英伟达 CEO 黄仁勋谈 AI 安全、开放模型与赢得 AI 竞赛

简要描述Brief Description

All-In Podcast 主持人与英伟达 CEO 黄仁勋展开了一场广泛的对话，讨论了人工智能安全叙事、Dario Amodei 的文章、开源模型与闭源模型之争、递归式自我改进，以及英伟达的资本与生态系统战略。采访期间，特朗普总统突然打来电话，将末日主义论调和对数据中心的反弹称为“骗局”；与此同时，黄仁勋认为，安全与美国领导地位并非非此即彼的选择，超级智能已经在狭窄领域中存在，而放慢脚步是错误的战略。All-In Podcast hosts sit down with Nvidia CEO Jensen Huang for a wide-ranging conversation on AI safety narratives, Dario Amodei’s essay, open source versus closed models, recursive self-improvement, and Nvidia’s capital and ecosystem strategy. A surprise phone call from President Trump mid-interview frames doomerism and data-center backlash as a “hoax,” while Jensen argues safety and American leadership are not false choices, that superintelligence already exists in narrow domains, and that slowing down is the wrong strategy.

目录Table of Contents

• 欢迎辞与达里奥的文章
  Welcome and Dario’s Essay
• 失败的预测与末日论叙事
  Failed Predictions and the Doomer Narrative
• 文化、举报者与正确开展工程工作
  Culture, Whistleblowers, and Getting Engineering Right
• 调节、根本原因与递归式自我改进
  Regulation, Root Cause, and Recursive Self-Improvement
• 开源、闭源模型与赢得人工智能竞赛
  Open Source, Closed Models, and Winning the AI Race
• 特朗普总统打来的意外电话
  Surprise Call from President Trump
• 通话之后：安全、工作岗位与数据中心
  After the Call: Safety, Jobs, and Data Centers
• 英伟达的资本配置与生态系统战略
  Nvidia’s Capital Allocation and Ecosystem Strategy
• 沿着技术栈向上、NeoCloud 与开放模型
  Going Up the Stack, NeoClouds, and Open Models
• 竞争、中国光刻、通用人工智能与超级智能
  Competition, China Lithography, AGI, and Superintelligence

欢迎词与达里奥的文章Welcome and Dario’s Essay

主持人：我们打断了每周节目。而我们只有在三个人的情况下才会暂停节目：特朗普总统、耶稣，以及 Jensen。全球排名第一的播客。这位就是 Jensen Huang。他是英伟达的创始人、总裁兼首席执行官。不管你是否意识到，他的决策都在塑造你的未来。英伟达是当前市场上最重要的股票。Jensen 可以说是历史上最出色的高管。营收同比暴增 97%。不仅需求已经非常强劲，而且还在加速增长。英伟达是唯一一个拥有完整技术栈的 AI 工厂计算平台。GPU 就像一台时间机器，因为它能让你更早看到未来。如果我们能够看见未来、预测未来，那么我们就更有可能把那个未来打造成为最好的版本。请欢迎 Jensen Huang。Host: We preempted the weekly show. And there’s only three people we preempt the show for: President Trump, Jesus, and Jensen. The number one podcast in the world. That’s Jensen Huang. He’s the founder, president, CEO of Nvidia. Whether you know it or not, his decisions are shaping your future. Nvidia is the most important stock in this market. Jensen is arguably the best executive in history. Revenue exploded 97% year-over-year. Not only is demand already strong, it is actually accelerating. Nvidia is the only computing platform that is a full-stack AI factory. A GPU is like a time machine because it lets you see the future sooner. And if we could see the future and we can predict the future, then we have a better chance of making that future the best version of it. Please welcome Jensen Huang.

黄仁勋：谢谢你。我也爱你。全球排名第一的播客。Jensen Huang: Thank you. I love you back. Number one podcast in the world.

主持人：当然。哇，我们喜欢这件新夹克。Host: Absolutely. Wow. We like the new jacket.

黄仁勋：嗯，你知道，你们以拍卖的方式开启了会议。我只是觉得你们需要一些活力。我知道我们现在谈论的是严肃的事情，但我们需要带着活力来讨论。Jensen Huang: Well, you know, you auctioned the open. I just felt you guys needed some energy. I know we’re talking about serious stuff here, but we need to talk about it with energy.

主持人：我们先从这周末的这篇文章——达里奥的文章——开始吧。实际上，有人把它通过 Pangram 检测了吗？我甚至不知道其中有多少是借助 AI 完成的，但那确实是一篇相当令人惊叹的文章。然后，我想让很多人感到意外的是，各家前沿实验室竟然围绕这篇文章本身形成了共识。Jensen，你先梳理一下发生了什么，你是如何阅读和解读这篇文章的，然后我们再深入讨论其中的一些细节。不过，开始时或许先谈谈你的总体看法。Host: Let’s start with this essay from this weekend — Dario’s essay. Actually, did anybody run it through Pangram? I don’t even know how much of it was AI-helped, but that was a pretty incredible thing. And then I think what a lot of people were surprised by was the coalescing of the frontier labs around the essay itself. Just Jensen — unpack what happened, how you read it, how you interpreted it, and then we’ll get into some details that were inside of it. But maybe just the high-level thoughts to kick it off.

黄仁勋：首先，里面有很多内容。第一部分是关于安全的，我们必须对此高度重视。安全至关重要。显然，安全与领导力并不是非此即彼的选择。你可以快速创新、快速执行，美国也能够发挥领导作用，同时确保安全。我认为这两者并非相互排斥，但安全显然很重要。Jensen Huang: Well, first of all, there was a lot of stuff in there. First there is a part about safety which we have to take very seriously. Safety is paramount. Obviously, safety and leadership are false choices. You’re able to innovate quickly, you’re able to execute quickly, and America’s able to lead and to do it safely. I think those are false choices, but safety is obviously important.

我认为他所谈论的是内部控制问题。显然，Cox 举报人的事情非常严重。无论何时出现举报人，都必须对此高度重视。我认为 Cox 有很大的勇气，把他的担忧公开出来。即便如此，其中还是有一些问题被混为一谈了。我认为举报本身没有问题。我认为，对未来的科学预测则不太妥当，因为它并没有以科学为依据。虽然这是由一位科学家表达出来的，但显然并非以科学为基础，因此我对此有异议。但显然，就举报人这部分而言，还有一大堆事情：暂停、放慢进度——如果他们觉得自己的公司已经失控，这些都是他们可以采取的自愿措施。There’s a matter of internal control that I think he was speaking to. Obviously the Cox whistleblower is a very serious matter. Whenever you have a whistleblower, you got to take it very seriously. I thought Cox had great courage to put out what his concerns were. And even then there were some issues that were kind of conflated within that. I think the whistleblowing is fine. I think the scientific prediction about the future is less aligned because it’s not grounded on science. It was expressed by a scientist but it was obviously not grounded on science, and so I take issue with that. But obviously the whistleblower part of it — there’s just a whole bunch of stuff: pausing, pacing — those are all the voluntary things that they could do if they feel that their company is out of control.

如果考克斯看到了什么——我们不知道考克斯看到了什么——但如果他看到公司已经失控了，而这或许是从研究向工程转型的过程。正如你所知，这些实验室正在从研究转向工程。非凡的人才，非凡的工程能力。但显然，工程与研究是不同的。也许这种转型并不顺利。我们不知道他看到了什么，而最终只有他自己知道。但如果问题在于缺乏控制，那就是另一个话题了。政府应该如何应对？现在突然之间，监管——一篇博客就把所有事情都涵盖了。If Cox saw something — we don’t know what Cox saw — but if he saw that the company was out of control, and maybe it’s a transition from research to engineering. As you know, these labs are transitioning from research to engineering. Extraordinary talent, extraordinary engineering. But obviously engineering is different than research. Maybe that transition is clumsy. We don’t know what he saw and ultimately only he knows. But if there was a matter of lack of control, that’s a different topic. How should the government deal with it? Now all of a sudden regulation — it just covers everything in one blog.

失败的预测与末日论叙事Failed Predictions and the Doomer Narrative

主持人：你能不能帮我们梳理一下？其实我们这周在播客里试着玩了这个游戏，但挺难的。也就是：你该怎么描述——我妈妈打电话给我，她会问，这个所谓的“文明灭亡”到底是怎么回事？我不知道该怎么向她解释。所以，当你有这样非常聪明的人把它量化、定量化时，我想这可能正是让一些人感到不安的地方。他们会想，这是什么意思，10%的灭绝概率？没人知道该怎么向普通人解释，这种事究竟怎么可能。Host: Can you just help us sort of unpack? We tried to play this game actually this week on the pod and it was difficult, which is: how do you describe — my mom calls me and she’s like, what is this whole civilizational death thing? I don’t know how to explain it to her. So when you have very smart people like that quantize it and quantify it, I think that’s probably what’s perturbing to some people. They’re like, what does that mean, 10% of extinction? Nobody knows how to explain that to the average person how that’s even possible.

黄仁勋：首先，我们不应该这么做，因为这都是凭空捏造的。这些人受过良好教育——他们被称为研究人员——显然是在实验室里工作。因此，把这些词汇联系起来，再加上这样的预测，是令人担忧和不安的，不应该这样做。这是不负责任的。Jensen Huang: Well, first of all, we shouldn’t, because it’s made up. These are well-educated — they’re called researchers — obviously they’re working in a lab. And so the confluence of these words and then the prediction is alarming and troubling and it shouldn’t be done. It’s irresponsible.

现在，事实是这样的：让我们回过头来看看真正的事实。事实是，曾经有人预测，五年后放射学将完全被人工智能接管，世界上将不再有放射科医生。事实证明，结果恰恰相反。如今，世界比以往任何时候都更需要放射科医生。然而，人工智能已经完全接管了放射学，这很棒——它实现了扫描影像阅片的自动化，这很棒。Now the fact of the matter is, let’s go back and look at the real facts. The facts are there was a prediction that in five years’ time radiology will be completely taken over by artificial intelligence and there’ll be no radiologists in the world. That has proven to be exactly the opposite. We need more radiologists than ever in the world. However, AI has taken over radiology completely, which is great — it automated scan reading, which is great.

曾有人预测，在六到十二个月内——难道不是就在去年吗？——在六到十二个月内，90%的代码就会已经由人工智能生成。事实证明这是错误的。去年有人预测，在六到九个月内，50%的入门级工作岗位将会消失。事实证明这也是错误的。我的意思是，所有这些预测都错了。There was a prediction that within six to twelve months — wasn’t it just last year? — within six to twelve months, 90% of code would already be generated by AI. That has turned out to be wrong. Within six to nine months, that was predicted last year, 50% of entry jobs will be wiped out. That has proven to be wrong. I mean, all of these predictions have been wrong.

主持人：没错。嗯，GPT-2 太不安全了，不能发布。Llama 3 太不安全了，不能发布。明年将有一半的白领工作消失。就业末日。Host: Right. Well, that GPT-2 would be too unsafe to release. That Llama 3 would be too unsafe to release. Half of white-collar jobs would be gone next year. The jobs apocalypse.

黄仁勋：我们必须承担责任。我们必须为所有那些愚蠢的预测负责。所以，我们应该把这些都记录下来。当然，人们确实会这么做，并提醒我们，那些预测与美国最终赢得人工智能竞赛的结果并不一致。Jensen Huang: We have to take accountability. We have to take account for all of the stupid predictions that were made. And so we ought to just keep track of all that. And of course people do, and remind us that those predictions are inconsistent with ultimately America winning the AI race.

主持人：简而言之，有些人会说，你知道，他们会说要相信专家，他们拿新冠疫情来类比——而新冠疫情同样始于一些研究人员、受过良好教育的人，他们对这件事的了解程度与我们其他人存在不对称，而他们所说的一些话，最终我们发现，从事实来看并不正确。因此，现在“相信专家”运动与另一种观点之间正在发生一场战争：好吧，我们还是看看这些预测的实际历史，更有条理地思考吧。Host: The short form for that is some people are saying, you know, they say trust the experts and they used the analog of COVID, which again started with people that were researchers, educated people that had an asymmetric awareness of the thing that the rest of us did not, saying things that ultimately turned out we find out in facts not to be true. And so there’s this war that’s happening right now between the trust-the-experts movement and, you know, well, let’s just look at the actual history of these predictions and let’s just think more methodically.

这从何而来？因为它来自那些真正制造它的地方。你觉得其心理构成是什么，或者真正的动机是什么？也许是商业动机，也许是政治动机。你能不能大致猜一下，或者说说你是如何看待他们为什么要这么做的？Where is this coming from? Because it’s coming from inside the places that’s actually making it. Like what do you think is the psychological makeup or what is the real incentive? Maybe it’s a business incentive, maybe it’s a political incentive. Can you just maybe guess or how do you think about why they’re doing this?

黄仁勋：首先，我得告诉你，这些都是历史上影响最为深远的公司之一。它们拥有非凡的工程师、非凡的研究人员，做出了真正出色的工作。一方面，我与这些公司之间开展着非常密切的合作；另一方面，我们却不得不在公开场合进行这样的对话。这实在令人遗憾。我认为，这些公司确实应该按照我们过去创办公司的方式来打造，也就是在沉默中进行。Jensen Huang: Well, first of all, I got to tell you these are some of the most consequential companies in history. Extraordinary engineers, extraordinary researchers, really fantastic work. On the one hand, I work very closely with them as companies to companies. On the other hand, we have to have conversations like this in public. And it’s really unfortunate. And I think that these companies really ought to be built the way that we used to build companies, which is in silence.

文化、举报者与正确开展工程工作Culture, Whistleblowers, and Getting Engineering Right

主持人：等等，Jensen，所以你不允许组织中的任何人代表整个组织发言，尤其是在他们周末过得很糟糕或愤而辞职的时候。他们不被允许代表你或组织发推文。Host: So wait, Jensen, you don’t allow anybody in your organization to speak for the entire organization, especially when they’re having like a bad weekend or they rage quit. They’re not allowed to tweet on your behalf and the organization’s behalf.

黄仁勋：不，因为这是他们来为我们工作时所做的决定，而我们也告诉他们：在我们公司工作时，你们应该这样行事。如果你喜欢我们公司的文化——正如你所知，英伟达的文化以及英伟达的员工群体都让人感到非常满意——他们喜欢公司始终如一、保持稳定这一点；他们也认同我们的核心价值观，即关照员工家人，为他们创造能够完成毕生事业的条件。我们从事有意义的工作，尽可能低调地完成，并为其他所有人的成功作出贡献，这让我们感到非常自豪。因此，人们会被这类核心价值观所吸引。Jensen Huang: No, because that’s what they decided when they came to work for us and we told them: these are the way you behave when you work in our company. And if you like the culture of our company — which as you know the NVIDIA culture and the NVIDIA employee base, incredibly happy — they like the fact that the company is consistent, that we’re stable, that our core values are consistent with taking care of the families and creating the conditions by which they can do their life’s work. That we do meaningful work, we do it as quietly as we can, and we contribute to everybody else’s success, which we’re very proud of. And so those kind of core values people are attracted to.

但是，当你加入我们公司工作时，你所做的一些事情我们也并不认可。比如说，我们不欢迎在公司内部进行政治讨论。把这些话题带回家去。你们可以在公司外讨论政治。我们是一家不涉政治的公司。我们不偏袒任何党派。我们希望美国取得成功。无论哪个政府执政，我们都会竭尽所能帮助美国取得成功。因此，关于种族、宗教、政治以及所有类似话题的讨论——我们会告诉员工，请在公司外进行。这些不适合在我们公司谈论。But when you come and work in our company, there are also some things that we don’t appreciate that you do. Like for example, we don’t welcome political discourse inside our company. Take it home. You guys talk about politics outside the company. We are an apolitical company. We’re bipartisan. We want America to succeed. And whatever government is in place, we’ll do everything in our power to help America succeed. And so the discourse about race and religion and politics and all of that stuff — we tell people do it outside the company. It’s not for us.

主持人：更具体地说，谈到人工智能监管——萨提亚今天上午在这里，他说，你知道，在我们讨论那些可能真正阻碍发展的监管措施之前，为什么不先把一些基础工作做好？为什么不把测量做好？为什么不把标准化做好？把工程做好。以一种更可预测的方式将研究成果转化出来，这样我们就不会制造恐慌。准备好之前，先把它控制在内部，不要对外公开。你认为正确的应对方式是什么？你知道，Demis 提议过成立一个更像 FINRA 的组织。至于 Dario 想要什么，目前还不清楚——某种跨国的、变异式的机构，拥有某种控制权。你对此持什么立场？我们现在需要什么？Host: In terms of AI regulation then more narrowly — Satya was here this morning and what he said is, you know, before we talk about regulation that could really stymie things, why don’t we just get some basics right? Why don’t we get measurement right? Why don’t we get standardization right? Get the engineering right. Translate the research in a more predictable way so that we’re not fear-mongering. Keep it inside until we’re ready to expose it. What do you think the right response is? You know Demis had a proposal which was sort of this more FINRA-like organization. It’s not clear what Dario wants — this transnational mutated thing that has some sort of control. Where do you land on this? What do we need right now?

调控、根本原因与递归式自我改进Regulation, Root Cause, and Recursive Self-Improvement

黄仁勋：你知道，监管应该解决实际问题。所以问题在于，我们遇到过哪些实际问题？如果你看看这些实际问题，到目前为止，它们全都来自各家实验室。而之所以会这样——为它们辩护一下——是因为它们拥有最多的算力。之所以拥有最多的算力，是因为它们正试图解决前沿问题。因此，前沿实验室会成为最大危险来源，这是合乎情理的。高中生不太可能造成什么问题，因为他们根本不会拥有足够的算力。初创公司也不太可能成为问题的源头，因为它们同样不会拥有足够的算力。事实上，你可以放眼全球，除了前沿实验室之外，其他所有人都不会拥有足够的算力。Jensen Huang: You know, regulation should solve actual problems. And so the question is what actual problems have we enjoyed? And if you look at the actual problems, all of the actual problems so far have come from the labs. And the reason for that — just in their defense — is because they have the most compute. And the reason for that is because they’re trying to solve the frontier problems. And so it’s sensible that the frontier labs will be where the most danger comes from. It is unlikely that a high school student did something because they just simply won’t have enough compute. It’s unlikely that a startup will be the reason because they won’t have enough compute. In fact, you could look across the planet and everybody won’t have enough compute with the exception of the frontier labs.

所以现在的问题是，如果你看看实际发生了什么——他们正在开展开创性的工作，这真的非常困难——他们正从研究转向工程。我可以想象，他们显然正在构建世界上最具重大影响力的一些技术和公司。他们在同时建设公司、文化、技术、工程和产品。因此我能理解，现场有些手忙脚乱。但无论如何，一个实验室发生了四起事故，另一个实验室发生了一起重大事故——你首先必须做的，就是从工程角度对问题进行根因分析：发生了什么，我们本可以做些什么不同的事情，以及我们将实施并制度化哪些措施，无论是技术、方法还是流程，并确保不让这种事情再次发生。So now the question is, if you look at what actually happened — and they’re doing pioneering work, it’s really very hard — they’re transitioning from research to engineering. I could imagine they’re obviously building some of the most consequential technology and companies in the world. They’re building their company, they’re building their culture, they’re building the technology, they’re building engineering, they’re building products all at the same time. And so I can understand it’s a little bit hair on fire. But nonetheless, the four incidents from one lab, the one giant incident from the other lab — the first thing that you have to do is just root cause the problem from an engineering perspective: what happened, what could we have done differently, and what are we going to implement and institutionalize, whether it’s technology or methods or processes, and make sure that we don’t let it happen again.

现在我敢拿钱打赌，在上述每一个案例中，未来他们都有能力控制局面，防止类似事件再次发生。因为如果事情不在他们的控制范围内，另一种可能就是——而我确信那四起事故不会再次发生。我确信他们已经找到了根本原因并解决了问题。我确信他们现在已经拥有了用于沙箱、运行时、监控器和持续监控的技术。所以我确信，他们现在的技术已经好得多、好得多了。另一种可能同样不太现实，那就是他们说：“听着，我们发生了这些事故，分析完成后得出的结论是，我们对发生的一切一无所知，也完全不知道该如何控制局面，所以我们正在向社会寻求帮助。”如果真是这样，那么我们这些拥有工程师的公司就应该派工程师进去。我的意思是，如果可以的话，我们也应该给他们提供建议，但我对此表示怀疑。我认为他们拥有非凡的人才。他们已经处理好了。Now, I would bet you money that in every single one of those cases it is within their control in the future to prevent it. Because the alternative if it’s not in their control — and I’m sure that those four incidents won’t happen again. I’m sure they root-caused it and fixed it. I’m sure they have now technology for sandboxes and runtimes and monitors and continuous monitors. And so I’m certain they have much, much better technology now. The alternative is also unlikely, which is for them to say, look, we had these incidents, after we’re done analyzing it we came to the conclusion we don’t know anything that happened and we have no idea how to control it and we’re asking society for help. Now, if that’s the case, then we ought to — a bunch of companies with engineers ought to send engineers in. I mean, and we should advise them if we can, but I doubt it. I think they have extraordinary people. They got this handled.

主持人：但我们的行动并不是在真空中进行的。David，昨晚你告诉我，有一家中国实验室，也就是 GLM 的开发者，准备投入 30 亿美元用于一次递归式自我改进运行。所以，也许你可以先向 Jensen 介绍一下这个情况。Host: But we’re not operating in a vacuum. David, last night you informed me that there is a Chinese lab, the makers of GLM, who are going to put three billion towards a recursive self-improvement run. So maybe you could tee that up for Jensen.

David：嗯，公告就是这么说的。是的，创始人刚刚筹集了50亿美元，并表示他们的优先事项之一将是努力实现能够训练下一个 AI 的递归式 AI，并尽可能将其中的流程自动化。David: Well, that’s what was announced. Yeah. The founder just raised five billion and said that one of their priorities is going to be trying to get to recursive AI that trains the next AI and to try and automate as much of that as possible.

Jensen Huang：嗯，这是现在很时髦的新说法，但正如各位所知，RSI 是一套理念的组合。它始于上下文中的内容，始于技能，始于反思，始于强化学习和合成数据生成。这些都是非常合理的理念，能够让 AI 随着时间推移，在解决问题方面不断变得更好。你们也可以采用低秩方法——所有这些都不涉及权重。实际上，你可以改进权重，这被称为 LoRA。LoRA 可以通过合成数据生成和强化学习得到改进——无需训练基础模型本身就能增强它——然后随着时间推移，你可以利用所有这些经验再次训练基础模型。Jensen Huang: Well, this is the new sexy phrase, but as you guys know RSI is a combination of a system of ideas. It starts with in-context stuff. It starts with skills. It starts with reflection. It starts with reinforcement learning and synthetic data generation. And these are all very sensible ideas that cause AI to get better at solving a problem over time. And you could also have low-rank — all of that stuff doesn’t include the weights. You could actually improve the weights and it’s called LoRA. LoRA could be improved with synthetic data generation, reinforcement learning — enhance it without training the base model itself — and then over time you could train the base model again with all of that experience.

所以我认为，利用这项技术来提升各种任务的生产力，包括构建 AI，是一件明智的事情。我认为这是一个非常合乎逻辑的想法，而且我确信每个人都在某种程度上使用它。只是现在，这个说法被用来以某种方式将这项技术武器化，也许还试图营造一种它将要失控蔓延的印象。And so I think it’s a sensible thing that you’re going to use the technology to enhance productivity of all kinds of tasks including building AI. I think that’s a very logical idea and I’m certain that everybody is using it in some degree. It’s just this phrase is now being used to weaponize the technology in some way and maybe to turn it as if it’s going to spiral out of control is the impression they’re trying to give.

主持人：但你不相信那是真的？Host: But you don’t believe that’s real?

黄仁勋：不，当然不是。原因在于，你可以整天在公司内部进行 RSI，但当你发布产品时，你得对它进行评估，对吧？你得再次测试它，对吧？你得确保没有发生回归，对吧？所以，核心就是控制流程。这些实验室从实验室阶段进入工程阶段后，将会拥有好得多的控制能力。而当它们具备更好的控制能力时——这种控制能力来自方法、知识、实践、工具和技术——所有这些都有助于实现更好的控制、验证和评估；这将使 RSI 能够在公司内部完成，并让优质产品得以对外发布。Jensen Huang: No. No, of course not. And the reason for that is because you could RSI all day long inside your company, but when you release a product, you’ve got to evaluate it, don’t you? You have to test it again, don’t you? You have to make sure that there’s no regression, right? And so the basic process of control. These labs, as they move from labs to engineering, they will have much, much better control. And when they have much better control — and control comes from methods and knowledge and practice and tools and technology — all of those things that lead to better control, verification and evals — it’s going to enable RSI to be done inside the company and for good products to be released outside.

开源、闭源模型与赢得人工智能竞赛Open Source, Closed Models, and Winning the AI Race

主持人：我们先来谈谈开源。我指的是 Hugging Face——我们之前就此进行过沟通，我说过，这将是最具深远影响的收购之一。我甚至不想把它称作一笔交易，因为我认为它的重要性远不止于此。请从第一性原理出发，解释一下开源、闭源和开放权重之间的区别，以及随着时间推移，这个生态系统应当如何协同发展。Host: Let’s talk about open source for a second. I mean this Hugging Face — we were communicating about this and I said it’s going to be one of the most consequential acquisitions. I don’t even want to call it a transaction because I think it’s more important than that. Give us your first-principles explanation of open source versus closed source versus open weights and how the ecosystem should fit together over time.

黄仁勋：这个世界既需要闭源模型，也需要开源模型。我尽可能多地使用闭源模型。这个周末我用了其中四个，它们的表现非常出色。它们处于前沿水平，使用体验很棒，运行效果极其出色，而且一直在不断进步。对我来说，闭源模型有点像瓶装水。你们知道，水是免费的，各位。我不知道以前有没有告诉过你们，但水是免费的。我不想打破大家的幻想，不过水确实是免费的。今天早上，我洗澡时就用了很多免费的水。所以，你要在合适的地方使用合适的水。这和电力没有什么不同，也和我们在这个世界上使用的各种商品没有什么不同。两者都需要。Jensen Huang: The world needs both closed models and open models. I use as much closed models as I can. This weekend I used four of them and they work terrifically. They’re frontier. They’re a great experience. They work incredibly well. They’re getting better all the time. And the way I think about closed models is kind of like bottled water. You know, water is free, you guys. I don’t know if I’ve told you guys, but water is free. I don’t want to burst everybody’s bubble, but water’s free. And this morning, I used a lot of free water taking a shower. And so you use the right water in the right places. And this is no different than electricity. This is no different than all kinds of commodities that we use in the world. You need both.

现在说到开放模型，你之所以需要它，是因为这可能出于主权、隐私或专有技术等方面的原因。看看事实吧。事实是，在过去六个月里，有4000亿美元的风险投资流入了以AI为核心的公司。其中80%使用开放模型。如果没有开放模型，它们如何实现自己的梦想？因为它们的梦想可能各不相同。显然，这些梦想会与前沿实验室的梦想不同。而美国拥有许多不同的创新方式，这正是我们的核心优势之一。伟大的想法不断涌现，源源不绝。因此，开放模型使这一切成为可能。开放模型让每一个——如果我们想赢得AI竞赛，这并不意味着少数几家科技公司赢得AI竞赛，而是意味着美国的每一家公司。每家公司、每个行业、每位研究人员、每位教师、每名学生、每家初创企业——所有人都能获胜。其中一些会使用闭源模型，很多则会使用开放模型。Now in the case of open, the reason why that you need it is because it could be for sovereignty reasons, privacy reasons, proprietary technology reasons. Look at the facts. The facts are in the last six months, $400 billion of venture funding went into AI-native companies. 80% of them use open models. If not for open models, how could they build their dream? Because their dream could be different. Obviously, it’ll be different than the frontier labs’ dreams. And America has so many different ways to innovate. That’s one of our core strengths. Great ideas just coming out of the fountain. And so open models enables that. Open models enables every single — if we want to win the AI race, it’s not about a few technology companies winning the AI race. It’s about every company in America. Every company, every industry, every researcher, every teacher, every student, every startup — everybody wins. Some of them will use closed models. A lot of them will use open models.

主持人：那么，我就想问一下，开放模型来自中国还是美国，这重要吗？Host: Well, let me just ask, does it matter if the open models come from China or the US?

黄仁勋：嗯，我们正在尽一切努力，为开放模型作出贡献。不过，一旦你下载——比如说，可能如今世界对开源的绝大部分贡献都来自中国。他们拥有更多的工程师。因为中国是一个更大的国家，所以他们能够大规模地生产各种成果。因此，他们能够大量培养科学和数学专业的学生。这是我们的劣势之一。他们通过清华大学这样的优秀高校，大批量地培养这些人才。Jensen Huang: Well, we’re doing everything we can to make a contribution in open models. However, the moment you download — like for example, probably the vast majority of the world’s contribution to open source today is coming from China. They just have a lot more engineers. They produce everything in large scale because it’s a larger country. And so they produce science and math students in volume. That’s one of our disadvantages. They’re manufacturing them through amazing universities like Tsinghua University in high volume.

嗯，如今他们也在为开源项目作出贡献。我们下载 Linux，下载 Kubernetes，下载各种软件。其中很多都曾有中国人的参与。一旦你把它下载下来，它就是你的了。我们可以将其分叉，改进它，把它变成自己的。所以，当你下载这些中国模型中的某一个时，它碰巧是由中国一些非常优秀的研究人员开发的，但现在它已经属于你了。你想拿它做什么都可以。Well, they contribute to open source today. We download Linux. We download Kubernetes. We download all the software. A lot of it has been touched by Chinese. And once you download it, it’s yours. We fork it. We improve it. We make it ours. And so when you download one of these Chinese models, it just happens to be made by some really great researchers in China, but it’s now yours. Whatever you want to do with it.

主持人：那么，这场竞赛究竟是什么？Host: So what exactly is the race?

黄仁勋：我认为这是一个非常好的观点。我的意思是，这场竞争真正关乎的是谁能最好地利用这项技术。你知道，上一次工业革命——那些发明家是麦克斯韦、伏打和安培。他们没有一个是美国人。上一次工业革命源自欧洲。但我们充分利用了它。我们在社会层面上比世界上任何其他国家都更好地发挥了它的优势。看看它给我们带来的结果吧。我希望确保下一代工业革命也能这样发生。Jensen Huang: I think that’s a really good point. My point is the race is really about who exploits the technology best. You know, the last industrial revolution — all of the inventors were Maxwell, Volta, Ampère. None of them were American. The last industrial revolution came from Europe. But we exploited it. We took advantage of it socially better than anybody else in the world. Look how it turned out for us. I want to make sure that this next generation happens just like this.

主持人：那么，为什么共产党人此刻能如此成功地在这里传播他们的信息？Host: So why are the communists getting their message out so successfully here right now?

黄仁勋：你知道，我首先认为，这种叙事要务实得多。在中国，没有人会说什么“这个要完了、那个要完了”，或者大肆渲染灾难、散布末日论。他们对此更加务实。他们把人工智能视为一种能够推动经济发展、促进社会进步的技术，也没有那些基本上宣称人工智能将终结人类文明的群体。Jensen Huang: You know, I think first of all the narrative is much more practical. Nobody in China is saying that there’s end of this and end of that and cataclysmic this and doomer that. They’re much more pragmatic about it. They see AI as a technology that’s going to advance their economy, advance their society, and they don’t have these groups who are basically saying it’s going to end civilization.

而且我们是在凭空编造。令人沮丧的是，如果这是真的，那我们就应该谈论它，并采取行动，对吧？即使这是真的，我们也应该花更多时间采取行动，而不是让一群对此无能为力的人忧心忡忡。建设它是我们的职责，对吧？And we’re making it up. The part that is frustrating is if it was true, then we ought to talk about it and go do something about it, right? Even if it’s true, we ought to spend more time doing something about it than worrying a bunch of people who can’t do anything about it. It’s our job to build it, right?

主持人：历史上是否曾有过这样的时刻：如此多的人如此激烈地宣称某件事，而那件事却如此不真实？Host: Has there ever been a point in history where so many people have so vehemently said something that is so untrue?

黄仁勋：而且这些说法从可衡量的角度来看——实际上可以明确证明——是不真实的，而且它们作为不真实的说法确实是说得通的。它们没有科学依据，也没有研究依据。所有基于科学和研究的结论都证明事实恰恰相反。Jensen Huang: And they’re measurably — they’re actually demonstrably untrue and it actually makes sense as untrue. It’s not based on science. It’s not based on research. Everything that’s based on science and research proves otherwise.

主持人：这是对未知领域的恐惧吗？人类从未涉足那里。我们从未见过它。因此，我们害怕它，也就很容易告诉所有人应该害怕它。Host: Is it a fear of the frontier? Humans have never been there. We’ve never seen it. Therefore, we’re scared of it and therefore it’s easy to tell everyone to be scared of it.

黄仁勋：这也可能是人生经历的影响，David。让我举个例子。我刚从学校毕业时是一名工程师，并没有做太多打字工作。原因在于，我属于软件普及之前的第一代人。那时，我们必须先制造出计算机，才能让软件成为可能。你能想象吗？在这一代人中，每一个进入工程领域的工程师，都把全部时间花在打字上。你找到工作后，实际做的就是这些：他们给你一台笔记本电脑，给你一把椅子，然后你开始打字。你整天都在打字，从睁开眼的那一刻起就开始打字。可是，在打字之前就已经有工程学了。所以你能想象吗？这个世界有堆积如山的工程工作，而其中大部分已经不再是打字了。当然，在打字出现之前，我们也有忙碌的工程师。我认为，在打字之后，我们还将完成大量伟大的工程工作。Jensen Huang: It could be life experience as well, David. So let me give you an example. When I first graduated from school, I was an engineer and I didn’t do that much typing. And the reason for that is because I was the first generation before software became popular. We had to go build the computers to make software possible. Could you imagine — in this generation every single engineer who came into the world of engineering, you spend all your time typing. Literally that’s what you do when you get a job: they give you a laptop, they give you a chair, and you start typing. You type all day long. You type from the moment you wake up. Well, there was engineering before typing. And so can you imagine that the world has a mountain of engineering work to do where most of it is not typing anymore. Sure, we had busy engineers before typing. I think we’re going to do a lot of great engineering after typing.

我说的“打字”，指的是编程。所以，即使是在 NVIDIA，当软件工程师和我交谈时，我也会告诉他们：你们只是在打字。我一直这么说，显然是为了开玩笑。我还会告诉他们，我最喜欢的按键是退格键。原因是，最好的软件就是最小的软件。所以我希望你们使用退格键。When I say typing, I mean coding. And so even at NVIDIA when software engineers talk to me, I tell them, you’re just typing. I’ve been saying that forever, but obviously for fun. And I tell them, my favorite key is backspace. And the reason for that is because the best software is the smallest software. So I want you to use backspace.

特朗普总统的意外来电Surprise Call from President Trump

主持人：我们来认真聊聊英伟达。我们来稍微拆解一下英伟达。这里说的“拆解”，就是解释一下各个组成部分，因为其中涉及很多战略考量。我们从最底层开始。Host: Let’s actually talk about Nvidia. Let’s do a little teardown of Nvidia. Tear down meaning just explain the pieces because there’s a lot of strategy at play. Let’s start at the absolute bottom.

黄仁勋：哦，不，总统先生。哦，是的，先生。我得告诉您，要不是您打来电话，我会——我正在和几位好友一起登台。我正在和几位好友一起登台。我正和萨克斯一起登台。您知道，就是那一群人。对，杰森在这里。查马斯在这里。大卫也在这里。对。我正坐在几千人面前，我们正在谈论——结果我们谈的正是您。干得漂亮，先生。干得漂亮。您能看透这一切——我的意思是，这里面有很多复杂之处，而事实上，您看透了这一切，我们都真的非常感激。Jensen Huang: Oh no. Mr. President. Oh, yes, sir. I gotta tell you something. If it wasn’t because of you calling, I would — I’m on stage with the besties. I’m on stage with the besties. I’m on stage with Sachs. You know, the whole group. Yeah. Jason’s here. Chamath’s here. David is here. Yeah. I’m sitting in front of a few thousand people and we’re talking — as it turned out we were talking about you. Good job, sir. Good job. The fact that you saw through all of that — I mean, there’s a lot of complexity and the fact of the matter is you saw through all of that and we’re all just really grateful.

特朗普总统：告诉他们我向他们问好。President Trump: Tell them I said hi.

主持人：您想跟大家打个招呼吗？杰森想把您的通话切换到免提模式。把他切到免提。直接对着麦克风说。请稍等。请稍候，先生。我们正在拿麦克风。总统先生，您现在是在对全世界讲话。Host: Do you want to say hi to the crowd? Jason would like to put you on speaker mode. Put him on speaker. Right into the microphone. Hang on a second. Hold on, sir. We’re getting a microphone. Mr. President, you’re now talking to the planet.

特朗普总统：你看，人生最妙的一点在于，Jensen 能开发出世界上最复杂、十年内无人能够仿制的计算机芯片。但他却搞不明白该怎么把我接到免提上。President Trump: You see, the great thing about life is that Jensen can develop the most complex computer chip in the world that nobody can copy for 10 years. But he can’t figure out how to put me on speaker.

我们必须记住这一点。人工智能真是太有意思了。简直就像一场阴谋——而最开心的群体是中国，中国非常高兴。我甚至可以说，在这个国家，很多原本什么也得不到的州现在也很高兴，因为大量想要来那里的人正涌入这些州。但现在突然间，你看到他们正在芬兰建设。他们想在那里建一个。谷歌想在芬兰建一个大型数据中心，而这让我很不高兴，因为他们没能获得许可。我告诉你，这全都是一场骗局。数据中心非常棒，它们能让人们变得富有，也能让各州变得富有，它们将成为未来20到25年的石油。它比互联网和人工智能更重要，你知道，重要得多。而他们正好顺了许多不想看到这件事发生的人的意。那些人可能是政界人士，也可能是中国。我们不会让这种事发生。这是一场骗局。We have to remember this one. So interesting the AI. It’s almost as if conspiracy — and the happiest group is China, and China is very happy. And I could even say in the country a lot of states are happy that weren’t going to get anything because they’re being inundated by people that want to be there. But now all of a sudden you see they’re building in Finland. They want to build one. Google wants to build a big one in Finland, which I’m not happy about because they were unable to get permitting. And I’m telling you, it’s all a hoax. The data centers are great and they make people wealthy and they make states wealthy and it’s the oil of the next 20, 25 years. It’s bigger than the internet and the AI, you know, much more so. And they’re just playing right into the hands of a lot of people that don’t want to see it happen. And that could be political people. It could also be China. And we’re not going to let that happen. It’s a hoax.

黄仁勋：您说得对，先生。我们不会让那种情况发生。Jensen Huang: You’re right. We’re not going to let that happen, sir.

特朗普总统：不，我们不会让这种事发生。机器人不会接管世界。这种事不会发生。你知道，我叔叔可能，坦率地说，是有史以来最优秀的人——在麻省理工学院当了41年、42年的教授，被认为是最聪明的人之一。他在那里待了41年，位居顶尖——他就像站在梯子的最顶端。做过很多事情。Jensen对此了如指掌。但确实做过很多事情。所以，如果你相信资源理论的话，我有一点遗传上的力量。但我是相信的。我有遗传上的——President Trump: No, we’re not going to let it happen. The robots are not going to be taking over the world. And that’s not going to happen. You know, my uncle was probably maybe the best of all time, frankly — professors at MIT for 41, 42 years and known as being one of the most brilliant men. And he was there for 41 years as the top — he was like at the top of the ladder. Did many things. Jensen knows all about it. But did many things. So I have a little genetic strength if you believe in the resource theory. But I do. I have genetic —

黄仁勋：这就解释了你为什么对人工智能了解这么多。Jensen Huang: That explains why you know so much about AI.

特朗普总统：嗯，我了解人工智能。我对人工智能也有常识。机器人不会接管一切。人工智能不会接管世界的其他部分。这整件事都是一场骗局。不过，即便如此，我们也必须稍微谨慎一点。我们必须非常——你知道，我们必须采取行动，而且要审慎行事。但这并不意味着我们要停止产业发展，因为，你知道，我们要用接下来的十年研究如何摧毁它。所以，我完全支持你们。我甚至不知道你们对此是什么看法。我还以为你们和我的想法一样。President Trump: Well, I know about AI. I also have common sense about AI. The robots will not be taking over. The AI will not be taking over the rest of the world. The whole thing is a hoax. Now, with that, we have to be a little bit careful. We have to be very — you know, we have to do things and we have to do them prudently. But that doesn’t mean we’re going to stop industry because, you know, as we work on the next 10 years about how to destroy it. So, I’m with you all the way. I didn’t even know how you felt about it. And I assumed you felt the same way as me.

黄仁勋：是的，先生。Jensen Huang: Yes, sir.

特朗普总统：如果我们要领先——我常说，谁赢得了人工智能，谁就赢得了一切。人工智能就是如此重要。它比互联网的影响还要大。谁赢得了人工智能，谁就赢得了一切。我们不能让这种事情发生。而这在很大程度上也包括数据中心。有些社区原本正在衰败，如今却拥有了数据中心。现在，它们成了富裕社区，非常富裕的社区。President Trump: And if we’re going to lead — and I have an expression, it’s whoever wins AI wins. That’s how big it is. It’s bigger than the internet. And whoever wins AI wins. And we can’t let this kind of stuff happen. And that includes very much includes data centers. There are communities that were dying that have data centers right now. And now they’re wealthy communities. Really wealthy communities.

黄仁勋：我们将确保美国人工智能竞赛中的每个人都能获胜。每个行业、每家公司、每个州、每一个人。Jensen Huang: We’re going to make sure that everybody wins in the AI race in America. Every industry, every company, every state, every people.

特朗普总统：好的。我对此有强烈的看法，而且我所处的位置能够对此采取行动。我们不会让那种事发生。所以，我不知道参加会议的都有谁。我根本不知道自己到底在和谁说话，不过我会看看。President Trump: Good. Well, I feel strongly about it and I have the position that can do something about it. We’re not going to let that stuff happen. So, I have no idea who’s at the meeting. I have no idea who the hell I’m talking to, but I’ll see.

黄仁勋：您听到了吗？成千上万的人正在为您鼓掌，先生。Jensen Huang: Did you hear that? Thousands of people are clapping for you, sir.

特朗普总统：我所知道的是，如果你们在那里听Jensen的讲话，他做得非常出色，David也做得非常出色。祝大家好运，我们将继续着眼于未来。这个国家从未有过如此出色的表现。我们有2万亿美元的投资正在进入这个国家，而在昏昏欲睡的乔·拜登执政期间，远远不到1万亿美元，而且那还是四年的总额。这是短短一年内的情况。所以，你们知道，这个国家从未见过这样的局面，我们会让它继续下去。非常感谢大家。President Trump: All I know if you’re there to listen to Jensen, but he’s done an amazing job and David has done an amazing job and good luck to everybody and we’re going to stay with the future. The country has never done better. We have 20 trillion dollars of investment coming into the country and that’s as opposed to much less than 1 trillion under sleepy Joe Biden and that was for four years. This is in one year. So, you know, the country has never seen anything like it and we’re going to keep it going. And so, thank you all very much.

黄仁勋 / 主持人：谢谢您，总统先生。我稍后再给您回电话。谢谢您，总统先生。Jensen Huang / Hosts: Thank you, Mr. President. I’ll call you back later. Thank you, Mr. President.

通话之后：安全、就业与数据中心After the Call: Safety, Jobs, and Data Centers

主持人：我还以为那只是个噱头。你知道那件事正在发生吗？Host: I thought it was a bit. Did you know that was happening?

黄仁勋：不，那是真的。我一开始还以为只是开玩笑，当时我说，把他切到免提上。Jensen Huang: No, it was real. I thought it was a bit at first when I was like, put him on speakerphone.

主持人：他会给你打电话。他怎么会——他会在夜里的任何时候给你打电话，对吧？那次我们在椭圆形办公室，他打电话时你正在睡觉。你睡着了，然后他就说，把他叫醒。Host: And he calls you. How do you — he calls you any hour of the night, right? We were in the Oval that time when he called you sleeping. You were asleep and he like said, wake him up.

主持人：我感觉糟透了，因为他就像是在问：“谁要来参加这顿晚餐？”然后我们把名单过了一遍。他说：“那 Jensen 呢？”我说：“不行，先生。他在度假。”因为他不得不把这次假期推迟了五年。然后他说：“把他叫到电话上来。”什么叫度假？Host: I felt so bad because he’s like, who’s coming to this dinner? And we go through the list. He’s like, well, what about Jensen? I said, no, sir. He’s on vacation. Because he had to postpone this vacation for five years. And he’s like, get him on the phone. What’s vacation?

但你为什么认为他能看穿这个骗局？这实在是一件非同寻常的事。民调支持率跌到了负80。所以，换作任何其他坐在椭圆形办公室的人，都会去做受欢迎的事。你代表的是人民。这是所有人都想要的。他们想关闭数据中心和人工智能。这似乎是当下受欢迎的做法。但他说这是一个骗局，还直截了当地称其为骗局。他是怎么做到的？But why do you think he sees through the hoax? It’s quite an extraordinary thing. It’s polling minus 80. So for anyone else that’s sitting in the Oval Office, you’re going to do what’s popular. You’re representing the people. This is what everyone wants. They want to shut down the data centers and AI. It seems to be the popular thing in the moment. But he says it’s a hoax and he calls it. How does he do that?

黄仁勋：我得告诉你，我不确定。原因在于，很多人都信以为真。所以，事实是，情况很复杂。你知道，起初，如果你看看这些报道，它们都基于两点。第一点，它们最初是以国家安全为依据的。而最近，这一依据被彻底击碎了。所以，这个说法不再以国家安全为依据，而是以安全为依据。Jensen Huang: I got to tell you, I’m not sure. And the reason for that is because a lot of people are falling for it. And so the fact of the matter is it’s complicated. You know, at first, if you look at the stories, it’s all anchored on two things. The first thing that it was anchored on was national security. And recently, that was all blown to bits. And so that story is no longer anchored on national security. Now it’s anchored on safety.

现在，如果你希望 AI 是安全的，首先我们需要确保开发它的实验室能够掌控局面，并且对它们进行良好的测试。如果我们希望有第三方来确保第三方评估机构确实可用——这和财务控制没有什么不同。你们都知道我们有审计师。而审计师其实——他们不必像我们一样精通我们的业务，但只需要提出正确的问题。我听有人说，拥有独立的审计师或评估人员是件好事，但必须有多个。我也同意这一点。正如评估人员和审计师需要有多个一样，这能确保不会有一家公司变得偏执，或因任何原因受到某种影响。所以，你可以通过许多不同的方式来解决这个问题。因此，我认为最重要的是：让我们安全地构建这项技术，确保对它的测试也是安全的。我完全认识到，我们正在构建的东西非同寻常。但这些都是非同寻常的公司，我们应当用非同寻常的标准来要求它们。而且它们也愿意接受这样的标准。Now, if you want AI to be safe, the first thing is we need to make sure that the labs that are building it are in control, that they’re good tests for them. If we would like to have third parties to make sure that third-party evaluators are available — that’s no different than financial control. You guys know we have auditors. And the auditors are quite — they don’t have to be as expert as we are in our business but they just have to ask the right questions. And I heard somebody say that it’s good to have independent auditors or evaluators but they just have to have multiple. I agree with that too. Just as there’s multiple evaluators and auditors, it makes sure that one company doesn’t become, you know, pilled or somehow influenced for whatever reason. And so there’s a lot of different ways that you could solve this. And so I think the number one thing is let’s build the technology safely. Let’s make sure that the testing of it is safe. And I recognize completely that what is being built is extraordinary. But these are extraordinary companies and we ought to hold them to extraordinary standards. And they want to be.

主持人：我想再回到开源这个话题上来。我们一年前还没有真正重视它。当时落后了两年，也就是18个月。Host: I wanted to go back to open source for a second. A year ago we weren’t taking it very seriously. It was two years, 18 months behind.

黄仁勋：正如各位所知，和特朗普总统通电话时，挑战之一就是很难说点什么。我这么说肯定会惹上麻烦。我敢肯定，他会打电话来找我算账。不过，无论如何，我本来想告诉他以及在座各位的是，人工智能正在创造数量庞大的就业岗位。政府执政初期，以及我第一次给他打电话、第一次与他见面时，他最想要的就是在美国创造就业机会。他希望让美国重新实现工业化。他希望确保美国拥有足够的能源，以支撑下一次工业革命。没有能源，就没有工业增长。因此，他希望确保能源增长、就业增长，并推动供应链重新实现工业化。Jensen Huang: The one thing that — as you guys know — one of the challenges when you’re on the call with President Trump is hard to say something. I’m going to get in trouble for that. I’m sure he’s going to call me up on that. But anyhow, what I was going to tell him and all of you is that AI is creating an enormous number of jobs. The thing that he wanted more than anything at the beginning of the administration and my first phone call with him, my first time I met him, is that he wants to create jobs in America. He wants to re-industrialize the United States. He wants to make sure that United States has the energy to support the next industrial revolution. Without energy, there’s no industrial growth. And so he wants to make sure that there’s energy growth, that there’s job growth, that they’re re-industrializing the supply chain.

看看我们现在正在做的一切。所有这一切都正在此时此刻发生。我们创造的就业岗位比以往任何时候都多。我们正在创造软件行业的就业岗位。我们刚才还谈到，最近仅仅六个月，就有4000亿美元的风险投资流入人工智能行业。嗯，这创造了大量就业岗位。当然，这也创造了巨大的算力需求，我对此感到高兴；同时还带来了大量的数据中心需求，而这正是我们应该谈论的话题。我曾与得克萨斯州州长阿博特交谈过，他希望呼吁整个行业，在我们在美国各地建设数据中心时，要体恤当地的小型社区——也就是说，要成为更好的倾听者。Look at everything that we’re doing right now. All of it is happening right now as we speak. We’re creating more jobs than ever. We’re creating software jobs. We were just talking about earlier: $400 billion of venture financing went into the AI industry just recently — six months. Well, that’s created a ton of jobs. It’s created, you know, obviously enormous amount of demand for compute which I’m happy about, which is also creating a lot of demand for data centers and we ought to talk about that. I was talking to Governor Abbott of Texas and he wants to appeal to the industry to make sure that we are empathetic to the small communities as we’re building data centers all across America — just to be better listeners.

英伟达的资本配置与生态系统战略Nvidia’s Capital Allocation and Ecosystem Strategy

主持人：我们不妨花点时间真正谈谈这一点。如果拆解英伟达的各个组成部分，你会发现，实际上，为了让整个生态系统运转起来，你不得不成为“AI银行”，而且必须在各个层面都做到这一点。比如，你刚刚与 Cloverleaf 合作开展了土地、电力和机房主体业务。你还与贝莱德、高盛以及其他众多机构做了一件很棒的事，从根本上创造了融资能力。请谈谈你的资本配置策略——要让更广泛的生态系统建立起来，让各方能够参与进来并为下一阶段提供承销，需要发生些什么？Host: Let’s actually talk about that for a second. What’s incredible about Nvidia if you break down the component parts is you’ve effectively had to become the bank of AI to get the ecosystem going and you’ve had to do it at all the levels. You know, you just did this thing with Cloverleaf where you’re doing land, power, shell. You did this great thing with BlackRock and Goldman and all these folks to essentially create the financing capability. Walk us through your capital allocation strategy — like what has to happen to get a broader ecosystem, folks to be able to come in and underwrite this next phase.

黄仁勋：嗯，正如大家所知，我们正在创造一个新的工业革命，而它的每一个方面都是真实存在的。这个新产业需要制造业，就像电力、互联网，以及如今的人工智能一样。我们可以为任何事物提供动力，可以找到任何东西。现在有了人工智能，我们可以提出任何问题，并获得答案。不是这样吗？所以，这就是我们的未来。我们接入以太，想问什么都可以问，而它能够向我们解释。Jensen Huang: Well, we’re creating — as you guys know — this is a new industrial revolution and every aspect of it is true. This new industry requires manufacturing just as electricity, internet, and now AI. We power anything, we can find anything. Now with AI, we can ask and know anything. Isn’t that right? And so that’s our future. We tap into the ether and we can ask it of anything we want and it could explain it to us.

现在，为了实现这一点，它必须生成智能。因此，这是一个生产过程，也正因如此，必须建设这些基础设施。但一旦基础设施建成，问题就变成了：美国各地其他所有层面又该如何？这个行业并不只是关于模型，也不只是关于芯片。它主要关乎上层应用，也主要关乎基础设施层——数据中心以及所有基础设施，包括建设、电力和发电——这一切都涉及其中。Now, in order for that to happen, it’s got to produce the intelligence. And so that’s a production process which is the reason why this infrastructure has to get built. But once you get the infrastructure built, the question is what about all of the other layers across the United States? This industry isn’t just about the model. It’s not just about the chips. It’s mostly about the applications on top. It’s mostly about the infrastructure layer, the data centers and all the infrastructure — the construction, the electricity, the power generation — all of that is involved.

因此，我会审视整个生态系统，寻找瓶颈；如果有些地方正在孕育非凡的公司——也就是存在约束，限制着非凡公司的发展——那么，也许需要扩大供应链规模，这样等我们准备好部署算力时，它们也能为我们准备好：土地、电力、数据中心主体建筑。归根结底，这与审视上游供应链没有什么不同。相比大多数人，我可能会更多地思考长期供应链，因为我们的公司规模非常庞大，而要想取得成功，就必须有一大批公司为我提供支持。你知道，康宁必须支持我——康宁的温德尔必须支持我——还有 Lumentum，当然也包括 TSMC 以及各家存储器公司。因此，早在这场革命发生、增长到来之前，我们就已经开始与所有这些公司合作，以便让增长能够实现。现在，我正在下游做同样的事情。And so I look across the entire ecosystem and look for bottlenecks and if there are places where extraordinary companies are being built — constraints — extraordinary companies being built, maybe it’s supply chain that has to get scaled up so that when we’re ready to deploy compute that they’ll be ready for us: land, power, shell. And so this is no different than looking at the supply chain upstream. I probably think about the long-term supply chain more than most because our company’s really large and in order for us to succeed, a whole bunch of companies has to support me. You know, Corning has to — Wendell at Corning has to support me, Lumentum, and TSMC of course and memory companies. And so we started working with all of these companies long before the revolution, before the growth came, so that the growth could happen. Now I’m doing it downstream.

向技术栈上层发展、NeoClouds 与开放模型Going Up the Stack, NeoClouds, and Open Models

主持人：竞争周期往往是这样的：随着时间推移，经过很长一段时间，收益通常会沿着技术栈向上移动，直抵应用层，在那里你可以在更长时期内获得超额收益。我的意思是，你们收购了 Hugging Face——现在你们某种程度上已经积极涉足服务业务了。我的意思是，像 OpenRouter 这样的产品似乎很自然地就会成为你们要做的事情。显然，构建类似 Bedrock 的产品有更好的方式。我相信你们也在考虑这些。自然的结论是什么？因为看起来，处在上层的那些公司完全不介意尝试向下扩展。而你们拥有最强的资产负债表、这些极其出色的工程师，以及已被验证的经验，能够把事情做好、把产品开发出来并推向市场。所以，你们如何看待抬头望向上方，并说：“我大概也能做到这一点？”Host: The competitive cycle tends to be though that the earnings over time over long stretches of time tends to move up the stack right towards the application layer where you can over-earn for larger periods of time. I mean you bought Hugging Face — now you’re sort of actively in the serving business. I mean it seems pretty natural that products like OpenRouter make a lot of sense. It seems pretty obvious that there are better versions of ways to build things like Bedrock. I’m sure you think about it. What’s the natural conclusion? Because it seems like the folks up here have no issue trying to move down. And you have the best balance sheet, these incredible engineers, and you have the proven experience to make it right and engineer the product and get it out. So how do you think about looking up and saying, I could probably do that?

黄仁勋：英伟达之所以能够运行全球每一个模型——我们是唯一一家——这太不可思议了。大约一年前，也就是一年半以前，我们运行的唯一模型是 OpenAI 的。而现在看看——真是令人惊叹。Meta 的模型也可以使用了。Grok 也已经可以使用了。Grok 非常出色。我们现在也运行 Gemini。而 Anthropic 也在我们的平台上不断扩大规模。从一年半以前开始，现在所有这些前沿 AI 模型都已经开放并可供使用了。所以，模型的数量正在不断增长——还有一大批公司也在构建前沿模型，我就不点名了。AI 实验室的数量也在增加。Inflection 这类公司、Reflections，还有很多很多。Physical Intelligence——这样的例子不胜枚举。Jensen Huang: The reason why Nvidia runs every single model in the world — we were the only — it’s incredible. Last year about a year and a half ago the only thing we ran was OpenAI. And now look — amazing models are available. The Meta ones are available. You’ve got Grok is available. Grok’s incredible. We now run Gemini. And Anthropic is scaling up on our platform as well. Since a year and a half ago, you got all these frontier AI models that are now open that are available. So the number of models that are growing — there’s a whole bunch of companies that I won’t mention that are building frontier models as well. And the number of AI labs are growing. The Inflection-types, the Reflections, the list goes on. The Physical Intelligence — the list goes on.

好的。所以，所有这些实验室都是建立在 NVIDIA 之上的。之所以这样做，是因为作为一家公司，我更希望我们帮助每个人取得成功，而不是从中分一杯羹。因此，我们会尽可能向上做到所需的程度，但同时尽可能保持在较低的位置。我们的战略就是：在满足需求的前提下尽可能向上延伸，同时尽可能保持在较低的位置。之所以这样做，是因为如果不是 NVIDIA 创建了 cuDNN，所有这些框架都不会存在；如果不是我们创建了 Megatron Core，也不会有如今的大规模训练。因此，我们会发明所需的全部技术，做到必要的程度，然后让百花齐放。所以，这种姿态坦率地说让我们成为唯一的——Okay. And so all of these labs are building on NVIDIA. And the reason for that is because as a company, I rather for us to help everybody succeed instead of taking a slice out. And so we would go up as far as we need to but as low as possible. Our strategy is go up as far as we need to and as low as possible. And the reason for that is because if I do that — if not for Nvidia creating cuDNN, all of the frameworks wouldn’t exist. If not for us creating Megatron Core, then all of the large-scale training wouldn’t have happened. So we invent all the technology necessary as far as we need to and then we let a thousand flowers bloom. And so that posture allows us to be quite frankly the only —

主持人：好吧，说实话，我同意你的看法。反对意见可能是，在超大规模层面拥有更多竞争确实会很棒。而且我认为你在支持新兴云服务商方面做得非常出色。确实有一些。顺便说一句，我想是你把我介绍给了 Nscale——太棒了，一切都很出色。他们令人惊叹。但我们需要大约 50 家这样的公司。我们需要 100 家。我们需要 1,000 家。而这可能需要一些——Host: Well, look, let’s be honest that I agree with you. The pushback would be that it really would be great to have more competition at the hyperscale layer. And I think you’ve done a great job supporting the NeoClouds. There are some. And by the way, I think you introduced me to Nscale — superb, great, everything. They’re amazing. But we need like 50 of these guys. We need a hundred of them. We need a thousand of them. And it just may take some —

黄仁勋：是的。你知道，我出奇地不争强好胜。这不是我的风格。你知道，我的想法是，比如说，有五家超大规模云服务商我就已经很满意了。不过，我之所以注意到所有 NeoClouds，也就是我们所说的 NCPs 的早期客户，都是超大规模云服务商，原因在于超大规模云服务商一年做一次规划，但当前的市场动态实在太不稳定，所以他们几乎总是判断错误。因此，面对这些灵活、行动迅速的区域云服务商，他们了解自己所在的州、国家或地区，能够以一种身处西雅图或帕洛阿尔托的人很难纵观全球的方式，拿下土地、电力和机房外壳。Jensen Huang: Yeah. You know, I’m surprisingly uncompetitive. That’s not my thing. You know, my thing is kind of like for example, I’d be more than happy with five hyperscalers. However, the reason I noticed the early customers of all the NeoClouds, all the what we call NCPs — all the early customers were the hyperscalers. And the reason for that is because the hyperscalers plan once a year, but the market dynamics is so volatile right now that they’re always almost wrong. And so with all these regional clouds who are agile and they can move fast, they know their state or they know their country, they know their region, they’re securing land, power and shell in a way that’s hard for somebody who sits in Seattle or sits in Palo Alto to be able to see the planet.

因此，我们现在基本上已经拥有了一个大规模的分布式企业网络，这些企业正在为我们建设基础设施、确保土地和电力供应，并提供机房空间。如今，各国都意识到这具有战略意义。因此，许多国家都在说：我要掌控自己的电力，只把它提供给本国企业。Nvidia 在那个国家也有业务，我们可以帮助该国的 NeoClouds 发展壮大。 所以，无论是澳大利亚的 Fermion——我们刚刚在澳大利亚完成了一大批工作，新增了两吉瓦的容量——还是东南亚，当然还有 Yotta 等公司新增的几吉瓦容量。我们正在建设数吉瓦级的容量，并不断扩大规模。And so we now have basically a large-scale distributed network of companies that are building, securing land, power, shell for us. And now countries realize it’s strategic. So many countries are saying I’m going to take my power and only give it to my own companies. Well, Nvidia is in that country as well and we could help the NeoClouds in that country grow. And so whether it’s Fermion in Australia — we just did a whole bunch of stuff in Australia — brought on two more gigawatts. Southeast Asia of course, Yotta and others bring on a few gigawatts. And so we’re building gigawatts. We’re scaling up.

主持人：不过，这一点还是很清楚的。我只是想把这一点说清楚。很明显，你们正在向高层次发展，并且高度聚焦于开源。当然，你们的 Nemos 做得非常出色。我经常使用它们。Hugging Face、Poolside 和 Lagoon——这些都是非常、非常扎实的产品，而你们现在正在通过“收购人才”来招募他们，或者不管这种做法叫什么。然后，你们用于自动驾驶的开源技术栈也极具颠覆性。所以——Host: It’s pretty clear, though. I just want to get this one thing in. It’s pretty clear that you’re going pretty high up and getting very focused on open source. Obviously, you have your Nemos doing exceptionally well. I use them often. Hugging Face, Poolside and Lagoon — very, very solid product that you’re now aqua-hiring, hiring, whatever it is. And then you have your open source stack for self-driving also very disruptive. So —

黄仁勋：我们是五个领域的前沿模型。Jensen Huang: We are the frontier model in five domains.

主持人：你似乎不是为了拿银牌才打造产品的。你似乎总是奔着金牌去的。那么，你是在冲击金牌吗？你会拥有无可争议的最佳开源模型吗？这件事的第二部分是：开源模型能否追赶上前沿模型，而你是否是能做到这一点的人？Host: You don’t seem to build products to get the silver medal. You seem to go for the gold. So are you going for the gold? And will you have the best hands-down open-source model? And then part B to that is can open source catch up to frontier models and are you the person to do it?

黄仁勋：所以，Jason，逻辑是这样的：我们会去做这件事，一方面是因为我们能做到——我们具备完成这件事的技能——另一方面是因为客户需要我们这么做。比如，Alpamayo 是世界上首款具备思考能力的自动驾驶汽车。所谓具备思考能力、能够推理，就是你不需要那么多数据——不必用几十亿小时的道路数据进行训练，因为你可以进行推理。把问题拆解开来：我以前见过这种情况。它不完全相同，但总体上与那种情况大致相同。明白吗？那么，为什么 Alpamayo 是必要的？世界上有大量汽车公司。世界上的每一辆车最终都将实现自动驾驶，但除此之外，每种农业技术设备、每辆卡车、每辆厢式货车也都将实现自动驾驶，而其中大多数企业的规模还不足以自行构建完整的技术栈。所以，我会为它们打造一套非凡的技术栈。它们只需针对自己的应用场景进行最后一公里的适配。未来，一切会移动的东西都可能实现自主运行。Jensen Huang: So the logic, Jason, is that we will build it because one, we can — we have the skills to do it — and because our customers need us to do it. So, for example, Alpamayo is the world’s first thinking self-driving car. And by thinking, by reasoning, you don’t need as much data as — you don’t have to train on a few billion hours of road data because you could reason about it. Break down the problem into: I’ve seen this before. It’s not exactly the same, but it’s largely the same as that. Okay? And so Alpamayo — why is it necessary? Well, there’s a whole bunch of car companies. Every car in the world is going to be autonomous, but beyond that, every ag tech, every truck, every van, and most of them aren’t big enough in scale to be able to build that whole stack. So I’ll build an extraordinary stack for them. They do last-mile adapting for their application. Now, everything that moves in the future could be autonomous.

如果不是我们构建了其中一些生物学模型，这个世界就不会拥有它们。ESM-2 蛋白质基础语言模型——那是我们创建的。ESMFold、OpenFold、AlphaFold 2——所有那些采用 SE(3)-等变技术的东西——所有这些技术，如果不是我们构建它们，根本不会存在。我最喜欢的项目之一是 Proteina Complexa，它正在合成下一代蛋白质及其结合结构——这是我们打造的突破性成果。因此，我们会去构建这些东西，因为礼来需要它，默克需要它，其他公司也需要它，而他们没有能力做到，或者还没有发展到那个阶段，所以我们可以作出真正的贡献。所以，我做的一切都是出于需求。我并不是想要颠覆什么。我们早上醒来时，并不是想要颠覆任何人。我们只是早上醒来，努力帮助每一个人。If not for us building some of the biology models, the world wouldn’t have it. The ESM-2 protein foundation language model — we created that. ESMFold, OpenFold, AlphaFold 2 — all the stuff with SE(3)-equivariant — all of that stuff, technology wouldn’t have existed if we didn’t build it. One of my favorites, Proteina Complexa, is synthesizing next-generation proteins and its binding — it’s groundbreaking stuff we built. And so we’ll build that because Lilly needs it and Merck needs it and others need it and they don’t have the capability to do it or they’re not yet there and so we can make a real contribution. So I do everything out of need. I’m not trying to disrupt. We don’t wake up in the morning try to disrupt anybody. We just wake up in the morning try to help everybody.

竞争、中国光刻、AGI 与超级智能Competition, China Lithography, AGI, and Superintelligence

主持人：Jensen，关于可能正在对你们核心业务构成威胁的竞争对手，你怎么看？能否请你谈谈？我只是想听听你的看法。你怎么看埃隆宣布的这座占地 1 亿平方英尺的设施？Host: Jensen, what about competitive threats that might be emerging to your core business? Can you just comment? I actually want to just get your take. What’s your take on the 100 million square foot facility Elon’s announced?

黄仁勋：如果有人能做到，那就是他。我和他曾一起乘飞机前往一个国家——同行的还有一个有时会给你打电话的人。那是一架很不错的飞机，我们聊了很多——你知道，埃隆喜欢谈这些事情——所以我们花了很多时间讨论这件事。Jensen Huang: If anybody could do it he can. And the two of us were on a flight together to a country — with a person who sometimes calls you on the phone. It was a nice plane and we had like — you know Elon likes to talk about these things — and so we spent a lot of time talking about it.

主持人：我的意思是，任何人都能做到这一点吗？因为你们是设计芯片的，并不负责制造。你们的芯片能在那里制造吗，还是说——Host: Is that anybody could do it because I mean you design chips, you don’t fab them. Could your chips be fabbed there or is it —

黄仁勋：嗯，我们对工艺技术了解很多，因为我们正在突破一切的极限。而且你知道，由于我们的扩展规模如此之大，公司内部拥有令人难以置信的内存技术。我们是全球最优秀的系统公司。你知道，我们有很多令人惊叹的技术——我们可以聊聊这些。而且你无法阻止埃隆去做这件事，这是他的——这就是他的超能力；一旦他决定要做某件事，就很难阻止他。Jensen Huang: Well, we know a lot about process technology because we’re pushing the limits of everything. And you know because we scale at such large scale we have incredible memory technology inside the company. We’re the world’s best systems company. You know we got lots of amazing — we could talk about it. And you can’t discourage Elon from doing it which is one of his — that’s his superpower and once he decides to go do something it’s hard to stop him.

主持人：您能谈谈您对中国先进光刻系统现状的看法吗——这些系统是本土研发的吗？Host: And can you give us your take on where China is with advanced lithography systems — native grown?

黄仁勋：他们将在 2030 年达到那里。Jensen Huang: They’re going to get there by 2030.

主持人：到 2030 年。Host: By 2030.

黄仁勋：是的。而且 2030 年转眼就到了。还有，到那时我们大家都已经死了这么久了。所以——Jensen Huang: Yeah. And 2030 is just around the corner. Also, that’s how long we’ll all be dead at that time. So —

主持人：那么对于中国来说，这是否意味着开关一拨动，所有这些就会几乎立即进入中国大陆的晶圆厂？Host: And for China, does that mean the switch is flipped and then that’s all going to go into mainland fabs almost immediately?

黄仁勋：你知道，理解中国的方式是，他们非常擅长大批量生产。而这只是时间问题。我认为还要以几十年为尺度来看——你知道，我已经在这个行业很长时间了，对于英伟达来说，我必须思考下一个十年以及再下一个十年会发生什么。所以两三年只是弹指一挥间，根本算不了什么。因此，就他们而言，他们已经在那里了。Jensen Huang: You know, the way to think about China is they’re really good at high-volume production. And this is just a matter of time. And I think in decades as well — you know I’ve been around a long time and for Nvidia I’ve got to think about what happens next decade and decade after that. So two or three years is just a click. It’s nothing. And so as far as they’re concerned they’re already there.

主持人：Jensen、Elon 和 Gwen——我们得治理美国。我们得快跑。速通。放慢脚步绝对是错误的策略。Host: Jensen, Elon, and Gwen — we’ve got to run America. We got to run. Speedrun. Slowing down is definitely the wrong strategy.

嗯，我的意思是，我想我们行业里的大多数人都明显感觉到，我们似乎正处于 AGI 时刻。而显然，这个定义就是和任何其他人一样聪明。Well, I mean it feels apparent, I think, to most of us in the industry that we’re kind of in the AGI moment. And it’s a definition obviously just as smart as any other human.

黄仁勋：我认为我们已经做到了。Jensen Huang: I think we’re already there.

主持人：我们已经到那里了，对吧？所以，根据你所看到的情况、你的客户群以及你在这里的发展历程，超级智能就是下一个里程碑。Host: We’re there, right? And so then superintelligence is the next waypoint based on what you see, based on your customer base, based on your history here.

黄仁勋：但是，Jason，我认为我们也已经达到那个阶段了。Jensen Huang: But Jason, I think we’re there too.

主持人：你觉得我们已经进入超级智能时代了吗？Host: You think we’re at superintelligence?

黄仁勋：是的。当你限定在一个狭窄的领域——比如我的自动驾驶汽车，我不需要你给我做煎蛋卷，我只希望你开车。这就是超智能。它比人类更好——事故率只有人类的十分之一。合成蛋白质、对蛋白质进行虚拟筛选——我们已经做到了。Jensen Huang: Yeah. When you take a narrow segment — I mean my self-driving car, I don’t want you to make me an omelette, I just want you to drive the car. That is super intelligent. It’s better than a human — one-tenth the accident rate. Synthesizing proteins, doing virtual screening of proteins — we’re already there.

主持人：站在人类发展的前沿，你觉得有趣吗？Host: Are you having fun being on the frontier of humanity?

黄仁勋：我喜欢。我喜欢。伙计们，那里的情况很棒。未来很美好，我们都想抵达那里。听着，骑上自行车。我们很多人其实不必工作。但我得告诉你们，未来太美好了，我们不能不去实现它。所以我希望每一个——我希望自己能在那里。我希望你们所有人都和我一起在那里。我们作为人类，将共同取得巨大成功。同时，我们必须鼓励他们，推动他们继续前进。正如你们所知，他们正在做非常、非常重要的工作。我希望他们能够成功。我也非常希望我们能降低戏剧性，更重要的是，我们需要让整个美国都和我们一起前进。只有这样，我们才能成功。Jensen Huang: I like it. I like it. And guys, it’s great there. The future is great and we want to get there. Listen, ride the bike. A lot of us don’t have to work. But I got to tell you, it’s too good not to be. And so I want every — I want to be there. I want all of you guys there with me. We’re all going to be there. We’re going to be enormously successful together as a humanity. And in the meantime, we got to encourage them, urge them on. They’re doing really, really important work as you guys know. And I want them to succeed. I also would love for us to tone down the drama and most importantly, we need all of America to come with us. That’s how we make it.

主持人：女士们、先生们，让我们欢迎黄仁勋。Host: Ladies and gentlemen, Jensen Huang.

黄仁勋：谢谢你，兄弟。很感激你。谢谢。Jensen Huang: Thanks, man. Appreciate you. Thank you.

主持人：太棒了。真是精彩。Host: That was awesome. That was great.