AI 对算力的无尽需求：瓶颈与地缘政治动态

在我的 AI 职业生涯中，我还没遇到过一个觉得自己算力够用的人。

In my career working in AI, I have yet to meet a single person that ever felt like they had enough compute.

我找不到更好的嘉宾了，Andrew Angie，全球公认的 AI 领袖。

I could not ask for a better guest, Andrew Angie, globally recognized leader in AI.

数据中心是构建数字经济的关键基础设施。我认为开放权重模型是地缘政治影响力的巨大来源。职业道德、速度。当中国政府做出全国性承诺时，那是全产业的承诺。这实际上是一股非常强大的力量，我不会低估它。

Data centers are the critical infrastructure for building the digital economy. I think that open way models is a tremendous source of geopolitical influence. The work ethic, the velocity. When China's government makes an all nation commitment, it's an all industrial commitment. That's actually a very powerful force that I wouldn't underestimate.

准备好了。Andrew，我仰慕你很久了，一直很期待这次对话。所以，谢谢你。非常感谢你今天能来。

Ready to go. Andrew, I've been an admirer for a long time, so I've been really looking forward to making this happen. So, thank you. Thank you so much for joining me today.

不，谢谢你 Harry。我看了很多节目。我很喜欢你最近和我朋友 Martin Casado 的那期，非常难忘。所以真的很激动能来这里。我爱 Martin，他是个非常特别的人。

No, yeah, thank you Harry. I watched a bunch of shows. I really enjoyed your recent one with my friend Martin Casado as well. That was very memorable. So actually thrilled to be here. I love Martin. Very very special man.

我想从你之前说过的话开始。你说 AI 是新的电力，当我想到电力以及我们今天的处境时，我想了解瓶颈所在，似乎每个人都认为瓶颈在于数据、算力和算法。这是我们应该考虑的三个方面吗？如果是，你认为哪个是最大的瓶颈？

I want to start with something that you've said before. You said AI is the new electricity and when I think about electricity and where we are today, I want to understand the bottlenecks and everyone seems to suggest that it really is about data, compute and algorithms. Is that the three parameters to which we should think about bottlenecks? And if so, which one do you think is the biggest bottleneck?

我认为目前最大的两个瓶颈……可能是……电力是其中之一。在美国，我确实担心许多数据中心运营商被许可审批卡住，我知道当地社区的支持很重要，有些人不想在那里建数据中心，但就像我们曾经为某一代人修建公路和铁路一样，数据中心是构建数字经济的关键基础设施。所以美国和一些西方国家缺乏电力是个问题。相比之下，我看到中国在左右开弓地建发电厂，包括核电站。所以这将是一个有趣的动态。另一个瓶颈是半导体。但 AI 非常复杂，我认为我们还需要更多数据，还需要更好的算法。所有这些都值得努力，但短期内，电力和半导体存在一些限制。

I would say the two biggest bottlenecks right now... it may be... I think electricity is one of them. So in the US, I am honestly worried that many data center operators are stuck in permitting, and I know that local community support is important and some people don't want a data center there, but once we built roads and railways as the infrastructure for a certain generation, data centers are the critical infrastructure for building the digital economy. So lack of electricity in America and in a number of western countries is a problem. In contrast, I see China building power plants left and right, including nuclear. So that will be an interesting dynamic. And then semiconductors is another bottleneck. But AI is so complicated. I think we also need more data, we also need better algorithms. All of it is worth working on, but in the short term, some constraints with electricity and semiconductors.

你能谈谈你认为最紧迫但大多数人没有意识到的半导体限制吗？

Can you talk to me about the constraints around semiconductors that you think are most pressing that most people don't realize?

首先，在我的 AI 职业生涯中，我还没遇到过一个 AI 从业者觉得自己的算力够用。所以，给我们多少算力，我们都会用光，然后说还不够。这已经是过去 20 年左右的约束了。但我看到的是，随着生成式 AI 的兴起，出现了非常有价值的工作负载。例如，AI 辅助编程，它太棒了，让我们生产力大大提高。但如果你用 Claude Code 太多，有时会受到很大限制，我发现许多公司都有过剩的需求，这是一个非常罕见的问题。这么多人都想要更多推理，想要生成更多 token，而我们就是没有足够的半导体和数据中心电力来满足需求。但 AI token 生成可以做很多事情，当我们无法向需要的人提供足够供应时，这很令人沮丧。在需求端，如果你用太多，就会受到很大限制。

First, in my career working on AI, I have yet to meet a single AI person that ever felt like they had enough compute. So, get us any amount of compute, we will use it all up and say we still don't have enough. So this is a constraint for the last 20 years or so. But what I'm seeing is with the rise of Gen AI, there are very valuable workloads. For example, AI assisted coding, it's fantastic. It's making us so much more productive. But if you use Claude Code enough, sometimes you get very limited, and I find that many companies have excess demand, which is a very rare problem to have. So many people want more inference, want more tokens generated, and we just don't have the semiconductors and the data centers electricity to meet the demand. But there's a lot we could do with AI token generation, and it's frustrating when we can't supply enough to people that want it. On the demand side, you get very limited if you use too much.

我该如何理解这种对更多算力的无止境需求以及随之而来的改进，同时认识到许多人说 GPT-5 是缩放定律已达到某种程度的例子，并且对效率的关注已成为一种转变？我该如何平衡这两种看似不同的观点？

How should I think about that insatiable need for more compute and the improvements that come from it with the recognition that many people say GPT-5 was the example that scaling laws have been reached to a certain extent and a focus on efficiency has been a transition. How should I balance these two seemingly differing opinions?

确实，token 生成正变得更高效、更便宜。事实上，看看 OpenAI 的开放权重模型，他们发布的模型运行起来非常高效。所以我认为他们做得很好……是 120 加 200 亿参数还是什么，有 57 亿活跃参数？所以运行起来确实非常高效。但尽管 token 生成成本在下降，我们的需求却是无止境的。AI 中发生的一件有趣的事情是，如果我们看价值桶在哪里，其中一个大的价值桶是 AI 辅助编程。我认为这可以追溯到更早的时代。在上一代，我认为谷歌主导了水平信息发现，比如网络搜索，但在互联网建设过程中，有很多垂直领域的机会。所以我们有了旅游垂直领域如 Expedia，一批人在零售领域竞争，另一批人在交通、社交媒体等领域竞争。我们现在看到的是 ChatGPT 拥有如此强大的消费品牌，ChatGPT 似乎是新一代水平信息发现的主导者，尽管我认为 Gemini 通过控制 Android 和 Chrome 的渠道优势也是一个重要的参与者。但如果水平信息发现就是这样，那么仍然有大量空间来构建许多垂直领域。其中一个明确且非常有价值的垂直领域是 AI 编程辅助，Claude Code 就是……我每天都在用，很喜欢。OpenAI 的 Codex 也很有势头。但它显然让开发者生产力大大提高，效率极高，以至于需求爆棚。让我们越来越多地使用它。我觉得兴奋的一点是，我经常把 AI 编程辅助看作其他工作职能可能发生变化的预兆，随着 AI 营销工具、AI 招聘工具、AI 金融工具变得更高效。所以我经常把 AI 编程辅助看作其他领域随着工具改进可能发生变化的预示。

So it is true that token generation is getting more efficient and cheaper. In fact, if you look at OpenAI's open weight model, they actually released models that are very efficient to run. So I think they did a good job with... was it like a 120 plus 20 billion parameters or something with 5.7 billion active? So it's actually a very efficient model to run. But despite the cost of token generation falling, our demand for it is insatiable. One interesting thing that's happened in AI is if we look at where the buckets of value are, one of the big buckets of value is AI assisted coding. And I think this hearkens back to an earlier era. In a previous generation, I think Google came to dominate horizontal information discovery like web search, but there was room for lots of verticals when the internet was being built. So we ended up with travel verticals like Expedia, a bunch of folks fought out in retail, a bunch of others fought out in transportation, social media, and so on. What we're seeing now is ChatGPT has such a strong consumer brand, ChatGPT seems to be the dominant player in the new gen horizontal information discovery, although I think Gemini with its channel advantage through control of Android and Chrome is a serious player as well. But if that's where horizontal information turns out to be, then there's still plenty of room for lots of verticals to be built out. And one of the clear buckets of really valuable verticals is AI coding assistance, where Claude Code is... I use it every day, love it. OpenAI Codex has a lot of momentum as well. But it's clearly making developers so much more productive and efficient that the demand is just through the roof. Let's use more and more of this. One thing I find exciting is I often look at AI coding assistance as a harbinger for what might happen to other job functions as well, as AI marketing tools become more efficient, AI recruiting tools, AI finance tools become more efficient. So I often look at AI coding assistance as a foreshadowing of what may happen to other sectors as the tools get better for them too.

我最近请了 Cohere 的 Joel Pino（前 Facebook 员工）上节目，她说 AI 编程助手在成熟度上可能处于 2016 或 2017 年图像生成的水平。你认为今天的环境公平吗？还是不同意？

I had Joel Pino from Cohere and formerly of Facebook on the show recently and she said that AI coding assistants are in the same place that maybe image generation was in 2016 or 2017 in terms of maturity. Do you think that's a fair state of the environment today or do you not think so?

我不知道。我认为它更先进。我认为 2016 年图像生成还不是特别有价值。

I don't know. I think it's further along. I think in 2016 image generation wasn't super valuable.

我不记得当时有那么大的价值，但我认为今天 AI 编程辅助真的很有价值。在 AI Fund，我的工程负责人最近说：‘我们来考虑一下工具标准化吧。’他基本上说：‘我需要这些工具，除非我死了，否则别想拿走。’我觉得我们的开发者感受非常强烈。我自己再也不想在没有 AI 编程辅助的情况下写代码了。我认为这些工具确实很好用，但还有很大的提升空间。

I don't remember it being that valuable back then, but I think today AI coding assistance is really valuable. At AI Fund, my head of engineering recently said, 'Let's think about standardizing on tools.' And he basically said, 'I need these tools, and you have to pry them out of my cold, dead hands.' I think our developers feel really strongly. I myself never want to have to code again without AI coding assistance. I think the tools are really working well, but there's still a lot of headroom for how much better they can get.

我想回到核心瓶颈：我们说过是关于电力和半导体。当我们看今天数据中心的建设时，正如你所说，监管在很大程度上阻碍了它。你认为特朗普在过去几年从基础设施角度对 AI 在美国的发展是帮助更多还是伤害更多？

I do just want to go back to the core bottlenecks: we said they're about electricity and semiconductors. When we look at the build-out of data centers today, as you said, regulation has been a big part of preventing that in a lot of ways. Do you think Trump has done more to help or to hurt the progression of AI in the United States from an infrastructure perspective over the last few years?

美国联邦政府做了一些好事，也做了一些不太有帮助的事。我觉得清除不必要的监管是一个非常好的举措。甚至去年，两党合作的舒默 AI 洞察论坛——我认为有很多人游说美国政府通过扼杀性的监管。有很多被炒作的 AI 安全叙事说 AI 可能导致人类灭绝，这是一种荒谬的说法，目的是为了通过扼杀性的反竞争监管，常常是为了关闭开源和开放权重。幸运的是，我们击退了很多这样的尝试。但我认为两党 AI 洞察论坛在挖掘真相方面做得很好，并得出结论：美国应该投资 AI，而不是通过不必要的监管来减缓其发展。我认为特朗普做得很好，他的整个团队——David Sacks、Christian 等——在清除不必要的监管方面做得很好。另一方面，美国的一大竞争优势是吸引人才的能力，包括高技能人才以及目前可能不具高技能但未来可能成为高技能人才的年轻人。所以我认为，如果美国在吸引人才方面投入不足，那将是一个非受迫性失误。最后，科学投资——帮助我们的高等教育机构拥有资源来培养研究生和投资科学技术——我认为这非常宝贵。所以任何损害这一点的事情也会非常不幸。

The US federal government has done some good things and some less helpful things. I feel like clearing out unnecessary regulations has been a very good move. Even last year, the bipartisan Schumer AI Insight Forum — I think there were a lot of people lobbying the US government to pass stifling regulations. There were a lot of hyped-up AI safety narratives saying AI could lead to human extinction, which is a kind of ridiculous statement, to try to get stifling, anti-competitive regulations passed, often to shut down open source and open weights. Fortunately, we beat back a lot of that. But I think the bipartisan AI Insight Forum did a really good job digging into the truth and concluding that America should be investing in AI rather than passing unnecessary regulations to slow it down. I think Trump did a good job, and his whole team — David Sacks, Christian, and so on — did a good job clearing out unnecessary regulations. On the flip side, one of America's huge competitive advantages has been its ability to attract talent, including high-skill talent as well as young talent that may not currently be high-skill but could be in the future. So I think to the extent that America is not investing as much in attracting talent, that would be an unforced error. And then lastly, investments in science — helping our institutions of higher education have the resources to train grad students and invest in scientific technology — I think that's really precious. So anything that damages that would also be very unfortunate.

如果我给你一根监管魔杖，Andrew，你会改变什么来产生最显著的推动作用？

If I gave you a regulatory magic wand, Andrew, what would you change that would have the most significant needle-moving impact?

美国很幸运，有很多非常聪明的人想来这里解决非常具有挑战性的难题。我们的许多诺贝尔奖得主是移民——爱因斯坦就是一个例子。我认为继续将美国培养成一个吸引优秀人才在尊重法治的民主国家中共同工作的地方，将有助于我们前进。我认为确保半导体供应链也非常有价值。我在台湾有很多朋友；我爱台湾。美国对台积电的依赖令人担忧，以防万一发生什么事。还有一件社会上非常有趣的事情：最近有一份皮尤报告显示，有多少美国人对 AI 充满热情，有多少人不热情。尽管很多 AI 技术是在美国发明的，但很多人不信任或不喜欢 AI。

America is fortunate to have a lot of very smart people wanting to come here to do really challenging, tough problems. Many of our Nobel laureates are immigrants — Einstein is an example. I think continuing to cultivate America as a place to attract great talent to work together in a democratic nation that respects the rule of law would help us move ahead. I think securing the semiconductor supply chain would be very valuable as well. I have a lot of friends in Taiwan; I love Taiwan. And America's dependency on TSMC is concerning in case anything happens. And then there's one very funny thing that happened in society: there was recently a Pew report showing how many Americans are enthusiastic about AI versus not enthusiastic. Even though a lot of AI technologies were invented in America, a lot of people don't trust or don't like AI.

我工作的乐趣，Andrew，在于我能和了不起的人交谈，并交叉验证他们说的话。红杉的 David Cahn 说：‘一个非常有用的有效性指标是：AI 能否取代劳动力最底层 5%的能力？’Cohere 的 Joel 说：‘不，那是废话。真正的问题是，它能否将人的能力提升 10 倍？忘掉最底层 5%。它能提升 10 倍吗？’考虑到这些，你如何看待 AI 对劳动力成功的衡量标准？

The joys of what I do, Andrew, is I get to speak to incredible people and cross-reference what they say. David Cahn from Sequoia said, 'A really useful barometer for effectiveness is: can AI replace the bottom 5% of capabilities of what a workforce does?' Joel from Cohere said, 'No, that's crap. The real question is, can it 10x people's ability? Forget the bottom 5%. Can it 10x?' How do you think about a barometer for success of the workforce with AI with those in mind?

在软件工程领域，它正在加速代码编写。有很多项目过去需要六个工程师花半年时间才能完成，而今天我自己或我的一个工程师可以在一个周末内完成。我希望我们再也不用回到没有 AI 辅助的编程时代，因为加速是不可思议的。例如，有一个周末我想：‘我想要一些闪卡让我女儿练习乘法。’她想要闪卡。所以我想我可以开车去商店买一堆闪卡，或者直接用 AI 帮我写代码生成并打印一堆闪卡。我选择了后者。这是一个非常低经济价值的 AI 辅助编程应用，但我能很快完成。

In the case of software engineering, it is accelerating the writing of code. There are so many projects that used to take six engineers half a year to build that today I or one of my engineers can build in a weekend. I hope that we never have to go back to coding without AI assistance again because the acceleration is incredible. For example, one weekend I thought, 'I want flash cards for my daughter to practice multiplication.' She wanted flash cards. So I thought I could either drive to the store and buy a bunch of flash cards, or I could just use AI to write code for me to generate and print out a bunch of flash cards. I did the latter. This is a very low economic value use of AI-assisted coding, but I could get it done very quickly.

你认为 vibe coding 是一个持久的市场吗？比如，你认为每个人都会想编程，可访问性很重要，还是你认为它只是让构建者更好、更高效地构建？

Do you think vibe coding is an enduring market? Like, do you think everyone will want to code and accessibility is important, or do you think it bluntly just allows builders to build better and more efficiently?

我认为我们需要以上所有。我对‘vibe coding’这个词有复杂的感受，但抛开术语吹毛求疵，我认为每个人都应该学习编程。我看到的是，对于很多不仅仅是软件工程的工作角色，会编程的人比不会编程的人能完成更多工作。例如，我的市场人员有一次想进行用户调查。她想要一个让人们提供实时反馈的东西。她在应用商店里找了一圈，没找到。所以她说：‘你知道吗？我打算花两天时间自己写一个。’花了两天时间，但我的市场人员随后构建了一个小型的移动应用，用户可以通过向左或向右滑动来对我们想要测试的一些营销信息提供反馈。正因为如此，我们能够进行用户实验并获得反馈，这帮助她更好地完成了市场人员的工作。相比之下，一个不会编写小型应用让用户滑动并提供反馈的市场人员，就无法做到这一点，无法获得反馈，也无法推进工作。如今，我最好的招聘人员不仅手动筛选简历，他们还会编写提示词，让 AI 帮助他们筛选简历。这很有趣。

I think we need all of the above. I've had mixed feelings about the term 'vibe coding,' but nitpicking terminology aside, I think everyone should learn to code. What I'm seeing is that for a lot of job roles that aren't just software engineering, people who can code can get more done than people who can't. For example, my marketer wanted to run a user survey once. She wanted something for people to give live feedback. She looked at the app store and couldn't find anything. So she said, 'You know what? I'm going to spend two days to code it up.' It took two days, but my marketer then built a little mobile app where users could swipe left or right to give feedback on some marketing messages we wanted to test. Because of that, we were able to run user experiments and get feedback, which helped her do her job better as a marketer. In contrast, a marketer who couldn't code a little app to let people swipe around and give feedback would just not have been able to do this, would not have gotten feedback, and would not have been able to move forward. Today, my best recruiters not only screen resumes by hand, they are writing prompts to get AI to help them screen resumes. It's been interesting.

但回到你的观点，比如，哦，人们不应该害怕，但他们确实害怕。你看到这会导致效率提升，意味着裁员，如果你能用 AI 筛选更多。我不喜欢这种危言耸听。但如果你能用 AI 筛选更多，我就不需要另外三个分析师了。我认为有一小部分工作，坦白说，确实有麻烦。但我认为对于绝大多数知识工作者来说，实际上关于炒作有一点。嗯，AI 很神奇。有很多事情它做不到。所以，这个幻想的 AGI 有一天 AI 能做人类能做的一切，我认为我们离那还很远。我会说没有，比如几十年，甚至可能更长。关键是如果 AI 能做招聘人员 30%的工作，谁知道呢，也许 50%，虽然那感觉有点高。还有另外 50%到 70%的事情仍然需要人类来做。但也很清楚，如果你用 AI 而别人不用，那你能完成的事情会有巨大差异。所以更好，你知道，用 AI 好得多。但因为 AI 不能做所有事，很多工作角色仍然需要人类来做。你不认为我们有一个白领人才管道问题吗？无论你是顾问还是初级法律助理，你能做的很多事情正在被 AI 取代。他们实际上在裁减初级员工。你到处都能看到这种情况。所以恐惧的是，10 年后我们将有一个人才缺口，没有初级员工晋升为高级员工，因为我们取代了他们。

But going to your point on like, oh, people shouldn't be fearful and they are fearful. You see that that would lead to efficiency gains which mean headcount reductions if you can screen so much more with AI. I'm not into this kind of fear-mongering. But like if you can screen a lot more with AI, I don't need my three other analysts. I think there's a small subset of jobs that, you know, frankly are in trouble. But I think for the vast majority of knowledge workers, actually here's one thing about hype. Um, AI is amazing. There's a lot of stuff it can't do. So, this phantom AGI someday where AI can do everything a human can do, I think we're very far away from that. I would say no, like decades away, maybe even longer. And the trick is if AI could do 30% of a recruiter's job, you know, who knows, maybe 50%, although that feels a little bit high. There's another like 50 to 70% of stuff that we still need the human to do. But it's also clear that if you use AI and someone doesn't, that's actually a huge difference in what you can accomplish. So better, you know, much better off using AI. But because AI can't do everything, there's still plenty of work that we still need humans to do for a lot of job roles. Do you not think we have a white collar talent pipeline problem though, which is whether you're a consultant or you're a legal associate in the junior ranks, a lot of what you can do is being replaced by AI. And they are actually cutting juniors. You're seeing this across the board. And so what's the fear is we're going to have this talent hole where in 10 years time there's no juniors to go up into seniors because we've replaced them.

是的，我不认为有那么严重。我认为有一个大问题，但我不认为正是那个问题。所以让我告诉你我在软件工程中看到的情况。

Yeah, I don't think it's as dire as that. I think there is a big problem, but I don't think it's exactly that problem. So let me tell you what I'm seeing in software engineering.

我认识的最有生产力的工程师，他们不是应届大学毕业生。他们是拥有 10 年或 20 年经验的人，并且真正精通 AI，了解 AI 工具，理解 AI。所以那些有经验且精通 AI 的人，比世界上一两年前见过的任何东西都快。低一个层级的是真正精通 AI 的应届大学毕业生。所以我雇用了不少人，应届大学毕业生，他们通过社交网络社区真正学会了 AI 工具，他们行动很快，但不如那些有经验且懂 AI 的人。应届大学毕业生再低一个层级的是有 10 年编码经验的人，但他们有一份舒适的工作，并且出于某种原因仍然像 2022 年 ChatGPT 之前那样编码。那些人，我不再雇用那样的人了。但有些人，你知道，舒适的工作让他们继续用旧方式编码，就是没学 AI。我认为那些人可能在某个时候会遇到麻烦。然后哦，但还有另一个层级，那就是有麻烦的层级，即不懂 AI 的应届大学毕业生。一个不幸的事情是大学课程变化缓慢。所以我实际上感到很难过，即使今天，有些大学培养的计算机科学本科生，从未在互联网上调用过一个 API。对吧？想象一下，一个计算机科学本科生毕业，从未听说过云计算。就像，什么是云？哦，我不需要运行东西。那很奇怪。你不能作为一个计算机科学专业的学生却不知道如何在云上做事。我越来越觉得这不对。我觉得我们不能在培养计算机科学专业的学生时，不确保他们知道如何使用 AI 来帮助编码，不让他们了解 AI 构建模块。但大学在意，而这正是进入就业市场的一群学生，他们真的很挣扎。但懂 AI 的应届大学毕业生，我们找不到足够多。所以很多企业喜欢雇用那些应届大学毕业生。

The most productive engineers I know, they're not fresh college grads. They are people with 10 or 20 years of experience or whatever and really on top of AI and know the AI tools and understand AI. So those people with experience and on top of AI move faster than anything the world has seen even one or two years ago. One tier down is actually fresh college grads that are really on top of AI. So I've hired quite a few people, fresh college grads that for whatever reason through the social network community really learn the AI tools and they move really fast, but they're not as good as people with experience that know AI. One tier down from the fresh college grads is the people with 10 years of coding experience, but who had a comfortable job and for whatever reason is still coding like it's 2022 before ChatGPT. Those people I just don't hire people like that anymore. But there are people that, you know, the comfortable job they kept coding the old way and they just did not learn AI. I think those people may get into trouble at some point. And then oh but there's one other one which is the tier that is in trouble which is the fresh college grads that don't know AI. One unfortunate thing is university curricula are slow to change. And so I actually feel pretty bad that even today there are universities graduating CS undergrads that have not made a single call to a single API on the internet. Right? Imagine graduating a CS undergrad that has never heard of cloud computing. It's like what is a cloud? Oh, I don't need to just run things. That's weird. You just can't be a CS major and not know how to do things on the cloud. And I'm getting to a point where I don't think it's right. I feel like we've got to not train CS majors without also making sure they know how to use AI to help them with coding, without also making them know the AI building blocks. But universities care, and that's a cohort of students entering the job market that's really struggling. But the fresh college grads that know AI, we can't find enough of them. So many businesses love to hire those fresh college grads.

我只想谈谈你说的那些 10 倍、100 倍的工程师，他们太棒了。我们看到薪酬包、补偿品牌，比以往任何时候都大，你知道，在某些情况下，一个工程师就达到 35 亿美元。这些薪酬包是否合理，考虑到他们对公司企业价值的影响，还是我们应该担心的泡沫式薪酬包？

I just want to touch on the 10x 100x engineers that you said are just amazing. We're seeing pay packets, compensation brands, larger than they've ever been, you know, three and a half billion dollars in certain cases for a single engineer. Are these justified pay packages given the impact that they are having on companies' enterprise value or is this bubble-like pay packages that we should be concerned by?

我不知道。真的很难说。我认识一些人，他们拿到了非常大的薪酬包。我实际上为他们感到高兴。我认为资金投入到，你知道，很好地支付 AI 人员，这很棒。

I don't know. It is really hard to say. I know a number of people that have gotten really huge pay packets. I'm actually very happy for them. I think it's great the funding going into, you know, pay AI people really well.

我是好意。你认为一个工程师值一亿美元吗？我担心你只是不会那么有生产力。比如如果我一夜之间给你一亿美元，天哪，你可能会买一栋好房子去度假，你知道，你会失去一点效率。

I mean it nicely. Do you think it's like a hundred million for an engineer? I worry that you're just not going to be as productive. Like if I give you $100 million overnight, god, you might buy a nice house and go on holiday and, you know, you lose a bit of efficiency.

我不知道。我有很多硅谷朋友，出于某种原因，赚了一点钱。他们中的很多人只是继续非常非常努力地工作。同样，之前和之后，你知道，他们最终赚了一点钱。所以我发现很多科技文化中，我们做事是因为有趣，因为它让我们希望能帮助别人，是改变世界的一种方式。我发现财富让人变得懒惰的程度远低于人们的猜测。

I don't know. I have a lot of Silicon Valley friends that, for whatever reason, have made a little bit of money. Many of them just keep working really really hard. Equally before and after, you know, they wound up making a little bit of money. So I find that a lot of the tech culture we do stuff because it's fun, because it lets us hopefully help other people, is a way to change the world. I find that wealth makes people become lazy much less than one might guess.

我很想知道你怎么看这个。你说了它可能影响许多不同垂直领域的所有不同方式。你说我们过度炒作末日场景和介于两者之间的一切。你，Andrej Karpathy 最近说 AGI 只会融入 2%的 GDP 增长，老实说，我觉得这听起来有点乏味，Andrew。我想要一些生产力提升的巨大转变。你认为融入 2%的 GDP 增长是你所期望的，还是你期望更显著的 5%、6%，就像孙正义在软银期望的那样？

I'm intrigued to see how you think about this. You said all the different ways that it could impact many different verticals there. And you said we overhype doomsday scenarios and everything in between. You Andrej Karpathy recently said AGI will just blend into 2% GDP growth, which I thought sounded a little bit unexciting to be honest, Andrew. I wanted some seismic shift in productivity increase. Do you think a blend into 2% GDP growth is what you expect or do you expect a much more significant 5, 6% like Masayoshi Son at SoftBank expects?

我希望我们能更接近 5%、6%或更高的 GDP 增长。展望未来，事实证明当今世界最昂贵的东西之一是智能。这就是为什么，至少在美国，雇用高技能医生为我们提供医疗建议或雇用高技能导师耐心教我们的孩子如此昂贵，因为那种智能，培养那位智慧的医生、智慧的教师、智慧的顾问非常昂贵。但有了 AI，我们终于有了一条让智能变得便宜的途径。

I hope we can get much closer to five, six or more percent GDP growth. When looking to the future, it turns out one of the most expensive things in today's world is intelligence. This is why it's so expensive, at least in the US, to hire a highly skilled doctor to advise us on a medical condition or hire a highly skilled tutor to patiently teach our kids because that intelligence, training up that wise doctor, wise teacher, wise adviser is very expensive. But with AI, we finally have a path to make intelligence cheap.

那么在未来，如果每个人都能被一群聪明、见多识广的助手协助，处理所有目前只有社会上相对富裕的人才能雇人处理的话题，那么个人将变得更有能力，能完成更多事情。那些高度赋能的个体，他们的生活将截然不同，GDP 增长也将是巨大的。我完全理解并同意这一点。这有点像在谈论知识的民主化及其带来的好处。你之前提到了一个词——‘开放’，关于我们看到的开放权重生态系统。在很多情况下，我们看到了一种回归封闭世界的趋势。你对这种回归封闭的趋势有何看法？你如何分析当今开放与封闭的格局？

And so in the future, if everyone can be assisted by an army of smart, well-informed staff on all of these topics under the sun that currently only the relatively wealthy in society can afford to hire people for, then individuals will be so much more empowered and able to get so much more done. And that highly empowered individual's life will be so different, and GDP growth will be massive. Totally get that and agree. Kind of speaking about that democratization of knowledge there and the benefits that come from it. You said a word before which was 'open' about the open weights ecosystem that we've seen. We've seen this reversion back to a closed world in a lot of cases. How do you feel about the reversion back to a closed world, and how do you analyze the state of play today in that open versus closed?

情况仍然非常动态。我认为对于许多美国公司来说，领先的前沿模型通常保持封闭，而次一级的模型，性能稍差一些，则以开放形式发布。我认为这总比没有好。我实际上很感激所有发布开源开放权重模型的团队。另一个动态是，中国尤其已经真正领先，或者说，在发布大量优秀开放权重模型方面走在前列。所以我会说，这并非我十年前所能预料的——中国 AI 最终会比美国 AI 更加开放。

It's still very dynamic. I think for a lot of American companies, the leading frontier model is often kept close, and then the one-tier-down model, not quite as good, is released as open. I think it's much better than nothing. I'm actually grateful for all the teams releasing open-source open-weight models. And then the other dynamic is that China especially has been really taking the lead, or well, taking a lead or getting up there in terms of releasing tons of really good open-weight models. So I would say it's not what I would have predicted a decade ago, that China AI would end up being more open than America AI.

你认为为什么中国想要一个开放的 AI 世界？

Why do you think China wants an open AI world?

事实证明，开放对一个国家的发展非常有利。当一个团队发布开源软件时，知识在邻近社区中的流通要快得多。所以我看到的是，当一个中国团队发布开放权重模型时，当然，美国人可以利用它。但中国经济从中受益更多，因为一旦某样东西是开放的，团队之间更容易互相联系说：‘嘿，老兄，这到底是怎么工作的？我在模型的这个部分遇到了麻烦。’这种知识流通对创新非常有价值。而当美国有更多封闭模型，团队试图支付上亿美元的薪水来挖人才时，知识流通就变得非常缓慢，这拖慢了美国和欧洲的创新速度。

It turns out that openness is great for a country's development. So it turns out that when a team releases open-source software, circulation of knowledge is much faster to the close-by community. So what I see is that when a team in China releases an open-weight model, then yes, of course, Americans can take advantage of it. But the Chinese economy benefits even more from it because once something is open, it's easier for teams to call each other and say, 'Hey buddy, how does this really work? I'm having trouble with this part of the model.' That circulation of knowledge is really valuable for innovation. And when the US has more closed models and teams are trying to pay these hundred-million-dollar salaries to extract talent, then that circulation of knowledge becomes very slow and it slows down the rate of American and European innovation.

然而，随着模型层的商品化和开放化，它实际上增加了制造业和规模化制造能力的溢价，而中国在这方面的能力远超美国。你不认为这实际上主导了他们很多想法，即为什么他们想要削弱美国模型的优势？

With the commoditization of the model layer though and the kind of opening of it, it actually increases the premium on manufacturing and the ability to manufacture at scale, which China has a much greater ability to do than the US. Do you not think that actually leads a lot of their thinking around why they want to remove the strength of US models?

除了开放权重模型有助于促进创新和知识流通之外，我认为开放权重模型还是地缘政治影响力的巨大来源。例如，如果有一天某个发展中国家的孩子问一个政治敏感话题，或者问‘在这种情况下国界在哪里？’或‘这个事件或那个事件的历史是什么？’，他们最终使用的模型的来源国将提供某种答案。而答案是否偏向某个国家的价值观，实际上是影响力和软实力的巨大来源。不管你喜欢与否，开放权重模型是 AI 供应链的关键部分。中国将免费、低成本或三种模型释放到供应链的这个关键部分，意味着他们真正开始建立领先地位和庞大的用户基础。这也将成为影响力的来源。这就是为什么我认为拥有强大媒体和娱乐产业的国家——事实证明，韩国因其领先的娱乐产业而拥有巨大的不成比例的影响力。人们听 K-pop 之类的，这为国家买来了大量影响力。好莱坞曾是美国的巨大软实力来源。它描绘了美国梦的某种愿景，谈论自由和民主的价值观。我认为这是传播和软实力的另一个前沿。

In addition to increasing innovation and circulation of knowledge, which the open-weight models help with, I think that open-weight models are a tremendous source of geopolitical influence. So for example, if someday some kid in some developing nation asks a question about a politically sensitive topic, or asks 'Hey, where are the national borders in this case?' or 'What is the history of this event or that event?', the country of origin of the model they end up using will be delivering some answer. And whether the answer is skewed towards one nation's values or another nation's values is actually a tremendous source of influence and soft power. Like it or not, open-weight models are a key part of the AI supply chain. And China releasing free, low-cost, or three models into that key part of the supply chain means they are really starting to build up a lead and build up a commanding user base. That too will be a source of influence. This is why I think nations with a strong media and entertainment industry—it turns out South Korea has vastly disproportionate influence because of their leading entertainment industry. So people listen to K-pop or whatever, and that buys the nation a lot of influence. Hollywood was a tremendous source of soft power for America. It paints a certain vision of the American dream, talks about the values of freedom and democracy. And I think this is another frontier of communications and soft power.

你有着最迷人的视角，显然在谷歌和百度都待过多年，因此在某些方面处于双方立场。我们有一种奇怪的二元对立：AI 竞赛中中国与美国。你同意这种中国与美国在 AI 竞赛中的定位吗？

You have the most fascinating perspective, having obviously spent many years at Google and then at Baidu as well, and so having been on both sides of the table in certain respects. We have this kind of strange binary polarization of the AI race: China versus the US. Do you agree with that positioning of China versus the US in an AI race?

我认为有很多合作空间，也有一些领域会是竞争性的。首先，虽然人们甚至我自己有时会谈论 AI 竞赛，但并没有单一的终点线。这不是一场比赛。AI 是一种通用技术，你可以在编码上更好或更差，在回答问题方面更好或更差，在帮助营销和金融等方面更好或更差。所以 AI 有很多不同的能力，没有单一的终点线。而 AI 是一种能力；我认为我们将在很长一段时间内不断改进。所以我觉得由于公关，AGI 被炒作得像一个终点线，但我不认为它是一个终点线。我们只是在未来几十年里拥有持续改进的能力。话虽如此，AI 能力更强的国家将更强大，其公民将更繁荣，经济增长更快。所以我认为，在不同国家激励不一致的程度上，AI 能力更强的国家将能够做更多事情。就像如果一个国家有极好的电网，而另一个国家经常停电，那么一个国家可以利用电网做更多的制造业、更多的工业工作，从而做更多的事情。

I think there's a lot of room for cooperation, and also some places that will be competitive. So first, while people sometimes even I talk about the AI race, there's no single finish line. It's not one race. AI is a general-purpose technology, and you could be better or worse at coding, better or worse at answering questions, better or worse at helping with marketing and finance, and so on. So AI has many different capabilities, and there's no one finish line. And AI is one capability; I think we're going to keep on improving for a long time. So I feel like because of PR, AGI has been hyped up as if it were a finish line, but I don't think it's a finish line. It's just we have continually improving capabilities for decades to come. Having said that, nations with stronger AI capabilities are going to be more powerful, their citizens will be more prosperous, economies will grow faster. So I think there is, to the extent that different nations' incentives are not aligned, nations with more powerful AI capabilities will be able to do more. Just like if a country has a fantastic electricity grid and another country has power outages, well, one country can just use the electricity grid to do more manufacturing, more industrial work, just do a lot more that way.

你不认为我们仍然低估了中国的能力吗？我的意思是，我认为在欧洲我们肯定低估了，但在美国，恕我直言，我看到很多美国人在定位上的傲慢。然后你去中国，你去过中国并在中国待了很长时间，你会意识到他们行动的速度和强度。这与欧洲和美国相比是不同层次的。

Do you not think we still underestimate China's ability though? I mean, I think we definitely do in Europe, but I think in the US, respectfully, I see a lot of US arrogance around your positioning. And then you go to China, and you've been to China and spent huge amounts of time in China, you realize the speed and the intensity with which they move. It's a different level to both Europe and the US.

是的。公平地说，我认为美国、欧洲、中国也都有各自的问题。

Yeah. Just to be fair, I think US, Europe, China all have problems as well.

但话虽如此，我认为当中国政府做出举国承诺或产业承诺时，其工作热情和速度实际上是一股非常强大的力量，体现在国家对半导体和教育系统的投资上。例如，K-12 学生接受 AI 使用培训，企业也使用 AI 并分享知识，有时还利用国家机器进行建设和国际销售。此外，还有对稀土元素的控制。所以我认为这种全经济、全国家的努力实际上是一股非常强大的力量，我不会低估它。

But having said that, I think the work ethic and velocity when China's government makes a whole-nation commitment or industrial commitment is actually a very powerful force, with state-level investments in semiconductors and the education system. For example, K-12 kids being trained to use AI, businesses also using AI and sharing knowledge, and sometimes building this stuff and selling internationally with state apparatus. And also control over rare earth elements. So I think that whole-of-economy, whole-of-country effort is actually a very powerful force that I wouldn't underestimate.

鉴于我们不应低估它，你认为我们对芯片实施出口管制是正确的吗？显然英伟达经历了很多来回的出口管制。你认为这是对还是错？

Given that we shouldn't underestimate it, do you think it's right that we have export controls on chips? Obviously Nvidia has had a lot of export controls back and forth. Do you think that's right or not?

我认为芯片出口管制在很大程度上适得其反。美国首先对华为实施限制，后来对英伟达、AMD 及其他半导体的出口进行限制，这实际上激励了中国。在出口管制之前，中国的半导体发展并不快。那是一个不错的领域，有一些投资。但当美国这样做时，中国真正加速了其半导体发展，美国激励了中国这样做，而这对中国来说正在产生回报。我认为许多中国公司正在构建产品，单个芯片性能较弱，但通过大量芯片的组合，试图提供与上一代英伟达竞争的产品，可能越来越接近当前一代。所以，如果纯粹从美国国家自身利益来分析，我认为这导致中国以可能对美国长期不利的方式加速了半导体产业。

I think the export control on chips has largely backfired. The way the US first put restrictions on Huawei, and later on export of Nvidia and AMD and other semiconductors, really incentivized China. Before the export controls, semiconductor development in China was not moving that fast. It was a nice area with some investment. But when America did that, China really accelerated its semiconductor development, and America incentivized China to do this, and it is paying off for China. I think a number of Chinese companies are building offerings where individual chips are less powerful but with a much larger number of chips, trying to build offerings competitive with the last generation of Nvidia, maybe increasingly the current generation. So if I were to analyze purely US national self-interest, I think that caused China to accelerate a semiconductor industry in a way that may not be helpful to the US long term.

我身处欧洲，你显然住在伦敦。你之前告诉我你出生在伦敦。我的问题是，明显感觉我们远远落后，有人说我们已经输了。你对欧洲在这个全新世界中的地位有何看法？欧洲能做些什么来重新获得与美国和中国之间的某种平等？

I sit in Europe, you obviously live in London. You told me you were born in London before this. My question to you is it transparently feels like we are very far behind and people say you've already lost. How do you feel about Europe's position in a very new world and what can Europe do to regain some semblance of equality between the US and China?

如果我对欧洲监管机构有一个愿望，我与不少欧洲监管机构交谈过。我听到诸如“我们想成为 AI 监管的领导者”之类的说法，并认为这是一种竞争优势。恕我直言，这不是竞争优势。所以我给欧洲的一个愿望是：停止过度监管，专注于投资和建设。AI 时代还处于早期阶段。比赛还早，欧洲有很多聪明人。让人们努力工作。不要强迫他们不努力工作。让那些想努力工作的人努力工作，停止过度监管，去投资和建设。

If I had one wish for the European regulators, I spoke with quite a few European regulators. I was hearing things like 'we want to be leaders in regulating AI' and that's a competitive advantage. With all due respect, that's not a competitive advantage. So my one wish for Europe is: stop regulating so much and just focus on investing and building. It's so early in the days of AI. It's still early in the game, and Europe has plenty of smart people. Let people work hard. Don't force them to not work hard. Let people that want to work hard work hard, and stop overregulating, and just go and invest and build stuff.

我们最需要在哪些方面进行投资，而目前投资不足？

Where do we most need to be investing where we are not investing enough?

大量资本正涌入数据中心和基础设施。我们可以争论是否存在泡沫。我们确实需要大量投资。我们是否到了人们使用如此深奥的金融工具来寻找资金，以至于会出现泡沫的地步？我们可以讨论。所以我们确实需要大量投资，但什么时候变成过度投资？这是一个有趣的问题。我认为我们还需要大力投资的另一个领域不仅是基础设施、数据中心、基础模型层，还有应用层。因为其他人已经花费了数十亿美元来训练这些 AI 模型，我们现在可以以几百、几千甚至几十美元的价格访问它们。所以构建大量以前不可能实现的应用是很好的。从风险投资的角度来看，我从多位风投那里听说，奇怪的是，尝试新事物的成本如此之低，以至于想法反而更少了。不太确定在应用层如何投入大量资本。事实上，如果你看看许多应用层的投资，有时感觉公司投入 1 亿美元，以便他们可以支付给 OpenAI 或 Anthropic，然后 OpenAI 或 Anthropic 再支付给英伟达，所有钱最终都流向了那里。话虽如此，在应用层有很多有价值的赌注可以下，就是去构建东西。但困境在于，你可以以非常资本高效的方式做到这一点。所以如果有人想说“我想投入 100 亿美元”，是的，你可以建造价值 100 亿美元的数据中心。我们知道如何花这笔钱。但如何在构建应用上花费 100 亿美元？问题在于，尝试一个想法几乎只需要 100 万美元。那么我如何花掉 100 亿美元？这既是问题也不是问题，但我认为我们应该去做。

There's tons of capital going into data centers and infrastructure. We can debate whether there is a bubble or not. We definitely need a lot of investments. Are we getting to the point where people are using such esoteric financial instruments to find cash for it that there'll be a bubble? We could debate that. So we definitely need a lot of investments, but when does it become overinvestment? That's an interesting question. The other place that I think we need to invest in a lot is not just the infrastructure, data center, foundation model layer, but the application layer. Because others have spent billions of dollars to train these AI models, we can now access them for hundreds or thousands of dollars, or even tens of dollars. So it's wonderful to build tons of applications that were just not possible before. From a VC investment perspective, I've heard from multiple VCs that bizarrely the cost of trying something out is so low that there are fewer ideas. It's not quite sure where to put massive amounts of capital to work at the application layer. In fact, if you look at a lot of the application layer investments, sometimes it feels like firms are putting in $100 million so that they can pay OpenAI or Anthropic, so that OpenAI or Anthropic can pay Nvidia, which is where all the money ends up. Having said that, there are so many valuable bets to be placed at the application layer to just build stuff. But the dilemma is you could do it in a very capital-efficient way. So if someone wants to say 'I want to put $10 billion to work', yes, you can build $10 billion worth of data centers. We know how to spend that money. But how do you spend $10 billion in building applications? The problem is almost it only costs me a million dollars to try an idea. So how do I spend $10 billion? It's kind of a problem and also not a problem, but I think we should.

嗯，是这样吗？因为当你看到 AI 利润率时，AI 应用层公司的利润率是多少？它们很糟糕。它们不赚钱。构建它们成本很高，因为你有大型工程团队。它们成本更高，而不是更低。

Well, does it? Because when you look at AI margins, what margins for AI application layer companies? They're terrible. They make no money. They cost a lot of money to build because you have large engineering teams that build them. They cost more, not less.

我认为情况仍然各不相同。我看到很多软件应用的萌芽，它们构建起来并不昂贵，而且如果你的大语言模型 token 使用不是你的主要支出。如果你看看 Rippling 或 Lovable，它们 80%的转嫁成本都流向了 Anthropic。

I think it still varies. I'm seeing a lot of green shoots of software applications that were not that expensive to build, and if your LLM token usage is not the majority of your expense. If you look at a Rippling or a Lovable, 80% of their pass-through is to Anthropic.

是的。

Yeah.

所以我感到兴奋的动态是，随着大语言模型 token 成本持续下降，我们将看到经济模式如何变化。现在 token 还很贵，但希望这种情况会改变，而创造的价值非常大。我记得早期外卖配送的时代，例如，我在美国和中国都看到过。有很多风险投资补贴的饮食。那很棒；我们可以吃送来的食物，基本上是风险投资补贴的。我认为我们现在看到的是大量风险投资补贴的 AI 服装。物理定律或金融定律表明，在某个时候这不能永远持续下去。但最终稳定下来的情况是，我认为会有一些非常有价值的企业，它们不会永远依赖风险投资补贴。但驾驭这个疯狂的风险投资补贴世界以取得良好结果需要很多技巧。话虽如此，我仍然想说，有很多较小的应用，它们还没有达到数亿美元的收入，也许只有数百万或数千万美元的收入，但构建和运营成本并不高，我认为我们会看到它们继续增长。

So the dynamic that I'm excited about is that as LLM token costs continue to come down, we'll see how the economics change. Right now tokens are just expensive, but hopefully that will change, and the value created is really large. I remember an earlier era in the early days of food delivery, for example. I saw this in both the US and China. There was a lot of VC-subsidized eating. It was great; we could eat food delivered, basically VC-subsidized. I think we're seeing that right now with a lot of VC-subsidized AI clothing. The laws of physics or the laws of finance say that at some point this can't go on forever. But where it settles down will be that I think there will be some very valuable businesses that are not perpetually VC-subsidized. But navigating this crazy VC subsidy world to get to a good outcome takes a lot of skill. Having said that, I still want to say there are a lot of smaller applications that are not yet doing hundreds of millions of dollars, maybe they're doing millions or tens of millions of dollars of revenue, that haven't been quite expensive to build and to operate, and I think we'll see them continue to grow.

说到那些持续增长的小众领域，你怎么看待大型单一模型与更小、更高效、更专业模型之间的世界？你的想法有没有改变，关于哪个会更占主导地位？

Speaking of the smaller niches that continue to grow, how do you think about the question of a world of large monolithic models versus much smaller, much more efficient, much more specialized models? Has your mindset changed around which will be more dominant?

我认为很明显会是所有类型并存。我们会有大型模型、中型模型和小型模型。我对此有信心，因为智能的本质是多样的。有时我们做简单的智力任务，比如有人让我教女儿拼写'butterfly'——那是低难度的智力任务。有时我会坐下来花几个小时思考复杂的技术问题。智能有范围，所以我们希望 AI 做的事情也有很大的范围。如果你想让 AI 做基本的语法检查和拼写检查，你不需要万亿参数的模型——只需要一个小模型，可能本地运行就行。但如果你想做复杂的推理来写代码，那么拥有一个强大的模型会更好。所以我非常有信心，我们最终会有各种规模的模型，从大到小，来完成各种任务，就像人类处理不同难度的任务一样。AI 也是如此。

I think it's clear it'll be all of the above. We will have large models, midsize models, and tiny small models. The reason I'm confident about that is because the nature of intelligence is diverse. Sometimes we do intellectually easy tasks, like if someone asked me to tell my daughter how to spell 'butterfly'—that's a low intellectual task. And sometimes I'm sitting down thinking for hours about a complex technical problem. Intelligence has a range, and so the set of things we want AI to do has a huge range. If you want AI to do basic grammar checking and spell check, you don't need a trillion parameter model—just a tiny model maybe running locally. But if you wanted to do complex reasoning to write a piece of code, then having a powerful model is going to do better. So I'm very confident we'll end up with a huge range of models, small and large, to do the huge range of tasks, just like humans do a range of tasks of difficulty. Same with AI.

这是否意味着你不同意 Andrej Karpathy 的说法，他认为有用的智能体还需要十年？

Does that mean you disagree with Andrej Karpathy when he said that useful agents are a decade away?

我不同意。我认为我们现在已经看到了有用的智能体式工作流。AI Fund 的团队为许多任务构建了这么多智能体式工作流，没有它们我们根本无法完成这些任务。

I disagree with that. I think we're seeing useful agentic workflows right now. AI Fund's team has built so many agentic workflows for so many tasks that we just could not even do the tasks without them.

能给我举个例子吗？我很感兴趣。

Can you give me an example? I'm fascinated.

一年多前，在拜登和特朗普的一次辩论之后，我们认为关税合规可能会成为一个问题。不幸的是，我们是对的。所以去年大约八月，我们开始构建帮助关税合规的技术。文书工作极其复杂——进口一辆自行车涉及成本、车轮尺寸等规格，还有很多法规。这让我想，'人类真的在做这个吗？'所以我们构建了智能体式工作流来阅读合规文件、获取规格、匹配建议等等。这现在成了一家投资组合公司叫 Guiga Dynamics，由于关税合规复杂性的增加，它表现不错。没有智能体式工作流我们根本无法做到。我们还有印度的医疗助手、Katus 的法律文档处理，以及许多其他工作流。所以我认为有用的 AI 智能体式工作流今天已经存在，不仅在我们的初创公司——大企业也有内部工作流离不开 AI 智能体。

Over a year ago, after one of the Biden-Trump debates, we thought tariff compliance might become an issue. Unfortunately, we turned out to be right. So last year, around August, we started building technology to help with tariff compliance. The paperwork is incredibly complex—importing a bicycle involves specs like cost, wheel size, and many regulations. It made me think, 'Are humans really doing this?' So we built agentic workflows to read compliance documents, get specs, match suggestions, and so on. This is now a portfolio company called Guiga Dynamics, which has been doing well due to increased complexity in tariff compliance. We just could not have done this without agentic workflows. We also have medical assistants in India, legal document processing with Katus, and many other workflows. So I find that useful AI agentic workflows exist today, and not just in our startups—large businesses also have internal workflows that couldn't be done without AI agents.

当我们考虑业务的核心时，就是利润率。这些业务大多没有利润率。你今天投资时关心利润率吗？还是你持乌托邦观点，认为它会随着时间和效率提升自行修正？

When we think about the core of a business, it's margins. Most of these businesses don't have margins. Do you care about margins when investing today? Or do you take the utopian view that it will correct itself with time and efficiency gains?

在某个时候，物理定律或金融定律——利润率确实重要。但 AI 的一个棘手之处在于技术会变化。所以我们不会假设技术停滞不前；我们假设它会进化。一个明显的点：token 价格一直在快速下降——根据你信谁的说法，每年下降 80%。当我们构建原型时，我们通常不担心 token 成本，因为首要任务是构建用户喜爱的产品。然后当我们构建了一些东西，用户开始使用，我们的 API 账单开始攀升。它变得非常昂贵——相当于一个工程师的薪水，然后超过两个，然后是一大堆工程师。但幸运的是，到目前为止，几乎每次我们都能用技术让成本曲线下降得比市场 token 价格下降的速度更快。所以绝对利润率很重要，但当你对技术走向有看法时，你就不会为今天的利润率而构建，而是为你预测的未来利润率而构建。这是一个重要的区别。但我们也不持盲目的乌托邦观点，比如'AGI 等等'——那也太简单化了。

At some point, the laws of physics or finance—margins do matter. But one tricky thing about AI is that the technology is going to change. So we don't build assuming the technology will be stagnant; we build assuming it will evolve. One obvious point: token prices have been rapidly falling—80% per year depending on who you believe. When we build prototypes, we routinely don't worry about token costs because the first priority is to build a product users love. Then after we've built something, users start using it, and our API bill climbs. It gets really expensive—costing the salary of one engineer, then more than two, then a whole bunch. But fortunately, almost every time so far, we've been able to use techniques to bend the cost curve back down even faster than the rate at which token prices are falling. So absolute margins are important, but when you have a view for where the technology is going, it lets you not build for today's margins but for what you can forecast them to be in the future. That's an important distinction. But we don't take a blind utopian 'AGI blah blah blah' view either—that's overly simplistic.

你怎么看待 AI 世界中的防御性？很多人认为复制的时间大大缩短，防御性本身受到质疑。你同意吗？

How do you think about defensibility in an AI world? A lot of people suggest that the time to copy is reduced significantly, and defensibility itself is questioned. Do you agree?

护城河正在改变。我发现护城河往往是行业的函数，而不是技术的函数。所以 AI 作为一种技术，并不能为大多数企业提供护城河的答案。如果你为无人机、法律或其他领域构建 AI，护城河更多是那个行业的函数。

Moats are changing. I find that moats tend to be a function of the industry rather than a function of the technology. So AI as a technology doesn't really offer an answer to the moats for most businesses. If you're building AI for drones, legal, or whatever, the moat is more a function of that industry.

嗯，但模式方面有一个变化：以前软件本身就是一个护城河，如果你花十年时间构建一个软件，别人很难复制。现在这个护城河弱了很多，但其他模式呢？比如你是否试图用 AI 加速构建一个双边市场，这可以很有防御性；或者你是在做消费者还是企业级产品，品牌和声誉效应也能帮你建立防御性。所以我觉得软件模式变了，但其他模式往往需要基于行业分析。好的，软件模式变了，太棒了。现在我们有了重要的利润率，但弹性也大了一些。软件模式在大型企业保持相关性方面发生了变化。那么，阻止大型企业积极实施 AI 并避免自身消亡的最大障碍是什么？

Um but one thing that is changing with regard to modes is previously software used to be a mode right if you had you know invested 10 years to build a software it's really hard to replicate that that one mode is much weaker than before but other modes like are you trying to use AI to accelerate to build a two-sided marketplace which can be very defensible or you know are you building for a consumer more for consumer enterprise are there on brand and reputational effects right that can help you build defensibility there. So I find that um the software mode has changed but other modes tend to be analysis based on the industry. Okay. So software modes have changed. Fantastic. And so we now have margins that matter but we have a little bit more elasticity there. The software mode has changed in terms of like the ability to stay relevant for large enterprises. What are the single biggest barriers that are preventing large enterprises from implementing AI aggressively and prevent themselves being extinct?

我认为大多数大型企业最大的障碍实际上是人和变革管理。

I think the biggest barrier in most large enterprises is actually people and change management.

不是数据。

Not data.

不是数据。我认为绝对不是数据。不是说数据不重要，但那绝对不是瓶颈。我觉得数据就是数据。关于 AI 炒作的有趣之处在于，炒作中几乎总有一丝真相，只是被夸大了十倍。让我举个例子，然后再回到数据。有一种说法是，有了 AI，我们会有单人独角兽公司。如果你想用一个人打造一个十亿美元的独角兽，那没问题，这是好事。但坦白说，如果你有十亿美元的估值，你完全雇得起两个甚至十个员工。为什么非要炒作成只用一个人呢？所以团队规模在缩小，用更小的团队做更多的事，这是真的。但炒作却说要用一个初创公司打造独角兽。我发现很多 AI 炒作都很难分辨，因为里面有一丝真相，只是被过度放大了。回到数据。数据很重要，但事实证明数据非常垂直化，你不需要像想象中那么多数据就能起步。例如，Landing AI 与金融机构和医疗行业合作很多。很多金融机构有大量交易数据。把 PDF 文件转换成 markdown 文本，处理它，从中找到价值。比如，我们可以把 SEC 文件中的复杂金融表格非常准确地转换成 Excel 表格，然后让分析师或 AI 分析并得出结论。所以只要有点拼劲，看看内部数据、公开数据，往往就能有所进展。而且很多互联网数据是通用数据，但世界上大部分数据实际上是私有的，很多商业数据是非常有价值的交易数据——销售数据、产品数据、制造数据、物流数据。一个知道如何利用这些数据的拼劲十足的团队，就能开始构建东西并从中获得价值。不是说更多数据不好，但你不至于连第一步都迈不出去或者缺乏数据。

It's not data. I think it's definitely not data. Not that data is not important, but that's definitely not the bottleneck. I think data has been data. So the interesting thing about AI hype is there's almost always a gem of truth in the hype. It's just that it's been hyped up 10 times more than the reality. And maybe actually let me give one example then I'll come back to data. There's been this buzz about oh with AI we'll have unicorns with one employee right it's like a thing and it's fine if you want to build a unicorn startup billion dollar with one employee go for it. It's a good thing to do, but frankly, if you're doing a billion dollar valuation, you could afford to pay two employees or even 10. So why do you need to hype it all the way up to say let's do this with just one employee, right? So it is true that team sizes are shrinking and get more done with smaller teams. So that is true, but the hype is then saying let's build a unicorn with one startup. I find a lot of AI hype. It's so hard to disentangle because there's a gem of truth in it. It's just been hyped up a lot more. It's on data. Data is important. But it turns out that data is very verticalized and you don't need as much of it to get started as you think. So for example, Landing AI does a lot of work with financial institutions and healthcare. Many financial institutions have plenty of transaction data. Take the PDF file, turn it into markdown text, process that, find value in it. For example, we could take SEC filings, large complex financial tables, very accurately turn those financial tables into Excel spreadsheets, then get your analysts or your AI to analyze that and draw conclusions. So you can often do that with a bit of scrappiness. Look at internal data, look at your public data, you can often get some stuff going. And it turns out that a lot of internet data is general purpose data. Most of the world's data is actually private, and a lot of business data is very valuable transaction data—sales data, product data, manufacturing data, logistics data. With a scrappy team that knows how to use it, you can actually start to build something and get value out of it. Not to say more data wouldn't be even better, but you're not stuck to even take the first few steps or lack the data.

Andrew，我和很多这类企业的 CEO 聊过，他们说：‘Harry，你在开玩笑吗？你觉得我们能为企业数据做好安全和权限管理？不。我们没有 Slack，没有 Notion。一切都是定制的。你看摩根大通、高盛这些公司绝对禁止使用 ChatGPT。他们在构建内部系统。这就是企业采用 AI 的现实世界吗？

Andrew, I speak to many CEOs of these size businesses and they say, 'Harry, are you kidding me? You think we can get security and permissioning for our data and our enterprise? No. We don't have Slack. We don't have Notion. Everything is custom-built. You're seeing the likes of JP Morgan, Goldman Sachs absolutely refuse any ChatGPT use. Building internal systems. Is that the world that we inhabit for enterprise AI adoption?

我认为我们会达到的。我发现很多企业正在采用 OpenAI、PIGB 等。今天仍有企业使用本地部署而非云端。但我们正在进步，这需要时间。关于 AI 炒作中说的两年内实现 AGI，我觉得这太荒谬了。对于大多数合理的 AGI 定义，这根本不会发生。就像我们进入云时代已经很久了，但仍然有大量本地部署的工作。我认为 AI 的采用会很棒，会带来巨大的 GDP 增长，但所需时间会比炒作说的长得多。我实际上认为十年后我们仍将致力于识别和构建企业中有价值的应用。话虽如此，未来一两年我们会取得很大进展，但即使十年后也不会完成。

I think we'll get there. I find that a lot of enterprises are adopting OpenAI, PIGB and many others. I think today there's still businesses that are on-prem rather than on the cloud. But we're making progress and it'll take time. One thing about AI hype that we have AGI in two years or whatever—I think that's just ridiculous. For most reasonable definitions of AGI, that's just not going to happen. Just as how long we are into the cloud era, but we still have an awful lot of on-prem jobs. I think AI adoption will be wonderful. There will be tremendous GDP growth, but it's also going to take much longer than the hype says it will. I actually think that a decade from now we will still be working to identify valuable applications in enterprises and building them. Having said that, we will make a lot of progress over the next one or two years, but we're not going to be done even 10 years from now.

还有哪些关于 AI 及其采用和实施的观点，大家以为是对的，但实际上错了？

What else does everyone think they know about AI and its adoption and implementation that they get wrong?

就在今年早些时候，我们看到一些资深商业领袖建议人们不要学编程，理由是 AI 会将其自动化。我们回头会认为这是有史以来最糟糕的职业建议之一。随着 AI 辅助让编程变得更容易，更多人应该学编程，而不是更少。我已经看到了市场例子，比如构建一个用于反馈滑动的应用。但我认为对于很多工作职能来说，知道如何精确告诉计算机你想做什么的人，会让计算机为你工作，他们会更强大。在可预见的未来，精确告诉计算机你想做什么的语言就是编程。这不意味着你应该手写代码。手写代码正在过时。但让 AI 为你写代码，能做到这一点的人会更高效、更强大、更有乐趣。

Even earlier this year, we saw some senior business leaders advise people to not learn to code on the grounds that AI will automate it. We'll look back on that as some of the worst career advice ever given. As coding becomes easier with AI assisting us, a lot more people should learn to code, not fewer. And I'm already seeing the market example just now with building an app for feedback swiping. But I think for a lot of job functions, people that know how to tell a computer exactly what they want it to do so the computer can do it for you, they'll just be more powerful. And for the foreseeable future, the language of precisely telling computers what you want them to do is coding. It doesn't mean you should write code by hand. Writing code by hand is becoming obsolete. But get AI to write code for you, and people who can do that will be more effective, more powerful, and have more fun.

如果我们还处于早期阶段，十年后我们仍在寻找和识别可以显著改进的领域，我们有足够的资金来支持这十年所需的能源和算力吗？Sam Altman 说他需要一万亿美元，需要整个日本的能源。如果十年后我们仍然没有那么多改进，我们有资金支持吗？

If we're that early where a decade's time we're still going to be looking and identifying areas where it can improve meaningfully, do we have enough money to fund both the energy and the compute requirements for that 10-year period? Sam Altman has said he needs a trillion dollars. He needs the energy of Japan. If we're 10 years out before we have still not that much improvement, do we have the money to fund it?

我认为未来两年我们会看到很多改进。但十年后我们仍然不会停止获得更多改进。一个非常有前景的领域是 AI 辅助编程。我们看到了真正的生产力提升和实际回报。它确实在改变软件的编写方式，非常棒。

I think we'll see plenty of improvement over the next two years. But I think we still won't be done getting even more improvements 10 years from now. One place where super promising is AI-assisted coding. So we're seeing real productivity gains, real returns. It's really changing the way software is written. It's really been fantastic.

坦白说，我很多朋友都说，有了 AI 帮忙，编程有趣多了。所以我们已经看到了回报，但 10 年后我们也不会停止增长。但如果你看 TAM，AI 投资成功的秘诀是：我们会看到从软件预算到人力预算的转变吗？抱歉，是从人力预算到软件预算的转变？如果实现了，那我们就找到了圣杯，你和我都能用基金赚大钱。好消息是，TAM 大幅增加了，支出也大幅增加了。但如果我们实际上不会失去任何人力，那我们就看不到从人力预算到软件的转变。你认为我们不会看到这种转变吗？

Frankly, so many of my friends say coding is so much more fun with AI to help us out. So we are seeing returns, just to be clear, but we still won't be done growing this, you know, 10 years from now. But if you look at the TAM, the secret to success in AI investing is: will we see a transition from software budgets to human labor budgets? Sorry, from human labor budgets to software budgets? And if we have that, then holy grail, me and you will make a lot of money with our funds. And fantastic news because the TAMs have massively increased, or the spend has massively increased. If we're like, we're not going to actually lose any people, then actually we don't see that transition from human labor budget to software. Do you think we won't see that transition?

所以对我来说，问题是：AI 主要是为了节省成本还是为了增长？我知道改变工作流程很难。很多公司倾向于考虑节省成本。但问题可能在这里。我确实看到了一个模式。假设我有一个工作流程，比如有五个步骤，每个步骤占我 20%的精力。比如我在做贷款审批，决定是否批准贷款。简单起见，五个步骤各占 20%。如果你能自动化其中一个步骤，那就节省了 20%的成本，这很好。如果你是一个低利润企业，这很棒，但感觉不像是一个游戏规则改变者。我发现，更有价值的 AI 用户实际上需要重新思考工作流程。我看到的模式是：与其节省 20%的成本，你可以做到，这没问题。实现增长的两种模式是：做得更多或做得更快。以贷款审批为例，如果我不节省 20%的人力，而是重新设计工作流程，缩短决策时间。这样，客户不用等两周让贷款官员处理，而是 10 分钟内就能得到初步答复，这就改变了产品，推动了增长。这是更快的模式。还有另一种模式：很多企业只能为昂贵的高端客户提供高接触服务，比如客户服务。但如果你现在能服务更广泛的群体，或者金融咨询，不再只为少数人提供高接触服务，而是将同样的服务质量带给更多人，这同样改变了产品，推动了增长。所以，不是节省成本，而是让 AI 帮你更快地做事，或者将一项任务做上千次。不再服务少数人，而是服务更多人，因为现在这样做在经济上可行。这就是我看到的推动价值增长的两个模式，我认为这对释放大量 GDP 增长至关重要。

So to me, the question is: is AI mostly for cost savings or is it for growth? And I know it's difficult to change workflows. A lot of companies tend to think cost savings. But maybe here's the problem. There's actually one pattern I see. Let's say I have a workflow that has, you know, like five steps, right? And let's say each step takes 20% of my effort. Like maybe I'm underwriting approvals, you know, do I approve this loan or not? So let's say for simplicity, five steps each takes 20% of my effort. If you can automate one of those steps, that's a 20% cost savings, which is really nice. It could be great if you're a low-margin business, but it doesn't feel like a game changer. So what I find is that the more valuable users of AI actually require rethinking that workflow. And the pattern I see is: instead of taking a 20% cost savings, which you could do, that's fine, nothing wrong with that. The two patterns to then getting growth is either do more or do it faster. So in the case of underwriting, making loans, if instead of saving 20% of my human labor, if I can now rework the workflow to turn around my decision-making time. So instead of someone needing to wait, you know, two weeks before a loan officer looks at it, but we can just give you an initial answer in 10 minutes, that changes the product and lets you drive growth. So that's a faster pattern. And then there's also the more pattern. So another example: there are a lot of businesses that could do high-touch, say, customer service only for expensive high-end clients, right? But if you can now serve a much larger group of people, or let's say financial advice: instead of giving high-touch financial advice to a small group of people, if you can now deliver that quality of service to a lot more people, then that again changes a product and lets you drive growth. So instead of cost savings, if AI lets you do something way faster or lets you take a task and do it a thousand times more. Instead of serving a small number of people, let's serve a lot more people because it's now economic to do so. These are the two patterns I've seen to drive value increases, and I think that will be important for unlocking lots of this GDP growth.

你说在经济上可行。你认为我们是否必须看到垂直整合，比如英伟达既拥有模型又拥有芯片层，我们看到 Facebook 建设的数据中心比谁都多。每个人都在建数据中心。拥有每一层堆栈很重要吗？还是我们会看到参与者各自拥有堆栈的水平层？

You said economic to do so. Do you think it's crucial that we see vertical ownership in terms of, say, Nvidia owning models as well as the chip layer, and we're seeing Facebook build out data centers more than anyone. We're seeing everyone build out data centers. Is it important that we own every layer of the stack, or actually will we see individual participants own horizontal layers of the stack?

我认为这是随着时间演变的。我来打个比方。在计算产业的早期，赢家是垂直整合的玩家，因为如果你想把键盘连接到带有 CPU 的电脑主板上，键盘是正负 5 伏，而 CPU 是其他电压，这能行吗？我们不知道 API 边界在哪里。或者你的 CPU 内存布局和计算方式，以及你的数学加速器需要相互协作。所以在我们对边界和 API 有清晰概念之前，像 IBM 这样的整合型玩家能解决所有问题，构建出有价值的产品。但随着产业成熟，我们开始有了标准，比如现在有 USB 标准，之前有其他标准。现在你造电脑，别人造键盘，插上就能用。所以当产业不成熟时，如何划定边界让不同参与者各司其职并保持互操作，这不太清楚。但随着产业成熟，有了更多标准，比如我想在互联网上发布一个压缩模型，文件格式是什么？是时候看到更多标准了。这会让单个参与者更容易做出东西，并融入更大的生态系统。

I think it evolves over time. I'm going to make an analogy. In the early days of the computing industry, it was the vertical players that won, because if you want to connect the keyboard to your computer motherboard which has a CPU, is it okay if your keyboard has plus minus 5 volts and your CPU has some other voltage? Is that okay or not? So we didn't know where the API boundaries were. Or if your CPU has memory laid out a certain way and compute, and your math accelerator needed to interoperate with each other. So before we wound up having a clear conception of where to draw lines and what the API boundaries are, the integrated players like IBM back in the day could solve all the problems and build valuable working products. But as the industry matured, we started to have standards, like for example now we have a USB standard. Before there were other standards. So now you make a computer, someone else makes a keyboard, we plug them together and it all works. So when an industry is immature, it turns out where to draw those boundaries to let different participants do their part and have it still interoperate is less clear. But then as the industry matures and there are more standards, like if I want to publish a compressed model on the internet, what's the file format for that? It's time to see more standards. Then that makes it easier for individual players to do something and still have it fit into the broader ecosystem.

那么你认为扎克和山姆花那么多钱在数据中心上是正确的吗？还是他们应该耐心等待产业成熟，然后成为水平玩家？

So do you think then like Zach and Sam are right to be spending as much as they are on data centers, or should they be patient and wait for the maturation of the industry where they can then be horizontal?

我认为很明显，OpenAI 的投资到目前为止已经得到了回报。有可能在某个点过度投资，但我不确定现在是不是那个点。而且我认为很多参与者用来转移风险的金融工具也很有意思。我发现过度使用复杂的金融工具来转移风险，有时会增加出现泡沫的风险，所以这值得关注。

I think clearly OpenAI's investments have paid off to date. It is possible to overinvest at some point, but I don't know if that is the point. And I think also the financial instruments being used by many players to shift risk around have been really interesting. I find that overly used complex use of financial instruments to shift risk is sometimes increases the risk of there being a bubble at some point, so that's something to watch out for.

你担心循环交易吗？

Do you worry about the circular deals?

这值得关注。我并不感到惊慌，但你知道，我认为事情可能更泡沫化或更不泡沫化。这些迹象表明事情有点泡沫化。

It's something to keep an eye on. I'm not alarmed by them, but it is, you know, I think things could be more frothy or less frothy. Things could be more of a bubble and less of a bubble. And these are signs of things feeling a little bit more bubble-ish.

这些迹象什么时候会变成你的一大担忧？

When does a sign turn into a big concern for you with these?

我想你提到了红杉那篇关于 AI 6000 亿美元问题的文章。我对此感到担忧，但有趣的是，我对不同堆栈层的担忧程度不同。我看到应用层的 ROI 非常清晰，我认为这很棒。别人训练了这些模型，他们能用 10 万或 100 万美元构建应用，却没有产生 ROI？然后我认为，校准到合适的基础设施投资水平很棘手。但话虽如此，同时也很清楚，我们需要更多的电力、更多的数据中心和更多的半导体。这也很清楚。所以我们应该大力投资。我很高兴我们正在这样做。但到底投资多少才合适？我认为这是个棘手的问题。不过应该很多。

I think you mentioned the Sequoia article on the $600 billion problem of AI. I am concerned about that, but it's interesting, my concern for different layers of the stack is different. So what I'm seeing is for the application layer there is very clear ROI. I think it's fantastic. So someone else trained these models, who can build applications for $100,000 or a million dollars and not generating ROI? And then I think it is calibrating to the right level of infrastructure investment that is tricky. But having said that, it is also at the same time very clear that we do need more electricity, more data centers, and more semiconductors. That too is very clear. So we should be investing a lot. And I'm glad we are. But what exactly is the right amount to invest? I think that's the tricky question. It should be a lot though.

你对泡沫讨论感到恼火吗？

Do you get annoyed by the bubble discussion?

我对泡沫讨论并不恼火。但我确实对炒作感到恼火。我不知道什么时候监管机构会打电话给我说：‘嘿，我们听说 AI 可能导致人类灭绝。’谢天谢地，现在这种情况比几年前少多了。然后对话应该是如何提升劳动力技能，在哪里投资，而不是如何减缓这件事。我认为炒作严重扭曲了公众对 AI 的看法。炒作的一个缺点是，如果没有公众对 AI 的支持，事情就会放缓。例如，我有一个朋友经常和高中生打交道，他告诉我他和一个女孩谈论从事 AI 职业的事。她说：‘你知道吗？我听说 AI 可能与人类灭绝有关。我不想和那有任何关系。’所以这种炒作让一个高中女生在 AI 如此有前途的时候远离了它。我认为这确实导致人们做出奇怪的决定，无论是个人学生层面还是社区层面，比如社区关闭一个可能对社区和世界都有益的数据中心。我认为这也令人遗憾。

I don't get annoyed by the bubble discussion. I do get annoyed by the hype. I don't know when regulators are calling me up and saying, 'Hey, we heard AI could lead to human extinction.' Thankfully, much less of that now than a couple years ago. Then the conversation should be how can we upskill the workforce, where can we invest, not like how do we slow this thing down. I think the hype has really distorted public perception of AI. One downside to the hype is that without public support of AI, things slow down. For example, a friend of mine works a lot with high school students, and he told me he was talking to a girl about pursuing a career in AI. She said, 'You know what? I heard AI could have something to do with human extinction. I don't want to have anything to do with that.' So this hype turned a high school girl away from working on AI at a time where it'd be so promising for them to leap into AI. I think this really causes people to make weird decisions, both at the individual student level as well as at the community level, where a community shuts down building a data center that could be good for the community and good for the world. I think that's also unfortunate.

你对教育机构最大的建议是什么，以确保他们为学生适应 AI 一代做好准备？

What's your biggest advice to educational institutions to make sure they equip students for a generation of AI?

拥抱它，更新课程，尽可能多地教他们 AI。学生将生活在一个使用 AI 并得到 AI 帮助的世界里。必须教学生做到这一点。我认为不同领域会有所不同，但有一点是明确的：让所有学生都学会编程。

Embrace it, update curricula, teach them as much AI as possible. Students are going to live in a world where they will be using AI and having it help them. Got to teach students to do that. I think it'll be different for different fields, but one thing that is clear is get all your students to learn to code.

在过去 12 到 18 个月里，你对 AI 的哪一点看法改变了？

What's one thing you've changed your mind about AI in the last 12 to 18 months?

我认为我最喜欢的工具一直在变。如果你在过去一年里每三个月问我最喜欢的编程工具是什么，我的答案会不断变化。

I think my favorite tools keep changing. If you ask me every 3 months over the last year what my favorite coding tool is, my answer would have kept on changing.

你认为 Anthropic 会在编程大战中击败 OpenAI 吗？

Do you think Anthropic will beat OpenAI in the coding wars?

很难说。OpenAI 拥有非常强大的消费者品牌，这很难被攻破。相比之下，开发者更可能随时切换编程工具。所以我喜欢 Claude Code，它很棒，但我发现过去一个月我使用 OpenAI Codex 更多。我认为 OpenAI Codex 实际上已经获得了真正的动力。同时我也在关注 Gemini CLI，我认为它也在变得更好，可能比人们认为的速度更快。所以编程开发工具和 API 工具市场，模式比拥有强大消费者品牌更弱。我认为这是公司需要解决的问题。

Really hard to say. OpenAI has a very strong consumer brand and that's very defensible. In contrast, developers are more likely to switch coding tools on a dime. So I love Claude Code. I think it's fantastic, but I find myself using OpenAI Codex much more over the last month. I think OpenAI Codex has actually gained real momentum. And then I'm also keeping an eye on Gemini CLI, which I think is also getting better, maybe at a faster rate than people have given them credit for. So the coding dev tools and API tools market, the mode is weaker than having a strong consumer brand. I think that's something that companies have to sort out.

告诉我你从比亚迪最大的收获是什么？它和我们西方习惯的任何公司都如此不同。你最大的收获是什么？

Tell me what was your biggest takeaway from BYD? It's such a different company to anything that we're used to in the west. What was your biggest takeaway?

我非常欣赏比亚迪以及中国生态系统的速度和强度。我认为很不幸的是，在美国某些地方，建议某人努力工作被视为政治不正确或类似的东西。

I really appreciated the speed and intensity of BYD and also of the China ecosystem. I think it's really unfortunate that in some parts of the United States, advising someone to work hard is viewed as politically incorrect or something.

在欧洲，我会因此受到指责。

In Europe, I'm chastised for it.

哦，好吧。希望欧洲观众不会因此讨厌我或我们俩。坦白说，我希望人们每周工作 4 小时就能取得巨大成功，但现实是，当人们努力工作时，他们能完成更多。我想承认，并不是每个人在人生的每个阶段都能努力工作。所以我的孩子出生后那一周，我没有那么努力。我请假陪孩子，超过一周。我认为我们需要尊重各行各业的人，包括那些因为各种原因暂时无法努力工作的人。但如果有人想努力工作，引用史蒂夫·乔布斯的话，‘在宇宙中留下印记’，让我们赋予他们力量并庆祝这一点。如果有人因为任何情况无法努力工作，我们也尊重这一点，也许也庆祝这一点。但我认为这是一个我们可以建造很多东西的时刻。努力学习并建造东西的人会取得很多成就。

Oh, okay. All right, great. Hopefully the European viewers won't hate me or hate us both for that. I think frankly, I wish people could work 4 hours a week and be wildly successful, but the practical reality is when people work hard, they get more done. Now I want to acknowledge that not everyone at every point in their life is in a position to work hard. So the week after my kids were born, I didn't work that hard. I took time off to spend time with the kids, right, for more than a week. I think we need to respect people in all walks of life, including people that for whatever reason are not in a position to work hard at that moment. But if someone wants to work hard, go, quote Steve Jobs, 'make a dent in the universe,' let's empower them and celebrate that. If someone for whatever situation can't work hard, let's also respect that and maybe celebrate that. But I think this is a moment in time where there's so much stuff we could build. People that work hard to learn a lot and build things will accomplish a lot.

你经历过 996 吗？

Did you do 996?

996 这个术语不是我明确使用的。我发现现在的工作，我真的很热爱我所做的。它真的不像工作。但在很多周末，我坐在咖啡店里编码，因为这是周六我能做的最有趣的事情。所以我实际上懒得记录我的工作时间。可能很多。

The term 996 wasn't an explicit term that I use. I find that these days I work, I just really love what I do. It really doesn't feel like work. But on a lot of my weekends, I'm sitting in a coffee shop coding away because it's the most fun thing I could do on a Saturday. So I actually don't bother to keep track of my hours. It's probably a lot.

从运营者转变为投资者最困难的是什么？

What's the hardest transition element moving from operator to investor?

关于 AI Fund，是的，我们自称是一个基金，但坦率地说，我们日常运营基金的方式更像运营者而不是投资者。AI Fund 是一个风险工作室，我相信我们的技能实际上在于建设，而不仅仅是资本资产配置之类的。所以我们非常努力地筛选想法。我们与客户交谈，我有时自己也会参加客户电话。然后我们引入创始人与我们合作。我们审查产品，提供产品反馈，讨论定价。所以我的日常生活更像是运营者。是的，最终我们必须进行财务尽职调查，我开支票并进行后续投资。我们做所有这些，但更多。

One thing about AI Fund, yes, we call ourselves a fund, but frankly the way we run the fund day-to-day, we act much more like operators than investors. AI Fund is a venture studio, and I believe our skill set is actually in building, not just in capital asset allocation or whatever. So we work really hard to screen ideas. We talk to customers, I'm sometimes on customer calls myself. Then we bring in founders to work alongside us. We're reviewing the product, giving feedback on the product, arguing about pricing. So my day-to-day life is much more operator. And yes, eventually we have to do the financial diligence and I write a check and do follow-on. We do all that, but a lot more.

真的很抱歉，Andrew。那你是一个基金还是一个孵化器？

I'm really sorry, Andrew. Then are you a fund or are you an incubator?

所以我们称自己为风险工作室或风险建设者。孵化器通常引入已经有想法的创始人。我们比那更早。我们经常与我们的投资者和合作伙伴一起提出一个想法，只有在有了想法之后，我们才去寻找最好的创始人与我们共同创办公司。所以我们不称自己为孵化器。

So we call ourselves a venture studio or a venture builder. The term incubator usually brings in founders that already have an idea. We go earlier than that. We often work with our investors and partners to come up with an idea, and only after we have an idea do we go and try to find the best founder to co-found the company with us. So we don't call ourselves an incubator.

那么当你进行初始投资并种子公司时，你拥有多少所有权？

How much ownership do you have then when you make those original investments and seed the company?

这取决于情况。我们最终会获得一些普通股，作为建设公司的汗水股权，然后我们的第一笔投资通常是在 400 万美元上限下投入 100 万美元，所以大约 20%的所有权或 SAFE。

It depends. We end up with some common stock for the sweat equity of building the company, and then we usually our first check in is at like a million dollars at a $4 million cap, so kind of 20% ownership or SAFE.

所以我们基本上在进入时获得 20%到 25%的所有权，外加一些普通股。

And so we're basically getting 20 to 25% ownership on entry with a couple of common.

是的，加上一些汗水股权的普通股。

Yeah, plus some common for the sweat equity.

完全理解。

Totally get you.

你认为最大的……

What do you think is the biggest...

但对我来说，我们这么做的原因是，我发现虽然有些风投通过竞争性交易流赚了很多钱，但我认为我的团队最大的贡献不是争夺热门交易。而是发现创意并创建那些如果没有我们和创始人共同创办就不会存在的公司。所以我认为我们通过创建新公司而不是仅仅发现热门公司来投资，为世界创造了更多价值。

But to me, the reason we do this is because I find that while there are VCs who do the competitive deal flow thing and make a lot of money that way, I think my team's biggest contribution is not fighting over hot deals. It is finding ideas and creating companies that would not exist but for the fact that we and a founder got together to co-found it. So I think we create more value in the world by creating new companies rather than only discovering hot companies to put money into.

安德鲁，今天你最担心什么？我喜欢你的乐观和开放心态。反过来，什么让你担忧？

What concerns you most today, Andrew? I love your optimism and open-mindedness. What concerns you on the flip side?

让所有人跟上我们的难度。在以往的经济颠覆方式中，比如我们的国家从以农业为主转向非农业，一个农民可以一直种地直到退休，但孩子们必须学习不同的技能，也许搬到城市。这次变化如此之快，我们需要今天活着的人学习新技能，而不是需要他们的孩子学习新技能。这实际上非常具有挑战性。从历史上看，我认为我们从未擅长过这一点。

The difficulty of bringing everyone along with us. In previous ways of economic disruption, like when our nations went from mainly agriculture to non-agriculture, someone who was a farmer could keep farming until they retired, but the kids had to learn a different trade, maybe move to city. The change is so fast this time around that we need people who are alive today to learn new skills, as opposed to needing their kids to learn new skills. And that's actually very challenging. Historically, I don't think we've ever been good at that.

你做了很多采访，安德鲁。你和许多记者交谈。我不是记者。实际上从未有过工作。你觉得向你提问的采访者质量好吗？

You do a lot of interviews, Andrew. You speak to many journalists. I'm not a journalist. Never actually had a job. Do you find the quality of interviewers that ask you questions good?

我认为媒体在策划和传播知识方面扮演着重要角色。我认为记者提问的质量随着时间的推移明显在提高。但仍然存在炒作元素，不断扭曲信息生态系统。不幸的是，有财务激励、监管俘获、立法利益类型的激励以及某些类型的炒作。这实际上是我看到的一种模式。我不会点名任何公司，但我发现那些有东西可输的公司，随着时间的推移，他们的言论变得更加温和。所以作为一家成熟的公司，你只会说更明智的话。但有些公司我认为面临更大的生存风险。我发现那些公司，一些我不想点名的，是最糟糕的炒作来源，因为他们没什么可输的。他们在很多方面只是胡说八道。这是一种绝望中的发泄。当你看到 Demis，显然是一位杰出的领导者，或者你看到 Sam 甚至 Dario，他们所有人随着公司的成熟都显著缓和了自己的立场。

I think media has an important role to play to curate and disseminate knowledge. I think the quality of questions that reporters are asking has been very clearly trending up over time. But there is still the hype element that keeps distorting the information ecosystem. Unfortunately, there are financial incentives and regulatory capture, legislative benefit types of incentives, and certain types of hype. That's actually one pattern I've seen. I won't name any companies, but I find that companies with something to lose have over time become more moderated. So as an established company, you just say more sensible things. But there are some companies I think are at greater existential risk. And I find those companies, some that I don't want to name, to be the worst sources of hype because they've got less to lose. They just say a bunch of random stuff in many respects. It's a lashing out in desperation. When you look at a Demis, obviously a brilliant leader, or you look at a Sam or a Dario, all of them have moderated their position significantly with the maturing of their companies.

是的。不对个人发表评论，但我认为当你有东西可输时，你会说更明智的话，但当你的公司面临更大的生存风险时，有时人们为了筹款会说奇怪的话。我喜欢以乐观的语气结束。展望未来十年，你最兴奋的一件事是什么？对我来说，比如我母亲患有多发性硬化症。我认为我们将在一些多年来没有太大进展的疾病上取得令人难以置信的医学发现。这让我兴奋。什么让你兴奋？

Yeah. No comment on individuals, but I think that when you have something to lose you say more sensible things, but when your company faces greater existential risk, sometimes people say weird things for fundraising. I like to finish on a tone of optimism. What single thing are you most excited for when you look forward to the next decade? So for me, for example, my mother's got MS. I think we'll have incredible medical discoveries in some diseases that we haven't really made much advancement in for years. That excites me. What excites you?

听到你母亲的事我很难过。让我兴奋的是，我希望赋能每个人构建 AI。我认为从你有一个想法到实现它之间的距离现在大大缩短了，而且我们需要不仅仅是软件工程师来创造。所以未来，我希望很多人不再说‘有这个应用吗？’，而是说‘我为此建了一个应用。’不再只是软件用户，而是成为软件创造者。当你达到那个境界时，世界各地的人们将更有能力，完成更多事情，享受更多乐趣。

I'm sorry to hear about your mother. What excites me is that I want to empower everyone to build AI. I think the distance between you having an idea and building it is now much shorter, and we need not just software engineers to be creating. So in the future, I hope that a lot of people, instead of saying 'Is there an app for that?', they'll say 'I built an app for that.' And instead of just being a software user, they'll be a software creator. When you get there, people all around the world will be much more empowered, get more done, and have more fun.

安德鲁，这真是太愉快了。非常感谢你忍受我的刨根问底和追问。你太棒了，我真的很感激你的时间。现在，我真的很喜欢你的节目。所以，能来到这里是我的荣幸，哈里。

Andrew, this has been such a joy to do. Thank you so much for putting up with my prying and my pressing. You've been amazing and I really appreciate the time. Now, I really enjoy your show. So, it's a privilege to be here, Harry.