Krishna Rao - Anthropic's CFO on Compute, Scaling to $30B ARR, and the Returns to Frontier Intelligence - [Invest Like the Best, EP.471]

但这只是我们四处寻找可用算力的一个例子，就像你所说的那样。

But that's just one example of us just, as you said, looking for near term compute wherever we can get it.

随着算力总量不断增长，这种短期算力在整体中的占比会越来越小。

As the compute base grows, that near term compute becomes a smaller and smaller fraction of what's available and what's out there.

我们关注的是，能不能把现有的算力有效利用起来？

We look at it as can we deploy that compute that's available productively?

有时候答案是肯定的，有时候则不是。

Sometimes the answer is yes and sometimes it's no.

但如果可以，我们就会根据它的价格、使用时长、地理位置、算力类型，以及我们能多高效地运行它，来评估回报率。

But if we can, then we look at the economic return on it based on what its price, what duration we have it for, where it's located, what type of compute it is and how efficiently we can run it.

所以我们有一套评估流程。

So we have kind of a process to assess.

顺便说一句，这套流程我们也用来评估长期合作项目。

And that same process, by the way, we use to assess longer term deals as well.

上个月，我们和谷歌以及博通签了一份五吉瓦的协议，从2027年开始使用他们的TPU。

So last month we signed a five gigawatt deal with Google and with Broadcom for TPUs starting in 2027.

我们还和亚马逊签了协议，采购最多五吉瓦的Trainium芯片。

We also signed a deal with Amazon for Trainium for up to five gigawatts as well.

这笔承诺总额超过一千亿美元。

It was an over $100,000,000,000 commitment.

其中大部分算力已经到位，剩下的将在今年晚些时候到明年陆续交付。

And a lot of that compute is actually already landing and will land in the rest of this year into next year.

这么说吧，这些算力就像一层层的蛋糕，各自在不同时间点上线，能力也不一样。

And so if you think about it, it's a bit of this layer cake of compute that's starting at different times with different capabilities.

我们一直在动态地比较这些算力资源。

And we're very dynamically comparing that compute.

对我们来说，最关键的是算力的性价比，什么时候能用上，以及我们内部能用它做什么。

It's price performance over time that's really, really important to us, when it lands and what we think we can do with it internally in the business.

要优化的因素太多了——算力类型、成本、使用周期，每一个都得权衡。

And so there's so many different variables you have to optimize for around what compute it is at what cost and over what time horizon.

不过，我们有一套灵活的方法，来评估短期、中期到长期的算力需求。

But we have a pretty dynamic way of looking at near term compute and then medium to long term compute.

但我们评估的核心内容其实差不多。

But the things we're assessing are largely the same.

不一样的地方主要在于时间跨度。

What is different is just the time horizon.

那关于

What about the

性能与价格之间的权衡呢？

trade off use of price per performance?

比如按每个token的成本、吞吐量和速度之间的平衡？

Trade off between like cost per token or something, throughput and speed?

这两点当然都很重要。

Both are important, of course.

从客户的角度来看，他们都很在意。

From the customer perspective, they care about both.

速度的提升可能会解锁一些我们现在还想不到的有趣能力和应用场景，尤其是随着技术越来越快。

Speed probably unlocks some capability and use cases that are really interesting that we don't know about yet as these things get faster.

你能稍微讲讲，在评估时你们是怎么权衡计算资源的吗？

Can you talk a little bit about that trade off in compute as you're assessing it?

我们在评估三个不同的芯片平台时，也会同时考虑每个平台的多个代际产品。

As we look across three different chip platforms, we also opt multiple generations of chips within it.

比如 TPU V5E、V6、V7，还有 Trainium 2、Trainium 3 这些。

It could be TPU V5E and V six and V7 and Trainium two, Trainium three.

这些芯片在性价比曲线上所处的位置都不一样。

All of them are at different places on the price performance curve.

更重要的是，我们会看怎么实际使用它们。

And then we importantly look at how we will utilize it.

性价比很重要，因为它关系到效率。

Price performance is important because of efficiency.

速度对某些应用场景也同样关键。

Speed is also important for certain use cases as well.

所以我们对计算资源的评估非常细致，具体到每种芯片能带来什么、在什么时机用。

So we look at the compute down to a very granular level in terms of what it can deliver for us and when.

这一点我们确实一直在做，由计算团队主导，但我们会和业务部门紧密协作，共同确定：我们需要这些算力在哪些地方、用来做什么。

And that's something that we do, again, our compute team leads that, but we closely collaborate across the business to say like, where do we need this compute and for what?

比如，我们可能需要CPU来支持强化学习任务。

Okay, we might need CPUs for RL.

那我们需要多少？用在哪儿？具体是什么样的需求？

Okay, how much of that do we need and where and what's the nature of it?

我们可能需要更前沿的计算资源，专门用于部署或训练我们最先进、速度最快的模型。

We might need this more leading edge compute and we're going to deploy it for our best and fastest models or for training them.

从我们的角度看，这既取决于客户需求，也取决于每种芯片最适合做什么，以及我们什么时候能用上它们。

From our perspective, it's customer demand, but it's also really down quite granular in terms of what is each chip best for and then what will we have when.

我总是很好奇，像Anthropic这样的公司，新算力的消化速度有多快——假如明天突然给你增加两倍的算力，你会怎么用？

I'm always so curious by the metabolism of, in this case, Anthropic for new compute, how fast you could take if I airdropped on you twice the compute that you have tomorrow.

你能消化得掉吗？

Like, would you consume that in?

你大概多久能用完这些新增的算力？

How fast would you consume that?

如果明天我一下子给你十倍的算力，你多久能用完？

If I airdrop 10 times the compute on top of you, how fast would you consume it?

你能给我们一个大概的估算吗？

Can you calibrate us on that?

在这三个用途上——训练、内部客户需求——需求好像根本停不下来。

It feels like demand's unlimited between these three uses, training, internal customer demand.

每个人说的都是一回事。

Everyone's saying the same thing.

哪里都缺货。

Shortages everywhere.

内存股票涨得飞快。

Memory stocks booming.

如果明天你手里的算力直接翻两倍、五倍、甚至十倍，你是不是立马就能用光？

Is it that extreme that if you 2x or 5x or 10x the amount available to you tomorrow, just more or less instantly consume it?

这其实跟我们怎么用、以及算力的通用性有关。

This goes back to how we use it and the fungibility of it.

所以答案是，我们在这些应用场景上都受到限制。

So the answer is we're constrained across those use cases.

一两年前要消化掉这些算力会难得多，尤其是像你提到的异构算力突然增加，因为这些芯片平台各不相同，差异很大。

A year or two ago would have been harder to consume, especially like a heterogeneous compute drop in your example, really quickly because these chip platforms are different and they are different.

有些平台用起来更复杂。

Some are harder to operate.

它们在使用方式上各有各的特性。

Some of them have idiosyncrasies in terms of how we use it.

我觉得现在要是能多拿到一批算力，应该能很快部署到各个应用场景里。

I would say today that getting a bunch more compute, I think it would be deployed very rapidly across those different use cases.

我们目前对算力的分配和调配方式，跟以前差不多。

We probably have the same allocation or calibration that we do with compute today.

但我们现在能更快地启动并部署几乎任何类型的算力了。

But it's become a lot easier for us to spin up very quickly and deploy almost any type of compute.

这一点，我们认为是个实实在在的优势。

And that's something, again, we think is a real advantage.

其中一个

One of

我特别想听听你怎么权衡的，就是平台模式——我把业务搭建在Claude之上，让它驱动我的产品——和你直接替我做出我想做的东西之间的张力和取舍。

the interesting tensions and trade offs that I'm fascinated to hear how you think through is between the platform approach where I build my business on top of Claude and it powers my thing versus you doing the thing that I wanted to build.

这就像Claude的设计思路和Figma这类产品的经典区别。

This is like the classic Claude design versus Figma or something like this.

你怎么看待该在应用层走多深，是只做一个纯粹的赋能层，比如提供推理引擎和智能能力，然后让别人自由发挥、通过API付费，还是深入去做具体的应用？

How do you think about the right balance of how deep into the application layer you should go versus just being a pure enabling layer of we're going to provide the reasoning engine and the intelligence and world go forth and build whatever you want, pay us through the API or whatever.

这看起来是个很有趣、内部也确实存在矛盾的讨论。

That seems like a fascinating internal discussion and tension to some degree.

我觉得我们目前做的大部分东西，本质上都是平台。

The way I would think about it is most of what we're building is platform.

我们相信，平台能积累巨大价值，但真正获得更大价值的，是那些在平台上构建产品的客户。

We think that there's so many examples of where a platform can accrue a lot of value, but the customers who are building on that platform actually accrue even more value.

我们今天正在做的，就是为这种局面打基础。

We think that's what we're setting up for today.

这有点像AWS的早期阶段。

It's maybe akin to the early days of AWS.

如果你想想Claude平台以及现在集成到其中的所有工具和服务——这不仅仅是直接访问模型，还包括提示缓存、使用虚拟机、在内部调用Claude Code或Dispatch，等等。

If you think about the Claude platform and all the tools and services that are now built into it, because it's not just the raw model access, it is prompt caching and the ability to use virtual machines and Claude Code being called within there or Dispatch, on and on.

Claude Agents SDK和托管代理。

The Claude Agents SDK, Managed Agents.

所有这些在我看来，都是让其他公司能把模型的智能融入自己产品中的各种途径。

All of these are effectively, I think of as vectors to access that model intelligence for other companies to build into their own products.

这正是我们目前最关注的方向，也是我们认为业务未来最主要的增长点。

That's most of what we're focused on and really most of where we think the business is going from where we are today.

我们也会在这个平台上开发自己的应用，因为有几件事是成立的。

We will also build our own applications on that same platform where a couple of things are true.

第一，如果我们对模型的未来方向有清晰的洞察，并且能借此展示价值、为用户创造实际收益，那可能就像Claude Code这样的产品。

Number one, if we feel like we have a vision into where the models are going and we can kind of demonstrate that and create customer value in that, that might be something like Claude Code.

我们发现，市场上很多东西其实都是由开发者推动起来的。

We are able to say actually a lot of what's out there in the market was developer led.

Claude Code 是由 Claude 团队主导的平台。

Claude Code is a platform that's Claude led.

当初一年多前推出时，我们觉得模型还做不到这一点。

And we think the models can't quite do that today when it was launched a little over a year ago.

但现在我们认为，模型已经做到了，而且还会继续进步。

But we think they'll get there and they have.

一是提前布局，顺应模型能力的发展方向。

One is kind of building ahead to model capabilities.

二是探索如何为整个生态创造可借鉴的价值。

The second is thinking about ways to demonstrate value for the ecosystem that others might emulate.

比如 Claude 针对金融、生命科学，甚至 Claude Security 这样的场景，都是我们对平台的组合应用。

If you think about Claude for financial services or Claude for life sciences or even something like Claude Security, these are ways in which we've composed the platform.

我们和客户用的是同一个平台。

Again, we're building on the same platform as our customers.

这就创造了一个公平的竞争环境。

That creates a level playing field.

我们还认为，这些领域将产生巨大的价值，客户能赢，我们也能赢，所以当你看到我们推出这些产品时，无论是安全、SCIM还是金融服务，我们都是以合作共生的方式推进的。

We also think that there's so much value that's going to accrue in these areas that our customers can win and we can win as well, which is why you've seen as we've launched some of these products, we've done them in a collaborative partnership oriented way, whether that be on the security side or SCIM or financial services.

我们一直与生态系统紧密合作。

We've partnered with the ecosystem.

我觉得我们的策略主要是横向的。

So I think of our strategy as mostly horizontal.

只有在我们认为能带来独特价值、独特视角，或者能向市场展示平台如何创造价值的时候，我们才会做垂直整合。

We'll build vertical where we think we have some value to add or a perspective that's useful or a way to demonstrate to the market how we think about our platform adding value.

大部分价值都会流向那些在我们平台上构建应用的客户。

A lot of the value is going to accrue to the customers that are building on top

的用户。

of it.

我们的目标是打造最优秀的模型，然后开发产品、工具和服务，让这种智能能力在客户内部广泛传播。

Our goal is build the best models and then build the products and tools and services that allow that intelligence to proliferate within customers.

你有多在意，人们其实害怕你这个事实？

How much do you care that it's just a reality, that people are scared of you?

你掌控了这些新应用最核心的部分——底层的智能推理引擎，这确实可能是事实，甚至现在就已经是事实了，大部分价值其实都落在了Anthropic平台之上，而不是被平台本身捕获。

There is a sense that because you control the most essential piece of these new applications, the underlying intelligence reasoning engine, that may totally be true, maybe already is true, that more of the value is accruing on top of the Anthropic platform than is being captured by it.

但即便如此，一想到这个还是让人心里发毛。

But nonetheless, it's still scary to imagine.

你或许也可以对云计算和AWS说类似的话。

And I guess maybe you could say something similar about cloud and AWS or something like that.

但你到底有多在意，你的潜在客户或现有客户，其实对你这个竞争对手感到害怕？

But how much do you think and care about the fact that some of your would be customers or existing customers are in fact scared of you as a competitor?

这个行业难就难在，它变化得太快了。

Part of what is hard in this business is it's changing so quickly.

模型的能力，有时候连我们自己都没想到。

So the model capabilities sometimes even surprise us.

所以我们发布模型或基于它的产品时，过去五到十年才发生的变化，现在几个月就全赶上了。

And so when we release models or products on top of that, there's an element of what's happened in prior waves over the course of five years, ten years, twenty years is happening in months now.

我们一推出新东西，用户也跟我们当初一样，感到意外。

When we release things, people are also surprised by it in some ways in the same way that we were surprised by it.

但我觉得，我们根本上想做的，是围绕整个生态系统做深度合作。

But I think fundamentally what we are trying to do is be very partner oriented towards the ecosystem.

所以我们会有早期试用计划。

And that means that we have early access programs.

我们会和客户紧密配合。

We work very closely with customers.

认真听他们说，希望我们提供哪些功能。

We listen to them about what capabilities they want.

但这并不意味着，我们发布的东西不会让人惊叹——‘天啊，这比我想的强太多了’。

That doesn't mean that the things we release are sometimes not moments where you're like, wow, that's way more powerful than I thought it would be.

或者‘我真没想到模型这么快就能做到这种程度’。

Or I didn't realize the models would be able to do that this quickly.

这其实反映了我们当前所处的这个智能发展周期的现实。

Part of that is a reality of where we are in this cycle, in this development of intelligence.

另一方面，我们也希望把这些能力真正普及开来。

Part of it is also like we want to make those capabilities really accessible.

这对客户来说也会带来巨大的价值。

And that should accrue a lot of value to customers as well.

那些早早介入、积极采用我们平台工具的客户，我们相信能帮他们加速成长。

And customers that are front footed on that and adopt and also ones that are building and using the tools that we offer on our platform, we think we can actually accelerate them.

这在一定程度上是前沿模型开发的现实，但我们的做法可能更不一样，也更注重合作。

And so I think some of it's a reality of frontier model development, but our approach to it is probably a little different and more partner oriented.

你之前提到第一季度从九到三十，进展快得惊人，这让我很想知道定价的问题。

You said before going nine to 30 in the first quarter, the pace is so insane, which makes me wonder about pricing.

关于如何为令牌或系统使用定价，这个动态真的特别有意思，因为一年前很多人会觉得价格肯定会一路下跌。

The dynamic of how to price tokens or use of the system is so fascinating to me because I think a lot of people a year ago would say price is going to constantly fall.

但实际情况是，很多地方的价格反而在上涨。

But actually what's happening is it's going up in many cases.

这在不同层面都成立，比如像Mythos这种功能强大、定价很高的产品，还有H100的硬件成本，以及H100的租赁价格，看起来都像一条微笑曲线。

This is true at different levels, whether it be the mythos pricing that is quite high because it's so powerful, the cost of an H100, the rental price for cost of an H100 looks like a smile curve.

我很困惑，既然大家都受算力限制，为什么没人干脆大幅提价，来试探真正的均衡点在哪里呢？

I'm very curious why if everyone is compute constrained, why everyone doesn't just raise prices a lot to try to find what the right equilibrium is.

所以我想听听你对定价的看法，你是怎么考虑的，其中有哪些权衡，为什么不多涨点价呢？

And so I'd love you to just like riff on pricing, how you think about it, what the trade offs are, why not raise prices a lot.

公司成立才刚过五年。

The company is only a little over five years old.

去年三月是我们第一笔收入到账的三周年纪念。

This past March was the third anniversary of the first dollar of revenue into the business.

直到2024年3月，我们才真正拥有了自己的前沿模型。

And we only had a frontier model for real for the first time in March 2024.

所以，从时间尺度来看，这很重要，也是我思考问题的背景。

So the timescale of these things, it's an important backdrop as I think about it.

在Haiku、Sonnet和Opus上，我们的定价一直相对稳定，而Mythos是新推出的，情况不同。

Our pricing has been relatively stable across Haiku, Sonnet, and Opus, and now Mythos is obviously newer.

我们几乎没怎么调整过价格。

We've made very few pricing changes.

我们最大的一次调价，是在推出Opus 4.5时，降低了Opus系列的价格。

The biggest pricing change we made was to bring down the price of the Opus family when we launched Opus 4.5.

如果我们想想为什么会这么做？

If we thought about why did we do that?

其实是因为我们发现OPUs系列模型的使用率远低于它们的实际能力。

It's really because we found that OPUs class models were underutilized relative to their capability.

很多人明明该用OPUs，却硬是把问题塞进Sonnet里去解决。

People were trying to often fit an OPUs problem into a sonnet.

从我们的角度看，用OPUs来服务这些需求效率很高，降价之后客户用起来也更轻松了。

We were able to serve that very efficiently from our perspective, but actually bring down the price, which made it more accessible to customers.

我们希望客户能从这些模型里获得大量价值。

We want our customers to generate a lot of value from it.

现在他们已经从我们的模型中获得了很高的投资回报。

They're generating a ton of ROI from our models today.

我们希望这种势头能持续下去，因为我们的目标是让这些技术在整个生态里普及开来。

We want that to just continue because our goal is to proliferate this throughout the ecosystem.

我们认为，这些应用场景才刚刚起步，还处于非常非常早期的阶段。

We think we're in the very, very early innings on all of these use cases.

要做到这一点，最好的办法就是让这种智能技术普及到各行各业，从初创企业到数字原生公司，再到全球最大的企业。

The best way to do that is get this intelligence in the hands of as many businesses from startups to digitally native businesses to the largest companies in the world.

这其中一部分意味着，我们必须把价格定在大家负担得起的水平，让他们能从中获得实实在在的价值。

Some of that means that you have to make it in a price point that's accessible and that allows them to get a lot of value from it.

在调整Opus的定价时，我们看到了杰文斯悖论的现象。

Changing pricing for Opus, you see this Jevons Paradox.

我们降低了价格，但使用量的增长远远超出了我们的预期。

We lowered the price of it, but the consumption went up way, way more than what you would have expected.

正因为我们在客户需要的点上找到了平衡，他们才得以更频繁地使用它。

And so because we hit that sweet spot for customers, they were able to use it a lot more.

我们自身也有足够的效率，能够大规模地为客户提供服务。

We had the efficiency to be able to serve it to customers at scale.

客户随后把这套系统整合进了自己的工作流程，所以当我们推出Opus 4.6时，这只是一个模型升级，他们可以直接无缝替换。

And then they were able to build that into their workloads such that when we released Opus 4.6, it's a model improvement, they can slot it in.

我们并没有调整价格。

We didn't change the price.

所以我们认为定价的稳定性很重要。

And so we think pricing stability is important.

而且我们也觉得，定价要能带来价值，让这种杰文斯悖论真正发生，这一点非常关键。

And we also think that pricing to get that value and to see that kind of Jevons Paradox happen is really important.

这方面另一个因素是利润率，以及企业如何思考利润率。

The other component of this is margins and how you think about margins as a business.

again，因为搭建这些前沿实验室需要巨额资本投入，我们之前提到的那些手段——效率和价格——都直接影响利润率。

Again, because this is so unbelievably capital intensive to build these frontier labs, you've got the levers we talked about, which is efficiency, price.

这两者都跟利润率息息相关。

Both those things relate to margin.

既然需要这么多资本投入，为什么不干脆定一个健康的利润率，再据此设定价格呢？

Given how much capital you need, why not just say we want to have a healthy margin and set the price accordingly?

当然，如果效率提升了，价格也有可能随之下调。

And maybe that price can come down if efficiency is better or whatever.

我只是想知道，你们是怎么把利润率和定价结合起来考虑的？

And I'm just curious how you think about margins as it relates to pricing in the business.

我们考虑的是整体计算支出的回报。

We think about what is the return on our compute spend writ large.

也就是我们之前提到的各种工作负载，比如服务客户。

So that is all of the different workloads that we've talked about, whether it's serving customers.

这些都可以看作是在不同时间周期内支持收入的活动。

If you think of all of those are kind of in support of revenue over different timescales.

如果我做的是推理服务，那是在为今天的收入做贡献。

If I serve inference, it's in support of revenue today.

如果我做模型开发，可能是为了六个月后能打开新市场、带来收入的能力做准备。

If I do model development, it might help for a capability that unlocks TAM that drives revenue six months from now.

中间的各种情况也都一样。

And everything in between.

比如我内部优化来推动新产品上线，这些也都服务于同一个目标。

If I do internal acceleration to launch a new product, all of these things are in support of that.

目前来看，我们的计算投入回报非常可观。

Our returns on that compute expense today are robust.

我们把它看作是对整个计算资源投入的回报。

We think of it as what is the return on that full envelope of compute?

从这个角度看，我们对自己的现状非常有信心。

We feel really good about where we are from that perspective.

我们在为客户创造价值的同时，也获得了非常强劲的计算投入回报。

And we're balancing delivering value to customers with also seeing a really, really strong return on that compute ourselves.

就像我们之前在第一季度提到的，收入增长时，并不是因为我们在这段时间新增了大量计算资源。

If you think about when revenue grows, as we mentioned kind of in Q1, it's not like we onboarded a bunch of new compute in that time period.

我们说过，计算资源的投入其实是基于一年前就规划好的增长节奏。

We talked about compute comes based on a ramp that might have been determined twelve months ago.

那种为服务单个客户而产生的可变成本概念，并不适合我们的商业模式。

This idea of variable cost that's on the incremental to serve a customer, it doesn't really fit our business.

这种说法是想把我们的业务硬套进软件行业的模式里。

It tries to maybe fit our business into a software paradigm.

实际情况并非如此。

That's not the case.

实际上，计算资源支撑着所有这些业务活动。

In actuality, compute is supporting all of these activities.

我们在计算资源上获得了强劲的回报，这正是我们的衡量标准。

We're really generating a robust return on that compute, and that's our measuring stick.

我们认为，所拥有的计算能力决定了我们短期内和长期内能推动多少收入。

I think it's something where we think of the compute envelope that we have as the thing that is able to govern how much we're able to drive revenue both over the short term and the long term.

如果你是这些

If you're this great customer of the

计算服务商的优质客户，那这些服务商需要做些什么，才能帮你实现这样的回报？

compute providers, what does that group need to do to be a great provider to you to help you drive that return?

我们很幸运的是

We're fortunate in that

我们和亚马逊、谷歌、微软都有非常好的合作伙伴，同时还有博通和英伟达。

we have really great partners in Amazon, in Google, in Microsoft, but also with Broadcom and Nvidia as well.

在我们的生态系统里，我们是唯一一家同时部署在三大云平台上的公司。

Our ecosystem, we are the only model that's on all three clouds.

现在，我们是唯一一个在三大芯片平台上都运行大型语言实验室的模型。

Today, we're the only model that has large language lab that's using all three of these chip platforms.

这些合作远不止是简单的采购关系。

These collaborations are much deeper than just procurement.

这一点常常被人忽视。

I think that's something that's often lost.

就拿我们和亚马逊的合作来说，我们的团队已经深度融入了Anthropic实验室的团队。

If you think about our relationship with Amazon, our teams are deeply embedded with the Anthropic Labs team.

我们大量使用Trainium芯片。

We are good users of Trainium.

我们花了很多时间和精力，和对方团队紧密协作，内部也做了很多工作。

We've spent a lot of time and energy and worked closely with the team and internally.

我们会一起规划算力容量。

And that's something where we plan capacity together.

和其他几家超大规模云服务商，我们也都是这么做的。

We do that with the other hyperscalers as well.

如果你看看这三大云平台，它们对我们来说也是极佳的分发渠道。

If you think about the three clouds, they're great distribution engines for us too.

我们自己的直接业务也同样非常强大。

We have a really, really robust first party business as well.

这些合作关系是多方面的，无论是芯片开发、部署算力、提供服务，还是最终把产品送达客户手中。

But these are multifaceted partnerships, whether it be on developing the chips themselves, landing that capacity, serving it, and then ultimately distributing it to the customers.

每一个合作伙伴关系都包含这些要素。

Each one of them has that element to it.

它们各有不同，各有优势，但我们在这几大云平台之间，以及与博通和英伟达的合作中，都配合得非常顺畅。

They're different and each of them has relative strengths, but we've been able to work really well across those clouds and then obviously across Broadcom and Nvidia as well.

我在想你们团队，比如财务部门，是怎么运作的。

I'm thinking about your function like the finance team.

我想象着在不同时间维度上，面对这么多复杂变量，计算算力的投资回报率，这让我好奇，你们自己内部是怎么用这些强大工具来管理团队和业务的？

I'm picturing this ROI on compute thing on different horizons with all these complex variables, which makes me wonder, how do you use these powerful tools yourself internally to run your group and the business?

在Anthropic的财务团队里，Claude Code和Claude整体是怎么应用的？

What is the deployment of Claude Code and Claude in general on the finance team at Anthropic?

这真的很有趣，因为我们大约一年前就开始用Claude Code了。

This is really interesting because we were using Claude Code about a year ago.

我开始问大家，是不是大家都只是在随便用它写代码，还是有别的原因？

I started asking people, is everyone just kind of taking up vibe coding or what is it?

我们渐渐把Claude Code当成一个助手，一个数字同事，不只是用来处理编程任务。

And we started to use Claude Code as almost like an assistant, a digital coworker, not just for coding tasks.

这其实正是后来CoWork的雏形。

And that actually was early in what eventually became CoWork.

我们想，既然它在智能软件开发上这么有效，那它也应该能帮上所有知识型工作的忙。

That was kind of an extension of Claude Code to say what it's done for agentic software development it should do for all of knowledge work.

后来我们就开始把它真正投入实际使用。

But then we started to productionize that.

我对此真的特别自豪。

And I'm actually really proud.

我们和产品团队也花了大量时间一起打磨这个东西。

We spend a lot of time with our product team too.

他们通过观察我们的使用方式，来提供意见和反馈。

They kind of see how we use it and get input and feedback from that.

如今，我们所有的法律实体都能用Claude生成法定财务报表。

But today, all of our legal entities, we can produce the statutory financial statements using Claude.

当然，最终还是有人复核，但所有这些报表都是由Claude生成的。

And yes, a human checks it, but all of those financial statements are produced with Claude.

我们还开发了一个更实时的平台，叫AntStats。

We also have a more real time platform called AntStats.

以前，要花大量时间梳理数据、得出结论、写备忘录，或者每天发布报告，说明当天发生了什么、背后的原因是什么。

And it used to take a lot of time to sift through all the data, get to conclusions, write a memo about it, or publish a regular report on what's happening over the course of the day, what's driving it.

现在，我们为Claude建立了一套专门针对财务的技能库。

We now have a library of skills for Claude that are specific to finance.

我上次统计时，已经有七十多个技能了，所有人都能通过这个共享库访问。

Last I checked, there were over 70 of them that everyone can kind of access through this common repository.

除此之外，我们还开发了一个名为MFR的月度财务复核技能。

And on top of that, we built an MFR, a monthly financial review skill.

它能生成我们的月度财务分析报告。

And it can produce our monthly financial review.

已经九成到九成五的完成度了。

It's 90 to 95% ready.

接下来我们的讨论就集中在该怎么做上。

And then all of our discussion becomes about what do we do?

这些变化背后意味着什么？

What are the implications?

不是去确认具体发生了什么，因为Claude不只是在报天气。

Not what exactly happened because Claude is not just reporting the weather.

它还在帮我们分析背后的原因。

It's also helping to think about drivers.

为什么数字会这样变动？

Why did the number change in the way it did?

这让我们对业务有了深刻的洞察，无论是做月度分析，还是日常运营都一样。

And that gives you tremendous insight in the business, both in terms of this like MFR that we do, but also on a daily basis.

以前做一份关于收入驱动因素或计算资源使用情况的周报要花好几个小时，现在只要三十分钟就能搞定。

And so what used to take hours to produce a weekly report for what's driving revenue or what's driving our compute utilization now comes down to thirty minutes.

这样一来，我们就能把时间花在真正重要的业务战略影响上。

And then we can spend our time on the actual strategic implications of the business.

而且我们也能更快地把分析结果交给业务负责人。

We can also get it in the hands of business leaders much more quickly.

这个洞察引擎的速度快多了。

The insight engine is a lot faster.

我有个仪表盘。

I have a dashboard.

我会查看令牌使用情况。

I look at token usage.

还有排行榜。

Leaderboard.

是的，我们并不根据这些数据来给员工发奖金。

Yeah, we don't compensate people on it.

没人

No one's

为了这个去最大化使用令牌。

trying to token max for that.

但有意思的是，财务团队里一些资历最深的人反而是使用令牌最多的人。

But it's really interesting because some of the most senior people within the finance team are actually the biggest users of tokens.

这不只是那些刚入职的22岁年轻人，他们有编程背景，周末自己玩，然后带到工作中来。

So it is not just the 22 year old who joined and has a coding background and was doing that on the weekends and brought it to work.

还有人用这些工具来改变自己的工作方式。

It's also people using the tools to change how they're working.

比如，我们头号用户是我们税务主管，他专注于税务政策引擎，自动化团队里大量工作流程。

Like I think our number one user is our head of tax and he's focused on tax policy engines and automating large parts of the workloads that are happening within the team.

我特别喜欢看到这种现象。

So I love seeing that.

我常跟人说，如果我们自己都不用、不挑战极限，怎么指望客户去这么做呢？

I tell people, if we're not super users of this, if we're not pushing the limits of it, how can you expect customers to do that?

作为一个人，你有没有觉得有点毛骨悚然？

Just as a human, does it freak you out at all?

我听过太多这样的例子了。

I've heard so many examples like this.

渐渐地，我们好像就开始听AI的安排了，不管是销售、日程还是别的什么。

It starts to feel like we just start doing the stuff that AI tells us to do in the sales example or the calendar or whatever.

也许这确实挺好。

And maybe that's great.

也许AI比我们更会协调、更有全局观、更擅长优化，我们确实该听它的。

Maybe it's just such a better coordinator and wide thinker and optimizer than we ever could be that we should do what it tells us to do.

但我觉得，这种未来正飞快地逼近，多少有点令人不安。

But it feels ever so slightly dystopian to me that that reality is coming quickly.

我自己也遇到过类似的情况。

I've had examples of it too.

感觉还挺酷的。

It feels kind of cool.

确实挺有用的。

Like it's helpful.

但与此同时，如果我仔细想想，我会觉得，哎呀，我好像只是在照它说的做，而不是我在指挥它。

But at the same time, if I really close my eyes, I'm like, Oh, I'm just like doing what it tells me versus me telling it what to do.

这真是一个很有意思的人际动态，我很想听听你的看法。

It's a really interesting just human dynamic that I'm curious for your take on.

我对此可能有点不同的看法。我觉得我们公司在招聘方面一直做得不错，招到了很多优秀人才，但AI让这些本来就很厉害的人效率变得更高了。

I maybe have a slightly different view on it in that I think we've been able to hire great people at the company, but it has made even those incredibly talented people so much more productive.

我觉得这有点像杰文斯悖论，只不过是在劳动力方面的体现，也就是说，人们的效率变得极高。

There's a little bit of this, I think of it again like Jevons Paradox, but for labor, which is that we have people who become incredibly more productive.

正因为如此，我们招了更多的人，因为根本不用担心活儿不够干。

We've hired a lot more people because of that, because there's no shortage of work to do.

现在有了 Claude 的协助，大家花在 MFR 上对账的时间少了，反而有更多精力去思考，比如怎么把省下来的资源再投入到业务里。

And now with the assistance of Claude, people are spending less time in that MFR trying to reconcile some number, but they're actually thinking, oh, how do we reinvest this in the business?

我们该怎么动态地分配资源呢？

How do we think about dynamically allocating resources?

以前我得花很长时间对账，比如在会计工作里，要花好几天才能结账。

Whereas before I'm working to tie out a number or I'm, in that accounting example, taking a long time to close the books.

所以我反而觉得这更像是大幅提升我们效率的加速器。

So I actually think of it as even more optimistically that it is an accelerant to our productivity.

这意味着我们能完成更多工作，而且即使团队扩大了，新成员在熟悉如何使用Claude之后，效率也会越来越高。

And that actually means that we can get a lot more done And that even as we grow the team, those people are more productive as well as they come up the curve on how to use Claude within our company.

我觉得这种情况现在在很多公司也开始出现了。

And I think that's starting to be true across many companies as well.

我很想聊聊投资者和资本筹集的事。

I'd love to talk about investors and capital formation.

当然。

Of course,

你之前得筹集了巨额资金。

you've had to raise tons and tons of capital.

但另一方面，如果我稍微换个角度看，以你们当前的收入来算，融资估值其实也没那么离谱。

At the same time, it seems as though if I just squint my eyes and think about the multiple on current revenue, it's not that crazy in terms of where you're raising money.

我特别想知道，你和投资人打交道的经历是怎样的？他们的理解是如何一步步加深和成熟的？

I'm so curious for you to teach us about what it's been like to interact with investors, how you've seen their understanding of the company evolve and mature.

你觉得现在投资人群体对公司的理解到了什么程度？

Where do you think the investors as a group understand it now?

他们对Anthropic还有哪些误解？

Where are their misunderstandings about Anthropic?

跟我们说说这一面的生活吧。

Tell us that side of your life.

我不知道这占你工作量的百分之多少，但确实是很重要的一部分。

And I don't know what percentage it is of your job, but it's an important part.

我大约两年前加入这家公司。

So I joined the company about two years ago.

那时候我们正在完成D轮融资。

We were closing our Series D at the time.

那轮融资并不顺利。

That was not a straightforward fundraising.

那时候，公司其实只有一款前沿模型，正好处在融资中期。

The company really only had a frontier model in the middle of that fundraising.

融资快结束时，正好赶上FTX事件，他们正在清算一批Anthropic的股份。

Towards the tail end of it, the FTX transaction was happening, which was liquidating a bunch of Anthropic shares.

那算是整个事情的起点。

That was kind of the starting point.

那时候，大家最常问的是：你们为什么非得搞前沿模型？

And at that point, the questions were around why do you need to have a frontier model?

这能带来什么回报？

What's the returns to this?

他们也对我们的使命和做事方式有很多疑问。

They're also around our mission and how we approach things.

有人觉得，AI安全和做大生意，这两者根本是矛盾的。

People said, well, AI safety and building a really big business, aren't those things at odds?

还有不少人误以为，你们的销售团队规模特别小。

And there were also a lot of other misconceptions on your Salesforce is really small.

难道你不该像那些企业级软件公司一样把它做大吗？

Don't you need to scale it like all these enterprise software companies?

当时总想把我们套进一个早就存在的固定模式里，有点勉强。

There was just a bit of a paradigm around trying to fit us into a particular mold that had existed before.

随着时间推移，情况慢慢变了。

Over time, it's evolved.

到了2024年，我们完成了E轮融资。

At the 2024, we raised the Series E.

那时候，公司的年化收入已经接近十亿美元了。

The business had scaled to close to a billion dollars of run rate revenue.

但我们首轮交割的那天，深海新闻刚爆出来。

But the day of our first close was the day the deep sea news came out.

交易是完成了，但市场瞬间剧烈波动，很多人开始重新思考：我是不是该彻底重新评估对AI的整体看法？

We got the close done, but certainly a ton of volatility as people then said, wait a minute, should I just totally re underwrite how I think about AI in the total?

这就是E轮的情况。

That was a Series E.

当然，我们吸引了许多优秀的投资人，但大家还是有些疑虑，他们看我们的预测，心想：你们这么快就做到了十亿美元收入，这怎么可能持续下去呢？

Obviously, brought on great investors across all of these, but people still had some of those questions, but they looked at our forecasts and they thought, I get it, you've grown to a billion dollars of revenue so quickly, but there's no way you're going be able to Yeah, that's keep just not possible.

这有物理定律的限制。

There's laws of physics.

你们是企业级公司，这很好，但用户采纳速度肯定慢得多。

You're an enterprise, which is great, but adoption is going to take so much longer.

想想云计算花了多久才普及，现在还有多少人用本地部署呢。

I mean, at how long it took with cloud and how many people are still on prem.

业务持续验证了我们的观点：前沿智能的回报率确实非常高。

The business continued to prove out the thesis that the return to frontier intelligence is really high.

我们真正专注的是，模型驱动的增长如何通过产品、销售团队和渠道实现。

We are really focused on what's really happened is model led growth enabled by products and our go to market team and our distribution.

他们也看到了，我们主张：打造这种变革性技术固然重要，但必须以正确且负责任的方式去做。

What they also saw was that this thesis of, hey, it's really important to build this transformative technology, but to do it in the right way and do it responsibly.

他们还发现，我们的商业模式和一个大多数人不理解、也不相信的观点紧密相连：我们不仅投资模型研发，也大力投入AI安全研究。

That had this really interesting interlink with our business that most people didn't really understand or really believe, which was that we invest in research, not just in model development, but also in AI safety research.

我们率先开发了可解释性技术，简单说就是给模型做核磁共振，看清楚神经网络内部是怎么运作的。

We pioneered interpretability, which is think of it as like an MRI for the model to see inside the neural network, how it works.

我们还开创了对齐科学，确保模型能准确执行你的指令，以及它在多大程度上会偏离预期。

We pioneered alignment science, which is you want the model to do what you tell it to and how often does it do that and how often does it stray from that.

这些工作对我们的使命至关重要，所以我们才去做。

Those things are important for our mission and that's why we did them.

这些努力还带来了意想不到的好处——一旦你能看透模型的内部，就能更好地设计和优化它。

They had these downstream effects where it turns out if you can look inside the model, you're better at building them.

最后一点，既然我们面向企业客户，现在我们的客户已经包括了财富十强中的九家。

And then the last linkage, if you're selling to enterprises, we now sell to nine of the Fortune 10.

这些企业都把客户信息、核心数据交给我们，还让模型与员工甚至客户直接互动。

All of those enterprises are entrusting us with customer information, their data, they're interacting with their employees, sometimes even interacting with their customers as well.

这些可是最敏感、最重要的业务场景。

Those are the most sensitive workloads.

越来越多的企业正在把关键业务部署在Claude和我们的Claude平台上。

More and more of these businesses are running on Claude and our Claude platform.

我们对安全、可解释性和对齐所做的投入，还会持续下去，这些其实也直接让企业客户和所有用户受益，因为他们正需要这些。

When you have this investment that we made and will continue to make in safety, interpretability, alignment, that actually inures to the benefit of the enterprise customers as well and all of our customers because they want that.

如果他们要把如此多的访问权限和数据交给我们，还让我们深入他们最敏感的业务流程，那他们肯定希望合作的是一家值得信赖的公司。

If they're going to entrust us with all that access and all that data and the ability to work in the most sensitive workflows within their company, they want a company that they can trust.

我们当初投入这些，并不是为了这个结果，但事实证明，它确实带来了长远的好处——我们不仅站在技术前沿，更是一家注重安全、值得信赖的公司。

That's not why we invested in it, but it did have this downstream effect that we've really seen prove out again and again to be a company that is both at the frontier, but one that is investing in safety and that you can trust.

自从我加入公司以来，我们已经融资了七千五百亿美元。

We've raised $75,000,000,000 since I joined the company.

此外，上个月我们和亚马逊、谷歌达成的协议，未来还会带来另外五千亿美元的资金。

We have another $50,000,000,000 that'll come in into the future from the Amazon and Google deals that we linked last month.

这是一笔巨额资本，但人工智能业务本身资本密集，我们必须靠这些资金来支撑增长。

So that's a tremendous amount of capital, but it's a capital intensive business and we need this capital to support that growth.

这一切都说明，我们的业务运转得非常高效。

It all goes to the fact that the business is running very efficiently.

所以我们融资，主要不是为了填补当前的亏损，而是为了应对未来的不确定性。

And so the reason we raise this capital is more because of that cone of uncertainty than it is to fund losses in the business today.

你对这种业务实现十倍增长的看法是怎么样的？

How is your own perception of this perspective for a 10x growth of the business?

第一次看到这种情况时，你真觉得这可能实现吗？

The first time that happened, did you personally believe that it was possible?

那时候不觉得这太荒谬了吗？

Did that seem absurd?

现在它变得越来越稳定，也许对你来说也渐渐习以为常了。

Now that it's becoming consistent and maybe it's becoming more commonplace to you or something.

但当初面对这个不确定性时，你心里觉得连续多年实现十倍增长的概率有多大？

But what was your own view staring at this cone about the odds of hitting a 10x type of growth so many years in a row?

我加入公司的时候，年化收入大概是2.5亿美元。

When I joined the business, had about $250,000,000 of run rate revenue.

当时的目标是做到10亿美元。

And the plan was to get to a billion.

我就问，那计划哪一年实现？

And I said, great, in what year?

那是线性思维。

That was linear thinking.

达里奥在预测收入方面，一直比我准得多。

Consistently, Dario has been a much better predictor of the revenue than I have.

随着时间推移，我们对业务的理解和预测能力提升，差距应该会慢慢缩小。

I think we're going to close the gap over time as we get better at forecasting and understanding the business.

但第一次看到这种情况时，我脑子里全是各种质疑：物理定律、大数定律，这根本不可能。

But definitely the first time I saw it, you have all these arguments about the laws of physics and law of large numbers and this can't.

收入到底从哪儿来的？怎么可能这么快就涨上去？

Where is the revenue coming from and how can it be added this quickly?

客户怎么可能这么快就切换过来？

And how can customers move this quickly?

企业市场里真能这么干吗？

And is this even possible in enterprise?

但随着你慢慢看清业务内部是怎么运作的，看到用户采纳曲线和指数增长的真实表现，这些疑问就一个个被打破了。

All of those things start to get broken down over time as you see how the business works internally and you see how the adoption curves and the exponentials that are happening.

同样，收入的增长是呈指数级的，但支撑它的还有许多其他并行的指数级增长因素。

Again, we have the exponential that's happening on revenue, but that underlies it are these many other exponentials that support that.

你开始越来越能看见、也越来越相信这一点。

You start to see and believe in that more.

但这并不意味着我们在预测和评估各种可能情景时不够严谨或不够深思熟虑。

Now, that doesn't mean we're not disciplined and thoughtful about the forecast and how we think about the range of scenarios.

而是说，我的思维方式已经从线性、渐进的模式，大幅转向了拥抱这种指数级增长，真正相信它的潜力——这和以往任何企业的演进方式都完全不同。

But it does mean that my thinking has shifted a lot more from linear and incremental towards leaning into this exponential and really believing in its potential and how this is just different than how other businesses have evolved.

你在每个融资阶段跟投资人交流时，我猜每个阶段、每一轮融资，都会遇到某个特别常见、最难解释清楚，或者投资人最难理解的问题。

As you've talked to investors at every stage, I'm sure every stage, every round that you've raised, there's something that is like the most common or hardest thing to explain to investors or that they're struggling the most to understand and get their heads around.

现在这个问题是什么？

What is that today?

我觉得是关于计算资源使用方式的这种新范式——不要只把它看作某个时间段内的可变成本，而要把它当成一种高度灵活、可随时调配的资源。

I think it is this paradigm of how compute is used, thinking of it as not just something that is a variable cost over some time period, but really this resource that's so fungibly utilized.

我们早上用同一块芯片做推理任务，下午或晚上又用它来开发模型。

We run workloads on one day in the morning on a chip for inference, and in the afternoon or evening, we use it for model development.

这种模式在软件公司或者工厂这样的企业里是不存在的。

That paradigm does not exist in a company like a software company or a factory.

如果你有一群人专门做研发，那这部分就是你的研发开支，没法随意重新调配。

You can't repurpose if you have a bunch of people doing R and D and that's your R and D expense.

在大多数传统公司里，员工没法在研发人员和生产工人之间互相转换。

They can't go and become cogs and vice versa in most traditional companies.

但在这里，这种资源的灵活调配是真正能做到的。

Here, you really have that fungibility that's possible.

我觉得，正是这一点让计算资源的回报变得如此关键。

And I think that's where the return on compute is so important.

而且现在越来越多的人开始意识到这一点。

And I think people are beginning to understand that.

但大家还是容易把这两项成本硬性分开，其实它们是相互促进的。

But there's still a tendency towards treating it like, oh, I have to separate these two costs, when in actuality, they're very self reinforcing.

而这种灵活性，才是真正推动短期和长期收入增长的关键。

And that flexibility is actually what helps to drive revenue short term and long term.

如果我强行把你从现在的岗位上拉出来，扔进一家大型投资公司的投资岗位，让你去深入考察那些公司，挑出最值得投的，

If I was to force you out of your role and into an investor seat at a great big investing firm, and then I said, job is to go grill these companies and invest in the best ones.

那你会对那些在研发大模型的实验室或公司提出哪些问题，才能真正戳中那些不确定、怀疑的点，找出它们未必能成为顶尖企业的关键问题？

What questions would you be asking of the labs or companies that are building models to really get at the heart of the points of uncertainty, of skepticism, of things that might not make these the best businesses of all time.

我想知道，从这个角度出发，你会怎么去分析？

Curious from that angle, how you would approach it.

首先，算力的整体投资回报率是多少？

First, what is the ROI on compute all up?

你们是怎么使用算力的？

How are you utilizing it?

现在能看到什么样的回报？

What return are you seeing today?

这个回报趋势是怎样的？

And how is that coming over time?

这确实是投资者经常问我一个问题，但我认为它非常关键。

That's certainly a question I get asked a lot by investors, but I think it's a really important question.

我们这样的公司正在投入巨额且前所未有的资金。

These are the massive unprecedented investments that companies like us are making.

那你们从这些投入中获得了怎样的回报？

What's the return that you're generating on that?

回报什么时候才会到来？它的增长曲线又是怎样的？

And when does it come and what is the shape of it?

我觉得这是第一个关键问题。

So I think that's one.

第二个问题是，你们的客户如何看待投资回报？

The second one is how do your customers see ROI?

我也经常被问到这个问题。

I get asked this question a lot as well.

大家只是拿来做测试吗？

Like, are people just using this for testing?

还是已经真正大规模地投入使用了？

Are they actually deploying this at meaningful scale?

对我们来说，这种现象非常明显。

For our business, we're seeing that in spades.

我们的年度净美元留存率超过了500%。

Our net dollar retention rate is over 500% on an annualized basis.

财富榜前十的企业里，有九家都是我们的真实客户，他们在做重大的采购决策。

Nine out of the Fortune 10, these are real customers making significant buying decisions.

他们

They

不再只是试用阶段了。

aren't pilots anymore.

没错。

Exactly.

来这儿的路上，我打了个车，二十分钟内就签了两笔上千万美元的订单。

On the way here, was in an Uber and I signed two double digit million dollar commits when the car ride, which was like twenty minutes.

从这个角度看，我们现在正接受全球最顶尖企业、最精明的买家和初创公司的评判。

From that perspective, we're seeing it when we're now being judged by some of the biggest companies in the world, the most sophisticated buyers and startups.

他们在市场上有选择，但最终选了我们。

They have choice in the market and they're choosing us.

但我经常被问到一个问题，尤其是从投资者的角度、怀疑者的角度：你们的客户到底怎么从中获益？

But I think one question I get a lot is, or I would ask from the investor seat, the skeptical investor seat, how are your customers getting return from this?

还有一个问题是，你对未来算力怎么看？算力将来会从哪里来？

Maybe a third one is how do you think about compute in the future and where does it come from?

毕竟，我们买算力的那些供应商，他们既把算力卖给人家，自己也会用。

Because obviously some of the places that we buy compute from, they sell the compute to others, they may use the compute internally.

长远来看，这两者怎么平衡？

What is the balance of that over time?

所以你的思路是，跟顶尖玩家合作，保持灵活性。

So your philosophy there is be involved with great players and have flexibility.

没错。

That's right.

有个挺离谱的数据：在普通民众眼里，AI这个概念的受欢迎程度，居然比国会还低。

There's this crazy stat about AI, just the generic concept being less popular than Congress amongst the general populace.

刚开始听到的时候，确实挺滑稽的。

And it's kind of funny when you first hear it.

但仔细一想，就觉得这事儿挺严重的。

But when you really think about it, you're like, this is kind of fucked.

我们得解决这个问题。

We need to solve this problem.

那些不从事科技行业、不住在湾区或纽约的普通人，还没真正感受到AI对他们的好处，至少从他们的看法来看是这样。

It doesn't seem like the general world that isn't in technology, doesn't live in the Bay Area or New York, does not yet feel or understand why this is good for them, just as measured by their opinion of it.

你觉得我们整个行业该怎么做，才能改变这种状况？

What do you think we need to do as an industry about that problem?

如果我们看看正在发生的这场变革，历史上其实有过不少类似的转折点，比如工业革命、互联网、云计算等等。

If we think about the transformation that's happening, there's been other transformative ways before, all the way back to the industrial revolution, the Internet, cloud, etc.

AI不一样的是，这一切发生得太快了。

One of the things that's different about AI is it's all happening so quickly.

过去需要几年甚至几十年才能完成的进展，现在几个月就搞定了。

You can have years or decades of progress that are being compressed into months.

回到人类对指数增长和线性增长的认知差异，这确实让人很不适应。

Going back to humans thinking in terms of exponentials versus linear, that can be jarring.

我们普遍对这项技术的潜力持非常乐观的态度。

We are very optimistic generally about the potential for this technology.

作为整个行业，我们还能更好地表达清楚它的价值。

We as an industry can continue to do a better job of articulating.

达里奥写过一篇叫《机器的爱与优雅》的文章。

Dario wrote this essay, Machines of Love and Grace.

文章讲的是这项技术如何改变我们的生活，比如加速药物研发，不仅针对常见病，也包括罕见病，推动生物医学的快速突破。

It's all about the potential for this technology to transform the way that we live, whether that be in drug development and curing diseases that are more mainstream, but also rarer diseases, the ability to accelerate that biological progress.

第二点，医疗健康领域，如何改善医疗资源的分配，提升发展中国家和资源匮乏地区的整体生活水平？

Number two, in health care and how health care is delivered to raise our standard of living in the developing world and in places where resources are not as plentiful?

我们怎么确保经济收益不只集中在少数人手里，而是真正惠及全球各地的人？

How do we actually make sure that the economic gains are also accruing to not just a small number of people in the world, but across the world?

我觉得，这些正是人工智能所承载的希望与可能性。

I think that all of those things are part of the promise and the potential of AI.

我们其实可以更生动地描绘出这个前景。

We could probably do a better job of painting that picture.

而且我们希望随着时间推移，能展现出更多实实在在的成果。

And we want to show more tangible results for that over time.

我觉得这些正在逐步实现。

I think that is coming.

这也是我最看好其中一点。

And that's one of the things I'm most optimistic about.

不过另一方面，我们也得把潜在风险说清楚。

I think on the other side, though, we do want to articulate the risks.

我们不能一味地告诉所有人一切都会很完美。

I don't think we should just tell everyone everything's going to be great.

前进的路上难免会遇到波折。

There are likely to be bumps on the road.

所以我觉得，大家更倾向于接受真实又平衡的分析。

And so I think people generally gravitate towards more honest and balanced assessments.

如果有人只跟我讲好消息，一点坏消息都不提，我就会想：我真的能信任这种观点吗？

If I feel like somebody's just telling me all the good news and none of the bad news, then I'm like, Okay, do I really trust this perspective?

这时候就需要平衡了。

That's where there's a need for balance.

你看，当变革在短时间内被压缩时，就会出现这些情况。

Look, these are some of the things that happen when change is compressed over a short amount of time.

我们该怎么在商业和政府之间协作，真正找到一些解决方案？

How do we work across commercial and government to actually come up with some of the solution sets?

所以我觉得关键在于清晰地说明这些机会。

So I think it's about a clear articulation of the opportunities.

也要思考这些解决方案可能是什么样子。

It's about thinking about what those solutions may be.

这可不是哪家公司单独能想出来的。

And that's not any one company that can come up with it.

我们并没有一个能解决所有问题的万能蓝图。

We don't have this blueprint that's going to solve everything.

但至少我们得展开对话，聊聊这些风险和负面因素，以及我们能做些什么来应对。

But to at least have that dialogue about some of the risks and downsides and what we can do to address it.

而且我们在讨论的时候，应该对这两方面都保持透明。

And then I think it's being transparent about both of those things when we talk about it.

我认为从长远来看，机会带来的收益肯定会远远超过那些可能出现的风险和损失。

I do think that over the long term, the opportunity is going be significantly higher and greater than some of the risks and the downsides that will happen.

但这并不意味着整个过程会一路顺风顺水。

But that doesn't mean it's going to be perfectly smooth on the curve.

你们的财务团队并不是因为大错而亏钱。

Your finance team isn't losing money on big mistakes.

而是那些没人留意的上千个小决定，一点一点地漏掉了钱。

It's leaking through a thousand tiny decisions nobody's watching.

Ramp 会在支出发生前就设置好管控机制。

Ramp puts guardrails on spending before it happens.

实时限额、自动规则，彻底杜绝救火式管理。

Real time limits, automatic rules, zero firefighting.

去 ramp.com/invest 试试吧。

Try it at ramp.com/invest.

随着业务发展，Vanta 会同步扩展，自动完成合规流程，为你提供安全与风险的统一信息源。

As your business grows, Vanta scales with you, automating compliance and giving you a single source of truth for security and risk.

了解更多，请访问 vanta.com/invest。

Learn more at vanta.com/invest.

Ridgeline 正在重新定义资产管理技术，它不只是软件供应商，更是真正的合作伙伴。

Ridgeline is redefining asset management technology as a true partner, not just a software vendor.

他们已帮助多家公司实现五倍增长，推动更快发展、更智能的运营，以及更强的竞争力。

They've helped firms 5x in scale, enabling faster growth, smarter operations, and a competitive edge.

访问 ridgelineapps.com，看看他们能为你的公司带来什么改变。

Visit ridgelineapps.com to see what they can unlock for your firm.

每家投资公司都独一无二，通用型人工智能根本不懂你的工作流程。

Every investment firm is unique and generic AI doesn't understand your process.

Rogo 懂。

Rogo does.

这是一个专为华尔街打造的AI平台，直接对接你的数据，理解你的工作流程，并产出实际成果。

It's an AI platform built specifically for Wall Street, connected to your data, understanding your process, and producing real outputs.

了解更多，请访问 rogo.ai/invest。

Check them out at rogo.ai/invest.

从OpenAI到Cursor再到Perplexity，最顶尖的AI和软件公司都靠WorkOS在一夜之间做好企业级准备，不用等上几个月。

The best AI and software companies from OpenAI to Cursor to Perplexity use WorkOS to become enterprise ready overnight, not in months.

访问 workos.com，跳过那些枯燥的底层基建，专心做好你的产品。

Visit workos.com to skip the unglamorous infrastructure work and focus on your product.

Mythos发布的时候，真是个特别有意思的时刻。

The release of Mythos was such an interesting moment.

那是很多人——包括我那些一直密切关注这领域的朋友——第一次说：这个东西真有点让人害怕。

It was the first time many people, friends of mine that are careful watchers of this stuff, said something like, This one kind of makes me scared.

这其实又回到了安全的问题上。

So it relates back to the safety question.

这也是你第一次公开表态：我们希望确保它不会被用在坏的地方。

It's also the first example of you coming out and saying, We want to make sure this isn't used for bad.

这可能是我们第一次担心它会被用在坏事上。

And it's maybe the first one that we are worried could be used for bad.

我想知道，在外界知道之前，内部关于这个决定的讨论是怎样的，你们是怎么做决定的。

I'm curious what that discussion was like internally before the world heard about it, the decision making process around it.

我们就拿这个例子来说说，随着技术继续进步、规模效应持续发挥作用，还有哪些事情真的让我们感到担忧。

And just using that as an example to talk about the things that do scare you as we continue to advance and the scaling laws continue to hold.

关于Mythos，有些人可能会误以为它只是一个网络安全模型。

One of the things about Mythos is that people may be misconstrued as just a cyber model.

实际上，它在很多方面都表现出极强的能力。

It is an incredibly capable model across many, many different dimensions.

我们发现，尤其是在网络安全领域，它的表现特别突出。

What we found was that cyber in particular was a place where it spiked.

这是我们第一次决定以不同的方式发布一个模型。

This was the first model that we decided to release in a different way.

我们采取的这种方式，始终与我们的使命和原则保持一致。

The way in which we did that, again, is consistent with our mission, our principles.

我们就是想这么做的。

We wanted to do it in that way.

我们采取分阶段的方式，是因为这个模型能力太强了，虽然大家最关注的是网络安全方面，但其实它在其他领域也能带来积极的应用，比如帮助修复代码库。

We have this phased approach to it because we think that when a model is this capable and again, cyber is the thing that people focus on, but there are other things as well we think it can be used in a positive way to patch code bases.

你可能已经看到过这样的例子：之前的模型在开源代码库中发现了22个安全漏洞，而Mythos则发现了250个。

You've seen these examples where we had an open source code base that a prior model found 22 secondurity vulnerabilities in and mythos then found two fifty.

这确实让人有点担心。

That is kind of scary.

但正是这些发现，影响了我们发布的方式。

But that informed the way in which we released it.

我们并没有说永远不发布它。

So we didn't say we're never going to release it.

我们决定，还是采取分阶段的方式。

We said, let's do it in a phased way.

我们选择先向一个群体逐步开放，这个群体将随着时间扩大，重点观察这项网络安全能力如何能被用于防御性目的，而不是攻击性用途。

Let's do it to a group that will expand over time where we can focus on this one cyber capability and how it can actually be used positively in a defensive way as opposed to in an offensive way.

我们认为这可以作为一个未来的参考模式。

And we think that's a template that could be used for the future.

但正因为这一特定领域，我们在发布时特别留意了这一点。

But because of this one particular area, we wanted to be cognizant of that in how we released it.

你现在规模这么大，几乎处处都能碰到你。

You're so big now that you run into everything and everyone.

比如几天前，政府就提出，或许未来在模型面向公众发布前，必须先获得政府的批准。

And one example of this is the government just a couple of days ago said maybe there'd be this new system where you have have to sort of pre approve the release of a new model with the government before it was released to the public.

你之前和国防部打交道的经历真是让人好奇，那过程到底怎么样？

Obviously, you had the crazy experience with the Department of War, which I'm really curious what that was like as you went through it.

现在人人都在关注你们公司、这项技术，还有其他几家正在开发类似技术的公司。

Now everyone cares about this company and this technology and the couple other companies that are building it.

你们是怎么应对这些的？

How do you navigate that stuff?

有些事确实超出了你们的掌控，但我相信你们一直在尽力和各方沟通协调。

Some of it is just, I guess, beyond your control, but I'm sure you're trying to work with people as best you can.

也许可以聊聊现在政府作为合作伙伴、参与者和监管者这两个例子。

Maybe talk about those two examples of the government now is a very relevant partner, player, overseer, etcetera.

我们非常重视与政府建立良好关系，因为我们确实认为监管是有作用的。

We prioritize having a strong relationship on this because we do think that regulation has a role to play.

我们一直公开表达这个观点。

We've been pretty vocal about that.

我们的立场是美国优先。

We are very America first in our approach.

我们希望这项技术能支持美国以及全球的民主国家。

We want the technology to support The US as well as democratic countries around the world.

这也是我们与政府密切合作的原因之一。

And that's one of the reasons why we've been working closely with the administration.

我觉得这其中需要一个平衡，对吧？

I do think that there's a balance, right?

你希望创新能快速推进，而不是被拖慢。

You want to be able to have innovation happen really quickly and have that not be slowed down.

但你也需要一套责任框架，来规范这些技术的部署方式。

But you also want to have this responsibility framework for how these things are deployed.

我们一直认为，这项技术影响深远，应该坦诚地讨论这些问题，包括和政府一起讨论。

Because we've long said that this technology has implications and we should have an honest conversation about them, that includes with the government.

所以我觉得神话流程就是一个很好的例子。

And so I think the mythos process is a good example of that.

你能跟我们多讲讲，你会怎么跟父母解释他们的文化准则吗？

Can you teach us a bit more about how you would describe their cultural tenets to your parents or something like this?

你觉得是什么真正塑造了这种文化的核心？

What feels like it really drives most of the culture?

我特别想了解关于写作这方面的情况。

I'm especially curious about the writing.

你经常听说达里奥会时不时地对外发布长篇文章。

You hear often that Dario publishes these long essays every so often externally.

但我了解到，他实际上发布得频繁得多。

My understanding is he does that way more frequently.

公司内部的写作文化非常浓厚。

And there's a lot of writing culture internally.

我想了解一下，身在其中的文化是什么样的？它和你以前待过的公司，或者同样在做类似事情的其他公司相比，最独特的地方在哪里？

I'm trying to get a sense of what the culture is like to be in and what makes it the most distinctive from other companies maybe that you've worked at or from other companies that are trying to do the same thing?

你感觉呢

And what's your sense

这些差异和独特性体现在哪儿？

of the differences and the distinctiveness?

文化是Anthropic真正独特的地方，我们对外也会谈到，但只有真正身处其中，才能体会到它的不同。

The culture is a real unique aspect of Anthropic and it is something that we do talk about externally, but it's different when you're in there living it.

我可以跟你分享一些我的观察。

I can tell you a little bit about some of my observations.

首先，我们有七位联合创始人，按理说这么多人很难合作，但实际运作得非常好。

First of all, we have seven co founders that shouldn't work on paper, but it really does in practice.

他们为公司文化树立了很好的榜样，也明确了什么才是真正重要的事情。

And I think they've really set the example for the culture and the things that really matter to the company.

我们对文化非常重视。

We take culture extremely seriously.

我们会做文化面谈，这可不是走个过场、应付了事。

We do a culture interview and it's not some pro form a thing we do just to check a box.

这是招聘评估中真正重要的一环。

It is a real part of the evaluation process.

一个人可能在其他所有方面都表现得无可挑剔，甚至是你见过的最聪明的人。

So somebody could be flying colors on everything else and really, really the smartest person you've met in this role.

但如果他过不了文化这一关，我们也不会录用。

We won't hire them if they don't pass the cultural bar.

我觉得这个说法挺贴切的。

And the way I would describe it, I like that frame.

如果要跟父母解释的话，那就是极度注重协作。

How would you describe it to your parents is it's one incredibly collaborative.

这意味着我们不能容忍山头主义、耍心机，或者非要抢功劳。

And this means that we don't really tolerate fiefdoms or the sharp elbows or the I need to take credit for this.

大家特别谦逊。

It's incredibly humble.

我们公司内部会在员工的笔记本上贴一个贴纸。

We have this sticker on their laptops internally.

我们的竞争对手能力很强，成功远非板上钉钉。

Our competitors are incredibly capable and success is far from guaranteed.

我觉得这正是公司运作的方式。

I think that's part of how the company operates.

就算达成了里程碑，有了好结果，地上也不会撒满彩纸。

If we reach a milestone and something good happens, there's not confetti on the floor.

大家想的都是：接下来该做什么？

It's like, what's next?

我觉得这正是公司文化中根深蒂固的：对使命的专注和高度一致。

And I think it's just that focus on the mission and the alignment that is imbued throughout the culture of the company.

另外一点是，我们鼓励深入的辩论、思想开放和真诚交流——大家会提出疑问，表达观点，但之后会进行建设性的讨论，最后共同决定下一步怎么做。

The other thing I would say is there's rigorous debate and intellectual openness and intellectual honesty that happens where people question things, people will express a point of view, but then there's dialogue around it that's productive And then we'll decide on a path forward.