor and it got worse.
而且我们身处一个竞争激烈的行业，当时还在努力摸索如何做好本职工作并打造成本效益高的产品。
So and we were in a really competitive industry and we were still trying to figure out how to do our job and build cost effective things.
所以这本来就已经很有挑战性了，而我们又在此基础上叠加了一个叫CUDA的架构，但当时没有任何应用，也没人愿意为此付费。
So it was already challenging as it is, and then we layered on top of this this architecture that was called CUDA that had no applications that nobody paid for.
是的。
Yeah.
这其实挺不可思议的，因为现在当我跟人工智能领域的人交流时，他们非常青睐NVIDIA显卡，原因之一就是CUDA，以及其出色的互联扩展能力。
It's kind of amazing because now when I talk to people in the AI world in terms of one of the reasons that they really love using Nvidia GPUs is because of CUDA, and then because of the ability to scale interconnect.
因此，你可以真正实现高度并行化，而这是当今市场上其他方法或架构难以做到的。
And so you can really like highly parallelize these things as well, which you can't necessarily do with other approaches or architectures that are in the market today.
对。
Yeah.
所以这个计算平台，CUDA，它很特别，因为它能实现一些奇迹般的事情。
And so this computing platform, LUT, it's strange in the sense that it performs these miraculous things.
我们靠着GeForce这款游戏显卡将它推向了世界。
And we carried it to the world on the backs of GeForce, which is a gaming card.
杰夫·辛顿为他的实验室拿到的第一块GPU。
The first GPU that Jeff Hinton got for his lab Mhmm.
埃拉德会说，杰夫进来时说：‘这里有几个GPU。’
Elad would tell you that Jeff came in and said, here's a couple of GPUs.
它叫GeForce，你们应该试着用它来做深度神经网络。
It's called GeForce, and you guys should try to use it for DNN.
所以他买的是一个游戏显卡。
And so it was a gaming card that he bought.
你们当时想到了哪些应用场景？
What applications did you have in mind?
因为正如你所说，你们最初是从游戏起步的，或者至少在九十年代公司刚成立时，游戏是你们的主要市场。
Because to your point, you started with gaming or at least you were very popular with gaming starting in the nineties when you started the company.
然后嗯。
And then Mhmm.
你知道，我开始越来越多地听到关于NVIDIA显卡的消息，尤其是在加密货币和挖矿的背景下，然后是在人工智能的背景下。
You know, I started hearing about Nvidia GPUs more and more both in the context of cryptocurrencies and sort of mining and then in the context of AI.
看起来这两个领域是很多人自发采用你们产品的市场。
It seemed like those were the two markets where a bunch of people were organically just adopting you.
你们有针对这些社区进行营销吗？
Were you marketing to those communities?
是因为人们开始意识到他们需要线性代数吗？
Was it just people started realizing that they needed linear algebra?
这就是计算平台的美妙之处。
That's the beauty of a computing platform.
一开始，你必须针对特定的应用场景。
In the beginning, you have to target the applications.
一开始我们确实这么做了。
And in the beginning, we did.
其中一个最早的应用是NAMD，嗯。
One of the first applications was NAMD Mhmm.
地震数据处理。
Seismic processing.
这两者分别属于粒子物理和图像处理，或者说逆物理问题。
Both of them are one of those kind of particle physics, the other one is image processing, if you will, and so inverse physics, if you will.
是的。
Mhmm.
在某个特定领域，我们主动去招聘研究人员。
And one particular domain, you know, we just went out to hire it to research.
我们去了科学计算中心，问他们：有哪些问题是你们目前无法解决的？
We went to scientific computing centers, and and we said, what kind of problems are just beyond your reach?
是的。
Mhmm.
这些应用包括量子化学和量子物理等等。
And the list of applications are include quantum chemistry and quantum physics, you know, so on and
如此类推。
so forth.
是什么时候你意识到，这个人工智能对我们来说真的很重要？
What was the moment when you said, wow, this AI thing is really important for us?
大概是在2012年左右。
It happened around 2012, I guess.
当时，安德鲁· Ng 主动联系了我们的首席科学家比尔·道利，希望合作找到一种方法，将他们正在开发的神经网络模型
And it was because simultaneously, Andrew Ng reached out to Bill Dowley, our chief scientist to work on a way to get the neural network model that they were working on Mhmm.
迁移到GPU上，这样他们就不用再使用成千上万的CPU服务器，而只需几块GPU就能完成训练。
Onto GPU so that they could, instead of using thousands of CPU servers, they could use a few GPUs to do training.
这是其中一个契机。
So that was one.
与此同时，杰夫·辛顿也联系我们，我们开始听说类似的事情也在杨立昆和他的团队身上发生。
Simultaneously, Jeff Hinton reached out to us and we started hearing about that and the same thing was happening with Yan LeCun and his lab.
是的。
Mhmm.
因此，在多个不同的实验室里，我们同时感受到神经网络正在崛起，这引起了我们的关注。
So simultaneously in several different labs, we're starting to feel that there's this neural network emergence and that attracted our attention.
是的
Yeah.
我想2012年也是AlexNet问世的那一年。
Guess 2012 was also the year when AlexNet came out.
我觉得那一年是深度学习整体转型的关键时刻，至少我记得当时心想，哇，一股令人兴奋的AI浪潮真的要来了。
I felt like that was a year of transition for deep learning in general in terms of really that was the moment in time at least that I remember thinking, wow, this really exciting wave of AI coming.
是的
Yeah.
然后我觉得在接下来的十年里，初创公司没什么太大动静，但许多现有企业开始采用这项技术。
And then I feel like for ten years, nothing really happened for startups, but a lot of incumbents started adopting this technology.
我们开始感受到它了。
We started feeling it.
我们早在那之前就听闻了，然后ImageNet就像大爆炸一样，吸引了我们所有的注意力。
We started hearing about it before that, and then ImageNet kind of it was the big bang, if you will, got all of our attention.
你提到早期的AI实验室从Nvidia获取游戏显卡，因为你们在解决别人无法解决的问题。
You talk about early AI labs as pulling this from Nvidia using gaming cards because you were solving a problem nobody else could solve Mhmm.
还有效率和规模。
And, like, efficiency and scale.
是的。
Mhmm.
Nvidia 是否会在某个时刻开始投资某个应用，因为它们认为这是一个正在增长的应用？
Is there a point at which Nvidia begins to, like, invest in an application because they think it's a growing application?
还是说它更像一个平台，由市场来决定它的走向？
Or is it more it's a platform and the market will take it from us?
不是。
No.
在每一个案例中，当某个应用找到用武之地时，我们都会问自己：我们怎样才能让它变得更好？
In every single case when an application finds use, we ask ourselves how can we make it even better?
是的。
Mhmm.
这一次，关于深度学习，我们获得的一个良好洞察是，将许多不同的观察结果综合起来，意识到这不仅仅是一个新的计算机视觉算法——尽管最初大多数应用都集中在计算机视觉上。
This time with deep learning, the good insight that we made, it was piecing together observations in a lot of different ways but realizing that this isn't just going to be a new algorithm for computer vision which is really most of the applications in the beginning.
但这一点会非常有帮助。
But which was going to be very helpful.
我的意思是，如果只是计算机视觉，我们就能用它来做各种有趣的应用，比如自动驾驶和机器人，我们确实这么做了。
I mean, just if it was just computer vision, we could use it for all kinds of interesting applications like self driving cars and robotics and we did.
但我们观察到，这可能是一种全新的软件编写方式，于是我们开始思考这对芯片设计、系统设计、互连、算法和系统软件意味着什么，真正去深入思考的不仅是为什么这令人兴奋、为什么如此有效——仅这一点就已经足够神奇了。
But we observed that this might be a new way of writing software altogether And asking ourselves what's the implication to chip design, system design, interconnect, the algorithm, the system software, to really reason about not just why is this exciting, why was it so effective, which that alone was plenty miraculous.
嗯。
Mhmm.
ImageNet在没有任何人工设计算法的情况下，一夜之间就达到了与三十年计算机视觉算法相媲美的效果。
That ImageNet without without specifically any human engineered algorithm would reach the level of effectiveness compared to thirty years of computer vision algorithms overnight.
这可不是一点点的差距。
It wasn't by a small amount.
所以第一个问题当然是：为什么它如此有效？这是否具有可扩展性？
And so the first question of course is why is it so effective and was this gonna be scalable?
如果它具有可扩展性，那对计算机科学的其他部分意味着什么？
And if it was gonna be scalable, what's the implication to the rest of computer science?
这种所谓的通用函数逼近器能够解决维度极高的问题，而你只需通过足够的数据就能学习到这个函数——当时我们开始相信，我们能够获得大量数据，并通过逐层训练系统性地构建出这样的模型。
What problems can't this universal function approximator, if you will, that can solve problems of dimensionality extraordinarily high, and yet you could learn the function using enough data, which at the time we were starting to believe we can get plenty of, and to systematically train this model into existence because you train them one layer at a time.
你能再多谈谈吗？我听说你在描述这种更广泛的平台变革时非常清晰，比如从网页的提供方式到生成方式，或者其他相关方面。
Can you talk a little bit more about I've heard you'd be very articulate in terms of how you view this as a broader platform shift, just even in terms of how pages are served versus, you know, generated or other aspects of that.
你能再多谈谈当前计算机科学领域向人工智能转型的总体情况吗？
Could you talk a little bit more about what's really happening right now more broadly in computer science with the shift to AI?
是的。
Yeah.
所以，现在让我们快进十年。
So you fast forward now a decade.
前五年主要是思考这对计算机科学整体带来的影响。
The first five years was about reasoning the impact to computer science altogether.
同时，我们也在开发各种新型模型。
At the same time, we're developing new models of all kinds.
对吧？
Right?
于是，从CNN到ResNets，再到RNNs、LSTMs，以及各种新型模型，不断增大规模，在感知模型方面取得了巨大进展。
And so CNNs to ResNets to RNNs to LSTMs to, you know, all kinds of new models and scaling them larger and larger, making great strides in perception models particularly.
当然，Transformer是一个重大突破。
And of course, the transformer was a big deal.
BERT也是一个重大突破。
BERT was a big deal.
你们都十分熟悉这个故事。
All of you know that story well.
你们有没有注意到，随着Transformer和BERT等技术的出现，数据量出现了显著增长？
Did you guys see like a step change in volume growth with transformers and BERT and such?
因为感觉拥有一个支持模型扩展的架构和注意力机制，确实为整个行业注入了强劲动力。
Because it feels like having a architecture and an attention mechanism that allowed for scaling of these models really was also a kick start in the industry.
能够从空间数据和序列数据中学习模式和关系，这种架构一定非常有效。
Well, the ability for you to learn patterns and relationships from spatial as well as sequential data must be an architecture that's very effective.
所以从基本原理来看，我认为Transformer将会是一个极其重要的突破。
So I think on its first principles, kinda think transformers is gonna be a big big deal.
是的
Mhmm.
不仅如此，你还可以并行训练它，并且能够真正地扩展这个模型。
Not only that, you could train it in parallel And you can really scale this model up.
因此，这非常令人兴奋。
And so that's very very exciting.
我认为，当Transformer首次出现时，我们意识到现在出现了一种能够克服RNN和LSTM局限性的模型，我们现在可以以非常大规模的方式学习序列数据。
I think that when transformers first came out, we realized that there's a model now that overcame the limitations of RNNs and LSTMs and we can now learn sequential data in a very large way.
所以这非常令人兴奋。
So that was very exciting.
BERT非常令人兴奋。
BERT was very exciting.
我们自己训练了一些早期的语言模型，并且看到了非常好的结果。
We trained some of the early language models ourselves and we saw very good results.
但直到强化学习与人类反馈的结合，以及检索模型和对话管理器中一些突破性工作的出现，才实现了有效的安全约束。
But it wasn't until the combination of reinforcement learning human feedback wasn't and of course some of the breakthrough work that was done with retrieval models, dialogue managers that does the guard railing.
直到所有这些因素开始融合，我们才迎来了如今广受欢迎的ChatGPT。
It wasn't until some of all of those kind of pieces started to come together that of course that we all enjoy ChatGPT.
埃拉德，你想要表达的观点是，计算机编程如今已被彻底颠覆。
And Elad, the point that you're trying to make is the observation that computer programming has now been completely disrupted.
这是计算历史上第一次，编程计算机的语言变成了人类语言。
That for the very first time in the history of computing, the language of programming a computer is human.
嗯。
Mhmm.
任何人类语言都可以。
Any human language.
甚至不需要语法正确。
It doesn't even have to be grammatically correct.
是的。
Yes.
任何人都能编程计算机，这相当令人惊叹。
It's fairly incredible that anyone can program a computer now.
嗯。
Mhmm.
所以这很重要。
And so that's a big deal.
你用不同的方式编程，它会写出不同的应用程序，这种新的计算模式能带来多大的影响？
The fact that you program it differently, it writes different applications, what is the reach of this new computing model?
显然影响相当大。
Apparently quite large.
这也是为什么ChatGPT成为历史上增长最快的应用程序的原因。
And it's it's the reason why ChatGPT is the fastest growing application in history.
嗯。
Mhmm.
我们邀请了Copilot的首席架构师亚历克斯·格拉维利。
We had Alex Gravely, who is the chief architect for Copilot Uh-huh.
他也上了我们的节目。
On the show as well.
而他最喜爱的，显然是拥有顺序代码预测功能，这非常强大。
And his favorite, like obviously it's, you know, very powerful to have sequential code prediction.
但他最喜欢的Copilot使用场景是，有人告诉他，他们原本不会编程，但现在会了。
But his favorite use cases of Copilot have been like people telling him that they don't code, but now they do.
是的。
Yeah.
没错。
Right.
这正如你所说，非常具有民主化意义。
Which I think is very democratizing as you said.
令人惊讶的是，你可以给ChatGPT一个要解决的问题，它能一步步推理，但最终却得出错误的答案。
It's quite amazing that you could give chat GPT a problem to solve and it reasons through it step by step but yet it arrives at the wrong answer on the one hand.
另一方面，如果你让它写一个程序来解决同一个问题，它却能写出完美解决问题的程序。
On the other hand you could tell it to write a program to solve the same problem and it writes a program that solves the problem perfectly.
有一个应用，一方面能推理并尝试解决问题，做得相当不错，几乎就要成功了。
The fact that there's an application that on the one hand reasons and tries to solve a problem and does a fairly good job at it, it's almost there.
另一方面，它可以完全编写一个程序来解决同样的问题。
On the other hand, it can write a program altogether to solve that same problem.
你真的得好好想想这背后的含义。
You gotta really wrap your head around the implication of this.
所以你觉得它是像
So do you view it as like
未来世界的一种机器意识吗？
the future world as some form of machine sentience?
首先，我甚至不知道‘Yeah’这个词在那种语境下是什么意思。
First of I don't even know what that word means in a Yeah.
某种程度上是的。
In a way.
是的。
Yeah.
是的。
Yeah.
我相当确定我是有意识的，但今天没那么确定了。
I'm fairly sure that I'm sentient, less so today.
所以某种程度上我们有
I'm kinda So we have
是的。
a Yeah.
这就是为什么继续吃吧。
That's why Keep eating.
这就是为什么我需要护士帮我启动。
That's why I need nurse to crank me
到这里来。
up here.
我打算试试。
I'm gonna try to.
是的。
Yeah.
知道。
Know.
我知道。
I know.
今天是艰难的一天。
Today was a tough day.
但我不确定。
But I don't know.
我相信我们现在拥有一种能够为多种类型问题进行推理的软件吗？
Do I believe that we now have a software that can reason through a problem for many many types of problems?
能够推理问题、解决问题，并持续提供系统性解决方案的程序吗？
Reason through a problem and solve and provide a solution or a program to systematically provide a solution on an ongoing basis?
答案是是的。
The answer is yeah.
嗯哼。
Mhmm.
是的
Yeah.
当你展望这个未来时，你如何看待英伟达业务线的发展方向？
And then as you look forward to that world, how do you think about where you wanna take Nvidia's lines of business?
但你过去也曾提到，英伟达做过诸如训练模型这样的事情，你们在这些方面做了一些非常有趣的工作。
But also, you mentioned in the past that Nvidia has done things like train models, and you've done some really interesting things there.
这在未来会成为你们越来越重要的工作内容吗？还是你们主要专注于芯片方面？
Is that gonna be an increasing part of what you do in the future, or are you mainly focused on the chip side?
或者你如何看待在推动研究与作为行业底层平台之间的这种平衡？
Or how do you think about that mix of helping to push forward some research as well as being the underlying platform for the industry?
我们是一家计算平台公司，必须根据需要向上延伸到整个技术栈，以便开发者能够使用它。
Well, we're a computing platform company and we have to go up the stack as far as we need to so that developers can use it.
所以问题在于，什么是开发者？
And so the question is what is a developer?
当然，最初开发者是指那些能够掌控自己操作系统的人员。
And in the beginning of course, a developer is somebody who controls their own operating system.
在那些年代，我们可能只需要向上延伸到设备驱动程序，或者稍微底层的一些层级，来帮助开发者使用。
And so in those days, we might only have to go as far up as device drivers or the layers slightly underneath that somehow to enable developers.
但对于科学计算和这些不同的领域，开发者实际上可能在使用某种求解器。
But for scientific computing and and all these different domains, the developer is actually using maybe a solver.
他们需要将该领域的算法以某种方式表达出来，以便能够被加速。
And they need the algorithms of that domain to be somehow expressed in a way that could be accelerated.
这就是为什么当我们进入这些多领域物理问题时，我们意识到必须自己开发这些算法。
Which is the reason why when we moved into these multi domain physics problems, we realized that we have to develop the algorithms themselves.
因为解决问题的算法与底层的计算机架构密切相关。
Because the algorithms of solving a problem relates to the computer architecture that's underneath.
如果架构是通过MPI和以太网等连接的CPU，那么这种算法肯定与数千个通过内部互连结构连接的处理器、以及数据中心内成千上万个GPU的架构大不相同。
And if the architecture is CPUs, connected through MPI and you know ethernet or whatever it is, That algorithm is surely very different than thousands of processors that's connected by a fabric inside one GPU and thousands of GPUs inside a data center.
因此，算法显然需要被重新构思和重构。
So obviously the algorithm has to be reframed and refactored.
所以，我们的公司非常擅长设计计算机算法。
So our company got very good at designing computer algorithms.
这可能是针对粒子物理或流体动力学的。
It could be for particle physics or fluid dynamics.
当然，有一天，它与深度学习和神经网络相关了。
And then of course one day, it was related to deep learning and neural networks.
CoDNN 本质上是一种用于加速深度学习的领域特定语言。
And cooDNN is essentially a domain specific language for accelerated deep learning.
因此，我们为深度神经网络做了这件事，也为计算机图形学中的光线追踪做了类似工作，这被称为 RTX。
And so we've done that for deep neural nets, we've done that for computer graphics with ray tracing, that's called RTX.
所有这些不同的领域库，本质上都是关于理解科学领域，然后重新设计算法，使它们变得极其快速。
All of these different domain libraries really is about understanding the domain of science and then redesigning algorithms that make them go incredibly fast.
那么，未来，什么是开发者？
Now, the future, what's a developer?
我认为，在未来，开发者很可能是那些与大型语言模型或基础模型互动的人。
Well, I think in the future, a developer is likely going to be somebody who engages large language models or foundation models.
如果有人能使用 ChatGPT 或 OpenAI 的模型，我非常鼓励这样做。
Now, if somebody could use ChatGPT or OpenAI's model, I really encourage that.
原因在于它们的表现极其出色。
And the reason for that is because they do such an incredible job.
如果有人通过微软使用它，我非常鼓励这样做。
If somebody could use it through Microsoft, I really encourage that.
如果有人通过谷歌使用它，我也非常鼓励这样做。
If somebody could use it through Google, I would really encourage that.
但如果有人需要为某个领域构建专有模型，比如创建一个新的基础模型，领域可能是蛋白质、化学物质，或者是气候科学、多物理场。
But if somebody needs to build a proprietary model for a domain, maybe create a new foundation model and let's say the domain was proteins or let's say the domain was chemicals or let's say the domain was climate science, multi physics.
这种基础模型非常专业化，但市场显然不小，因为药物发现领域、气候科学领域、气候科技领域都非常庞大。
That foundation model is pretty nichey and it's not a small market obviously because the field of drug discovery is large, the field climate science is large, climate tech is large.
它不太可能对每个人都有普遍的适用性。
It's not likely to be horizontally useful for every human.
因此，我们可能会决定为三维图形、虚拟世界开发一个基础模型，因为它们对我们至关重要。
And so we might decide to go do something like a foundation model for three d graphics, virtual worlds because they're super important to us.
我们可能会决定为机器人技术构建一个基础模型，因为这正好是我们擅长的领域交汇之处。
We might decide to build a foundation model for robotics because it's at an intersection of the things that we do very well.
即使如此，我们也会尽可能深入，但不会超过这个限度。
And even then, we'll probably take it as far up as necessary, but no further than that.
我们并不是想成为一家AI模型公司。
We're not trying to be an AI model company.
我们的目标是帮助各行各业创建AI模型。
We're trying to help industries create AI models.
主要是，我们希望帮助开发者。
Mostly, we're trying to help developers.
是的。
Yeah.
这很有道理。
That makes a lot of sense.
你基本上是跟着客户的需求，走到他们需要的任何层面。
You're basically following your customers up to whatever level they need you to.
没错。
That's right.
然后你
And then you
在那个合适的时机把它交给他们。
hand it off to them at that proper point.
我们尽量偷懒。
We're trying to be as lazy as we can.
是的。
Yeah.
你知道的，尽可能少做，但又足够必要。
You know, do as little as possible as much as necessary.
嗯。
Mhmm.
计算机科学的第一性原理，我们都清楚。
The first principles of computer science, we we all know well.
对吧？
Right?
尽快拒绝工作。
To reject work as quickly as you can.
尽可能推迟你剩下的工作，直到你被拒绝为止。
To defer whatever work you are left for as long as you can until you could be rejected.
然后剩下的工作你就必须做了。
And then whatever remains you have to do.
我们尽量少做，但又要做到足够多。
And we try to do as little as we can as much as possible.
这可以说是公司的原则。
That's kind of principles of the company.
懒惰，就是公司的原则。
Laziness, the principles of the company.
这就是关键所在。
That's the takeaway.
它就在上面
It's up on the
墙在外面，是的。
wall outside of the Yeah.
能多懒就多懒。
See as lazy as you can.
是的。
Yeah.
我正试图将这一点与公司做出的一些长期承诺协调起来。
I'm trying to square that with some of these very long term commitments that the company makes.
对。
Yeah.
对吧？
Right?
比如，Kuda 是一个长期的投入。
Like, Kuda is a very long term bet.
是的。
Yeah.
展开剩余字幕（还有 329 条）
我们十年前相遇时，英伟达的估值只有今天的百分之一，当时正面临积极投资者的压力。
And we met a decade ago when Nvidia was valued at one one hundredth of its value today and was facing like activist investors and such.
那时候，做出长期投入可能要更困难一些。
And it was probably like, let's say a little harder to make long term bets.
你如何平衡作为一家大型上市公司的压力——既要应对当下的机遇，又要兼顾架构性承诺或长期投资，并思考如何优先排序？
How do you balance the the pressures of being a large public company and the opportunities of today with sort of architectural commitments or long term bets and sort of think about that prioritization?
投资未来与实现当前的可持续发展并不冲突。
Investing in the future and being sustainable now are not in conflict with each other.
因此，所有初创公司首席执行官和所有首席执行官的挑战在于，找到一种方式来践行你的信念。
And so the challenge for all startup CEOs and for all CEOs is to find a way to to be able to do what you believe in.
这家机构的根本核心信念，并且有能力去实践它。
The the fundamental core belief of the institution and to be able to afford doing it.
这就是公司的使命。
That is the purpose of the company.
这既是一种信念。
And it's part conviction.
这还需要技巧。
It's part skill.
赚钱不是信念问题。
Making money is not a matter of conviction.
赚钱是技巧问题。
Making money is a matter of skill.
这是一种可学习的技能。
And it's a learnable skill.
我花了很长时间才学会。
And it took me a long time to learn it.
我承认这一点。
I'll admit that.
我已经从事这个行业三十年了，而前二十年——既然你提到十年前，那应该是从那时往前推二十年——
I've been at this for thirty years and for the first for well, apparently for the first twenty years since you went back ten years.
在头二十年里，我还在摸索如何做到这一点。
For the first twenty years, was still trying to figure it out.
但这是一种技能。
But it's a skill.
学习如何赚钱、如何高效地运营公司，这些都是技能，公司必须培养这些能力。
Learning how to make money and learning how to run a company efficiently, those are all skills and the company has to develop the skills.
而我们最终的做法是问自己：我们是否真的相信它？
And the way that we we ultimately do it is we ask ourselves, do we really believe it or not?
如果我们真的相信某件事，那么这就是企业的使命。
And if we really believe in doing something, then it it is the purpose of the enterprise.
这家机构的唯一使命就是追求它的信念。
It's the singular purpose of the institution to go pursue its beliefs.
其余的一切则取决于公司的智慧，比如尽力做好本职工作，打造人们愿意购买的产品，尽可能降低成本，提高公司效率。
And the rest of it is up to all of the cleverness of the company and, you know, try to do our jobs well and build things that people wanna buy and try to make it as cost effective as possible and make the company as efficient.
这些都是技能。
Those are all skills.
事实证明，最难的部分不是技能本身。
The hard part as it turns out is not the skill part.
我花了很长时间，但很多公司显然都知道如何赚钱。
It took me a long time but a lot of companies know how to make money obviously.
所以，有不止一家公司能赚钱，这说明这并不难。
So the fact that there are more than one company that makes money suggests it's not that hard.
别人能做到，哈考特也可以。
Somebody else can do it, Harcourt can be.
因此，我们专注于推进一种我们称之为加速计算的新计算模式。
And and so singularly advancing a new computing model we call accelerated computing.
我们相信，有一天，加速计算不仅能帮助我们解决普通计算机无法处理的问题，还会让我们接触到所有这些令人兴奋的应用，比如我今天热衷的数字生物学、我们关注的气候变化，以及机器人和自动驾驶汽车。
And we believe that someday that on the one hand, accelerated computing can help us solve problems and tackle problems that normal computers can't, and it exposed us to all of these amazing applications like digital biology that I'm excited about today, like climate change that we're excited about, like robotics and self driving cars.
如果不是因为我们追求那些用普通计算机无法实现的应用，我们怎么会发现这些事物呢？
If not for the fact that we're pursuing applications that were impossible with normal computers, why would we have discovered all of those things?
我们怎么会发现人工智能呢？
Why would we have discovered artificial intelligence?
我们怎么会成为大型语言模型的主力呢？
Why why would we be the workhorse of large language models?
因为大型语言模型几乎是不可能实现的。
Because large language models are barely possible.
如果你在做一件几乎不可能的事，你会找我们。
And if you are doing something that's barely possible, you call us.
我们就是解决这些问题的那匹马。
We're the horse you call to solve those problems.
所以我非常喜欢这一点。
And so I I love that aspect.
我喜欢我们能够发现这些未来技术的事实。
I love the fact that we get to discover those future.
另一方面，我们深信总有一天所有事物都会被加速。
On the other hand, we deeply believe that someday everything will be accelerated.
原因非常明确，那就是CPU的发展将到达极限。
And the reason for that is very clearly that that the CPU will run its course.
通用计算的扩展能力是有极限的，但你始终需要它。
And there's a limit to how far you could scale general purpose computing and you'll always need it.
你始终需要CPU。
You'll always need CPUs.
但我们所有人将要运行的应用类型，加速才是真正的最佳前进方向。
But the type of applications that we're all gonna run, acceleration is really the best way forward.
在我们的核心理念中，从第一天起，三十年前就是这样，这也是我们创立公司的原因。
And at our core we believe that from day one, Thirty years ago, that's the reason why we started the company.
因此，这是一种真正的信念。
And so it's the true conviction.
你在这一长达三十年的信念上得到了极大的验证。
You have been enormously vindicated on this thirty year belief.
在三十年经营公司、学习管理技能的过程中，你一定曾感受到这种信念受到过挑战。
You must have felt that conviction challenge at some point in thirty years of running the company and learning the skills to run the company.
你最接近失败的一次经历是什么？或者你最担心的时刻是什么？那时你是否想过，也许我错了？
What was the nearest death experience or the most concerned where you're like, maybe I'm not right?
或者这样的情况曾经发生过吗？
Or has that ever happened?
是我不适合这份工作吗？
The I'm not right for the job?
不是。
No.
你对加速计算及其重要性理解错了。
You're not right about accelerated computing and how important it will be.
第二个问题，是的。
The second one's yes.
首先，我认为任何人都不应当假设自己天生就适合这份工作，所以你几乎每天都要自我反省。
First of all, I I don't think anybody should assume that they're right for the job and so you should be gut checking on that almost every day.
澄清一下，那不是我要问的问题。
To be clear, that wasn't the question.
但我很乐意回答这个问题。
But But I'll I'll more than happy Very to answer well.
回答了这个问题。
That answer that question.
我曾经认为那是错的吗？
Did I ever believe that it was wrong?
没有。
No.
嗯哼。
Mhmm.
我相信加速计算是解决那些不可能问题的唯一途径，嗯哼。
I believe that accelerated computing is the absolutely the only way to solve problems that are impossible Mhmm.
根据定义，
By definition of
自身。
itself.
好的。
Okay.
公理上。
Axiomatically.
是的。
Yeah.
对吧？
Right?
另一方面，如果你能解决今天看似不可能的问题，而将来你需要这个应用具有广泛的适用性，那么加速计算会是最佳方案吗？
And on the other hand, if you can solve problems that are impossible today and someday you need that application to be broad based, would accelerated computing be the best approach?
答案是肯定的。
The answer is yes.
是的。
Yeah.
你认为CPU什么时候会达到其极限？
When do you think the CPU hits its limits?
你提到，最终你认为所有东西都会迁移过去，或者至少未来很大一部分会迁移过去。
You mentioned that, you know, eventually you think everything will move over or at least big chunks of the future will move over.
这是五年后，还是十年后？
Is that five years away, ten years away?
对于某些应用来说，这种情况已经发生了，没错。
For certain applications it happened That's right.
十二年前。
Twelve years
前。
ago.
是的。
Yeah.
是的。
Yeah.
是的。
Yeah.
对吧？
Right?
嗯。
Mhmm.
杰夫·辛顿、扬·勒昆和安德鲁。
Jeff Hinton and Jan LaCun and Andrew.
对吧？
Right?
安德鲁·杨，他们十二年前就发现了这一点。
Andrew Yang, they discovered that twelve years ago.
这是唯一的前进方向。
It was the only way forward.
在计算机图形学领域，这也是唯一的前进方向。
And computer graphics, it's the only way forward.
是的。
Yeah.
随着人工智能变得越来越重要，你管理和运营公司的方式有改变吗？
Has the way that you organize and run the company changed as AI has gotten more and more prominent?
比如，你有没有围绕它重新调整业务的某些方面？
Like, have you realigned aspects of the business around it?
或者，你如何看待在这种变化如此迅速、这个领域充满令人兴奋的进展的环境中，整体的管理方式？
Or how do you think about management in general in this environment where things are changing so rapidly and there's so many exciting things happening in this area?
你问了一个非常好的问题，也许我应该先退一步说，公司的架构不应该是通用的。
You're asking a really good question and maybe if I just take a step backwards, the company's architecture should not be generic.
世界上每一家公司都不应该像美国军队那样构建。
Every company in the world should not be built like the US military.
事实上，如果你看看世界上每一家公司的组织结构图，它们都差不多像美国军队。
And in fact, if you look at every company's org chart in the world, they kinda look like the US military.
上面有一个人，然后层层向下。
There's somebody on top and then it comes down.
然而，CEO的直接下属人数非常少。
And yet, the number of direct reports of CEOs are very few.
而那些刚刚开始学习如何管理一线经理的人，其直接下属人数却非常多。
And the direct reports of the people who are just learning how to manage first level managers are very large.
这恰恰与应该被设计的架构方式相反。
It's exactly the opposite of how it should probably be architected.
嗯哼。
Mhmm.
你可能会认为，直接向CEO汇报的人根本不需要管理。
You would think that the people that report to the CEO requires no management at all.
事实上，这通常是正确的。
And in fact, it's generally true.
我的直接下属都很成熟，才华横溢，工作能力极强，是出色领导者，具备卓越的商业头脑和远见。
My direct reports are sophisticated, they're really talented, they're incredibly good at their job, they're excellent leaders, they have great business acumen, they have excellent vision.
他们非常了不起。
They're incredible.
每一个都是如此。
Every single one of them.
这么说，你的管理团队人数比管理书籍里建议的六七人还要多？
I guess that means you have more than the management book except at six or seven or whatever.
是的。
Yeah.
我有大约40名直接下属，但没有一对一沟通，也没有职业指导，你知道的。
I have 40 somewhat direct reports and no one on ones, no career coaching, you know.
那么，你希望自己的人生怎样度过？
So what would you like to do with your life?
这些对话通常是和刚毕业的大学生或职业生涯初期的人进行的，我们当然很喜欢这样的对话，帮助他们规划职业、提供指导，让他们接触新的经验。
Those are conversations you have with new college grads and early career, and we love those conversations of course, and and helping them shape their career and mentor them, give them access to new experiences.
但在高管团队层面，我们的组织结构是为了同时推进大量不同的事务。
But at the executive staff level, we're organized so that we can pursue a whole lot of different things at the same time.
然而，对于一家软件公司来说，最重要的一件事是你要理解计算机架构。
However, one of the most important things about a software company is you have to understand computer architecture.
而关于计算机架构最重要的一点是，你只能负担得起一种。
And one of the most important thing about computer architecture is you can only afford one.
就像世界上一些最大的公司只有两种操作系统一样，全球最大的公司也仅有两种。
Just as some of the largest companies in the world only have two operating systems, the single largest company on the planet only has two.
怎么可能有这么多公司拥有如此多不同的计算机架构，还维持着七种、八种甚至九种指令集？
How is it possible that so many companies have so many different computer architectures and they have seven or eight or nine instruction sets that they're keeping around?
我们只使用一种指令集。
We have one instruction set.
我们只采用一种计算机架构，并且对此非常严格。
We have one computer architecture and we're super disciplined about that.
因此，在我们需要专注的地方，我们做到了。
And so where we need to be focused, we are.
在允许高级层面进行创新和探索的地方，我们给予了空间。
Where we allow for innovation and discovery at the senior level, we allow that.
所以我认为，公司这种上下有别的组织结构，与我们工作的本质是一致的，你知道的。
So I think the company is tapered and organized in a way that is consistent with the nature of our work, you know.
所以这是最重要的，也是我在打造我们公司过程中学到的关键一点：没有一种通用的架构适用于所有公司。
So that's the most important thing and and that's probably the takeaway for what I've learned building our company is, there is no one generic architecture for every company.
公司的架构应该契合其职能、目标，当然还有领导者的领导风格。
It should fit the function of the company, its purpose, and of course the leadership style of the leaders.
是的。
Mhmm.
是的
Yeah.
我认为这是一个非常重要的观点，大多数人并没有意识到：公司应该几乎是一个量身定制的结构，以支持首席执行官及其团队以及公司向客户提供的产品，而不是总是千篇一律。
I think that's a really important note that most people don't really realize is that a company should almost be a bespoke structure supporting the CEO and their staff and what the company is delivering to customers versus it's always the same thing.
没错。
Exactly.
我觉得
I think
这一点经常被忽视。
that gets lost a lot.
没错。
Exactly.
是的
Yeah.
是的
Yeah.
你需要某些特定的首席，需要某些首席，确实有些首席是你必须有的。
There's some particular chief that that you need and a chief that that you need and a chief that, you there's some chiefs that you do need.
是的。
Yeah.
但除此之外，你应该从第一性原理出发，为领导者以及职能设计出合理的架构。
But aside from that, you should start from first principles and architect something that makes sense for the the leader and as well as the the the function.
是的。
Yeah.
当我还在谷歌时，他们有一个著名的80-20-10法则，嗯。
When I when I was at Google, they had the famous eighty twenty ten Mhmm.
也就是说，80%是核心业务，20%是与核心相关或新方向的项目，另外10%是高度实验性的。
Where it was like 80% is core, 20% is like core adjacent slash new stuff, and then 10% was hyper experimental.
嗯。
Mhmm.
你有没有什么框架或方法来思考这类问题？
Do you have any frameworks or ways to think about that stuff?
或者我们就只是看看，基于我们构建的通用平台——比如CUDA以及其他内置功能——哪些用途是自然涌现出来的，当新需求出现时，我们就说好，去支持这个新东西。
Or it's just kind of like, let's see what organically is used in terms of this generic platform that we've built CUDA and other things that are built in to help support a lot of use cases and as they emerge we say okay, let's go support that new thing.
我没有类似那样的明智框架。
I don't have any wise framework like that.
我们公司有一些特定的方面，是专门为此设计和组织的。
There are a couple of things that our company is shaped and structured to do.
公司有一部分，是非常大的一部分，其使命就是完美地建造极其复杂的计算机。
There's one part, a very large part of our company is designed to build very very complicated computers perfectly.
因此，这是它的目标之一。
And so that is one of its missions.
这种架构、这种组织形式，是一种发明与精进型的组织。
That kind of architecture, that kind of organization is a invention and refinement organization.
我们有一大堆类似臭鼬工厂的团队。
We have a whole bunch of skunk works if you will.
这样做的原因是，我们正在尝试发明十年后才可能实现的东西，而我们并不确定它们是否真的能成功。
And the reason for that is because we're trying to invent things ten years out that we're not exactly sure whether it's gonna work or not.
而且有很多适应和调整，所以我们的公司实际上有两种不同的工作方式。
And there's a lot of adaptation, a lot of pivoting and so, you know, our company actually has has two different ways of working.
其中一种是相当有机的，时刻在变化形态。
One of them is rather organic, shape shifting all the time.
如果某个投资没有成效，我们就放弃它，把资源转移到其他地方。
If a particular investment is not working out, we give up on it, move the resources somewhere else.
所以这是公司中敏捷的部分，而另一部分则不是僵化的，而是经过精心打磨的。
And so that's the agile part of the company, and then there's a part of the company that's not rigid, but it's really refined.
因此，这两种系统必须并行运作。
And so these two systems have to work side by side.
你能谈谈H100吗？
Can you talk a little bit about the h 100?
下一个主力产品？
Next workhorse?
以及最重要的创新是什么？它的设计和交付流程是怎样的？
And what the most important innovations are and like what the design and ship process for that looks like?
我认为Hopper的重大突破在于认识到量化——即数值量化和数值格式——具有相当大的创新空间，能够显著降低需求。
I would say the big breakthrough for Hopper is recognizing that quantization, the numerical quantization, the numerical formats has a fair amount of innovation and ability to to reduce.
因为本质上它就是统计性的。
Because it's statistical in the first place.
现在的问题是，可以创建和训练什么样的模型，我们相信八位浮点数比当今科学计算中常用的六十四位浮点数更合适。
And now the question is what kind of models could be created and trained and we believe that eight bit floating point rather than if you look at scientific computing today, 64 bit floating point.
因此，只需将64位拆分为8位，就能通过不使用64位浮点数，将AI超级计算机的性能提升八倍。
And so just by breaking up 64 into eight, you could increase the performance of an AI supercomputer just by a factor of eight by not doing 64 bit.
所以，这几乎相当于在短短几代内，仅通过意识到64位浮点数并非必要，就实现了近十倍的性能提升。
So that's almost a factor of if you will, factor of 10 almost in just a couple of generation just by recognizing that 64 bit floating point wasn't necessary.
因此，其中一件大事就是这一点。
And so the one of the big things is that.
第二件事是Transformer。
The second thing is is transformer.
Transformer引擎如此通用且有效，以至于我们可以设计出专为学习和推理Transformer而优化的流水线。
The transformer engine is so universal and so useful that it's possible for us to design a pipeline that is shaped for learning and inferencing transformers.
所以，这可能是最重要的两点。
And so those are probably the two biggest things.
否则，它是世界上制造过的最大的芯片。
Otherwise, it's the largest chip the world's ever made.
它是，你知道的，世界上最快的芯片，超级节能，并且使用了世界上最快的内存。
It's, you know, the fastest chip the world's ever made and super energy efficient and uses the fast memories of the world's ever made.
然后我们将大量这样的芯片连接在一起，使其既快速又节能。
And then we connect a whole bunch of these things together so that it's fast and energy efficient.
但这些都是一些，你知道的，纯粹靠蛮力的东西。
But those are all, you know, kind of brute force y things.
但核心的架构理念是FP8、流和Transformer引擎。
But the big architecture idea is FP eight and flow and transformer engine.
当你思考这一点时，这就是公司主要的项目优化部分。
And when you think about then, so that's the big project refinement part of the company.
我们还考虑更灵活的部分。
We think about the more agile piece.
你们正在研究什么不可能的应用，以实现十年后可能变得重要的目标？
What's the impossible application you're working on to get a day that's ten years out you think is likely to be important?
我相信肯定有很多这样的应用。
I'm sure there are ton of them.
我们正在研究一大堆目前还无法实现的项目，但我很有信心它们最终会成功。
There's a whole bunch we're working on that don't work at the moment, but I've got a lot of confidence it will work.
好的。
Okay.
例如，自动驾驶仍在取得进展。
So for example, autonomous driving is still making progress.
但我坚信它一定会成功。
But I have every confidence that it will work.
我坚信会发现一种机器人基础模型。
I have every confidence that a robotic foundation model will be discovered.
通过使用人类语言进行表达，你可以让一个包含多种类型肢体和灵活性的巨型系统学会如何弯曲和调整自身，以完成特定任务。
And that through expressing yourself using human language, you could cause a megatronic system of almost different types of limbs and agility to be able to figure out how to bend itself, articulate itself to do a particular task.
你认为目前阻碍这一目标的因素是什么？
What do you think the blockers are to that today?
哦，我完全不知道。
Oh, I have no idea.
但我无法告诉你。
But I I can't tell you.
我只是不知道。
I'm just I don't know.
我不知道。
I don't know.
是的。
Yeah.
因为我们必须自己探索出通往那里的道路。
Because we have to discover our way there.
但我们确实知道的一件事是，我们已经懂得如何从非结构化信息中学习结构，比如语言、图像，而下一个重大突破将是视频。
But one of the things that we do know is that we do know how to learn structure from unstructured information, language, images, and of course the next big thing is video.
如果我们能仅仅通过观看视频来学习其中的结构，或许就能理解我们如何表达，并将这种能力泛化，从而为机器人构建一种去表达系统。
And if we could just watch video and learn the structure from the video we're watching, we might be able to learn how we articulate, and we might be able to generalize that and be a de articulation system for robots.
因此，我认为这些路标表明，各个部分正在逐渐整合。
And so I think the road signs, if you will, would suggest that the pieces are coming together.
但当我们真正到达那一步时，完全不知道会是什么样子。
But when we get there, have no idea.
但我认为这可能不到十年，我的猜测大概是五年左右，你会看到一些非常惊人的机器人。
But I think it's it's probably less than, my guess is gonna be less than ten years, probably about five years, and I think you're gonna see some pretty amazing robots.
这真是太令人兴奋了。
And so exciting.
是的。
Yeah.
在这些方面也有一些相关进展。
There's some things along those lines too.
比如，谷歌的保罗最近就发布了一些内容，这算是朝这个方向迈出的一步，我想这仍然属于变换器架构的范畴。
Guess, like, Paul from Google came out recently that's sort of a step in that direction, and I guess that's still in the transformer architecture.
你提到了Transformer管道，并将其融入到你们正在做的事情中。
And you mentioned the sort of transformer pipeline and sort of baking that into what you all are doing.
在AI方面，还有其他新的架构是你在关注的，或者你特别认为会发展出有趣成果的吗？
Are there other new architectures on the AI side that you're watching or especially you think will develop into something especially interesting?
嗯，Transformer有很多衍生版本，它们通常都统称为Transformer。
Well, there's a whole bunch of derivatives of transformers and they're all just kinda generally called transformers.
但这种基础架构正在被不断优化和改进，一方面如此。
But that basic architecture is being refined and deltaed and on the one hand.
另一方面，我们非常兴奋的一些工作，我们最初基于Ian Goodfellow的GAN研究，做了很多关于风格迁移和高分辨率图像生成的出色工作，这又催生了大量变分自编码器的研究，进而演变为所谓的扩散模型的近亲，这条路径我们发挥了非常重要的作用。
On the other hand, some of the stuff that we're really excited about that we we did a lot of work in, we started with Ian Goodfell's work on GANs and we did some really great work on a style transfer and high resolution generation of images and which led to a whole bunch of work in variational auto encoders, which then became, you know, if you will, a bit of a cousin of the diffusion models that And came so that entire path we played a very large role in.
从这项工作中将涌现出大量衍生成果。
And there's a whole bunch of derivative works that's gonna come out of that.
在从海量数据（无论是视频还是多模态学习）中学习结构的能力方面，这无疑将成为一个非常重要的方向。
Between the ability to learn structure from a giant amount of data, whether it's video or, you know, multimodality learning is going to be a very big thing of course.
而接下来的部分则是内容生成。
And then the next part of it is generating content.
如果你能生成图像，能生成二维和三维图像，那为什么不能生成蛋白质和化学物质，生成各种各样的东西呢？
And if you can generate images and you can generate two d and three d images, why can't you generate proteins and chemicals and you can generate all kinds of stuff.
几乎没有其他企业家能从三位创始人起步，三十年后打造出七千亿美元的市值。
There are almost no other entrepreneurs that have gone from three founders to CEO, thirty years and 700,000,000,000 of market cap.
你对听这个节目的企业家有什么建议？
What advice do you have for entrepreneurs that listen to the show?
这是一份非常艰难的工作。
It's a really hard job.
我不是说CEO这份工作。
I don't mean the CEO job.
建立一家公司本身就很难。
Just building a company is hard.
你们两人与许多公司从最初阶段的创立密切相关。
The two of you are associated with a lot of companies being formed from the very very beginning.
创办初创公司没有任何容易之处。
There's nothing easy about building a startup.
我甚至不明白为什么有人会第二次创业。
And I don't even understand it that anyone would build a startup twice.
这简直是一场磨难。
It is such a ordeal.
是的。
Yeah.
试着
Try to
劝别人别这么做。
talk people out of it.
比如第二次创业的人，因为我创办了两家公司，第二次的时候我会问：你真的确定还要这么做吗？
Like second time founders because I started two companies and the second time like, you sure you wanna do this?
毫无疑问，你不该这么做。
Oh, there's no question you shouldn't do it.
是的。
Yeah.
你绝对不应该这么做。
There's no question you shouldn't do it.
这一定是一种类似生孩子后的遗忘机制。
It has to be some sort of like forgetting mechanism, like with having kids.
你会觉得，第一次也没那么糟糕。
You're like, it wasn't that bad the
第一次。
first time.
明白了。
Got it.
说得太对了。
It's exactly right.
比如，你必须忘记它有多艰难。
For example, you have to forget how hard it was.
我不知道自己是怎么做到的，但我就是做到了。
And I don't know how I do it, but I just do.
我忘记了做这件事所伴随的痛苦和煎熬。
I I forget, I forget the pain and suffering that goes along with doing something.
一旦你达成某个目标，就会立刻转向下一个目标；当你达成下一个目标后，又会继续转向再下一个，生活就是这样。
Once you achieve something, you just move on to the next thing and once you achieve that, you move on to the next thing and it's just like life.
你只需要一步一个脚印地往前走。
You put one step in front of the other.
有什么建议？
What advice?
我不太愿意给他们什么建议，原因就在这里。
I'm reluctant to give them any advice and and the reason for that is this.
几乎任何建议都可能让你打消做这件事的念头。
Almost any advice would probably discourage you from wanting to do it.
我认为，无知是创业者的一项超能力，而你再也无法重获这种无知。
I think ignorance is one of the superpowers of an entrepreneur and you'll never get it again.
你再也无法重获这种无知了。
You'll never have it again.
我真的很喜欢我们公司的一点是，我们一直在不断重塑自己。
And so the thing I really love about our company is we're reinventing ourselves constantly.
我们在这家公司里就像创业者一样，我参加的所有会议都像是初创公司的会议，而且都很痛苦。
We're kind of entrepreneurs inside this company and all the meetings that I go to are really startup meetings and they're all painful.
这些会议都很痛苦，因为你又要从零开始，没有任何势头，基本上一切归零，每次都会让我想起创业有多痛苦，但当你打造出某样东西，而为你打造它的人们真正欣赏它时，这种回报也无比丰厚。
They're all painful because you're starting something from the ground up again, you have no momentum, you're you're basically at zero and it reminds me every single time how painful it is, but it's also so rewarding when you when you build something and the people you built it for appreciate it.
而且它确实带来了改变。
And somehow it made a difference.
然后你把这种能力与其他技能和能力结合起来，就能做得更加出色。
And then you combine that skill with some other skills and some other capability and also you can do something even greater than that.
一方面，我会告诉他们，创办公司是极其有回报的，你能够与之共事的那些人，这确实是事实。
On the one hand, I I would tell them that building a company is extraordinarily rewarding and all the people you get to work with, that's genuinely true.
但另一方面，创业过程中的痛苦和煎熬是任何你无法想象的。
On the other hand, the pain and suffering of doing it is unlike anything you can imagine.
所以，你知道，你有一天是英雄，第二天就成了混蛋，你不断经历这些循环，但你必须设法超越这一切，专注于你真正想做的事。
And so, you know, you're vulnerable, you're a superhero one day, you're a jerk the next day, you go through these cycles and somehow you have to look beyond all of that and focus on what you're trying to do.
所以我不知道是否给了他们什么智慧，除了如果你决心要做这件事，就别等太久，直接去做吧。
So I don't know if I gave them any wisdom aside from if you're determined that you wanna do it, don't wait too long, just go do it.
在你失去无知之前？
Before you lose your ignorance?
是的。
Yeah.
在你失去无知之前。
Before you lose your ignorance.
因此，如果非要说有一个特质，我会说，你必须有足够的决心，坚持自己的信念。
If therefore, there's one attribute, would say, you have to be determined enough to stay with your conviction on the one hand.
但另一方面，你也不能固执到失去灵活性，必须能够持续学习。
On the other hand, you can't be stubborn so that you can have agility, so that you can continue to learn.
所以，在一方面我相信我所做的事情，另一方面我又相信自己可能是错的，这种平衡很奇怪。
And so somewhere in that balance of I believe in what I'm doing on the one hand and I simultaneously believe that I could be wrong on the other hand, that is weird.
你必须同样坚定地相信这两点。
And you have to believe both equally hard.
我公司的名字叫信念，你可以有灵活性。
My firm's name conviction, you can have agility.
好的。
Okay.
我会把它当作一个糖果品牌来启动。
I'll start that as a candy brand.
是的。
Yeah.
对。
Yeah.
我们见过一些极具才华的初创企业首席执行官，他们几乎都对了。
And we've seen startup CEOs that are incredibly talented and they're almost right.
但他们太执着于正确，反而忘记了保持灵活、在过程中学习、调整和适应，我认为这就是一方面。
But they were so determined to be right, they forgot to be agile, to learn along the way and pivot and adapt and so I think that's on the one hand.
所以这是需要记住的一点。
And so that's that's one thing to remember.
而另一点是韧性，它伴随着遗忘而来。
And then the other is resilience which comes along with forgetting.
你必须忘记痛苦，继续前行。
You have to forget the pain, move on.
这有点像教练说的：别在意上一个得分。
And it's a little bit like coaches saying, don't worry about the last point.
你知道，你刚被狠狠揍了一顿，还错失了一个季度，就像我错失季度时，当你提到加密货币，我的手就开始冒汗。
You know, you just got your face kicked in, you know, and you missed a quarter, know, like when I miss a quarter, when you mentioned crypto, my hands started to sweat.
我知道。
I know.
我的心跳也开始加快。
My my heart started to beat, you know, faster.
因为我记得错失季度的时候，当我们加密货币季度没达成时，我们跌得很惨。
Because I remember missing the quarter and you when we missed a quarter during crypto, we missed it hard.
加密货币很难预测，我们曾从供不应求一下子变成供过于求。
Crypto was hard to predict and we went from having no supply to too much.
你知道吗，谁会错过一个季度，还差了20亿美元？
You know, who misses a quarter by, you know, $2,000,000,000?
我的意思是，这是个巨大的数字。
I mean, that's that's a big number.
嗯哼。
Mhmm.
大多数时候，你听到的都是CEO们只差了1500万美元。
Most of the time you hear CEOs miss it by $15,000,000.
是的。
Yeah.
不是20亿美元。
Not $2,000,000,000.
我认为莎拉提出了一个很好的观点，你现在已经打造出了科技界最顶尖的公司之一，并且正在积极推动可能是有史以来最重要的技术方向之一——人工智能。
I think Sarah had a great point in terms of you've built now really one of the marquee companies in the tech world, and you're really pushing forward what is potentially one of the most important ways of all time in technology, which is AI.
十年后、二十年后回望，你希望在公司层面或更广泛的范围内实现哪些具体目标？或者你希望二十年后回看时，有哪些事情真的发生了？
Ten years from now, twenty years from now, looking back, are there any specific things that you wanna accomplish either through the context of the company or more broadly, or other things that you, looking back twenty years from now you really hope happen?
这是个很好的问题。
That's that's a good question.
事实上，这是一种很好的思考方式。
And in fact that's a good way to think.
思考今天该做什么的最好方法，就是走向远方，站在未来回望。
The best way to think about what to do today is to go out into the distance, stand in the future and look back.
你们可能也这么做。
You guys probably do the same.
所以我会展望十年后，想想那时我后悔没做什么，然后现在就去做。
And so I'll go out ten years and look back at what did I wish I had done then, then do it now.
这就是答案。
That's the answer.
因此，有几项我们坚信自己能做出贡献的行业。
And so there are a couple of industries we really believe we can make a contribution to.
其中之一是医疗和药物研发。
One of them is healthcare and drug discovery.
这是一个在计算和数值上极其复杂的问题。
This is a problem that is computationally, numerically insanely complex.
可能的组合数量超过了宇宙中原子的总数。
The number of combinations is beyond the number of atoms in the universe.
这是一个非常庞大的问题空间。
It's a very large problem space.
而我们终于拥有了可能逐步攻克这一难题的必要工具。
And we finally have the necessary tools to maybe chip away at that.
至少，我们现在有能力理解氨基酸、序列、蛋白质、化学物质等的语言，甚至可能理解它们的含义。
And at the very minimum, we now have the ability to understand the language of and now potentially the meaning of amino acids and sequences and proteins and chemicals and such.
因此，如果你能理解结构、掌握语言、洞悉问题空间的含义，你就有可能解决它。
And so if you can understand the structure, you can understand the language, you can understand the meaning of the problem space, you might have a chance of solving it.
所以，我认为第一点，我们对此非常兴奋。
And so I think one, we're very excited about that.
我真心希望我们能为气候科学构建一个多物理场的基础模型。
I'm really really hoping that we go create a foundation model for multi physics for climate science.
这样我们就可以向它提问。
And so that we can ask it questions.
如果这些人为因素和人类行为产生影响，十年、二十年、三十年后地球会怎样？
If these human factors and these human drivers and we make these impact, what would happen to the earth ten years, twenty, thirty years from now?
这是一个极其复杂的问题。
It's an insanely complicated problem.
从计算量来看，人们估算这比当今地球上最快的超级计算机还要多出十亿到一万亿甚至十万亿倍。
Computationally, people have estimated it it's anywhere from a billion to 10,000,000,000 to a 100,000,000,000 times more computation than the fastest supercomputer on the planet today.
这基本上意味着我们永远无法实现。
That basically says we'll never get there.
嗯。
Mhmm.
另一方面，借助人工智能，我们或许真有机会将计算量减少十亿倍、一万亿倍。
On the other hand, with artificial intelligence, we might have a real chance of reducing that computation by a billion times, 10,000,000,000 times.
因此，我希望我们这一代人有机会在这两个领域做出巨大贡献。
So I'm I'm hoping that that we have the opportunity in our generation to make a huge contribution to these two areas.
所以我们正在做这件事。
So we're doing it.
地球二号是我们气候科学系统，而克拉拉是我们医疗与健康系统，旨在更好地理解如何在这一领域做出贡献。
Earth two and is our climate science system, and Clara is our medical and health care system to understand better how to contribute in that space.
非常令人兴奋。
Very exciting.
我还有一个最后的问题，是的。
I have one last question Yeah.
这和解决世界上最复杂、规模最大的计算搜索空间以拯救人类和地球同样重要。
That's as important as attacking the most computationally intensive, largest search spaces in the world to save humanity and the Earth.
从我们的观众那里来的问题：这些皮夹克是从哪儿来的？
From our audience, where did the leather jackets come from?
我妻子买的。
My wife.
好的。
Okay.
所以你不知道？
So you don't know?
我们得测试一下你妻子。
We have to test your wife.
我有
I have
一点概念。
no idea.
太神奇了。
Amazing.
这仍然是个谜。
I It remains a mystery.
我妻子和女儿总在为我找夹克。
My wife, my daughter, they're always hunting for jackets for me.
我得承认，我挂起来的大多数夹克都太前卫了，我根本不敢穿。
Most of the jackets, I have to admit, that I have hanging up are too fashion forward for me to carry out, you know.
这些是比较低调的款式，但有些衣服你得真的够潮才能穿得出去，所以我只是不想显得格格不入。
And so these are more modest ones, but but some of them are just you have to actually be cool to wear them, and so I just don't wanna look out of place.
那到底谁会穿这种衣服呢？
So no foundation model Who wears this?
不该穿那些夹克。
Should not wear those jackets.
非常感谢你参与这次对话，詹森。
Well, thank you so much for doing this, Jensen.
这是一场令人深受启发的对话。
It's been an inspiring conversation.
谢谢。
Thank you.
真的很享受这次交流。
Really enjoy it.
继续加油。
Keep up the good work.
感谢您收听本期的《No Priors》。
Thank you for listening to this week's episode of No Priors.
关注《No Priors》以每周获取新嘉宾，并在网上告诉我们您想听到哪些AI领域的人物。
Follow No Priors for new guests each week, and let us know online what you think and who in AI you wanna hear from.
您可以通过关注Sarah enormous来与我保持联系。
You can keep in touch with me and conviction by following at Sarah enormous.
您可以在Twitter上关注我，账号是Elad Gil。
You can follow me on Twitter at Elad Gil.
谢谢收听。
Thanks for listening.
《No Priors》由POD People联合制作。
No Priors is produced in partnership with POD People.
特别感谢我们的团队：Cynthia、Elad和Pranav，以及POD People的制作团队。
Special thanks to our team, Cynthia Elad and Pranav Ruddy, and the production team at POD People.
Alex Vigmanis、Matt Saab、Amy Machado、Ashton Carter、Danielle Roth、Carter Wogan和Billy Libby。
Alex Vigmanis, Matt Saab, Amy Machado, Ashton Carter, Danielle Roth, Carter Wogan, and Billy Libby.
还有我们的父母、孩子、学术界，以及tyranny.ml，不过是再普通不过的友好型AGI世界政府罢了。
Also, our parents, our children, the academy, and tyranny.ml, just your average friendly AGI world government.
关于 Bayt 播客
Bayt 提供中文+原文双语音频和字幕，帮助你打破语言障碍，轻松听懂全球优质播客。