Naveen Rao：推理为何可能成为 10 万亿美元市场，软件投资又将如何改变

简要说明Brief Description

Naveen Rao 探讨了能源与算力限制、模型与硬件协同设计、验证、强化学习、推理经济学、软件市场、机器人技术，以及新兴 AI 基础设施背后的投资假设。Naveen Rao discusses energy and compute constraints, model–hardware co-design, verification, reinforcement learning, inference economics, software markets, robotics and the investment assumptions behind emerging AI infrastructure.

目录Table of Contents

• 简介
  Intro
• 为什么能源将成为人工智能发展的最大瓶颈
  Why Energy Will Be the Biggest Bottleneck for AI
• 美国为何可能耗尽电力备用容量
  Why the US Could Run Out of Excess Power Capacity
• 计算领域投资者对效率仍有哪些误解
  What Compute Investors Still Get Wrong About Efficiency
• 为什么 AI 硬件需要与模型协同设计
  Why AI Hardware Needs to Be Co-Designed With the Model
• 芯片专业化会威胁英伟达吗？
  Does Chip Specialization Threaten Nvidia?
• 为什么 AI 芯片不会沦为大宗商品
  Why AI Chips Will Not Become Commoditized
• 为什么 AI 将瓶颈从创作转移到了验证
  Why AI Has Shifted the Bottleneck From Creation to Verification
• 我们会比信任人类更信任 AI 智能体吗？
  Will We Trust AI Agents More Than Humans?
• 为什么数据公司将变得更加专业化，而非商品化
  Why Data Companies Will Become More Specialized, Not Commoditized
• 为什么强化学习如此重要
  Why Reinforcement Learning Is So Important
• 为什么我们将超越 Transformer
  Why We Will Move Beyond Transformers
• AI 技术栈中哪些部分真正会变成商品？
  Which Parts of the AI Stack Actually Get Commoditized?
• 推理会成为一种商品吗？
  Will Inference Become a Commodity?
• 为什么纯软件投资者“要完蛋了”
  Why Pure Software Investors Are “F***ed”
• 为什么 20 年后 SaaS 的重要性将大幅降低
  Why SaaS Will Matter Far Less in 20 Years
• 为什么 AI 未必会摧毁开发者的工作岗位
  Why AI Will Not Necessarily Destroy Developer Jobs
• 机器人技术何时才能真正成为一个成熟的市场？
  When Will Robotics Actually Become a Real Market?
• 为什么推理将成为一个 10 万亿美元的市场
  Why Inference Will Be a $10TRN Market
• 为什么机器人技术的规模将是编程智能体的 100 倍
  Why Robotics Will Be 100x Bigger Than Coding Agents
• Naveen 如今在机器人领域的投资方向
  Where Naveen Is Investing in Robotics Today
• 为什么传统风险投资指标不再适用于深科技 компании
  Why Traditional VC Metrics No Longer Work for Deep Tech
• 当今最被高估的 AI 市场
  The Most Overrated AI Markets Today
• 为什么 80% 的 AI 令牌可能会通过开放模型流转
  Why 80% of AI Tokens Could Flow Through Open Models
• 为什么欧洲诞生的雄心勃勃的初创企业较少
  Why Europe Produces Fewer Ambitious Startups
• 为什么 AI 能够开辟全新的科学前沿
  Why AI Could Unlock Entirely New Frontiers of Science

引言Intro

未来几年，我们就会能源耗尽。这只是个短暂现象，并不是什么泡沫破裂。我们不会把模型奉为神圣。我们的思维方式是能源优先。Naveen Rao 是一位连续创业者。我的意思是，他把上一家公司 MosaicML 以 13 亿美元的价格卖给了 Databricks。如今，他带着 Unconventional AI 再度出发，这家公司已经筹集了数亿美元，最近一轮融资后的估值接近 50 亿美元。Navine 对许多复杂领域都有着广泛而深刻的见解，表达也十分清晰、流畅。We're going to be out of energy in the next few years. This is a blip. It's not some kind of bubble being burst. We don't treat the model as sacrian. And we're energy first in our mindset. Naveen Rao is a serial entrepreneur. I mean, he sold his last business to data bricks for $1.3 billion. Mosaic ML. Now he's back with unconventional AI, which has raised hundreds of millions of dollars. Their last round close to $5 billion in valuation. Navine is one of the most articulate and clear thinkers across such a wide range of very complicated topics.

他太棒了，参与这档节目也很愉快。在我看来，智能的需求将是无限的。我确实认为，未来10年，硬件才是值得关注的领域。我想说，如果你是纯软件投资者，那你就有点[ __ ]。未来10年，推理市场将达到10万亿美元。He was fantastic and this show was a joy to do. Intelligence to me will have infinite demand. I actually think the hardware is the place to be over the next 10 years. I would say if you're a pure software investor, you're kind of [ __ ] The inference market is going to be a 10 trillion market within 10 years.

我不会那样看待它，你知道，就是“天哪，人工智能要杀光所有人了”。这是我听过的最愚蠢的世界观。准备好了，出发吧。I don't look at, you know, oh my god, AI is going to kill all humans. That's the stupidest worldview I've ever heard. Ready to go.

哈里·斯特宾斯：纳文，我真是太期待这次对话了，老兄。我跟你说过，我特别特别崇拜你。马特·伯恩斯坦、马特·吴，还有很多其他人。老兄，非常感谢你今天来参加节目。Harry Stebbings: Naveen, I am so excited for this, dude. I told you I sto the [ __ ] out of you. Matt Bournestein, Matt Oo, many others. Thank you so much for joining me today, man.

当然。Absolutely.

为什么能源将成为人工智能发展的最大瓶颈Why Energy Will Be the Biggest Bottleneck for AI

是啊，我一直很期待。Yeah, I was looking forward to it.

是啊，我觉得这会很有意思，因为我看了你们很多节目。我把我觉得最有争议的一些言论都记了下来。顺便说一句，里面有几句相当犀利。不过我其实想先从这套基础设施的底层基础谈起，也就是能源。你之前谈过能源会成为瓶颈，以及它的重要性。你如何评估当下的能源供应状况？有哪些我们应该看到、却还没看到的问题？Yeah, I think it's going to be a lot of fun because I went through a ton of your shows. I wrote down what I thought were the most controversial quotes. There were a few zingers, by the way, but I actually wanted to start on like the foundational layer of this infrastructure, which is energy. You've spoken before about how it will be the bottleneck and the importance of it. How do you evaluate the state of energy supply today and what are we not seeing that we should see?

纳文·拉奥：能源供应设施仍然很难建设，对吧？我觉得大家没有意识到这一点。我的意思是，我看到过一些行业普遍数据，建造一吉瓦的供电能力大约要花 500 亿美元。所以大家动不动就把“吉瓦”挂在嘴边，好像这很容易似的。拜托，这可不容易。打个比方，建一座晶圆厂，比如半导体晶圆厂，成本大约是 200 亿到 300 亿美元。所以，单从资金和复杂程度来看，建造一吉瓦项目比建一座晶圆厂还要难。Naveen Rao: Energy supply continues to be difficult to build, right? I don't think people appreciate it. I mean, I've seen, you know, numbers from general industry is about $50 billion to build one gigawatt. So, people are throwing around the around the word gigawatt like it's like it's easy. And it's like, guys, this is not easy. Just to put that in perspective, building a fab like a semiconductor fab is on the order of 20 or 30 billion. So, this is one gigawatt is more than building a fab like if you if you look at it in terms of dollars and complexity.

所以，这很难。我认为过去几年我们看到的情况，是一直在消化现有的过剩需求。也就是说，我们通常会过度建设，而且基础设施的产出不会达到 100%。因此，我们一直以来为数据中心获取能源的主要方式，是看看能不能提高现有设施的利用率，从中再挤出接下来的几吉瓦电力。So, it's hard. I think what we've seen over the last several years has been soaking up the excess demand that's been around. So we overbuild generally and we don't run infrastructure at 100% output. So what we've been doing where we've been getting energy for data centers is primarily been from this like can we run things at higher utilization and you know and get that ne next couple of gigawatts out of it.

所以，是的，我觉得如果你看看美国电网，平均输出功率大约是900吉瓦，但我认为它的峰值能力可达1.2到1.3太瓦，也就是1300吉瓦。所以确实有一些富余容量，而这主要就是我们一直在利用的部分。这是你能快速获得的资源。但基础设施建设很难，而且我们需要建设更多。我们已经落后了。So yeah, I think if you look at the US power grid, we're on average something like 900 gawatts of output, but I think it's capable of peaking up to like 1.2 1.3 terowatts. So 1,300 gawatt. So there is some excess capacity and that's primarily what we've been going after. That's what you can get access to quickly. But building infrastructure is hard and we need to build more. We're behind.

我们落后多少？嗯，我觉得过去 20 年里，中国在总产能方面超过了我们，我认为这发生在 2023 年。但过去 20 年，他们的建设速度是我们的 6 倍。他们原本落后很多，但他们确实在这件事上大步快跑、全力以赴。我们对此批评了很多，因为我们说，哦，他们用的是肮脏的手段，烧煤啊之类的。没错，但他们也在努力以尽可能快的速度发展。How far behind are we? Well, I think over the last 20 years China surpassed us in terms of total amount of capacity that happened I think in 2023. but they were building 6x faster than us for 20 years. They were coming back from far behind but they they really went fast and hard at this. And we criticized that a lot because we said oh they're using dirty methods to do this coal burning this and that. True, but they're also trying to grow as fast as possible.

而我觉得现在他们正在回过头来，把其中一些问题收拾妥当。所以，你知道，与此同时，我们并没有真正投资能源基础设施，而这会让我们吃亏，因为要真正发展到那一步需要时间。And I think now they're sort of going back and cleaning up some of this. So, you know, meanwhile, we didn't really invest in energy infrastructure and that's going to hurt us because it's going to take time to really get there.

老兄，中国比我们领先这么多，我们能从中学到什么，以至于应该在政策和监管上放宽得多？我们能得到哪些启示？Dude, what can we learn from China being so much further ahead of us that we should be like way more loose on policy and regulation? What are the takeaways?

嗯，我觉得其中一些问题很难处理，因为它触及了我们和他们文化的核心。我觉得中国可以进行集中规划，而我们做不到；这两种方式各有优势，对吧？在那些大规模基础设施建设项目上，确实有优势，这一点毋庸置疑。不过，我认为在创新领域，尤其是最前沿的东西上，美国要强得多。人工智能是在这里诞生的，不是在那里。你知道，所有这些最新的创新都来自这里。当然，中国也做了很多很出色的工作，我不想贬低这一点，因为他们做的不只是技术蒸馏，也确实有真正的创新。Well, I think some of this is hard because it hits to the very core of our culture and their culture. I think China you know can centrally plan. We don't and there are advantages to both of these things right and in these things where you have these giant infrastructural rollouts yeah there's an advantage no doubt about it however I think when it comes to innovation tip of the sphere sort of things like the US is far and away better so AI was born here not there you know like all of these these latest innovations have all come from here now they're doing some really great work in China also I don't want to diminish that because it's not just distillation they're really doing some some true innovation.

不过我觉得，真正开拓新领域、冲在最前沿的那些事，是从这里来的，不是从那里来的。However, I think that tip of the spear stuff where you really open up, you know, new areas comes from here, not from there.

所以你知道，我觉得归根结底，我们必须发挥自身优势。So, you know, I think we have to leverage our advantages is what it comes down to.

看看如今稀缺的东西，你觉得现在还不稀缺、但三到五年后会变得稀缺的是什么？When you look at what's scarce today, what do you think is not scarce today that will be in 3 to 5 years?

美国为何可能耗尽富余电力容量Why the US Could Run Out of Excess Power Capacity

我认为这种富余产能很快就会变得非常稀缺。你知道，如果我们看看增长率，即便是 AI 增长率中保守得多的那种，接下来几年我们也很快就会面临能源不足的问题。而且，也不是说只要简单地说“好，我再找一些产能”就行了；那些富余产能都会被用完。这样一来，我们就会处于一种产能增长速度逐渐与建设速度趋同的状态。你知道，我听人说过：“哦，我们一年要建设 40 吉瓦。”这根本不可能，我们从来没做到过。我就是觉得这不可能，至少现在还不行。I think that excess capacity is going to start getting very scarce. you know if we look at growth rates even even the kind of much more conservative growth rates of AI we're going to be out of energy pretty soon in the next few years and it's not easy just to say okay I'm going to find some more capacity you will have used up all of that excess so then we're in a we're in a place where we kind of converge to the rate of building and you know I've heard people say oh we're going to build 40 gawatts in a year there's no way we have never done that I I just don't see as being possible. at least not right now.

也许再过几年，我们会有一些新发明，然后靠建设摆脱困境。但就目前而言，我们不可能一年建成 40 吉瓦。也许能建成 10 吉瓦。Maybe in a few years we'll come up with some new inventions and we will build our way out. But as it stands today, we're not going to build 40 gawatts in a year. Maybe we build 10.

如果我们建成10个，那就太惊人了。换个角度看，这相当于5000亿美元。所以，当我们审视这个债务周期时，它已经被认为处于极高位，而我们现在看到，这种情况开始在核心领域连锁显现或逐步展开，围绕这一点也有很多担忧。If we build 10, that's amazing. That's half a trillion dollars, just to put it into perspective. So when we look at the debt cycle that is already being talked about as being incredibly toppy and we're seeing that now starting to cascade or unpack in core and a lot of concerns around that.

是的。它是……它实际上只是在一开始吗，还是派对上的音乐会在晚上7点关掉？Yeah. Is it a is it actually just at the start or does the music turn off at the party at 700 p.m.?

我觉得，最大的决定因素是：对 AI 服务有没有需求，进而对 GPU 有没有需求，也就决定了这些新增能源建设是否会过剩。显然，我就在这个行业里。我算是很看好 AI 的人。我觉得需求很充足，而且我们甚至还没触及表面。你知道，AI 领域还有很多问题需要解决，但随着我们在未来几年逐步解决这些问题，就会催生出新的需求类型——而这些需求，我们今天其实还想象不到。I I think the biggest the biggest turner of that will be is there demand for AI services for therefore for GPUs and therefore for this excess this new energy buildout. I mean obviously I'm in this industry. I'm I'm pretty AI pill. I think there's plenty of demand. I don't think we've even scratched the surface. you know there's a lot of problems we have to solve in AI still but as we solve those and we will over the next couple of years it's just going to open up new new types of demand that we don't even really know about today.

所以我的感觉是，这只是一个小插曲，不是什么泡沫破裂。我的意思是，这有点像你问：“嘿，人们需要有人治愈癌症吗？”当然需要，对吧？如果你能治愈癌症，人们肯定会想要这种疗法。So my sense is that this is a blip. It's not some kind of bubble being burst. I mean it's sort of like if you said hey is there demand for people curing cancer? Of course there is right? Like if if you cure cancer people are going to want that.

比如说，如果你造出一台能做任何事的智能机器，人们肯定会想要它。Like if you build an intelligent machine capable of doing anything, yeah, people are gonna want that.

我总记得 Vin 或 Kosa 对我说过：“我希望尽可能多地承担技术风险，尽可能少地承担采用风险。”I always remember Vin or Kosa saying to me, I want to take as much technology risk and as little adoption risk as possible.

是啊，Yeah,

正是如此。我有一种治愈癌症的方法。很难做到，但采用风险为零。我完全理解这一点。但这是否意味着，如果我们眼看着供应只剩18个月就要耗尽，那么今天对过剩供应的定价就错了？我的意思是，你可以这么说。我想这取决于能源市场，对吧？能源市场就应该这样定价，价格也应该往上走。至少从一些个案中，我确实看到了证据。关于这一点，我没有具体数字可以引用，因为每一笔表后交易都是量身定制的，而且人们每千瓦时支付的费用，远高于接入标准电网时的价格。it's exactly that. I have a cure for cancer. Very difficult to do. but zero adoption risk. I totally get that. But then does that mean that we're mispricing the excess supply today if we're literally staring down the barrel of 18 months running out? I mean, you could argue that. I suppose that's up to the energy markets, right? that's that's how they should be pricing and moving up. I do see evidence through anecdotes at least. I don't have specific numbers to quote around this because every one of these behind the meter deals is bespoke and people are paying way more per kilowatt hour than they would off the off the standard grid.

所以，我认为我们已经看到了一种两级电价：数据中心用电的价值似乎高于普通用途的电力，对吧？So, I think we're already seeing a two-tiered pricing occurring where the value of electricity for data centers is sort of higher than it is for general use, right?

所以，现在表后有很多事情正在发生。一些公司正在、正在、正在开发新型技术，比如 Bloom Energy。Bloom Energy 有一项协议，要在亚利桑那州建设大约几吉瓦的项目。如果能成，那就太棒了。这或许就是我们需要的、能让发展继续下去的技术进步。不过，你知道，这些技术还没有真正、真正得到验证。顺便说一下，这些项目的经济性和你在电网上看到的情况大不相同。So, there's a lot of behind the meter stuff going on right now. U companies are are are building new sort of technologies like Bloom Energy. there's a Bloom Energy deal for building, you know, a couple gigawatts in in Arizona. if that works, amazing. That might be the technological advances we need to keep going. But, you know, it hasn't really really been proven out yet. And those economics, by the way, are very different than what you see on the grid.

怎么说？How so?

我的意思是，我觉得归根结底，如果看数据中心能源每千瓦时的价格，人们愿意支付的费用就是更高。你知道，因为利润结构要好得多，所以承受更高的价格也没问题。比如说，我有一吉瓦的算力，想把它卖出去，价格可以是成本的五倍。数字是我随口编的，但大致就是电费的五倍。所以，我在这方面有很大的回旋余地。I mean, I think if you just boil it down to how much people are willing to pay per kilowatt hour for energy for data centers, it's just higher. You know, you can tolerate a higher price because the margin structure is just so much better. Like basically, you know, if I have a gigawatt of compute and I want to sell that, I can sell it for let's call it 5x the price. I'm making numbers up, but it's on the on the order of 5x the price of what I pay for that electricity. So, you know, I'm I got a lot of room wiggle room there.

算力投资者对效率仍有哪些误解What Compute Investors Still Get Wrong About Efficiency

所以，我可能会为这种电力多支付 30% 到 40%。当我们从能源这个核心基础动力，稍微往下游转向使用它的计算环节时。So, I could pay 30% 40% more for that electricity. when we kind of go slightly further down from energy which is kind of the core foundational power to the compute that use it.

是的。Yeah.

现在，对于算力，我们有哪些认识是你希望自己两年前就知道的？我想，有一点我希望自己早就知道：我们实际上已经看不到物理效率的提升了。我们可以把计算机的构建方式分成三个层次。第一层是物理层。比如说，如果我能把一个比特从这里传到那里，而且成本更低——假设我用这里表示 0、那里表示 1，而我的手机有一定的重量。What do we know now about compute that you wish you had known 2 years ago? I think one thing I wish I'd known is the fact that we don't really see physical efficiency gains anymore. So you can separate out the way we build a computer in let's call it three levels. So the first level is the physical. So if I can send a bit from this point to that point and do it cheaper like you know let's say I represent zero as being here and one as being here and my phone has a certain weight.

所以，在这两种状态之间切换需要能量。如果我把手机做得更轻，所需的能量就会改变，对吧？所以这纯粹是物理层面的事情。我们通常所说的摩尔定律，或者你可能听说过的丹纳德缩放（Dennard scaling），也就是把器件做得更小，本质上也是一件物理层面的事情。So there's energy required to move between these two states. if I make a lighter phone it changes the amount of energy required right? so that's just a pure physical thing. And Moor's law as we generally refer to it or Dennard scaling you may have heard where you make devices smaller is really a physical thing.

所以，你是在让物理层变得更高效、更密集。这些创新在很大程度上已经停滞了。你知道，如果我们看看最近几代 GPU 或其他芯片，就会发现它们在效率上的提升已经不多了。效率提升来自更上一层，也就是计算架构。因此，我认为过去几年或几代产品中，我们看到的大多数效率提升，都来自数据类型变得更小：从 32 位降到 16 位、8 位，再到 4 位。So you're making the physical layer more efficient and denser. so those innovations have largely stopped and you know if we look at the last few generations of GPUs or other chips you don't see efficiency gains much anymore. You see efficiency gains from the next layer up which is kind of the computing architecture. So I would say most of the efficiency gains we've seen over the last several years or several generations have come from smaller data types going from 32-bit to 16 bit to 8 bit to 4bit.

所以你能获得相当不错的效率提升，因为你搬运的数据位更少了，计算所用的数据位也更少了。再往上一个层次，是在模型层面开始强制实现局部性并减少数据搬运，而这一点还没有得到很好的利用；我们在 Mosaic 稍微尝试过一些。你知道，五六年前我们不会这么想，因为模型还不是计算的主要驱动力。甚至，嗯，也许六年前，计算仍然是通用用途的：我运行网站、数据库之类的各种东西。So you get pretty good efficiency gains because you're moving fewer bits, you're computing with fewer bits. then the next level which is not really exploited that well and we started this a little bit mosaic is about how do I start enforcing locality and reduction of data movement at the model level and you know 5 6 years ago we didn't think about it like this because the model was not the primary driver of compute or even yeah maybe six years ago it was still compute was a general purpose thing I run websites and databases and all that kind of stuff.

不过，现在如果我只把计算机看作一种旨在提供智能的物理设备，思维方式就会有所改变。这样一来，我就能从模型、架构一直到物理层进行协同设计。我真希望自己早些知道，顶层还有这么多潜在的超额收益，而不只是物理层。我们长期以来都把重点放在物理层上。我想问，从关注模型之前转向关注模型之后，贯穿整个技术栈时，这种设计思维会发生怎样的变化？However, now if I look at the computer solely as a physical device designed to to deliver intelligence, it changes the mindset a bit. Now I can co-design all the way from the model through the architecture to the physical. And that's what I wish I had known how much sort of alpha there's there is left in that in the top layer, not just the physical. We focus on the physical for a long time. Can I ask how does that design mindset change when moving across the stack with that transition from premodel focus to postmodel focus?

是的。我的意思是，我们不会把模型奉若神明，对吧？我们的思维方式是能源优先，也就是说，我们做的每件事都必须让我们在能源方面占据优势。这听起来很简单，但过去没人以这种方式设计这些东西。一直以来，人们关注的都是速度。其实，这种思路可以追溯到 ENIAC，追溯到 20 世纪 40 年代。因为计算机的设计初衷，就是比其他方案更快。Yeah. I mean it means we can we don't treat the model as sacrian right and we're energy first in our in our mindset meaning that everything we do has to give us an energy advantage. It sounds like a simple thing but it's not how anyone has designed these things in the past. It's always been about speed. And this actually has its roots that go all the way back to ENIAC, you know, in the 1940s. because, you know, computers were designed to be faster than the alternative.

一直都是这样。顺便说一句，关于 ENIAC，你知道它为什么会被造出来吗？ENIAC 是个首字母缩略词。我想它代表的是“电子数字积分计算机”（electrical numerical integrator and calculator）。That was always the case. And in the ENIAC's case, by the way, actually, do you know why ENIAC was built? It's an acronym. electrical numerical integrator and calculator, I think is what it stands for.

听起来不错。Sounds good.

是的，这是一台由美国政府于 1945 年建造的计算机。它是最早的大型数字计算机之一。和所有曾经建造出来的机器一样，它也有自己的用途：计算炮弹弹道，而且要比当时的替代方案更快、更准确。那时的替代方案就是由人亲自计算。所以，机器必须更快，这才是关键。下一代产品问世时，也必须比这一代更快。你知道，GPU 问世时，必须比 CPU 更快地进行渲染。每一代产品都试图比前一代更快，这出于多种原因带来了效率提升。Yeah. So, this was a computer that was built by the US government in 1945. it was one of the first large-scale digital computers built and its purpose like every other machine that's ever been built has a purpose. Its purpose was to do artillery trajectory calculation and do it faster and more accurately than the alternative. The alternative at that time were humans who literally did the maths. So you have to be faster. That was the key. Next generation comes out, it's got to be faster than that. you know GPUs came out, they had to be faster at rendering than a CPU. each thing successively has tried to be faster than the previous thing that has led to efficiency for several reasons.

我的意思是，在相同的功耗预算下让东西变得更快，并不算更高效。但与此同时，摩尔定律也一直在发挥作用，你知道，我们从真空管发展到了晶体管。I mean making things faster and the same power envelope isn't more efficient. but also Moore's law was chugging along and you know we went from vacuum tubes to transistors.

晶体管一直在变得越来越小、越来越小、越来越小。而现在，我们已经来到这样一个阶段：从物理层面来说，变化并不大。比如晶体管并没有太大的改进。不过，在模型和架构层面，我们还有很大的提升空间。这是否意味着模型正在走向商品化？还是说，模型质量正在拉开差距——有些模型能够节省能耗、提高效率，而另一些则不能？Transistors kept getting smaller and smaller and smaller. and now we're in a place where that physical side of things is not really changing a whole lot. Like transistors haven't gotten much better. However, we have a lot of room left on the model and architectural levels. Does that mean we're seeing a commoditization of models or does that mean we're seeing actually a separation of model quality in those that are able to save energy and have efficiency gains versus those that aren't?

我觉得完全正确。所以，你知道，有一件事我一直做得挺不错，就是大致判断出接下来几年大家会把注意力放在哪里。我觉得这会是一个很大的……I think that's exactly right. So, you know, one thing I've been pretty good at is kind of picking where people are going to be focusing in the next couple of years. And I think this is going to be a big

Naveen，你真是太出色了。真的得给你点赞。你在选择市场时，会不会这样推演：我觉得世界会朝这个方向发展，所以我要在这里布局？exceptional Naveen. And really like credit to you. Do you walk through when you're picking a market and go like this is where I think the world's going to go and so I'm going to position myself here?

是的，当然。而且你知道，这归根结底也只是兴趣使然。对我来说，打造一台更好的计算机是最有趣的问题之一，你知道的。Yeah, absolutely. And you know it also comes down just to interest. You know, this is for me one of the most interesting problems there is to build a better computer, you know.

所以你当时坐在这里，在想法迷宫里还没变得非传统，然后你说：“我要专注解决的问题是……”那是什么？嗯，我的看法是：如果我要解决一个问题，会带来什么影响？也就是说，它能开启什么？事实上，这有点像做科学家。做一名优秀的科学家，关键就在于提出正确的问题——回答了这个问题，就能引出更多问题。这几乎就是一篇优秀论文的标志。So you sat down here pre- unconventional in the idea maze and you were like problem set that I'm going to focus on is dot dot dot. What was that? Well, I kind of look at it like if I were to solve a problem, what is the effect? Like what what does it unlock? In fact, it's kind of like being a scientist. Like being a good scientist is all about asking the right question that if you answer it, it unlocks many more questions. That's that's almost the hallmark of a good of a good paper.

所以对我来说，在这项事业中，很明显，我们会耗尽能源。这一点很清楚。我们现有的范式并没有真正聚焦于此。我认为这将成为关注重点，因为我们需要继续扩大规模，对吧？我们需要以许多新的形态提供智能，并大幅降低成本，才能让它无处不在。So for me in this in this endeavor, it was clearly like, okay, well, we're going to run out of energy. That's clear. The paradigm we have hasn't really been focused on this. I think this is going to be where the the focus will be because we need to keep scaling, right? Like we need to deliver intelligence in many new form factors and and greatly disrupt the cost in order to make it ubiquitous.

所以目前还没人考虑这件事。So no one's thinking about this yet.

为什么 AI 硬件需要与模型协同设计Why AI Hardware Needs to Be Co-Designed With the Model

我来开个头。可以吗？我们刚才在聊模型效率，以及它在能耗方面能带来什么。如今我们看到模型的效率越来越高。我们该如何看待自己在模型效率曲线上的位置？这方面还有很大的提升空间吗？还是说我们已经接近渐近点，只能一点一点地挤出最后的提升了？I should start it. Can I we were talking about like model efficiency and how it can deliver in terms of energy. We see models getting more and more efficient today. How do we think about where we are on that model efficiency curve? Is there a lot of room to run there or are we kind of at the asmtotic point where you're like we're ekking out the final increments?

不，不，不。我们还有很长的路要走，我认为原因在于，我们还没有真正对模型和其物理层进行协同设计。所以，你可以提出一个概念，或许仅靠软件，就能让标准 GPU 的效率提升，比如说 5%。这听起来很不错。然而，由于硬件本身的构造方式有限，你能从这个概念中获得的收益也有限。看看我们的大脑：智能与物理结构是共同演化的。我们让智能层和物理层共同演化，因此能够在物理层中充分发挥智能层的优势。No, no, no. We have a lot to go and I think the reason is we've not done this true codees between the model and the physical side of it. So, you can come up with a concept and you may be able to get say let's call it 5% efficiency gain just from software on a standard GPU. that sounds great. However, you can only exploit whatever concept so much because the physical thing is built in a certain way. If you look at our brain, they co-evolved. We co-evolve the intelligence layer and the physical layer and we're able to exploit things in the intelligence layer in the physical layer.

比如说稀疏性，稀疏性就是指计算单元之间没有连接。利用 GPU 中的稀疏性，我肯定能获得一定的收益。但如果使用标准的片外随机存取存储器，这些收益就会受到限制。So, sparity for instance, sparity means let's say just not having connections between computing units. I can I can get some kind of gain out of using sparsity in a GPU 100%. But when I have a standard offchip random access memory, the limit of those gains is not is capped.

我不能走得太远，因为这种物理硬件未必就是为利用这类稀疏性而设计的。所以，如果我能构建出能够利用它的东西，就能获得巨大的提升。这正是大脑所做的，也是它共同进化的方式。想到共同进化，或者说架构针对任务进行专门化，这是否意味着我们会始终看到芯片不断走向专门化？比如 Etched、Fractile、Olix，显然还有 Cerebras，以及许多专注于特定工作负载的公司。这就是计算的未来吗？你怎么看？I can't go too far because that physical thing is not designed necessarily to exploit this kind of sparity. So if I can build something that exploits it, I can get massive gains. That's really what the brain did. That's how it co-eolved. When we think about that like co-evolution or that specialization of architecture to the task almost does that mean that we'll unwaveringly have the specialization of chips we see etched and we see fractile and Olix and a number cerebrous obviously but a number of people going after specific workloads is that the future of compute how do you think about that

百分之百是这样，而且你知道，这其实可以追溯到图灵最初的论文：他想要造出一台智能机器。没错，我的意思是，这从一开始就是他明确提出的目标。所以我们沿着这条路走下去，把他的机器——一个理论构造——做成了实体版本。我们造出的任何东西都不能算是真正“图灵完备”，因为它们的内存有限，诸如此类。不过，我们确实能够构建出非常实用的应用，对吧？100% it is and you know this actually goes back to even Touring's original essay he wanted to build an intelligent machine Yeah, I mean that was that was his stated goal right at the beginning. so we went down this path of we took his machine which was a theoretical construct and we built physical versions of it. So anything we build is not truly quote unquote touring complete because you know they have finite memory and stuff like this. but we were able to build applications that were super useful, right?

我的意思是，我们可以用它们来运营企业。我们可以编辑视频、图片、文本，还能做各种很棒的事情。所以，过去60年来，生产力有了巨大的提升。但它偏离了最初的目标，而现在我们算是又回到了正轨。我认为，计算机的用途是提供智能；对我来说，它就是智能的一种物理体现。I mean we could run businesses off of them. We could we could edit videos and pictures and text and all kinds of great things. So that's a huge productivity gains over the last 60 years. but it deviated from that original goal and now we've kind of come back. So the purpose of the computer I think is to provide intelligence is to me it's it's just a physical manifestation of intelligence.

所以现在我们必须想办法改进它。有趣的是，如果你能打造出更好的智能物理体现形式，就创造出了某种通用的东西。从某种意义上说，我们的大脑是通用的。不过，硬件的工作负载是固定的，但它并不像我们谈论 CPU 时那样具有可编程性。因此，GPU 已经走上了为特定工作负载构建定制架构的道路。我认为我们会在这方面进行超优化，而这是一件好事，因为这样一来，我们最终就能打造出真正有能力创造任何东西的智能机器。And so now we have to figure out how to make that better. And what's interesting is that if you make a better physical manifestation of intelligence, you made something general. Our brains in some ways are general. However, hardware is fixed in its workload. but it doesn't do programmability in the same way that we think about it in a CPU say. So the GPU is already down this path of building customized architectures for workloads and I think we're just going to hyper optimize around this and that's a good thing because now we can build intelligent machines that are actually capable of building anything at some point.

芯片专用化会威胁英伟达吗？Does Chip Specialization Threaten Nvidia?

听着，原谅我这颗只会算钱的脑袋。你是个很棒的技术专家。很多方面来说，我就是个不太靠谱的交易员。不过我们刚才提到，芯片会坚定地走向专业化。这在多大程度上会大幅缩小英伟达这类横向供应商的 TAM？还是说，哥们儿，别这么目光短浅，现在还早得很，我们还有 50 倍的增长空间。是啊，我的意思是，市场会非常庞大。瞧，我认为英伟达显然在引领我们走到今天这一步的过程中执行得非常出色。Listen, forgive my financial mind. You're a wonderful technologist. I am a delinquent trader for a living in in many respects, but we mentioned like yes, unwaveringly specialization of chips. To what extent does that massively reduce the TAM for a horizontal provider Nvidia specifically versus dude do not be so smallminded. It is still so early. We have 50x the room to run. Yeah, I mean it is that the market is going to be huge. I mean look I think Nvidia obviously executed extremely well on the path to getting us here.

我是说，你知道，我在2014年创办 Nirvana 时，我们并不真正清楚这个市场会有多大。我很清楚，我们需要一种围绕神经网络构建的新型计算架构。这是一种描述计算的新方式；对我来说，这是计算机能够完成的最重要的事情。I mean you know when I started Nirvana in 2014, you know we didn't really understand how big the market was going to be. It was clear to me that we needed a new computing architecture around neural networks. That was a new way to describe compute and to me it was like the most important thing a computer could do.

我想，Jensen 也相对较早就意识到了这一点。我的意思是，他在 2014 年的 GTC 上就开始谈人工智能了。我对此记忆犹新，因为当时很多图形领域的人都对此很恼火。他们说：“你为什么在人工智能上花这么多时间？这根本不算什么。它不是一项重要的工作负载。图形才是。”而他正确地认识到，人工智能将会变得很重要，于是投入了大量资源朝这个方向发展，而且确实执行得非常出色。所以，我们不得不经历计算架构的演变：从这些机器最初的用途，到它们最终将成为的样子。I think Jensen recognized this relatively early on as well. I mean he he started to talk about artificial intelligence at the GTC in 2014. I remember it vividly because a lot of the graphics guys were pissed off about this. They were like why did you spend so much time on AI? It's nothing. It's it's it's not a important workload. And graphics is right. and he correctly recognized that this was going to be important, invested a ton in in in in that direction and actually executed very well. so we had to go through this evolution of computing architectures from what we use those machines for to what they were eventually going to become.

但我觉得现在我已经很清楚了，我不需要在意其他任何事情。我只需要关注如何非常高效、非常可扩展地运行神经网络。But I think now it's pretty clear to me that I don't need to care about anything else. I only need to care about running a neural network very very efficiently, very very scalably.

为什么 AI 芯片不会走向商品化Why AI Chips Will Not Become Commoditized

这将是衡量优劣的指标。and that's going to be the figure of merit.

你觉得芯片在多大程度上正在商品化？你知道，微软昨晚发布了他们的芯片。显然，谷歌凭借 TPU 取得了惊人的成功，亚马逊也有自己的芯片。Anthropic 正在打造自己的芯片。还有那个名字特别棒的——我超喜欢这个 Jalapino。我想知道下一个版本会叫什么。是啊，是啊。但我们会看到芯片就这样不可避免地走向商品化吗？首先，我觉得下一个肯定得叫 Habanero，因为它会更辣。我想最终可能会叫 Ghost Pepper 之类的，对吧？To what extent do you think we see the commoditization of chips? You know, Microsoft last night released their chips. You've obviously got you Google with amazing success with TPUs, Amazon with theirs. You got Anthropic building their own. You've got the wonderfully named I love this Jalapino. I wonder what the next version will be. Yeah. Yeah. but do we see that just inherent commoditization of chips? Well, first off, I think the next one's got to be habanero because it's gonna be it's gonna be hotter. I would think ghost pepper or something eventually, right?

我觉得你应该把它注册为商标，否则 Sam 可能会抢先注册。I think you should trademark it otherwise Sam might do it.

我会问他。所以我其实不觉得芯片会变成大宗商品。我有点把它看作与我们在进化过程中看到的情形并行发展。所以我对这一切如何运作有一个非常简单的世界观。对我来说，智能的需求是无限的。对智能的需求没有上限，因为我们生活的世界、这个宇宙，是充满竞争的。起初，两只细菌可能在争夺一处阳光充足的地方，或者类似的东西。拥有感知能力、能够移动或以某种方式适应环境的那一只就会胜出，并不断繁殖。然后，这些东西组织成真核细胞，再形成多细胞生物；每一次，本质上都是在提高能量层级和竞争赌注，但它仍然是充满竞争的。I'll ask him. so I actually don't see chips being commoditized. it I I sort of look at it as it runs in parallel to what we saw in evolution. So I have a very simple worldview of how all this stuff works. So intelligence to me it will have infinite demand. There is no limit of demand of intelligence because the world that we live in the universe is adversarial. So at first two bacteria might be trying to find a spot of sunlight or something like that. the one that has the sensory capabilities to be able to move or or somehow adapt itself into that environment wins and keeps proliferating. then those things organize into ukarotic cells and multi multi-ellular organisms each each time you're basically upping the level of energy and the stakes but it's still adversarial.

而现在，我们在人类群体之间就能看到这种情况，对吧？我们谈论美国对中国，或者不管是什么对什么。你知道，在某种程度上，我们开始更多地利用太阳能，就像卡尔达肖夫等级所说的那样。我们在打一场星际大战之类的，不管那是什么；要做到这一点，你需要智能。智能就是这些对抗性战斗的燃料和经济基础。因此，对智能的需求是无限的，而要创造智能，你就需要能源。And right now we see this between human populations, right? We talk about US versus China or whatever it'll be something versus something and you know at some level we start to harness more of our sun like this is like the Cardesev levels you know and we're fighting some intergalactic battle whatever that is you need intelligence to do that. Intelligence is the is the fuel the economy of these adversarial battles. So it has infinite demand and in order to create intelligence you need energy.

将能量转化为智能，靠的是硬件。我说的硬件，是指整个技术栈：芯片、系统、云、网络，所有这些东西。所以对我来说，如果只从经济角度来看，我要卖的就是将能量转化为智能的能力。The conversion of energy to intelligence is hardware. And when I say hardware, I mean the whole stack. I mean the chips, the systems, the cloud, the networks, all of those things. So to me, if I look at it just from an economy, like I'm going to be selling the capability to convert energy to intelligence.

而我越擅长做这件事，积累的价值就越多。所以就市场而言，我不认为它短期内会商品化。你会不会担心自己贪多嚼不烂，结果在从能源转向智能的过程中降低产品质量？在两者之间的过渡期，你能承接的越多，赚的钱就越多。And the better I do that, the more value I acrew. So I don't see it commoditizing anytime soon as far as the market goes. Do you worry that you bite off more than you can chew and then deprecate product quality in that transition from energy to intelligence? The more that you can eat in the in between, the more money you're going to make.

是啊，不过昨天我请来了 Modal 的创始人 Eric 上节目。Yeah, but I had the founder of Modal on the show yesterday, Eric.

我是说，我很喜欢他。但这家伙就是想什么都做。我是说，什么都想做。我当时就在想，产品广度到什么程度会影响产品深度？And I mean, I love him. But the dude just wants to do everything. I mean everything. And I was like, at what point does product breath damage depth?

你会考虑这个吗？Do you think about that?

我——我们一直都在做这件事。专注非常重要。我确实认为，随着 AI 的发展，世界正在发生变化。我们可以拓展更多领域。我们能够探索的范围比以前更广了。I we do all the time. Focus is super important. I do think the world is changing now with AI. We can take on more breath. Our surface area that we can explore is greater than it was before

仅仅是因为开发速度快得多。比如，为什么现在每家公司都可以说是复合型初创公司？你知道，在我们的第一款芯片中，我们构建了一组耦合振荡器。它们有一个耦合矩阵，还有一大堆彼此影响的振荡器。所以在做这件事时，我会问，比如说，在过去的世界里，也就是 AI 出现之前，我可能会问：好吧，如果我们不采用全连接，噪声会如何传播？just because of like speed of dev is much faster. Like why is that now every company's a compound startup so to speak? You know, in our first chip, we built a set of coupled oscillators. Okay, they they have a coupling matrix. They have a bunch of oscillators that sort of influence each other. And so when we're doing this, I would ask I you know, in in the old world, say before AI, I might have a question like, okay, well, how does noise propagate if we don't do all toall connectivity?

也就是说，我有一串级联的振荡器，起始端带有某种噪声特性。我想看看信息经过四到六级传递后会衰减成什么样。在过去，我得去问研究团队：“嘿，能不能帮我把这个建模？加上某种噪声模型。”So meaning I have like a daisy chain of oscillators and I have some noise characteristic at the beginning. I want to see how that degrades information over say four or six hops. In the old world, I'd have to ask my research team, hey, can you go model this? Put some sort of noise model in.

我这里只是想先有个初步的认识，对吧？好让自己的思路有个方向。这可能要花上一周左右，因为他们会说：“我还在忙别的事，下周再回来处理。”也许写代码、验证代码并给我结果就要一周。所以，我大概得等上几周才能得到答案。我完全可以直接把这个问题问模型，它会生成代码。我可以检查代码，也可以跟它来回交流几次。And I just want a zero with order understanding here, right? Just to kind of guide my own thoughts. That might take a week or so cuz they're like, I'm working on this other stuff. I'll I'll come back to it next week. Maybe it takes a week to write the code and verify it and give me an output. So, I got a couple of weeks to get an answer. I can literally ask this question to a model. It spits out the code. I can introspect the code. I can go back and forth with it with it a few times.

我一个小时就能得到答案。所以，原本需要我花两周时间思考的问题，现在只需要一个小时。因此，借助 AI，我单位时间内能够思考更多事情。现在I get an answer in an hour. So something that would have taken me two weeks to think through has now taken me an hour. So I can think through more things per unit time because of AI. Now

为什么 AI 已将瓶颈从创作转移到验证Why AI Has Shifted the Bottleneck From Creation to Verification

我觉得我们还是有个问题。昨天我刚发帖说，我们面临着 AI 垃圾内容泛滥的问题。这些模型产出的垃圾内容太多了，而且很难核实。其中有些不错，有些则不怎么样。我们得想办法弄清楚如何大规模处理这个问题。不过我可以做得更快，覆盖的范围也比以前更广。I think we still have a problem. I actually just posted on this yesterday that we have an AI slop problem. We have too much crap coming out of these models and it's very hard to verify it. Some of it's good, some of it's not. And we need to figure out how to how to do that at scale. But I can go faster. I can cover more surface area than I could before.

你为什么说我们有太多垃圾内容？是不是就像在 X 上刷到那些愚蠢的观点，比如现在刷屏的那个东西？我不是不尊重美国，但你们的政客嘛，脸部结构相当有趣——用英语可以这么说。那都是我看到的，不过还有他们不同 AI 生成版本。还有外星人，还有从国会出来的……我是说，这太疯狂了。你说的 AI 垃圾内容就是这个意思吗？What makes you say we have too much crap? Is it just like looking on X and seeing like the the dumb takes on, you know, the thing that is just covering my timeline right now? No disrespect to the US, but it's your politicians who are rather in interestingly structured faces. would be an English way of saying it. that's all of mine, but there's like different AI generations of them. There's aliens and coming out of Cong. I mean, it's nuts. Is that what you mean by AI slob?

不是。不是。It's not. it's not.

那只是娱乐而已。That's entertainment.

是啊，那更像是娱乐。我顺便说一句，英国人在自嘲式幽默方面可谓独占鳌头。这比我们在美国能做到的好太多了。所以，Yeah, that's more entertainment. I will say, by the way, the English have cornered the market on self-deprecating humor. It is so much better than we can ever do in the US. So,

说实话，顺便说一句，我们大家从英国这边看着这一切，都在想：“哇，我们之前还一直抱怨自己的政客呢。”[ __ ] 这些家伙。I honestly, by the way, we're all looking at this from the UK being like, "Wow, we've been moaning about our own politicians." [ __ ] these guys.

是啊，我知道。不，我们已经达到新的高度了。毫无疑问。我跟你说过，我们是创新者。好吧。Yeah, I know. No, we've taken to new levels. No doubt about it. I told you we're innovators. All right.

所以，实际上，这是个很好的例子：你看到一个输出结果，会说：“好吧，这种结果首先能通过你的直觉检验吗？”然后你就会觉得：“嗯，可能不行。”所以，我觉得特别有意思的是，如今的老年人看到图片时，并没有这种直觉检验。他们会说：“哇，这是真的。”我是说，我爸妈就曾经让我给他们看特朗普弹吉他的 AI 生成视频，他们还会说：“哇，他会弹吉他。”我就会说：“妈，他根本没在弹那把吉他。”And so, actually, this is a perfect example of you see an output and you say, "Okay, does that kind of does it meet your gut check first?" And you're like, "Huh, maybe not." So, I think what's really interesting is older people today see images and they don't have that gut check. They're like, "Wow, that's real." I mean, I've had my parents say like, you know, show me these AI generated videos of Trump playing the guitar and they're like, "Wow, he plays the guitar." I'm like, "Mom, he's not playing that guitar.

“那是 AI 生成的东西。”她说：“哦，好吧。”所以对我来说，很明显那是 AI 生成的。我觉得，我们在企业里看到的验证问题，在我们公司也存在。我们可以构建各种各样的产物。我们可以……我们可以编写代码，也可以构建模型。That's an AI generated thing." She's like, "Oh, okay." So, for me, it's quite obvious that it's AI generated. So, I think that verification problem we see in companies, I see it in our company. We can build all kinds of artifacts. We can we can build code. We can build models.

我们可以做各种各样的事情。但你必须形成闭环，确保每一项都正确，才能真正信任结果。而这实际上会逐渐成为一个巨大的瓶颈，因为你会产出那么多成果。所以我认为，我们已经把问题从生产转移到了验证。We can do all kinds of things. But you have to close the loop and make sure each one of those things is correct to actually trust the result. And doing that actually starts to become a big bottleneck because you're producing so many artifacts. So I think we've moved the problem from producing to verification

而验证的价值在于，如果你掌握了验证层，就掌握了自动化层，因为你有了可验证性，之后就能以此为基础不断推进。是这样吗？是的，我想是这样。我认为这是利用 AI 开发事物时必不可少的一环。但如果你掌握了验证层，就掌握了生产层吗？我不太确定。我觉得你还得掌握规格定义。也就是说，要产出有用的东西，比如造一辆车或一座桥之类的，你必须能够定义它的规格，还要想清楚：我解决的是一个有价值的问题吗？and the value of verification is if you own the verification layer then you own the automation layer because you have like verifiability that you can then progress there on. Is that it? yeah, I suppose I think it's a necessary part of using AI to develop things, but if you own the verification layer, do you own the production layer? I'm not entirely sure. I think you have to own the spec as well. So, producing the a useful output like building a car or a bridge or something, you have to you have to be able to spec it, you know, figure out am I solving a problem that's useful?

然后，生成的每一项产物都必须经过验证，确认其正确，并且逐步朝着规范要求推进。最终还必须能够确认：我确实按预期实现了这份规范。因此，这是一个端到端的流程，而不仅仅是验证。Then each one of these artifacts that's produced has to be verified to be correct and then marching towards that spec and eventually have to say yes I did indeed implement the spec as intended. So it's endto-end process not just verification.

所以，验证只是其中的一部分。我不知道该如何衡量它的价值。我想，我会把它看作整个过程的一部分，就像最终得到结果一样。So verification is a piece of it. I don't know how one would value that. I think I value it as part of the whole like getting to the end result.

我们请来了 Atlassian 的 Mike 上节目。他人特别棒。他聊到了思维链和理解的重要性，以及在公司不同职能中，由智能体主导决策和由人类主导决策之间的区别。我一部分觉得，我完全理解，这也完全说得通；但另一部分又觉得，这么想似乎有点过时了，这到底意味着什么？我们对智能体决策的信任不如对人类决策的信任吗？这听起来也有点奇怪。We had Mike on the show from Atlassian. Awesome guy. and he was talking about kind of the importance of kind of chain of thought and understanding, you know, agent-led decision-m versus humanled decision-m across different functions in in companies. And part of me was like, I totally get that and that makes absolute sense. And part of me was like that feels kind of old to think of it like that and like what does that mean? We don't trust agent decisions as much as we trust humans. That feels kind of weird too.

我们没有。We don't.

你怎么看？是的，我是说，听着，我同意你的看法。我觉得……How do you think about Yeah. I mean look I agree with you. I think

我认为，在很多情况下，我们暂时还不应该像信任人类那样信任人工智能的决策。我们要经过多年的训练，才能培养出判断力。比如成为一名神经外科医生，对吧？医学院毕业后还要接受七年的培训，你知道，这是有原因的。学会缝针并没有那么难，对吧？我读研究生时就做过。问题不在于外科医生所做的那些身体上的机械操作。I don't think we should trust Asian decisions as much as humans in a lot of cases just yet. We train for certain things for many years to build judgment. I mean becoming a neurosurgeon, right? It's a seven-year training after after med school, you know, and there's a reason for that. It's not that learning how to throw a stitch is that hard, you know? I've done that in grad school. It's not the physical mechanical part of being a surgeon that's the problem.

关键在于要有判断该朝哪个方向发展的能力。你必须见过足够多的案例，见过各种输出，也见过人们如何思考这些问题，才能逐步培养出这种判断力。我认为，我们现在对智能的定义是：我们拿一个模型，它是随机性的，对吧？我们可以，呃，我们可以多次提示它，然后得到各种不同的输出。It's having the judgment on which direction to go. You have to have seen enough cases, seen the outputs, seen how people think through these problems to then build that judgment up. And I think the way we're defining intelligence right now is we take a model that's it's stochastic, right? We can u we can prompt it several times and we'll get different sorts of outputs.

你所做的就是探索各种可能解决方案的分布。然后我们需要从中挑选出好的方案。你知道，你会得到很多不同的输出，而其中大多数都不好。你也可以说，人类也会这样做，对吧？我们会说，好吧，我该怎么做决定？嗯，我可能会仔细考虑所有输出，并用某种启发式方法来判断什么是好的。我们在内部会这样做，也许是通过内心独白，或某种内部流程。但模型会输出一条思维链，然后我们会得到许多条思维链，再判断哪条好，并沿着那条路走。对于每项任务来说，这种决策环节实际上都还没有得到充分理解，所以我认为我们不应该信任这些东西。And what you're doing is you're exploring the distribution of possible solutions. And then we need to pick which of those solutions is good. you know, you'll get a lot of different outputs and most of them are not good. And you could argue that a human does this too, right? We say, okay, what how do I make a decision? Well, I might think through all the outputs and I have some kind of heristic of deciding what's good. We do this internally, maybe it's an internal monologue or some sort of internal process. but a model spits out a chain of thought and then we have many chains of thought and we decide what's good and we take that path. that deciding element is actually not well understood for every task and so I don't think we should trust these things.

我认为，你构建的决策链越拉越长——一个决策接着一个决策，再接着一个决策——就越容易偏离正轨，最终可能得到看起来非常、非常稳健，实际上却完全错误的结果。I think the longer and longer deci decision upon decision upon decision chain you build the more off the rails it can go and you can get things that look very very robust but they're completely wrong.

这就是我们目前的情况。未来可能会有所变化。So that's where we are today. That will probably change in the future.

我们会比信任人类更信任 AI 智能体吗？Will We Trust AI Agents More Than Humans?

我觉得我们还需要不少突破才能走到那一步，但这就是我们今天所处的位置。所以，我个人很喜欢这样一个问题：你认为如今看起来古怪、奇怪、疯狂的事情，五年后会变得非常普遍，并且得到大家的认同？我记得我把信用卡信息填进 PayPal 时，父母还责备我；还有 OkCupid 之类的网站，我当时会想：你竟然能在网上找到丈夫或妻子？这想法也太荒唐了，当时简直疯狂。我认为，我们会比信任自己更信任我们的智能代理。I think we need a lot a few more breakthroughs to get there but that's where we are today. So I personally I love a question like what do you believe that today looks kooky weird nuts that in 5 years time will be very common place and agreed. I remember putting my credit card into PayPal and my parents telling me off and like remember like okay cupid and being like you find your husband or wife online what a ridiculous idea nuts at the time. I think that we will trust our agents more than we trust ourselves.

我觉得某种程度上确实如此。是的。I think at some point that's true. Yeah.

但你觉得我们今天离那个目标还很远吗？But you think that we are far from that today?

是的。我们今天还没到那一步。而且专业能力其实永远不会过时，只是会以不同的形式呈现。我认为，未来会是这样：我们开始构建某种东西，我们姑且称之为“模型”，但我其实不认为它只是一个模型。我认为它是整台机器。我认为——我认为智能是由物理实体和智能层合而为一的。它们无法分割。所以，我有一台机器，它几乎就像一个有机体，我可以教它成为各种任务方面的专家，而它的价值将来自不断积累这些专业能力。Yeah. We're not there today. And expertise doesn't really go out of style. It's just going to take on different forms. And what I think a future will look like is that we start building, we'll call it models, but I actually don't think it's a model. I think it's the whole machine. I think of I think of intelligence as having a physical and having the intelligence layer as one. They're they're not separable. So I have a machine. It's almost like an organism that I can teach to be an expert at various tasks and the value will come from building that expertise up.

就像，你知道，你上的是哪所学校？你是怎么学会当神经外科医生，或者厨师之类的？这种师承渊源很重要，对吧？比如你跟着某个人学习，学到了他们的方法论和思维方式。It's like you know which school did you go to? How did you learn how to be a neurosurgeon or or a chef or whatever that providence matters right? Like you went and studied under this person and you learned their methodology their thinking.

价值正是由此而来。就像你，或者任何人类，从头到脚都是一台智能机器，但真正让你有价值的，是你接受教育的方式。我认为，我们也会开始在机器身上看到类似的情况。我们可以复现这种智能以及所有相关能力，但真正重要的其实是机器受教的方式。因此，其中一部分在于知道如何得到一个真实可行、经过验证的最终结果。That's where the value came from. like you or any human is an intelligent machine top to bottom, but was it was the way you were taught that actually made you valuable. And I think we're going to start to see something similar in machines. And we can replicate that intelligence and all of that, but but the way you were taught is actually going to be the important thing. And so part of that is actually knowing how to get to an end result that is real, that will work, that is verified.

我其实没怎么受过教育。Naveen，我接受教育的时间很少，不过还是谢谢你。我希望人本身还是有些价值的。不，我完全理解你的意思。想到这种学习的价值时，很多价值都体现在一个人多年从事某件事、随着时间推移积累的数据中。是啊。I wasn't really taught much. I spent very little time in education, Naveen, but thank you. I hope there is some value in the human being. no I get you totally when you think about the value in that learning. A lot of that is in the data that one occurs over time in the years that you're doing something. Yeah.

如今我们看到数据公司——我是说，我们看到几十家公司营收达到天文数字般的水平，而这些水平放在以前简直大得惊人。当然，现在依然惊人，只不过它们在这么短的时间内就达到了。Naveen，老兄，你真是太聪明了。跟你聊天真是件乐事。我就是想问你一大堆关于这个世界的问题。这个市场会变成什么样？它会随着时间推移不断整合吗？还是会越来越细分，而你也因此建立起更强的防御壁垒？会发生什么？我……We see data companies today I mean but we see tens hit astronomical levels of revenue that previously were like astonishingly large. I mean they're still astonishingly large but in such short time. Naveen, dude, you're such a smart dude. It's such a joy talking to you. I just kind of want to ask you lots of questions about the world. What happens to this market? Does it just consolidate over time? Does it unbundle more and more and you build more defensibility? What happens? I

为什么数据公司将变得更加专业化，而非商品化Why Data Companies Will Become More Specialized, Not Commoditized

我认为，坦率地说，它会走向专业化；而且我觉得，如果你专注于某个领域，并真正开始解决重要问题，就能开拓一个规模庞大、难以被竞争者撼动的市场。你知道，这又是一个技能的幂律分布。你可以说，所有人类都能学习许多技能，但有些技能比其他技能更有价值。你构建的数据和课程也是如此，对吧？I think it specializes to be honest and I think if you specialize and actually start to work on on important problems, you have a big market that's defensible. You know, it's it's again it's a power law for skills. You know, you could argue that all humans are capable of learning many skills, but certain skills are more valuable than others. And the same thing would be true with data and curricula that you build, right?

完全理解。所以我们看到数据公司正在走向专业化。你不觉得它们会变成同质化商品吗？Totally get that. And so we see the specialization of data companies. So you don't think they're commodities?

是的，我不认为它会变成大宗商品。我觉得它一开始之所以像大宗商品，是因为它某种程度上只是大量收集观察数据，对吧？但当你开始构建质量更高的数据集时——在 AI 领域，我们会称之为更适合编程、或者更适合思维链之类的高质量数据——数据本身就开始变得很有价值。从某种意义上说，模型几乎只是它所学内容的一种体现。所以我认为，未来数据层会相当重要，我不认为它会沦为大宗商品。Yeah, I don't think it's commodity. I think it started as a commodity because it was sort of like you know at volume gathering of observations right but as you start to construct higher quality data sets that's what we call it in the AI world like higher quality for coding higher quality for chain of thought or whatever you actually start to have a lot of value in the data the model is almost just a manifestation of what it learned in a sense so the data layer I think is going to be pretty important going forward I don't see that commoditizing.

我认为，现在真正的关键在于构建模型，以及懂得如何选择数据。对于大型模型公司来说，问题很大程度上在于数据。我的意思是，他们确实在模型本身上不断取得进展，但大多数能力并非来自模型本身。I think building models and knowing how to choose that data is actually the game right now. The big model companies, it's largely a data problem. I mean, are they they they are making advances on the model itself. But that's that's not been where most of the capabilities come from.

这一直都是通过使用不同类型的损失函数来筛选数据。比如，我们之前有预训练，它采用的是一种非常特定类型的损失函数。之后，我们转向现在所说的中期训练，其中包括强化学习、监督微调，以及各种不同的组合。因此，从预训练到各种不同类型的训练、再到改变模型行为的整个训练课程，才是真正的秘诀。It's been from curating data using different sorts of loss functions. Like we had pre-training before which was a very particular kind of loss. Then we move to what we call mid-training now which is reinforcement learning supervised fine-tuning all these different sort of mixes. So that curriculum how you went from pre-training to all these different sorts of trainings to modify the behavior of the model is actually the secret sauce.

我想问一下，我们看到你提到了预训练、RL，然后显然重心已经转向后训练。随着重心从预训练转向后训练，你认为这是模型改进效率方面一种持久、永久的转变吗？我们会一直这样走下去吗？还是说我们会重新回到以 RL 为主，或者以预训练为主？随着时间推移，这些环节的重要性会如何变化？Can I ask you we've seen you mentioned there kind of pre-training RL and then obviously a real shift has moved to post-training. Yeah. With that shift of emphasis from pre-training to post-training, do you think that is an enduring permanent shift in the efficiency of model improvements and how we get there or will we move back to RL being the primary or pre being the primary? How does that look in terms of importance over time?

为什么强化学习如此重要Why Reinforcement Learning Is So Important

嗯，其实我不确定它是不是更高效。我觉得强化学习的效率更低。如果从每单位算力或每单位美元能让模型提升多少来看，它实际上效率更低，但它确实一直在推动进步，对吧？我认为，我们的目标是让模型在所有重要任务上的表现都变得更好。所以我觉得，在接下来的几年里，它可能会非常重要。我们可能会开始在中间层，甚至在预训练阶段，构建专业化能力。所以很难准确说未来的范式会是什么，因为这些场景其实都非常人为，是我们精心设计出来的，目的是让自己能继续解决下一个问题，对吧？Well, I don't know that it's more efficient actually. I think RL is more inefficient. if you look at it in terms of you know how much better the model gets per unit compute or per unit dollar it's actually more inefficient but it does keep moving forward right I think the idea is that we want to make the models better at all at the tasks that are important and so I think it's it's probably for the next several years yes very important I think we'll start to build specialization in those middle layers maybe even in the pre-training layer so it's hard to say exactly you know what will be the paradigm of the future because these aren't these are these are very contrived scenarios that we just engineer our way to the next problem, right?

这就像是，我在这里，想去那里，于是我设计出某种办法到那里；然后我又在这里，想去这里，于是我设计出某种办法到那里。最后你就会得到这种点的随机游走，而这就是这个行业演进的方式。It's it's like I'm here and I want to get to there so I engineer some way to get there and then I'm here and I want to get to here so I engineer some way to get there and you end up with this kind of random walk of points and that's what that's how the industry evolves.

我觉得模型公司在性能方面并没有考虑得特别长远，比如说“我要在思考 AI 系统的方式上取得某种重大突破，从而获得优势”。我会说，大型实验室如今没那么常这么做了。I don't think the model companies are thinking very long terms in in terms of performance like oh I'm going to make some massive breakthrough on the way I think about a an AI system that's going to give me an advantage. I would say the big labs are not doing that as much.

你觉得，在当前这个生态系统日复一日的竞争中，他们有能力这么做吗？Do you think they can afford to in the daybyday race that is this current ecosystem?

问题就在这里。That's the problem.

是的。Yeah.

这发展真的太快了。我的意思是，你必须紧跟步伐。比如说，如果你的模型晚了一个月，那就是灾难，对吧？如果你是 OpenAI、Anthropic、Grok 或 Google 的话。所以，是啊，他们现在还承担不起那么大的风险。你知道，像埃隆创立 xAI 的时候，你也看到了，他基本上是这么想的：好，这就是现在的范式。我们开始吧，去把这些人招来。然后砰的一下，他很快就达到了前沿水平，因为他有资源。It's really fast. I mean you've got to be on top of it. I mean like if you're one month late on a model that's that's disaster, right? if you're OpenAI or Anthropic or Grock or Google. So, yeah, they they they can't afford to take those big risks just yet. And, you know, you're seeing this like, you know, when Elon spun up XAI, you know, he he basically took, okay, here's the paradigm. Let's go, let's go hire the people. And boom, he got to the frontier pretty fast because he had the resources.

所以，这让我意识到，真正非常深入的创新并不多。你知道，他们只是把各个部分组合起来，交给一支懂得如何打造这些东西的团队，然后迅速执行。我觉得这和试图开创一种全新范式很不一样。你看杨·勒昆的公司，还有李飞飞的公司，他们都在尝试进行更深层次的创新。我觉得这或许就是初创企业领域的情况，而且开始有点像药物研发：所有重大风险都由初创公司承担，而大公司做的都是渐进式的工作。So, what that points out to me was that there wasn't a ton of like very deep innovation. you know it was taking the pieces putting them together with a team that knows how to build them and executing very fast. I think that's very different than trying to come up with a new paradigm and you know you look at like Yanukun's company and you know Few Wei's company they they were trying to do deeper innovations and I think that's what that might be in the realm of startups and it sort starts to look like drug development right like all the big risks are taken by startups and all the incremental stuff is done by big companies.

大公司会在初创公司展现出发展潜力时收购它们。The big companies buy the startups when they show promise.

我有时会问一些简单的问题，但我不知道答案，而你又特别聪明。为什么现在大家都对 RL 这么兴奋？就像你提到的，它某种程度上处于学习过程的中间阶段，而我身边有那么多人……我只是觉得，大家谈的全都是 RL。我就想，哦，我同意。对，我完全同意。I sometimes ask simple questions but I don't know the answers and you're very smart. Why does everyone get so excited by RL today? Like you mentioned it kind of the mid learning there and I just have so many people I just it's all about RL. I'm like oh I agree. Yeah, I agree 100%.

我的意思是，原因有好几个。如果你想想我们学习的方式，就会发现它主要是通过强化学习。如果我要教你一项新技能，比如学会怎么发网球，你知道，这其实就是一个强化学习循环。你知道，我会给你一个示范，让你看看它应该是什么样子：你的发球应该是这样的。也许我会亲自演示，或者给你看一段视频之类的，这样你就能建立一个关于什么是好的、什么是出色的发球的内在模型。I mean there's several reasons. I if you think about the way we learn it's primarily reinforcement learning. If I were to teach you a new skill, you know, learn how to serve a tennis ball or something, it's it's really an RL whoop. you know, you're I would give you a paradigm of what it should look like. Here is how your serve should look and you know, maybe demonstrate that or show you a video or something and it will allow you to build an internal model of what good is or great is.

然后你会不断根据那个内部模型来评判自己，并说：“是的，我正在逐渐接近目标，我在进步”，或者“我没有进步，我需要去求助。”这就是强化学习。我认为它之所以重要，是因为我们不可能收集到足够的数据来解决世界上的每一项任务。我们就是做不到。你不能只靠收集数据，通过预训练来解决这个问题。And then you keep judging yourself against that internal model and say yes, I'm getting there, I'm getting better or I'm not and I need to like ask for help. So that is RL. And I think the reason it's important is we cannot gather enough data to solve every task in the world. We simply can't. You can't pre-train your way out of this by just gathering data.

这实在太难了。坦白说，这些模型的效率还不够高，无法利用它。所以你需要一种范式，让你可以开始说：好，这就是最重要的 10 项任务。然后我会有某种我们称之为“可验证奖励”的东西。我有某种东西，也就是那个内部模型。比如，一个好的发球就是我用力击球，而且球落在界内，对吧？这非常可验证。我可以测量球速，也可以测量球是否在界内。It's just too hard. And the models frankly aren't efficient enough to make use of it. So you need to have a paradigm where you can start to say, okay, these are the 10 most important tasks. And then I have some sort of what we call a verifiable reward. I have some something that's that internal model. Like a good serve is I hit it really hard and it went in, right? That's very it's verifiable. I can measure the speed. I can measure whether it was in or out of bounds.

而当我有了这个可验证的奖励后，我就可以开始判断：我采取的行动是否带来了那个良好的结果？这样一来，我就有了一个能够让我解决任何一组通用任务的框架，这就是为什么……And when I have that verifiable reward, then I can start to say, okay, did the actions I took, did those actions lead to that good outcome or not? So now I have a structure that allows me to solve any set of general tasks and that's why

为什么我们将超越 Transformer模型Why We Will Move Beyond Transformers

强化学习真令人兴奋。你认同 Yan Lun 关于后 Transformer 时代世界的理论吗？这会如何改变一切？RL is so exciting. Do you buy Yan Lun's post transformer world theory and how does that change everything?

我确实认为，我们会进入一个后 Transformer 时代。我不认为 Transformer 是一切问题的终极答案。直到大概几年前，我才真正理解这一点，因为 Transformer 把很多东西混为一谈了。我们真正关心的是推理，也就是能够一步步推演问题。我不知道是谁说的，但大意是：当你没有信息时，你能做什么才体现了智能；而不是当你掌握信息时，你能做什么。所以，能够基于某种机制层面的理解或第一性原理来推演问题，才是智能。I do think we will be in a post transformer world. I don't think the transformer is the be all end all. I don't think I truly understood this until maybe a couple of years ago because transformers conflate a lot of things. what we care about is reasoning. Being able to reason through a problem. I don't know who said it, but it's like intelligence is what you do when you don't have the information, not what you do when you do have the information. So, being able to reason through a problem from some sort of mechanistic or first principles understanding is what intelligence is.

所以，Transformer 算是有那么一点这种能力，但它们实际上是通过这种智能体式的流程来实现的：获取多个样本，然后判断哪个更好。比如，我拿一条执行轨迹，再和另一条进行比较，然后说：好，这条轨迹更好。So, transformers kind of sort of have that, but they really do it through this agent, agentic kind of flow where you're taking multiple samples and you're deciding what's good. So I take one trace and then compare it to another trace and I say okay this trace is better.

所以我会把它称为我的推理轨迹。我们实现这种推理的方式，是通过一种检索机制。注意力机制基本上有这样一个上下文，它本身就是一种检索，然后我们可以从中采样，最终得到类似推理的东西。我觉得这种推理方式极其低效。我们没必要这样构建推理。我们的大脑并不是这样运作的。So I'm going to call that my reasoning trace. And so the way we kind of implement this reasoning thing is through a retrieval mechanism. So attention basically has this this context which is a retrieval thing and then we can we can sample off of that and eventually get to something that looks like reasoning. I think this is extremely inefficient for how we do reasoning. We don't need to build reasoning this way. Our our brains do not work in this way.

我们不会构建一条庞大的轨迹，把它输入另一个模型，然后再生成另一条轨迹。我们的大脑不是这样运作的。因此，有更高效的方法来实现这一点；从算法角度来看，也有性能更高的方法。我们会超越 Transformer。We don't build a huge trace, pipe pump it into the input of another model and then produce another trace. That's not how our brains work. So there are way more efficient ways to do this and way more high performance from an algorithmic st standpoint ways to do this and so we will move past the transformer.

你觉得，成为一名优秀创始人所需要具备的条件变了没有？我不觉得变了。我认为基本原则还是一样的。我觉得，对未来要有自己的看法——也就是你的愿景——现在可能比以前更重要。你知道，在 AI 之前的世界里，只要执行力够强，就能成为一名相当不错的创始人。你不需要有多么出色的愿景，只要知道如何执行就行。你知道，你只管出去执行，做一家大公司也没那么难，因为大公司行动迟缓，步履蹒跚。Do you think what it takes to be a great founder has changed or not? I don't think it's changed. I think the basic principles are still the same. I think having a viewpoint on the future, the your vision actually may matter more now than it did before. You know, I think in the in the preAI world, you could be a pretty good founder by just being able to execute. You don't have to have a very good vision. You just have to know how to execute. And you know, you just go out execute a big company, which isn't that hard because big companies are slow and you know, lumbering.

我觉得你确实可以创办一家初创公司，然后把它卖掉，或者以此为基础发展成一家公司。我觉得如今如果你只靠执行力，是走不了多远的。I think you could actually build a startup and you could you could sell it or or build a company out of it. I think these days if you're just based on execution I don't think you're going to get very far.

我非常赞同。能问你一个问题吗？我们之前聊过不同层次，你说你认为数据不会被商品化。我觉得专业化是有价值的。I agree incredibly. Can I ask you we spoke about the different layers and you said I don't think it's going to get commoditized in data. I think specialization I think there is value there.

AI 技术栈中哪些部分实际上会变成商品化服务？Which Parts of the AI Stack Actually Get Commoditized?

随着时间推移，哪一层会逐渐商品化？Which layer does get commoditized over time?

是的，这是个好问题。我的意思是，可以说能源已经是大宗商品了，对吧？所以它已经商品化了。我认为，一定程度的智能确实也开始变得商品化。说实话，我们已经看到了这种趋势。比如说，已经发布六个月的模型，像中国的开源模型，你知道，就是那个水平的模型，已经相当实用了，对吧？发布六个月的模型依然相当不错。所以我认为，这就是商品化的一个标志，对吧？Yeah, it's a good question. I mean arguably energy is already a commodity, right? So this is already commoditized. I think a certain amount of intelligence does actually start to become commoditized. We're already seeing it to be honest. Like models that are 6 months old, let's say the Chinese open models, you know, that level, they're pretty useful, right? Something that's 6 months old is still pretty darn good. so I would argue that's that's a hallmark of commoditization, right?

前沿并不是唯一重要的东西。对于某些问题来说，它确实重要，但六个月后，前沿模型就会出现在开放模型中。那么，帮我想想：如果交付机制或所有基础设施的产出都变成了商品化服务，会发生什么？The frontier is not the only thing that matters. For certain problems, it is, but in 6 months that frontier will be in the open models. So then help me out. What happens if the delivery mechanism or the output of all the infrastructure is then commoditized?

嗯，我觉得商品化并不一定意味着坏事。它只是意味着利润率确实会大幅下降，但同时也意味着市场规模非常大。所以你可以说番茄已经商品化了，但我们还是会吃番茄。我们仍然需要它们，你知道吧。因此，智能领域会有一个巨大的市场，而且市场还会继续增长，只是利润率可能不会很高。其实我不认为模型公司会是整个技术栈中掌握利润率的那一方。事实上，今天也不是这样。比如说，现在掌握整个技术栈利润率的是英伟达，对吧？Well, I think commoditization doesn't necessarily mean it's bad. It just means that margin yeah falls quite a lot, but it also means that the market is very big. So you could argue that tomatoes are commoditized, but we still eat them. We still need them, you know. So we will have a huge market for intelligence and the market will continue to grow, but the margins may not be huge. I actually don't think the model companies will be the ones who own the margin of the whole stack. It's not even true today. Like Nvidia owns the margin of the stack today, right?

推理会成为一种商品吗？Will Inference Become a Commodity?

我想说，他们可能是拿走利润空间中最大份额的人。I would say they're probably the one who take a lion share of the mar of the margin.

很多人告诉我，未来几年推理成本实际上会大幅下降，我们可能会看到推理市场走向商品化。A lot of people say to me the cost of inference will actually go massively down over the next few years and we'll see the commoditization potentially of the inference market.

你会同意这一点，还是不同意这一点？不，不同意这一点？Would you agree with that or not not agree with that?

好吧，我们来拆解一下。推理市场商品化是什么意思？这意味着，负责运行推理服务的运营商——我想应该是这样。对。Well, let's break it down. What does what does commoditization of the inference market mean? What that means is that the operator who's who's running inference I assume. Yeah.

是的，是的，当然。你们的烟花、你们的模型、你们的基础晒黑效果，以及它们提供的服务。Yeah. Yeah. Absolutely. Your fireworks, your model, your base tans and the services that they provide.

是的。Yeah.

站起来。Stand up.

因为如果你看看他们的价值主张，他们做的事情，而且做得非常出色，就是拿一个现成的模型，比如说开源模型，然后让它运行得很好，提供正常运行时间保障，并且让它在基础设施上的运行成本相对较低。所以他们走的是基础设施路线，对吧？基本上就是把模型映射到物理硬件上，并且把这件事做好。我认为未来这里会有价值，但推理的价格，也就是人们愿意为每个 token 支付的费用，会下降。Because if you look at their value prop, what they've done and they've done very well is take a model that exists an open model say and then make it work well, you know, with uptime guarantees and make it, you know, relatively costefficient to run on infrastructure. So they're an infra play, right? They're basically mapping the model onto physical hardware and doing that well. Now, I think there will be value there in the future, but the price of inference, like what people are willing to pay per token will drop.

所以基本上，他们得更充分地利用硬件的能力，在交付每个 token 的过程中挤出更多利润。而我认为，这一层能榨出的油水会减少。所以我觉得，这会导致他们的利润率降低，而且模型本身也会变成一种商品。And so they're basically they're they have to squeeze more margin out of the hardware's capabilities to the delivery of that token. And that, you know, the amount of juice in that layer, I think, will decrease. So I think it'll result in them having lower margins and I think the model itself like is commoditized.

现在有好几种不同的方式可以获得它。但硬件这一侧可能不会、不会崩塌。我认为这部分仍然会有优势，因为你还可以从中榨出更多价值，比如说，嘿，我能更高效地把能源转化为智能。在我看来，相比基础设施层，我们在这方面还有更多、很多价值可挖。所以我、我确实认为，未来10年硬件才是值得布局的领域。You have several different ways to get it now. But the hardware side of it probably won't won't collapse. I think that part of it is still going to have an advantage because you can still squeeze more juice out of like hey I convert energy to intelligence more efficiently. The we have more a lot more juice to squeeze there versus the infrastructure layer in my view. So I I actually think the hardware is the place to be over the next 10 years.

我需要改变自己的投资者定位吗？我知道这听起来很奇怪，Naveen，但我一直都是软件领域的投资者。过去 10 年里，我一直在投资 SAS，而且做得不错。Do I need to change who I am as an investor? I know that sounds strange, Naveen, but like I'm I'm a software investor. I've invested in SAS for the last 10 years. I've done it well.

是啊，Yeah,

哥们。做硬件投资真的挺[ __ ]难的。我……这是一种非常不同的思维方式。这一代投资者具备成为硬件投资者的条件吗？有什么建议？dude. It's quite [ __ ] hard doing hardware investing. I It's a very different mindset. Is this generation of investors equipped to be hardware investors? advice.

为什么纯软件投资者“彻底完蛋了”Why Pure Software Investors Are “F***ed”

我会说，如果你是纯粹的软件投资者，那你现在算是有点[ __ ]，说实话。我这么说相当有把握，因为我不认为 SaaS 还能成为你提升利润率的来源。从 2010、2011 年左右到现在，我们一直处于一种兴奋过头的状态：你知道，我可以很快在 AWS 云服务上搭建出一个东西，迅速测试市场、尝试各种方案，然后去争取一个规模达 1 亿或 2 亿美元的市场。SaaS 让市场出现了大规模碎片化，因为交付方式变得非常简单、没有阻碍。如今，我们进入了这样一个阶段：应用开始更多地围绕智能层展开，而我实在不觉得 20 年后软件还会是一个巨大的市场。I would say if you're a pure software investor, you're kind of [ __ ] honestly. And I'll say that pretty confidently because SAS I don't think is going to be where you gain margin anymore. And we had this this sugar high for the since what 2011 2010 time frame till now of you know I can very quickly spin something up on an AWS cloud service test the market quickly try things out and then go after a h 100 million or $200 million market. SAS was like massive fragmentation of the market because the delivery mechanism became very easy and frictionless. now we're in this place where applications start to become more about the intelligence layer and I just don't see software being a huge market in 20 years.

你难道不觉得、不觉得 SAS 会变成 Lorra 吗？我的意思是，Lora 算是 SAS 吗？嗯，我是说，有点算是面向企业的 SAS，但又不完全是。不是。所以它的语义难道不会改变吗？Do you not do you not just think SAS becomes Lorra? And what I mean by that is like is Lora SAS? Well, I mean kind of yeah enterprise SAS but kind of not. No. So does like the semantics not just change?

为什么 20 年后 SaaS 的重要性将大幅降低Why SaaS Will Matter Far Less in 20 Years

我觉得没人真的会从头开始就说：“哦，我们是家 SaaS 公司，对吧？”是的。我……所以，没错。我觉得你说得对。所以，我想我当时采用的是 SaaS 的经典定义吧。I don't think anyone really is like starting fresh being like, "Oh, we're we're a SAS company, right?" Yeah. I So, exactly. I think you're right. So, I was taking the classic definition of SAS, I suppose.

但我认为，有些公司交付的是一种成果，并通过软件与服务相结合的方式来实现这一点。没错。你知道，我们在好几家公司都看到了这种情况。我也相当积极地投资初创公司，通常是种子轮阶段的公司。我看到一批这样的公司正在涌现：它们把一些存在已久、过去一直以传统方式完成且必须准确无误的工程任务，借助 AI 加上人工协作，做得快得多、便宜得多。but I think companies where you deliver an outcome and you do so through a combination of software and services. Absolutely. And you know we're seeing that in in several companies and I invest pretty actively in startups generate the seed stage and you know I'm seeing a crop of companies come where they're taking engineering tasks that have existed for a long time that were done in a traditional way and have to be done correctly and they're making those tasks much more much faster much cheaper by using AI plus humans.

这很棒。我是说，我会说，Harvey 是法律事务所领域这类做法的第一个例子。你知道，过去的做法是，我有一批律师，他们会去审阅合同、起草合同之类的东西。然后他们基本上说：“嗯，我可以用人工并核验——不，抱歉，是用 AI，再由人工核验输出结果，用更少的人完成同样的工作，而且交付得更快，质量相同或接近相同，这样我就能颠覆这个市场。”It's great. And that's I mean I would say Harvey was a first example of this for law firms where you know you the old way was I have you know a set of lawyers who go and look at contracts and you know write contracts that kind of thing and they basically said well I can use humans and verifi or sorry use AI and verify the outputs with humans and deliver the same thing with fewer humans and deliver it faster at the same quality or near the same quality and I will disrupt that market.

所以我觉得我们会看到，这有点像新的 SA。我觉得你知道，我把它叫作“文字手淫”——就是风投们发明一套关于世界的说辞，比如那个“万亿美元服务业”的说法。你知道，我觉得那篇文章写得很棒，我也很喜欢写它的朱利安。非常好。但我们是不是只是在稍微包装一下？这不就是未来吗？人们用 AI 更高效地做事，结果把整件事都美化得天花乱坠。有点、有点过头了，不是吗？So I think we're going to see this is sort of the new SA. And I we see you know I call it word wank which is where like you know VCs come up with a coin theory of the world and it's like you know the trillion dollar services thing you know I think did a brilliant piece on and I love Julian who did it. Very good. But like are we not just slightly jazzing it up? That's just the future. People using AI to do things efficiently this like whole glamorization of it. Bit bit overkill, isn't it?

我，我觉得你说得对。I I think you're right.

你不就是，你不就是在投资一家律师事务所吗？Aren't you just Aren't you just investing in a law firm when you do that?

说到底，没错，但你做的不过是一个更好的律师事务所，对吧？就这样。可你也会愿意这么做，因为我遇到很多公司都在做你说的那种事，我就会想：我不想开律师事务所。我很乐意做 Lorra 或 Harvey 这样的公司，向律师事务所销售产品，但它们本身并不是律师事务所。你懂我的意思吧？Effectively, yeah, but you're just it's a better version of a law firm, right? That's all. And you but but you'd be okay to do that cuz I meet a lot of companies doing exactly what you say and I'm like I don't want to do a law firm. Like I'm happy doing a Lorra or a Harvey which sell to law firms but they're not actually a law firm. You get me?

不过，我们看到一些大型律师事务所现在正直接投资 AI。所以我认为，我认为在一定范围内，你基本会看到 Harvey 把产品卖给律师事务所，因为律师事务所有渠道。Well, but we are seeing that like some of the large law firms are now investing in AI directly. So I think I think in its limits you'll see basically Harvey sells to the law firms because the law firms have the channel.

你知道，这个渠道能长久吗？我不知道。不过我也能想象他们会说：“嗯，你知道吗？我们完全可以直接卖给企业。”就我个人而言，如果 Harvey 跟我说：“嘿，我有个解决方案，能处理你所有的保密协议，还能审查你的合同，诸如此类。我可以让你确信它确实有效。”那我会把 80% 的工作量，也就是我的法律事务工作量，交给他们处理。you know is that channel going to be enduring? I don't know. But I could also see them saying, "Well, you know what? We could just sell directly to companies." Like, personally, I would buy that if Harvey said like, "Hey, I have a solution that does all your NDAs. It does it reviews your contracts, blah, like all these things, and I can gain you know, confidence that it works. I would send 80% of my workloads, my legal workloads to them."

我很感兴趣，因为你谈到了一些正在打造自有模型的公司。我想你也看到，法律领域的 K& 承诺为此投入了大约 5 亿美元。你知道，Fireworks 的投资人 Lynn Roman 曾非常坦率地谈到，拥有自己的智能，以及企业利用自身数据打造自有模型的重要性。你认为这会是一个非常可行的未来，还是说这其实是一种略带技术乐观主义色彩、且立足于非常特定视角的预测？I am intrigued because you spoke about some firms that are building their own. And I think you saw K& in the legal space commit like $500 million to it. you know, Lynn at Fireworks Roman investor spoken very openly about owning your own intelligence and the importance of companies leveraging their own data to build their own models. Do you think that is a very viable future or or actually a slightly techno optimist kind of prediction from a very specific standpoint?

我觉得这有点过于技术乐观了，因为我认为你得达到一定的规模，才能构建并拥有自己的智能。就拿律师事务所来说吧。如果你是 Cooley 或其他大型律所之一，没错，你可以投资这些东西，也确实可以把它做成。I think it's a bit techno optimist because I think you have to have a certain number a certain amount of scale in order to build your own intelligence and own it. Keep going with law firms for instance. If you're Kulie or one of these big firms, yeah, you can invest in these things and you can you can make it happen.

如果你是在内布拉斯加州中部经营的一家夫妻店，那你就做不到这一点。所以你会使用某种服务，对吧？你可以借助这项服务来增强律师团队的能力，这样就能为客户提供服务了。所以我不认为每家公司都会利用自己的历史数据来改进自有模型。他们不会这么做。这对他们来说太难了。他们想要的是开箱即用的解决方案。if you're a mom and pop shop in the middle of Nebraska, you can't do that. So what you're going to use then is a service of some sort, right? you may augment your attorneys with that service and that's how you'll you'll provide service to your clients. So I I don't see this as every company will use AI in a way of leveraging their past data to make their own models better. They're not going to do that. It's just it's too hard for them. They want an out-of-the-box solution.

所以我认为，销售那些开箱即用的解决方案是有市场的。但这会是一个长尾市场。大型玩家会开始定制，因为他们的规模足以让定制变得合理。这是有道理的。这又回到了这个观点：你知道，我们世界的每个部分都是对抗性的。So I think there is a market for selling those out of the box solutions. but it's it's going to be this long tale. the big the big players will start to customize because they have the scale to justify it. It makes sense. And it goes back to this idea that every you know every part of our world is adversarial.

这取决于你在和谁竞争，对吧？如果你的对手是另一家大型律师事务所，而对方利用 AI 大幅提高利润率，那你多少也得这么做。小型家庭式事务所可能不会这么做，他们的竞争对手，你知道，也不会这么做。所以有一段时间，他们会继续留在另一个领域。Like it depends on who you're fighting, right? If you're fighting another big law firm and that big law firm goes and uses AI to supercharge their margins, well, you kind of have to do the same. the mom and pop shop probably isn't doing that and the people they're, you know, fighting against are not doing that. so they're going to stay in this other other realm for some time.

我认为，法律和编程领域的许多变化，都有一个共同基础：我们会把薪资支出的多少比例转向技术支出。你看，如今美国法律市场规模大约是 3000 亿美元，那么其中有多少会被这些服务提供商抢走？我认为，编程领域也面临同样的问题。你听过 Benioff 说，他们会从 60 亿美元的开发人员薪资预算中拿出 3 亿美元，大幅投入 Claude Code 和编程工具。I think one thing that underpins a lot of law coding is the percent of salary that we will then transition to technology spend. You know when we look at lawyers today you know I think it's about a $300 billion legal market in the US and you're looking at that going well how much of that will be eaten by these providers. I think you see that same question there with coding. You saw Ben off say that they'll spend $300 million of a $6 billion developer salary budget on claw code and coding tools significantly.

我想问你的是，对开发者来说，这个比例会维持在 5% 吗？会涨到 30%、50% 吗？还是会降到 1%？My question to you is will that stay at 5% for developers? Will that go up to 30%, 50%. Or will it come down to 1%.

为什么人工智能未必会毁掉开发者的工作岗位Why AI Will Not Necessarily Destroy Developer Jobs

因为这才是核心中的核心问题。我其实觉得这些比例可能会保持相对稳定，因为最后我们会遇到验证问题。说起来确实很难，因为真正难的部分，是训练人们判断什么是好的。开发者嘛，重要的其实不是你今天直接产出了多少东西，不是代码行数，而是代码的质量。那些经验非常丰富、水平顶尖的开发者，也就是所谓的 10 倍开发者，他们培养出了判断什么是好、什么看起来不错、什么在审美上恰到好处的能力，而且他们能很快把事情做成。Because that is the mother lode question. I actually think the percentages will remain maybe sort of stable because we end up with this verification problem. So it actually it's hard because training humans to know what's good is actually the hard part. So developers, you know, it's actually not about what your direct output is on this day. It's not like lines of code. It's actually like quality of code that matters. So very experienced, very top-notch developers, those 10x developers, whatever. you know, they developed a sense of what is good, what looks good, what's aesthetically right, and they can they can make things work fast.

在这个新世界里，你知道，那些程序员，那些懂得什么才是好的人，实际上能够验证模型的输出，从而变得更高效。所以我认为，结果就是我们只会产出更多。假设我们为人类加上 AI 付费，市场规模不会有太大变化，只是他们的产出会大幅提升。In this new world, you know, those coders, those people who know what's good can actually verify the output of models and become more efficient. So I think what happens is we just produce more and the market size let's say that we're paying for humans plus AI doesn't change drastically just their output greatly changes.

我之所以这么看，是因为我们可以稍微推演一下。假设我每年花 1000 万美元开发软件。现在，软件客户突然开始提出各种不同的需求。你知道，10 年前，他们还觉得一年发布一次就挺满意了。现在他们却说：“嘿，我希望我的问题能在一周内修好。”而要做到这一点，唯一的办法就是让 AI 和人类协同工作。所以我做的并没有减少，实际上反而多了很多。That's the way I look at it because let's let's play it out a little bit. So I'm spending in a company let's say $10 million a year to develop my software. now all of a sudden my customers of that software start having different demands. you know, 10 years ago they're like happy with one year release cycles. Now they're saying, "Hey, I want my bug fixed in a week." And so the only way to do that is actually to use AI with the humans. So I'm not doing less, I'm actually doing way more.

我现在要花一周修复那些 bug，最终能在一小时内修好，但我仍然要投入 1000 万美元来继续开发整个项目。我还需要有人来验证这些输出，并真正把精力转向有用的目标，比如修复 bug、添加功能。所以我认为，交付速度会加快，因为需求端开始提出更多需求。I'm fixing those bugs in a week and eventually fixing them within an hour, but I'm still paying $10 million to keep developing this whole thing. And I need the humans that can verify those outputs and actually move to a goal that's that's useful, fixing the bug, adding features. So I think the the pace of delivery goes up because the demand side starts to demand it.

我觉得这里被忽略的关键是，并不是因为同样的任务可以由 AI 加上人类以更低的成本完成，就意味着所有人的工作都会消失。现在，没错，这确实如此，但我接下来会做多得多的这类任务，因为需求增加了。这两者是相辅相成的。供给成本降低时，需求也会随之改变。所以我觉得最终会发生的就是，我们做的事情会更多。And I think that is the key thing here that's being left out is that it's not that everyone's job goes away because the same task could be done for less with AI plus humans. It's that now I yes that is true but I will I will now do way more of those tasks because the demand side is increased. You don't get one without the other. When the supply gets cheaper you get it changes the demand side as well. So I think that's what's going to happen is that we just end up doing more.

我们的发布周期更多，变更也更多。比如，这里有个实体例子，我觉得大家比较容易理解。假设我们今天有一套建造建筑物的流程。要是我想建一栋摩天大楼，假设需要两年的施工时间，而这取决于参与建造的人数，以及他们具备的技能组合。We have more release cycles, more changes. Like here's a physical example I think that people can get their heads around. You know, let's say today we have a rate of building a building. If I want to build a skyscraper, it takes two years of construction, say, and that's based upon x number of humans with, you know, y sets of skills.

假设突然间，我有了能完成其中 80% 工作的机器人。会发生什么？好吧，短期内体力劳动可能会受到冲击，但接下来你会说：“你知道吗，突然间，建造同一栋摩天大楼所需的成本和时间都变了。原来要两年，现在 80% 的工作都由机器人完成，我可以把整个工期缩短到 8 个月。”所以，并不是说你只会建造同样数量的摩天大楼，只不过把工期缩短到 8 个月。So let's say all of a sudden I have robots that can do 80% of that. What's going to happen? Well, okay, there might be a disruption in physical labor for a moment, but then you'll say, you know what, all of a sudden, my cost and time to build that same skyscraper, it was two years, but now 80% of those tasks are done by robots. I can bring the whole thing down to 8 months. So, it's not that you're just going to build the same number of skyscrapers and just do it in 8 months.

你们要建造的摩天大楼数量会翻一番，对吧？市场上会有更多活动，更多建筑被拆除，也会有更多劳动力投入这个领域。因此，项目总数会增加。投入这些项目的资金总额可能不变，但建造这些项目的将是人类和机器人共同组成的团队。You're going to build twice as many skyscrapers, right? there's going to be more things churning, more buildings being torn down, more more labor going into that market. So, the total number of projects will go up. The amount of dollars going into all of those projects may be the same, but it'll be a combination of humans and robots to build them.

Naveen，你说你显然会做天使投资，而且做得非常成功。你有点轻描淡写了。我现在投资时，一个与众不同的看法是：我不喜欢机器人领域的投资。我觉得我们离目标还很远，而且我确实认为时机很重要。我不是在说为机器人提供动力的基础设施。我觉得围绕机器人构建的数据集非常有意思。我们看到很多人，比如 Mecca 或 microagg，都在为此建设基础设施。Naveen, you said that you obviously angel invest, you angel invest very successfully. You kind of downplayed it. I think where I invest today a contrarian thought for me is I do not like robotics investments. I think we are so far from where we need to be and I do think that timing is important. Now I'm not saying the infrastructure to power robotics. I think the data sets around them are very interesting. We see a lot of people like Mecca or microagg infrastructure buildout for it.

但就实际应用层面来说，比如扫地机器人、建筑机器人，我们离理想状态还差得很远。我不想涉足那个领域，也不想承担时机上的风险。But like the actual like application layer where we see like robot cleaners, robot builders, we are so far away from where we need to be. I don't want to be there and I don't want to take the timing risk.

是我错了吗？是我太天真了吗？你比我聪明多了。Am I wrong? Am I naive? You're much smarter than me.

机器人何时才能真正成为一个成熟市场？When Will Robotics Actually Become a Real Market?

嗯，再说一次，我在这件事上的立场相当坚定，所以我会强烈反对。呃，好吧。你说我们还差得很远，具体来说是时间上还要多久？你觉得时间线会是怎样？Well, again, I'm pretty pilledled on this one, so I am going to disagree strongly on it. U Okay. When you say we're so far away, put that in put that in terms of time. What do you what do you think is the timeline?

五年多。five plus years.

好的。我们在这件事上的分歧没那么大。我认为，未来两到三年内，我们就能为其中一些任务开发出可运行的原型，而市场会在五到七年内出现。我觉得这个判断是对的。Okay. We're not so far apart on this. I think we'll have working prototypes for some of these tasks in the next 2 to 3 years and the market will emerge in 5 to seven years. I think that's right.

我觉得人形和非人形之间也有一道很明显的界线。没错。比如说，非人形这边，我绝对能看到一条更清晰的路径。至于人形……我，我真的只是……每当我觉得某件事特别明显时，我总会想：我肯定错得很离谱，因为人形显然还差得远。I think there's also a clear line of delineation between like humanoids and non-humanoids. Correct. Like non-humanoids I absolutely see that clearer pathway. the humanoids. I'm I'm really just str whenever I think something's really obvious, I'm always like, I must be very wrong because the humanoids is so obviously so far off.

你知道，我觉得我们将身处一个奇怪的世界，发展周期呈指数级增长。要套用我们旧有的模式真的很难，这就是问题所在。you know, I think we're going to be in this weird world where we live in exponential development cycles. It's really hard to apply our old models. That's the problem.

确实如此。所以，今天看起来还遥不可及的事情，沿着指数增长曲线发展，最终会比你想象的快得多。It is. So something that looks like so far away today with an exponential curve ends up being much faster than you thought

为什么推理将成为一个10万亿美元的市场Why Inference Will Be a $10TRN Market

还有像 TAM（总潜在市场）这类东西，比如市场规模是什么。我现在知道、但希望自己 24 个月前就知道的一点是：推理市场大得无边无际，简直没有上限，规模可达数万亿美元。10 年前，如果你问我哪些市场能达到万亿美元规模，我会说那是基本需求，比如住房、食品和交通，对吧？你绝不会想到科技领域会出现一个万亿美元规模的市场。但我认为，光是 AI 模型本身，未来几年就会成为一个万亿美元规模的市场，而推理市场将在 10 年内达到 10 万亿美元。and like tams like what you know a market size is you know like I think the inference market what do I know now that I wish I'd known 24 months ago that the inference market was just like unbounded it's unbounded it's multi- trillions like if you asked me 10 years ago you know what were trillion dollar markets it's basic needs it's housing food transportation right you would never have thought something in tech would be a trillion dollar market and I think just AI models themselves will be a trillion dollar market in the next few years and the inference market is going to be a $10 trillion market within 10 years.

所以，想想这件事简直不可思议，对吧？正如你所说，我们正在把劳动力市场中的一部分，以及价值创造链条中的一部分，转移到技术上。这就是我们能看到如此庞大的 TAM（总潜在市场）的原因。所以，你知道，我看待非常规领域的方式是：我认为这是一个绝佳的市场，因为如果我们能把 token 的交付效率提高一个数量级，也就是提升十倍，那么从某种意义上说，这实际上就把市场规模扩大了十倍，对吧？So this is insane to think about right and to your point we have we're moving some of that labor market and part of the de of the of the value generation stack to technology. So that's where we're getting these giant TAMs. So, you know, the way I look at like unconventional, the reason that I think it's a it's a fantastic market to be in is that if we knock off a zero, one order of magnitude efficiency on delivery of a token, that effectively 10xes the market in some ways, right?

为什么机器人技术的规模将是编程智能体的 100 倍Why Robotics Will Be 100x Bigger Than Coding Agents

所以，原本这是一个万亿美元的市场，现在突然间，它的适用范围大大扩大了。潜在用途的基础规模增长了十倍以上。这就是杰文斯悖论。所以我认为，我们正处在这样一个阶段：可以打造这些指数级增长的事物，从而释放巨大的价值并开拓庞大的市场。因此，我确实认为，机器人产业的规模将是编程产业的 100 倍。So, something that was a trillion dollar market, now all of a sudden you have way more applicability. The base of potential uses just increases more than 10x. This is Jeban's paradox. So I think we're in this place where we can build these exponentials and just unlocks massive value and market. So I actually think robotics is going to be 100 times larger than a coding.

哇。嗯，这跟我的看法稍微有点不一样。Naveen，在我们开始快速问答之前，能不能先聊聊这个？我不想显得特别愚蠢。你为什么这么说？我觉得我们对如何观察智能的成果，还只是刚刚触及表面。对智能来说，物理世界是我们完全还没有涉足的领域。比如我会说，迄今为止的机器人技术，像工业机器人，其实算不上真正智能。Whoa. Well, that is slightly disagreeing with me. Naveen, can we just like before before we do a quick fire, can we just unpack that because I don't want to be completely moronic. Why do you say that? I I think we're still scratching the surface of how to how to see the output of intelligence. So the physical world is something we've just not been in at all for intelligence. Like I would argue robotics to date like industrial robotics are not really intelligent.

他们写道，我们正在把我们拥有的智能赋予它们，比如教它们如何完成一项任务，而我们只是在照着做。写道，这并不算真正新颖的问题解决方式。在不久的将来，我们就会进入解决物理问题的领域。想象一下这样一个世界：我有数十亿个机器人四处活动，它们能动态地解决问题。比如，那就是个问题。那个东西刚坏了。我要去把它收拾干净。我要……我要把它换掉。They're wrote we're imbuing them with intelligence that we have like how to do a task and we're just doing it wrote. That's not really novel sol problem solving. we will be in the realm of physical problem solving not in the not distant future. Like imagine a world where I have billions of robots roaming around and they actually solve problems dynamically. like that's a problem. That thing just broke. I'm going to go clean it up. I'm going to I'm going to replace it.

就像在我们的世界里，清理所有东西、建造新建筑——你知道，解决现实问题的成本不会降到零一样。Like the cost of delivery of cleaning everything up in our world, building new buildings, you know, solving physical problems will will not drop to zero.

我讨厌那种声称会跌至零的说法。它会大幅缩减。所以我认为那个市场会非常庞大。这就是为什么我认为，虚拟世界最终会比我们今天看到的规模大得多。I hate those drop to zero kind of statements. It will greatly reduce. So, I think that market will be enormous. And that's why I think it's just going to be much bigger than what we see today in the virtual world.

Naveen 目前在机器人领域的投资方向Where Naveen Is Investing in Robotics Today

我能直说吗？那你具体会投资机器人领域的哪些部分？比如为机器人提供动力的模型公司、为模型公司提供动力的数据公司，还是消费应用层？我觉得现在还没到应用层。我更关注技术栈。说实话，我觉得这些做模型的公司还没真正想清楚，等它们发展成熟后会是什么样子。这没关系。我不是说它们必须……比如说，去构建……我不认为价值会在模型里。Can I be blunt? Where specifically then do you invest in robotics? Like in the model companies that power them, the data companies that power the model companies, the the consumer application layer where I don't think the application layer yet. I'm more in the technology stack. So I actually think these guys who are doing the models haven't really figured out what they're going to look like when they grow up. and that's okay. I'm not suggesting they need to like building I don't think the value will be in the model.

我正在投资那些开发模型的公司。我认为它们最终很可能会打造端到端的设备。它们会造出配备模型的机器人，让它特别擅长做某件事。它们会以此来销售产品。我的个人看法是，这些东西会结合在一起。I am investing in companies that are building models. I think they are probably going to end up building end-to-end devices. They're going to build a robot with a model that does something very, very well. That's how they're going to sell it. My personal take is that it's going to go together.

你们不会分居的。You're not going to have a separation.

如果 TAM 确实如你所说的那样，你还在乎价格吗？Do you care about price when the TAM is what you said it was?

不太是。不是。我只是觉得，呃，我需要把我财富中的一定比例投到这个领域，因为它很重要。而且这里，这里将会有大量增长。所以挺有意思的，因为我们现在处在这样一个阶段：你知道，价值投资者，或者你想怎么称呼都行，比如传统风投，会说：“哦，你知道，这价格也太高了吧。”各位，当我们谈论的是规模超过万亿美元的可寻址市场（TAM）时，价格几乎无关紧要。你知道吧，Not really. No. I I just see it as like I need to put I need to put some percentage of, you know, my wealth into that space because it's important. And that's where that's where there's going to be a ton of growth. And so it's interesting because we're in this place where you know value investors or whatever you want to call it like traditional VCs are like oh you know that's way overpriced. Like guys when we're talking about trillion plus dollar TAMs the price is almost irrelevant. you know,

你知道吗，我完全同意你。难处在于，我们往往也有客户，或者说老板，而他们就是 LP。要是你认为持股情况和价格非常重要，那 LP 甚至还是一个滞后指标。你到了年度股东大会上说：“嘿，持股情况变差了，价格涨了。”他们就会说：“[ __ ]你。”直到他们开始看到大型公司也出现同样的趋势，才会想：“啊，好吧，也许这就是发展方向。”对吧？如果你能把所有的钱都投进一家不是你自己的公司，你会选哪家？do you know I completely agree with you. The challenge is very often we have a customer too or a boss too and it's LPs and the LPs are an even lagging indicator if you think that ownership and price are very important and when you rock up to your AGM and you say hey ownership's gone down and price has gone up they go [ __ ] you. They do until they start to see the big firms also have the same trend and they're like ah okay maybe maybe that's where the way it's going right if you could put all of your money into one company that is not your own what would it be

是的，可能就是像物理智能这样的东西，你知道，他们把它当作一种模型来思考。我觉得也有好几家公司在做这件事，比如我觉得 ROA 和其他几家公司也在从这个角度思考。我认为在我看来，他们正在解决一个关于如何定义任务并执行任务的根本问题，也就是把任务转换成运动输出。对我来说，这就是当下物理世界需要解决的问题。所以，除了我们正在做的、打造一台新计算机之外，我觉得这就是下一个重大问题。yeah probably like a physical intelligence you know they they are thinking about as a model I think there are several companies doing that as well like I think roa and a few others are doing are thinking about it like this I think they are solving in my view something fundamental about how you define a task and then execute it so conversion of a task to motor output. To me, that's that's the problem to solve right now for the physical world. So, outside of what we're doing and building a new computer, I think that is that's the next big problem.

你在各家公司都取得了非凡的成功。关于拥有财富这件事，有什么是没人告诉你、但你希望当初有人告诉你的？You've been incredibly successful across companies, what does no one tell you about having money that you wish you had been told?

我觉得这不会改变你的价值观。你的资源依然有限，但与此同时，你还是会以某种方式看重某些东西。比如在某些事情上，我还是挺抠门的，其实没必要这样。I think it's it doesn't change your values. You still have a finite amount of resources, but also at the same time, you still value certain things in certain ways. Like I'm still kind of cheap when it comes to certain things. It there's no reason for it to be.

你在哪方面比较吝啬？What are you cheap with regards to?

嗯，要是你问我妻子的话，你知道，可能就是衣服之类的东西。你知道，我就会想：什么？我才不愿意花那么多钱买一条牛仔裤呢。well, if you ask my wife, you know, probably clothes, that kind of a thing, you know, I just like what? I'm not going to pay that for this for a set of jeans.

我觉得旅行这件事还是……你知道，我现在到了一个会尽量优化时间的阶段，但总的来说，我就会想，啊，这也太贵了。我就不做了。不是说我花不花这笔钱有什么区别。我只是，我只是看不到它的价值。所以我觉得问题就在这儿。我会把钱花在我认为有价值的事情上，而且在那些事情上我可以多花很多钱；但对于那些我看不到价值的事情，我确实不会再多花钱。I think travel still like I you know, I'm getting to a point where I will try to optimize my time, but overall I'm like ah that costs too much. I'm just not going to do it. It's not a you know, it wouldn't have mattered if I spent the money or not. I just I just don't see value. So, I think that's the thing. I I I put money into things where I see value and I can put a lot more money into those things, but the places where I don't, I actually don't spend anymore.

哥们，我们来一轮快速问答。我说一个简短的话题，你立刻说说你的想法。过去12个月里，有什么事让你改变了看法？Dude, we're going to do a quick fire around. So, I say a short statement, you give me your immediate thoughts. What have you changed your mind on in the last 12 months?

为什么传统风投指标不再适用于深科技Why Traditional VC Metrics No Longer Work for Deep Tech

是的，我觉得在投资方面，我对衡量进展的有用指标改变了看法。我们过去总是觉得，哦，我们需要把市场验证作为主要指标来看待，但我认为这是一种过时的看法。事实上，我觉得这种观念很大程度上源自 SAS。因为在没有深层技术路线的情况下，市场验证是有道理的。比如说，你要是说：“嗯，我正在开发一款能满足某种需求的产品。”那一切就都取决于市场验证。Yeah, I think in terms of investing, I've changed my mind on on useful metrics for progress. We always looked at it as like, oh, we need to look at market validation as as the main metric and I think that's an old way of looking at it. In fact, I think it comes a lot from SAS. because market validation makes sense when there's not a deep technology route. Like when you're saying, well, I'm building a product that serves a purpose. It's all about market validation.

但如果我有一件乍看之下几乎可以确定是真的事，比如“我要治愈癌症”，你就不需要验证市场，只需要验证技术问题上的进展。现在投资者分成两派。有些人说：“好吧，我们就……我们就重金投资这些公司。”But when I have something that is on its face almost true by inspection like I'm going to cure cancer. You don't need to validate the market. You just need to validate the progress on the technical problem. Right now investors are in are in two camps. Some are saying well let's go let's let's invest heavily in these companies.

你知道，管它估值呢。你知道，我们就是那类公司之一。另一派人会说：“天哪，这也太不守规矩了。你们还没有经过市场验证。”我其实觉得这种看法从根本上就是错的，而且很多成长型投资者仍然属于那一派。我算是读着那些讲如何看待风险和投资的书长大的。而我在这件事上的看法已经发生了相当大的改变。这不只是因为我的公司正在这么做，我觉得现在我们正处在这样一个阶段：我们在解决非常深层的技术问题，涉及机器人、计算，以及生物学。You know valuation be damned. You know we're one of those companies. And the other set is like oh my god that's undisiplined. It's like you don't have market validation. I actually think that's fundamentally wrong and a lot of growth investors are still in that bucket. So, and I I kind of grew up, you know, reading all the books about how you look at risk and investing. And I've changed my mind quite significantly on this. Not just because my company is doing it, but I think now we're in this place where we're solving really deep technical problems, robotics, computing, you know, biology.

这不是市场验证的问题。市场确实存在，而且非常明确。你只要解决了这个问题，百分之百的人都会买。关键在于解决这个问题的进展。And it's not about market validation. The market is there. The market is very clearly there. you solve that problem, 100% people are going to buy it. It's about progress of solving the problem.

是啊，尤其是在欧洲，这种思维方式很少见。如果完全不受资源限制，从职业角度来说，你会做出哪些不同的选择？Yeah, that's a rare mindset to have in Europe especially. if you were completely resource unconstrained, professionally speaking, what would you do differently?

我会开始建立自己的制造能力，因为我想越来越接近原材料的源头。我觉得我们现在身处这样一个世界：从硬件生态系统的角度来看，供应受限的情况非常多。如果不受限制，而且我可以随心所欲地建造任何东西——老实说，这有点像埃隆正在做的事。是的，我们应该建造自己的晶圆厂之类的设施，这样我就能针对自己的需求，对这一层进行极致优化。我不需要拿一个为了另外一百种用途而制造的通用产品，再想办法让它满足我的需求。I would start to build my own manufacturing because I want to go closer and closer to the source of raw materials. I think we're in this world now where there are so many supply constraints from from the hardware ecosystem point of view. If I were unconstrained and I could build anything like this is kind of what Elon's doing to be honest. yeah, we should build our own sorts of fabs because I can hyper optimize that layer for what I need. I don't need to take a thing that is a commodity that's been built for a hundred other purposes and make it fit my needs.

我会完全按照自己的需求来打造东西，从而做出更好的产品。坦白说，在我看来，计算机的全部价值都来自于提供智能。I would build exactly the thing I need and build a better product. And frankly, I see it as all the value of a computer comes from delivering intelligence.

当今最被高估的 AI 市场The Most Overrated AI Markets Today

如果我能把整个流程都优化到极致，甚至细化到硅锭从工厂运来后我该如何处理、如何切割，那我当然想这么做。回答得真好。你觉得如今 AI 的哪个领域最被高估、最不值得投资？比如我觉得，客户支持 AI 是目前最泛滥、也最缺乏差异化的领域之一，不过只是一个试图颠覆 Salesforce 的切入点，这相当没意思。If I can hyper optimize that all the way down to how I process, you know, the silicon ingot when it comes, you know, from the factory and how I slice it, I absolutely want to do that. Great answer. What area of AI do you think is most overrated today where you would not invest? So like for me I think customer support AI is like one of the most overp proliferated nondifferentiated spaces that's just like a an entry point to try and kill Salesforce which is quite uninteresting.

是的，我觉得那些快速见效的项目——也就是原本由人来完成的标准工作，我现在只想把它自动化——是目前最不值得投入的方向。我觉得这确实被高估了。实际上，我觉得智能体编程这类东西作为一个市场也相当被高估了，因为我认为价值会流向模型提供商。我不觉得其他公司能创造多少价值。我知道 Cursor 实现了很好的退出，这当然很棒。但大部分价值都来自模型提供商，来自模型本身。Yeah I think those those quick hits where you have a standard thing that's done by humans I'm just going to try to automate it. That's the worst place to invest right now. I think I think it's very overrated. I actually do think that like aentic coding stuff is quite overrated in terms of a market because I think it moves to the model provider. I don't think you get much value from the others. I know cursor had a great exit and that's wonderful. but the model providers, the model itself is where most of the value is coming from.

我不太认为关键在于运行框架或 IDE。它们的用户黏性并不强。我在自己的公司里也见过这种情况：人们可以很轻松地在这些工具之间切换。所以，重要的是模型以及它能提供哪些能力。因此我认为，撇开模型提供商不谈，智能体编程作为一个市场，我今天不会投资。I don't think it's the harness or the IDE so much. it's not super sticky. I've seen it in my own company. People move between these things pretty fluidly. So it's the model that matters and what capabilities it delivers. So I think aentic coding as a market you know modulo the model providers is something I wouldn't invest in today.

为什么 80% 的 AI Token 可能会通过开放模型流转Why 80% of AI Tokens Could Flow Through Open Models

三年后，多少比例的 token 流量会通过开放式与封闭式渠道？What percent of token traffic will flow through open versus closed in 3 years time?

我认为它会有 80% 是开放的。I think it would be 80% open.

这换算成美元流量是多少？What does that translate to in terms of dollar traffic?

这其实是个更好的问题。我认为，最终的格局可能会很像手机市场。粗略地说，iPhone 占整个市场的 20%，却拿走了 80% 的利润。情况可能也会类似。我们可能会发现，只有少数几个非常重要的问题需要前沿模型来解决，而人们也愿意为此支付高得多的费用。不过，你日常处理的所有常规任务，都会由开放模型来完成。that's a better question actually. I think it's going to probably converge on something that looks very much like the cell phone market. Just roughly speaking, iPhone is 20% of the total market, but 80% of the of the margin. Something similar to this. We'll probably see that there's a relatively small number of very important problems that Frontier models are needed for, and you'll pay much more for those things. However, all of the like routine tasks that you do will be handled by open models.

所以，我其实觉得支出大致会五五开，但按 token 数量算，只有 20% 会用在前沿模型上。你经历过的最棒的一次投资人会面是哪一次？为什么？So, I actually think there'll be a rough 50/50 split on spend, but in terms of tokens, it'll be only 20% to the to the frontier models. Who is the single best investor meeting you've ever had and why?

我可以先简单讲一点故事。I can give a bit of a story for first.

好。继续。Yeah. Go on.

我想你知道，自从我创办 Nirvana 以来，我大概参加过一千场投资人会议吧？所以，参加过第八百场之后，第一场会议也就不太会给人留下什么印象了。很早的时候，我和 Steve Jervson 见过面，他后来成了我的好朋友。当时我紧张得要命，因为听说过太多关于他的精彩传闻：这人超级聪明，会把我问得招架不住——我当时听到的说法大概就是这样。后来我和他聊了起来，我们一开始就兴致勃勃地讨论数字表示，我心想：“这人是个投资人啊。”之后我给他发了几篇论文，他在上面做了标注，还提出问题；我再给他发回复。我从来没想过投资人会是这种样子。你知道，他简直就像是在和一位技术同事交谈。I think you know since I started Nirvana, I've probably had I don't know a thousand investor meetings, right? So after you know the 800th one to the 800 the first one don't really leave an impression anymore. So very early on who you know I had a meeting with Steve Jervson who has now since become a good friend of mine and I was nervous as hell because I'd heard so many like amazing things about this guy super smart he's going to take me apart you know that's kind of what I was I was told I had a conversation with him which we just started geeking out on numeric representations and I was like this guy's an investor and then I send him a couple papers he like marks him up and asks questions I'm sending responses back to him like this is not what I expected from an investor ever, you know, this guy's like talking to, you know, like a technical u you know, colleague in a sense.

所以对我来说，那是我参加过的最棒的投资人会面之一，因为我们做的事情太新了，还没有人向投资人推介过。2014年，已经有10年没人推介过硬件创业公司了。这个人简直迫不及待地想投资这样的项目，因为他自己也一直在琢磨这件事。所以当时就像是：砰，咱们干吧。So to me that was one of the best the best investor meetings I had because we were doing something so new that no one had pitched it yet. No one had pitched a hardware startup in 10 years in 2014. And this guy was almost champing at the bit to invest in something like this because he had been thinking about it himself. And so that was just like boom, let's go.

这是个很棒的回答。我喜欢史蒂夫。我记得请他上过节目。他当时特别棒。你说他愿意持非共识立场，这正是我喜欢他的地方。It's a great answer. I love Steve. I remember having him on the show. He was he was fantastic. You talk about like willing to be non- consensus is what I love about him.

没错。你必须特立独行，才能下那些最终会有回报的大赌注。按定义来说，共识已经错了，对吧？它已经过时了。所以你必须不从众。That's right. You have to be to make to get make big bets that will pay off. By definition, the consensus is already wrong. right? It's already old. So you have to be non-conensus.

不过，我觉得身处欧洲面临的挑战是，创意实际上已经商品化了。我的意思是，现在当创始人已经成了一件很时髦、像是来体验一番的事。每个人都想当创始人，体验一下这种生活。创意的质量下降了。我不觉得美国人天生就更优秀、更聪明或更有才华，但我确实认为，你们在选择创意方面更有创新性。这可能是因为你们接触到的东西不同、看到的不同、思考方式不同，也更清楚在整个格局中该选择哪个方向发力。你知道，我们现在还在用 AI 做费用管理。The challenge I find with being in Europe though is actually you have like the commoditization of ideas. And what I mean by that is actually now is being a founder is quite a touristy thing to do. Everyone wants to be a founder for the experience. The quality of ideas has reduced. And I think I don't think Americans are inherently better, smarter, more talented. But I do think you are more innovative around idea selection. And that's probably because of exposure, what you see, how you think, where to play in the structure. You know, we're still doing expense management with AI.

是啊。没错，我觉得我们完全可以放开思路，把格局想得更大一些。Yep. No, totally. I think I think we are given the permission to think bigger here.

是的。Yeah.

在欧洲，你可得不到这种反应，因为整个投资者圈子以及人们的思维方式都非常保守。你知道，如果你带着一个疯狂的点子找上门，他们就会说：“那家伙疯了，我才不会投资他。”而在这里，他们会说：“那家伙真棒，眼光真远。”对吧。In Europe, you're not because the whole investor network and the way people think is so conservative. You know, if you come in with a crazy idea, they're like, "That guy's mad. I'm not going to invest in him." Whereas here, it's like, "That guy's awesome. He's thinking big." Right.

为什么欧洲孕育的雄心勃勃的初创企业更少Why Europe Produces Fewer Ambitious Startups

是啊。不，我完全同意你的看法。绝对没错。难就难在——我也遇到了这个问题——我确实会这么想：好吧，那就这么做吧。现在有一家 A 轮公司，他们在做航天相关的东西。我他妈一点头绪都没有。团队很棒，他们很出色，但哥们儿，这是从一只 4 亿美元的基金里投出去的 2000 万美元。这相当于 5%。你可不能轻易把这笔投资搞砸啊，哥们儿。这是在航天领域做的一笔风险极高的投资，而我完全看不懂。我多少会把这类投资看作是处于技术周期早期阶段的项目。你知道，我们这些技术人都会给你讲一个故事，而我们要做的就是构建一个叙事。这个故事肯定会错，百分之百会错。我不知道别的，但我知道它是错的。我不知道具体错在哪里，也不知道会错得多离谱，但它就是错的。我觉得，在那个阶段，你投资的是一个掌舵的人，而不是那辆车本身，对吧？所以，理念是对的，方向是对的，而你真正看重的是团队。Yeah. No, I completely agree with you. It's absolutely right. The hard thing is, and I've got this problem, I've got this genuine, okay, let's do this. There's a series A now, and they are building space stuff. I've got no [ __ ] clue. The team is great. They're awesome, but dude, it's 20 million check from a 400 million fund. That's a It's 5%. You can't lose that easily, dude. It's a super risky play in space and I do not understand it at all. I sort of look at these as you know in the early part of the technology cycle like you know all of us in who are technologists are going to give you a story and that's what we need to do is build a narrative that story is going to be wrong 100% will be wrong I don't know anything else but I do know it's wrong I don't know exactly where it's wrong I don't know how wrong it'll be but it is wrong and what you're investing in I think at that stage is someone to drive the bus not the bus itself right So the idea is right, the direction is right and the team is what you care about.

他们能推动这件事吗？他们能做出必要的调整和正确的技术决策，真正把这辆车开向有意义的目标吗？Can they drive that? Can they make the requisite pivots and and right technology decisions to actually drive the bus to something that matters?

实际上，这里的挑战在于，在这个市场上，你没有时间和他们打交道。The challenge actually there becomes in this market you don't have the time with them.

是啊，我倒是很想说“是啊”，但我不知道。你大概会有两到三场会议。Yeah, I would love to say yeah I don't know. you probably get two to three meetings.

你说得对。你得尽可能多做些工作，跟你认识的人聊聊之类的，但这很难。嗯，我同意。You're right. And you got to do as much work as you can. Talk to people that they know and whatnot, but it's hard. Yeah, I agree.

为什么人工智能能够开辟全新的科学前沿Why AI Could Unlock Entirely New Frontiers of Science

最后一个问题，哥们。你最期待什么？我喜欢乐观的态度。我不喜欢大家对泡沫和[ __ ]抱有的那种愤世嫉俗。你呢？我就是[ __ ]特别期待这件事。Final one for you, dude. What are you most excited for? I like optimism. I don't like the cynicism that we have around bubbles and [ __ ] What's your like I'm just [ __ ] pumped for this.

Naveen Rao：我把 AI 看作一种工具和能力，它能让我们提出关于宇宙的更宏大的问题。科学领域有太多我们尚未理解的事情，而这是因为这些问题难以企及。你知道，能够思考这些问题并真正取得进展的人数有限，所以这些问题才难以企及。我们能够着手解决并开始攻克的问题范围正在扩大。对我来说，这令人无比兴奋。我不会去想什么“天哪，AI 会杀光所有人”。这是我听过的最愚蠢的世界观。对我来说，这简直太荒谬了。说真的，我现在已经完全听腻了。Naveen Rao: I look at AI as a tool and capability that enables us to ask much bigger questions about the universe. There's so many things we don't understand in science, and it's because they're un approachable. they're unapproachable from you know just the number of humans who are who could think about that problem and actually make progress. The surface area of problems we can go after and start to solve starts to increase. And to me that's extremely exciting. I don't look at you know oh my god AI is going to kill all humans. That's the stupidest worldview I've ever heard. It's just it's just idiotic to me. Actually completely tired of it now.

这就像是在说，不不不，各位，AI 才是真正能让人类开始攻克这些更大难题的东西。比如，我们怎样才能探索更多宇宙？当我们开始深入理解物理学的更深层面时，就能做到。你知道，我们有虫洞之类的想法，这些都只是理论构想。现在我们可以开始借助 AI 取得进展：我能不能真的造出某种东西，利用这种物理现象？它到底是不是真的？这些问题我们现在都可以开始提出来，而过去我们根本做不到。And it's like no no no guys AI is the thing that enables humans to actually start to crack these bigger problems like how can we explore more of our universe we can do that when we start to understand deeper parts of physics you know we have these ideas of like wormholes and these things these are all like theoretical constructs we can now start to use AI to start making progress can I actually build something that exploits that that physical phenomenon or is it real or not like these are questions we can start to ask in the past we just couldn't.

我们没有足够多有能力探讨这些问题的人。所以对我来说，推动科学进步、推动人类认知向前发展，实在太令人兴奋了。我、我甚至都不知道该怎么形容。这让我感觉像个孩子。We didn't have enough humans that were capable of approaching these questions. So, to me, the ability to move science forward and to move human understanding forward is so exciting. I I can't even tell you. It makes me feel like a kid.

你知道，外面有一个奇妙无比的世界，我们以前甚至想都不敢想。我们被压得只敢相信，自己能看到的就只有这些。而现在，突然之间，整个世界都向我们敞开了。You know, there's a world of wonder out there that we just couldn't even think about. We got browbeaten into our own capabilities of like this is all we can look at. And now all of a sudden, that whole world is open to us.

我是说，正是这样的节目让我做了十年之后依然热爱主持节目。你的知识面之广真的太棒了。说实话，能亲眼见证这一切，对我来说简直不可思议。所以，谢谢你这么乐于涉猎如此广泛的话题。你真的太棒了，兄弟。I mean, shows like this are why I still love doing the show after 10 years. like the breadth of your knowledge is just like wonderful. it's it's amazing for me to like witness honestly. And so thank you for being so willing to engage across such a broad spectrum. And you've been amazing, dude.

这真的很有意思。我喜欢，喜欢这种对话，我们可以畅聊各种不同的话题。It's a lot of fun. I love I love these conversations where we can riff on so many different topics.