与 Exa 首席执行官 Will Bryk 探讨构建 AI 代理的搜索功能

Exa 正在深度聚焦于市场情报这一块。

That's something Exa's leaning very deeply into, like go to market intelligence.

因为我们非常重视这一点。

Because we we care a lot about it.

这真的让人很兴奋。

It's very exciting.

而且在公司内部用起来也非常实用。

It's also very useful to use internally.

我们既要寻找潜在客户，也要招聘人才。

We we have companies to sell to, and we have people to hire.

所以能和它交流真的特别好。

So it's been great to talk to it

关于如何更好地利用这一点，其实有一些想法。

around getting But, this with some

没错，这就是一个很好的例子。

yeah, that's an example.

我觉得你只是想要全面的，就是所有可能跟你有关联、又相关的人。

I think you just want comprehensive like, you you just want all all the all the people that that are that could be connected to you that are that are relevant.

顺便说一句，这其实特别美好。

And that by the way, it's a very beautiful thing.

没错。

Like Yeah.

你没问，但我认为搜索的一个美妙之处在于，世界上很多重要问题，本质上都是披着不同外衣的搜索问题。

You didn't ask this, but I think one beautiful thing about search is that a lot of important problems in the world are actually search problems dressed up in a different way.

再详细说说。

Say more.

举个例子？

What's the example?

对。

Right.

好吧。

Okay.

政治极化。

Political polarization.

我认为这本质上是个信息检索问题。

I would argue that's a search problem.

因为很多人都想理解这个世界，但接收到的信息却常常带有误导，甚至 outright 错误。

Because there are people out there who wanna everyone wants to understand the world, or most people wanna understand the world, but they're getting fed information that's just, like, you know, misleading in some way or or straight up wrong.

如果每个人都能接触到准确、可控且全面的信息，大多数理性的人自然会保持理性。

And if if everyone had, like, information that was accurate and, like, controllable and, like, comprehensive, I think most reasonable people would be reasonable.

我觉得确实如此。

And I think Yeah.

因为现在的信息环境太混乱、太分裂了，让本该理性的人变得不理性。

Because our information environment is so chaotic, it's so polarized, it's causing reasonable people to be unreasonable.

顺便说一句，我也在其中。

By way, I'm part of this.

我肯定在很多政治问题上持有错误的观点，因为我的信息并不完整。

Like, I'm sure I have incorrect beliefs on all sorts of political things because my information is not perfect.

这真是我想解决的问题。

It's something I really wanna solve.

孤独也是一种信息缺失的问题。

Loneliness is a search problem.

真奇怪。

Weird.

没人觉得孤独。

No one thinks it's lonely.

再说详细点。

Say more.

对。

Yeah.

在现代社会里，很多人其实都感到孤独。

A lot of people are feeling lonely in in modern society.

这是因为他们找不到可以一起相处、建立关系的人。

Well, it's because they're not finding why they're not finding people to hang out with or to be in relationships with.

对吧？

Right?

是的。

Yeah.

所以啊，孤独确实是个寻找问题，尤其是在城市生活里，很难找到志同道合、能建立联系的人。

And so, like, yeah, lonely is a search problem, and and it's in in especially in a city life, like, it's hard to find other people with similar interests or who might bond with.

如果有完美的搜索引擎，再加上人们愿意分享的信息，就能更容易找到这样的人。

And this is like, you know, with a perfect search engine, with whatever information people are willing to share, it would help to find other people.

比如说，我对飞行汽车有很多疯狂的想法。

So for example, I have a lot of crazy ideas about flying cars.

我真的很想要飞行汽车。

I really want flying cars.

我讨厌路上的汽车。

I hate cars on the road.

我真希望能找到一群讨论飞行汽车的人。

I would love to, like, go to a a group of people talking about flying cars.

我肯定能跟那些人成为好朋友。

I'm sure I would be great friends with those people.

我总不能直接搜一下，找找旧金山所有喜欢飞行汽车的人吧。

I can't just be like, find me all the flying car enthusiasts in in San Francisco.

我真希望能有办法做到这一点。

I would love to attempt to be able to do that.

明白了。

Okay.

然后这就回到你之前问的问题了，到底是什么让这种差异变得可能？

And then and then and this gets in your earlier part of your question, was like, what what allows for this differentiation?

这个问题的答案其实挺长的，关于‘不’这个部分。

So I it's a long answer to how No.

我太喜欢这个观点了。

I love it.

你开始能意识到，搜索其实是个更大的问题，但很多人还没意识到这一点。

You can start to see that search is like a bigger thing that people don't Yeah.

你想啊，搜索就像1998年那样，看到一个文本框，输入几个关键词，出来几条结果。

Think of search as like, you know, 1998, you see a text box, you type in a a few keywords, you get a few things.

那时候才叫搜索。

Like, that was search.

不是的。

No.

现在的搜索范围要广得多。

Like, search is way broader.

搜索其实是把整个人类协调起来，围绕我们想做的任何事情。

Search is coordinating the human species around anything we're trying to do.

那搜索怎么才能不再变得像日用品一样普通呢？

How does search become less and less commodity?

归根结底，还是得靠出色的检索能力和高质量的数据。

Well, it's it's always it's always about really good retrieval and really good data.

如果你在这两方面都做到极致——拥有全部数据，同时具备最完美的检索能力，那才是完美的搜索。

So if you if you do perfectly on both those things, you have all the data and you have all the the best possible retrieval, that that that is perfect search.

所以它既要收集整个网络的数据，也要收集那些不在网上的信息，再训练出超强的模型来搜索这些内容。

And so it is, like, both accumulating all the web's data, you know, accumulating data that's not on the web, and then training extremely powerful models to search over it.

从五年前开始，Exa在工程上就一直聚焦在这两件事上：索引和检索。

Those are those have always been the two like, index and retrieval have always been the two pushes at Exa on the engineering side since five years ago.

你能再多说说数据这块吗？

Can you say more about the data element?

当然，不用透露什么机密，但确实感觉网络的某些部分正变得越来越封闭。

And, you know, obviously, no need to share any secret sauce there, but it it does feel like the web is getting increasingly closed to some extent, parts of web.

对吧？

Right?

很多数据提供方都很担心，怕被海量请求压垮。

But there's a lot of fear out there from data providers on, oh, we don't wanna be stack overflowed.

对吧？

Right?

尤其是如果你的商业模式依赖的是用户访问你的网站时展示的广告曝光量。

Especially if your business model is around impressions you serve to humans visiting your site.

我觉得，这种担忧在这些商业模式中是最强烈的。

I think the fear is is the greatest among those business models.

你如何看待与数据提供方之间的这种互动？如何在确保实现完美搜索的同时，与那些可能越来越封闭的数据提供方合作？

How do you think about just sort of that interplay with data providers and making sure you can get to the path of per perfect search, but, you know, work with these data providers that are maybe becoming more closed?

是的。

Yeah.

我希望理想的情况是，搜索能做到极致，同时每个人都有动力去创造更出色的内容，甚至比以前更积极。

So I wanna get to the ideal world where you have perfect search and then every everyone's incentivized to create amazing content, even more so than before.

我觉得这里其实有个机会，我可以谈谈如何建立一个系统，让内容提供方的收入能增加。

I actually think there's an opportunity here, and I'll talk about how, to create a system where, like, content providers are making more revenue Mhmm.

因为他们参与了这个庞大的智能代理经济。

Because they are participating in this massive agent economy.

没错。

Yep.

对吧？

Right?

所以我喜欢从第一性原理出发来思考，这里的根本点在于，智能代理经济将会非常庞大。

So the way like, I I like to think from first principles, you know, the first principles idea here is that the agentic economy is gonna be huge.

说白了，我们每个人都会拥有代理，这些代理都会参与其中。

Like, basically, like, we're all gonna have agents, so those agents are gonna be participating.

你可以想象，这就像一个代理经济与赛博空间的结合，它们本质上在进行各种商业活动。

You can imagine it's like agentic economy and cyberspace, where basically they're doing, like, commerce.

它们会阅读信息。

They're, like, reading information.

我们平时在互联网上做的所有事，它们都会做，而且规模大上千倍。

They're it's like everything we do on the Internet, they're doing, but, like, a a thousand times bigger.

对吧？

Right?

所以这个代理经济里蕴含着巨大的价值，也就是钱。

So there's a massive amount of of value in this agent economy, meaning money.

既然有这么大的价值，与其让几百亿美元都流进一家公司，为什么不考虑每年让一家公司拿500亿美元，剩下的1500亿美元分给所有内容提供者呢？

And and so if there's so much value, like, instead of, like, you you know, hundreds of billions of dollars going to one company, what if, you know, $50,000,000,000 a year went to one company, and the other 150,000,000,000 went to all the content providers.

对吧？

Right?

这种新型智能代理经济其实有更多方式，能让内容提供者分到更多价值。

Like, there's ways to distribute the value in this new agent economy that are more favorable towards providers of content.

这正是我设想的路径——虽然不会是两千亿，但每年一万亿美金是完全有可能的，对吧？

And, like, this is kinda how I and by way, won't be 200, but it'll be a trillion dollars a year of Right.

没错，完全就是这样。

So there there's Totally.

如果我们能搞清楚这套机制，每个人都有机会大有作为。

If we could figure this out well, like, there's opportunity for everyone to to just, like, do amazingly.

你刚才提到的那些应用场景，我都特别喜欢。

I loved all the use cases you talked about.

这关系到市场推广的问题。

There's go to market.

对吧？

Right?

还有一点，我可以归到孤独感这一类，就是找到能真正连接的人。

There's also the you know, I'll put it in the loneliness bucket, but finding people you can connect with.

你没提到但我想稍微停一下的是编程。

One that you didn't talk about that I just wanna pause on really quickly is coding.

你能再多说说，为什么网页搜索能让编程类智能体在这个场景下更强大吗？为什么Exa特别适合编程这个用途？

Can you share more about why web search makes coding agents more powerful for this use case in particular and why Exa is such a perfect fit for the coding use case?

是的。

Yeah.

对。

Yeah.

当然。

For sure.

每个智能体在某些时候都像人一样。

So every agent at some point agents are like humans.

不管是谁写代码，总会需要去查资料。

Like, at any point when you're coding or when humans used to code, you would have to look up information.

对吧？

Right?

因为你需要最新的技术文档，或者去博客找点灵感。

Because you want, like, the most recent technical documentation, or you might look at a blog for inspiration.

所以智能代理也是一样的。

And so agents are very similar.

尤其是它们特别需要最新鲜的信息，确保写出来的每一行代码都不会有严重错误。

In particular, they really want the freshest information so that every every line of code they write does not have some critical error.

特别是对于编程代理来说，风险太高了，每提升一点质量都至关重要。

And, you know, especially with coding agents, like, the stakes are so high that, again, every every extra nine of quality matters.

现在的编程代理确实很聪明，但在信息检索质量上，它们还停留在上世纪初的水平，搜索能力太落后了。

So these coding agents are very intelligent right now, but in terms of retrieval quality, they've been in the dark ages, or the dark ages of, like, the early two thousands in terms of search quality.

但搜索技术在编程领域完全可以做得更好。

But, like, search can be way better over coding it.

我们说的都是技术文档、SDK这些，简直再合适不过了。

You know, we're we're talking technical documentation, we're talking SDKs, just like as like, perfect.

对代码库的检索要做到完美。

Perfect retrieval over the sinks.

这正是我们的目标。

That was our goal.

我们还没做到完美，但在检索各类代码相关资料方面，已经非常出色了。

And so we're not yet perfect, but we're extremely good at search over any sort of coding material.

所以，没错。

And so, yeah.

当很多编程代理尝试使用我们时，比如Cognition，我们之前提到过，现在他们已经在用Devin了。测试结果显示，我们的系统让Devin变得强大得多，准确率大幅提升，错误也少了很多——这太关键了，而且这种趋势还会持续下去。

So, you know, when, you know, a bunch of coding agents try us, like Cognition, for example, We've talked about tried us, like, we powered Devin now, and they've just found when they tested it that it just makes Devin way better, way more accurate, make way fewer mistakes, which really matters, and this will just continue.

顺便说一句，你也可以把编程代理看作是那种什么都想自己搞定的助手。

By the way, like, you can also think about coding agents as just like like, every agent is is gonna wanna do everything.

所以它要做的不只是检索代码。

And so it's not just searching over code.

它最终还得能获取全球最新信息，时刻掌握新闻动态，因为到时候，这个编程代理就会真正成为你的专属助手。

It's also at some point searching over the world's information, just being up to date with the news because, like, if you the coding agent will will become your agent.

我觉得编程助手和通用助手最终会融合在一起。

It's like, think coding agents and and just general agents are gonna, like, merge.

对。

Yep.

这是一个挺有意思的趋势。

And so that's an interesting trend.

不过呢。

But yeah.

没错。

Right.

不对。

No.

我的意思是，Codex 明显不是给开发者用的。

I mean, it's clearly Codex is not for for developers.

对吧？

Right?

这就像是把所有功能都整合到一个应用里的那种趋势。

It's sort of like the everything app to to that point.

我想提一下这个，因为它现在在推特上到处都是，我觉得跟Exa可能有某种关联。

I wanna bring in mostly because it's in the it's all over Twitter right now, and I think it might be an interesting tie to Exa.

这其实牵扯到一个话题，就是代币上限和代币经济。

And that's sort of this topic of token maxing, tokenomics.

你看，Uber都开始说他们花的钱太多了。

Just the fact that, you know, Uber is talking about how they're spending too much.

ServiceNow今年的预算早就用光了。

ServiceNow hit their budget for the year already.

我记得微软还提到要取消他们的四重代码授权。

You know, I think Microsoft talked about pulling their quad code licenses.

走向代币末日。

To tokenpocalypse.

代币末日。

Tokenpocalypse.

是的

Yeah.

好的

Okay.

就这样了

There we go.

没错

Yeah.

对，就是这样

There and yeah.

正是如此

Exactly.

换个更贴切的词

A better word for it.

不过我们以前聊过这个了

But we we've talked about this before.

但从你如何看待搜索如何让令牌使用更高效、更节省的角度来说，能再深入分享一下吗？

But in terms of how you think about search actually making token consumption more effective and efficient, can you share more perspective on that?

如果可以的话，你在这一块看到哪些实际成果了吗？

And to the extent you can, like, are results that you're seeing on that front?

是的。

Yeah.

对。

Yeah.

当然可以。

Sure.

检索其实能缓解令牌危机，因为我们没必要每次处理都动用超大模型。

So, like, retrieval can help solve the token apocalypse because, like, we should not be using gigantic models for every test.

现在越来越多的人也开始意识到这一点。

We should be using like, and people are starting to realize this.

应该用不同规模的模型组合，大的模型负责决策，然后把任务分配给小模型去执行。

Like, you should use a family of models of different sizes, the big model decides what to do, and it dishes out commands to the small models.

如果这些小模型能用上检索，它们的准确性和可靠性会高得多。

And those small models can be way more accurate and reliable if they're using retrieval.

检索能让小模型表现得像大模型一样。

So retrieval helps small models act like big models.

对。

Mhmm.

这相当于一种低成本的解决方案。

It's kinda like in a cheap way.

我们帮客户节省了大量令牌，因为他们可以用更小的模型配合检索功能。

And so we do save our customers a huge amount of tokens because they could use smaller models and use retrieval.

我们非常重视这一点，也投入了很多资源。

They could also we we have all we we care a lot about this.

我们投入了大量研究，专门优化如何从文档中精准提取最相关的信息，让模型不必消耗过多令牌——毕竟，任何多余的输入令牌都会显著增加成本。

So we have, like we've put a lot of research effort into how to, you know, extract only the most relevant information from documents so that these models can just, like, not have to consume too much tokens because, like, a lot of you know, any sort of input tokens can dramatically increase spend.

相比其他供应商，我们能帮客户节省大约二十倍的成本。

So we could, like, we could save, like, 20 x on cost for customers compared to other providers Mhmm.

通过高效地筛选，确定代理到底该从网上获取哪些信息。

By, like, being very efficient in, what information does the from the web does the agent actually see.

但总的来说，小模型加上检索机制效率高得多。

But, yeah, in general, smaller models using retrieval is much more efficient.

安德烈·卡弗里有条推文提到这个，我老在推特上说安德烈·卡弗里。

And, like, Andre Carverthee had a tweet about I keep mentioning Andre Carverthee on Twitter.

我不确定。

Don't know.

是的。

Yeah.

我知道。

I know.

我们都懂。

We all do.

对。

Yeah.

他很棒。

He's great.

几年前他发过一条推文，说趋势是朝向更小的、原始智能模块搭配工具的方向发展。

He he had a tweet about this a couple I think a couple years ago where it was like the the trend is towards, like, small smaller raw intelligence modules using tools.

这个趋势确实很重要。

And that trend will like, that's an important trend.

因为模型的成本主要取决于参数数量，这直接决定了推理开销。

Because, like, you have, you know, you have a limited like, the cost of the model is determined by the number of weights that that, you know, determines the cost of inference.

如果你把参数都用在存储无关信息上，比如法国的首都是哪里，或者你随便读过的一篇博客，那其实是在浪费参数。

And and if those weights are if you're wasting those weights on, like, all sorts of information about the world, like the capital of France, or, you know, this random blog that you read, like, you're just wasting tokens sorry, you're wasting weights.

这些参数应该专注在智能处理上。

Those weights should be focused only on, like, intelligent processing.

你完全有可能做出参数少于十亿、甚至只有十亿级别的模型，它们极其聪明，但对世界一无所知。

And you could you could probably get to models that are, like, 1,000,000,000, even less than a billion parameters that are extremely hyper intelligent and completely unknowledgeable.

对，没错。

Like, they Yeah.

这就像是爱因斯坦，根本没亲眼见过这个世界。

It's like Einstein, like, who never saw the world.

你可以这么理解。

That's kinda, like, the way to think about it.

然后它会使用一些成本很低、效率很高的工具。

And then it uses tools that are very cheap and and efficient.

这样会形成一个高效得多的模式，有助于缓解当前的算力短缺问题，对吧。

And that that that's a much more efficient world that will help solve this, like, this compute shortage Mhmm.

这个问题影响着每一个人。

That that is affecting everybody.

你觉得这是不是一种一时的激进观点？你认为这种现实会在2026年下半年开始出现吗？

Do you think this is more of like a hot take moment, but do you think that reality will start to take place second half of twenty twenty six?

因为现在我们正处于一个阶段，大家都在用最新、最大的模型，而正如你所说，可能是在过度投入。

Or because right now, we're in this phase where everyone's playing with the biggest best model that just arrived, and to your point, probably overspending.

那你觉得这种转变什么时候才会真正到来？

So when do you think this reality kinda sets in?

我的意思是，变化确实已经开始出现了。

I mean, reality is definitely starting.

我没法给出确切数据，但趋势已经无处不在了。

I don't have the exact it's like trends are everywhere.

是的。

Yeah.

对。

Yeah.

什么时候才会变得明显呢？

When does it become noticeable?

没错。

Yeah.

我觉得到2026年底，就会非常明显了。

I I would say by end of twenty twenty six, it's very noticeable.

哇。

Wow.

好的。

Okay.

对。

Yeah.

其实是个挺有争议的观点。

Of a hot take, actually.

你刚才提到你们做的研究工作。

So you talked about doing just the research that you're doing.

我觉得Exa本质上更像一个研究实验室，而不是别人常说的应用层公司或者基础设施公司，对吧？

And, you know, I think Exa, you sort of famously structured as more of a research lab, honestly, than, you know, what people are calling app application layer companies, infrastructure companies.

没错。

Right?

这本质上是一个专注于搜索的研究实验室。

You know, this is sort of a research lab focused on search.

从这个角度出发，我们最兴奋的一点，就是你们正在做的这些前沿工作。

And coming out of that, one of the things that we were most excited about, frankly, is just the exciting cutting edge work that you guys are doing.

其中一个方向是搜索在强化学习中的应用，这里面有不少效率上的提升空间等等。

One of those things was actually search as it pertains to RL, and there's a lot of efficiencies there, etcetera.

我想特别提一下，因为这个发现挺有意思的。

I wanted to just flag that because it was an interesting finding.

我觉得你们用的是 Tinker。

I think you use Tinker.

所以给 Thinking Machines 点个赞。

So shout out to Thinking Machines.

不过你能不能再多讲讲，你们在那边发现了什么，还有你们觉得哪些研究方向特别重要。

But say more about what you guys are finding there and and also just sort of what research directions you guys think about as being important.

对。

Yeah.

从宏观上看，训练大语言模型的很多核心思路，同样适用于训练搜索模型。

It's a at a high level, like, lot of the the big ideas in training LMs apply equally well to training search models.

比如，我们会对嵌入模型进行预训练。

So for example, like, we do pre training of embedding models.

我们还会对嵌入模型进行后训练。

We do post training of embedding models.

我们还会对搜索工具进行强化学习。

We do RL on on, like, search tools.

对吧？

Right?

很多在大语言模型上有效的方法，在检索系统里也同样适用，这点挺有意思的，但很少有人提到。

So, like, a lot of these things that are working in LLMs work in retrieval too, which is kind of interesting, and you don't hear a lot of people talking about.

所以，在那篇关于强化学习的博客里，我们其实尝试了让很多人用强化学习来优化搜索工具，但很少看到有人系统地比较不同的搜索工具。

So, yeah, in that RL blog post, we just we just basically try like, lot of people RL on a search tool, but we haven't seen as much many studies, like, testing different search tools that you

你用的是哪个。

are on.

没错。

Right.

所以我们直接拿SERP做强化学习，比如谷歌封装版和Exa，结果发现用Exa的效果好得多。

And so we simply RL ed on SERP, so, like, Google wrapping versus versus Exa, and found that, you know, RL ing on Exa does way better.

它不仅调用次数更少，效率更高，性能还更好。

Like, it both, like, uses fewer calls, so it's more efficient, and then it's, like, higher performance.

这其实很合理，因为Exa本来就是为智能代理设计的，能支持更复杂的查询——只要你真正捕捉到用户的真实需求，而不是像传统搜索引擎那样，非得把需求压缩成简短的关键词。

And this makes sense because, again, like, Exa was designed for agents to use, and so, like, it just it just allows agents to make more complex queries, like, if you really, like, capture what they actually want, as opposed to having, like, to compress what they want into, like, shorter phrases that are more for traditional search engines.

这篇博客挺有意思的，也确实很有启发性。

So that that was a cool blog post to explore, and think it was really helpful for that.

总的来说，我们的研究遵循的是那个‘苦涩的教训’——在很多方向上都靠规模效应取胜。

In general, our research is, like, the bitter lesson, so just, like, scaling laws in lots of different directions.

我之前提到的那些，比如微调、预训练、强化学习，我们其实一直在低调地做，很少被人注意到。

Some of the ones I mentioned, post training, pre training, RL, we've been pretty under the radar.

我觉得很多人根本不知道我们做了多少研究。

Like, I don't think people don't realize how much research we're doing.

当然，我们不是所有成果都会公开。

We don't publish all of it, obviously.

我们公开的确实不多，但在搜索领域，我们的工作量其实很大，只是大家都没意识到。

We don't publish much of it, but there is a lot to go in search, I and I don't think people realize that.

而且我觉得我们之所以明白，是因为我们一直对此着迷。

And I think we realize it because we've just been obsessed with it, one.

我们简直疯狂到极致，差距已经拉得太大了。

We're gonna crazy enough, like, that's just as far apart.

但更重要的是，我们之所以走到这一步，是因为我们的商业模式和服务对象。

But then also, I think the biggest thing is actually we're in because of our business model and who we serve Mhmm.

我们没有服务二十亿个差不多的消费者，也就是二十亿个普通人。

We've just been pushed in all these crazy directions because we're not serving 2,000,000,000 consumers who are kind of all the same, like 2,000,000,000 humans.

我们服务的是五千多家企业，它们正把我们推向各种意想不到的方向。

We're serving, you know, now, you know, over 5,000 businesses that are pushing us in all these crazy directions.

没错。

Right.

每天都在想，为什么这个不能为公司或用户做得更好？

Just like every day, it's like, why can't this be higher quality over companies or people?

为什么这个不能更快一点？

Like, why can't this be faster?

为什么信息提取的质量不能更高呢？

Like, why can't the information extraction be even higher quality?

所以我们被逼着往各种奇怪的方向探索，这也正是我们研究这么多方向的原因。

And so, we're being pushed in all these crazy ways, and that's why we're exploring all these research directions.

需求总是驱动着方向。

Like, always follows need.

是的。

Yeah.

对。

Right.

所以我们在Exa面临的需求多到离谱，必须不断改进，这就是我们做这么多疯狂研究的原因。

And so we have insane amounts of needs at Exa to do better, and so that's why we do all this crazy research.

对。

Yeah.

不对。

No.

这说得通。

That makes sense.

所以，跟这个相关的是，我想问问你对基准测试和渐进优化怎么看。

So I guess kind of tied to this is I wanted to ask you about how you think about benchmarks and hill climbing.

是的。

Yeah.

我这么说有点开玩笑的意思，但我们注意到，尤其是在你们这个领域的人，以及其他领域的人当中。

And I'll say this sort of tongue in cheek, but we've noticed that especially maybe among folks in your space, but also in other spaces.

对吧？

Right?

好像大家都拼命刷高基准分数，不管你怎么叫它。

There seems to be like benchmark maxing or whatever you wanna call it.

当然也不奇怪，每个人都在自己的基准测试里排第一。

And and, know, of course, no to no surprise, everyone is always at the top of their own benchmarks.

是的。

Yeah.

这说明肯定哪里不对劲。

That's how you know it's not some something's wrong.

这就是关键所在。

That's how you know.

对。

Yeah.

没错。

Exactly.

而且你显然一直在执着地追求真实情况。

So and and obviously, you have this relentless pursuit of ground truth.

没错。

Mhmm.

还有就是，你一直注重自我提升。

And and also, you you know, self improvement.

对吧？

Right?

我觉得，你本人也最先承认，嘿。

I think, like, you're the first to admit, hey.

这些是我们还能改进的地方。

Here are the areas we could be better on.

我们一直在持续不断地努力提升。

We're trying to improve on a, like, continuous basis.

但你内心是怎么看待哪些指标才是关键的呢？

But how do you internally think about what benchmarks matter?

对你们来说，什么是真正的基准？比如，哪些方面我们确实进步了，哪些方面反而退步了？

What is ground truth for you guys in terms of like, hey, we're actually better on this front, but worse on this front?

是的。

Yeah.

对。

Yeah.

不是的。

No.

评估指标在检索方面已经完全达到上限了。

A 100% the the the evals have been bench maxed in retrieval.

检索的评估项目其实没几个。

There aren't too many evals in retrievals.

这是个问题。

That's one problem.

确实没多少。

There's aren't that many Sure.

第三方的检索评估指标都已经被刷到满分了，但它们根本不能真实反映智能体搜索的需求。

Third party retrieval evals, and they've been, like, bench maxed, and they're and they're not really actually good representations of agentic search.

对。

Mhmm.

完全没错。

Totally.

智能体真正需要的是什么。

Like, what what agents actually need.

所以整个行业都存在一个问题，客户根本没法分辨什么是真的，这真让人遗憾。

And so it it is a problem in in the industry where, like, customers can't really know what is true, which is sad.

这也说明了我们需要一个真正出色的搜索引擎，来帮我们分辨真假信息。

It also demonstrates the need for a really good search engine to distinguish what is true.

对。

Yeah.

这不过是另一个例子罢了。

It's just another example.

但说实话，对客户来说，真正的标准其实是他们自己做的A/B测试。

But yeah, I mean, the ground truth for customers is their own AB test.

当他们直接拿我们和其他供应商，在自己的实际场景里做对比测试时，如果数据量足够，那才是最靠谱的。

And so when we do a like, right, like, literally they are testing us the versus other providers on on their use case, and if they have enough data to do AB tests, that's the best.

很多时候，他们会自己设计一套评估标准。

If they have their there's something often they make their own eval.

所以，成熟的客户会自己设计评估体系，特别专业的客户会做A/B测试，而那些只想找个简单方案的，可能就直接看网上公开的评估报告了。

So so sophisticated customers will make their own eval, super sophisticated customers will run AB tests, and then, like, customers who just want something might just, like, look at evals that are published online.

但对于那些成熟的客户来说，当他们测试我们时，谁真正领先就变得清晰多了。

But certainly for the sophisticated customers, when they test us, it becomes a lot clearer who who's who's, like, on on the top.

不过，我们确实希望改善这个生态体系。

But, yeah, I I I we wanna improve this ecosystem.

我们想成为那个推动整个生态进步的研究机构，定期发布成果。

We we wanna, like, we wanna be, like, the research lab that, like, helps improve the ecosystem and, like, publishes things.

即使我们不是第一名，我们也想把真实情况展示出来。

And even if we're not at the top, we wanna show.

所以你们接下来会看到更多这样的内容发布出来。

And so you'll you'll see more coming out there.

你曾预测，到2030年代，智能代理搜索的市场规模会超过谷歌搜索。

So you predicted that agentic search will be a bigger business than Google search by the twenty thirties.

能再详细说说吗？

Say more about that.

你观察到哪些趋势，让你相信这一点？

What are what trends are you seeing that lead you to believe this?

是的。

Yeah.

我的意思是，你其实只要估算一下搜索次数就能看出来。

I mean, just you could get this from basically, like, estimating the number of searches.

比如，LLM调用的总数，其中有多少比例需要搜索，每次搜索的成本是多少，然后把这些数据推演一下就行了。

So the number of LLM calls and the number percentage of those LLM calls that require search, then the cost of the search, and then you just, like, play it out.

这个趋势其实一直很明确，我们几年前的融资材料里就已经大致预测过这些走向。

And the trend has actually been pretty clear, and we've we've have, you know, pitch decks from, like, years ago where we kinda predict where things would go.

可能具体时间上会差个季度，但整体趋势是清晰的。

You know, maybe we're off by a quarter here or there, but it's, like, pretty, you know, it follows a trend.

所以，哪怕你保守地看这个趋势，也能看出智能代理搜索的市场规模会在20年代中期就变得非常庞大，更别说20年代末到30年代初了。

And so, yeah, if you if you follow that trend even conservatively, you get to a massive TAM for agentic search in the '20 I mean, even before, like, late twenty twenties and then early twenty thirties.

说真的，要准确描述代理会带来多少搜索量，其实很难量化。

Basically, like, the the number it's hard to express, like, how many searches will come from agents.

对吧？

Right?

对，就是这样。

Like Yeah.

普通人平均每天会进行几次搜索。

Humans on average make, you know, a couple searches a day.

但当每个人都有个人助手，而且你用的每一个软件工具都会通过检索来核对工作时，那情况就完全不同了。

But agents, when everyone has a personal assistant and every single software tool tool you use is gonna be checking its work with retrieval Totally.

我们说成几千次，是因为这个数字容易理解，大家也能接受，但其实最终会达到数百万次。

Like, the number of searches is gonna be, we say thousands because that that is, like, understandable and grok both people, but really, it's gonna be millions at some point.

搜索会像电力一样，遍布整个世界。

It's just gonna be like the world's gonna be filled with search in a way that the world is filled with electricity.

它就像一种基础基础设施，支撑着一切运作。

It's like like, it's a fundamental infrastructure that powers everything.

我觉得搜索的意义不在于提供完美的信息。

Like, I think less of search is, like, perfect information.

未来的世界会充满最高质量的信息。

So the world will be filled with, like, the highest quality information.

所以，当整个世界都充斥着某样东西时，通常意味着它的市场规模非常大。

And so, yeah, I mean, there is a lot of when the world is filled with something, it's usually a a large TAM.

如果你仔细算一算数字，我们觉得到2030年，这个市场会超过谷歌广告。

And and, yeah, if you just play out the numbers, we we think it will be bigger than Google Ads in 2030.

当然，广告本身也还是会占据很大一部分。

Not not that ads won't be a huge part of the world too.

这是显而易见的。

It's course.

广告对商业很重要，它也可能成为智能代理搜索经济中的一块重要组成部分。

Ads are important for commerce, and that might be a percentage of the agentic search economy too.

但这里的数字真的大得离谱。

But, yeah, it's just the numbers here are are insane.

归根结底，这取决于你是否相信大语言模型会渗透到每一个角落。

And it really comes down to a belief, like, do you think LLMs will eat the world, will eat all sulfur?

而我们一直坚信这一点。

And, like, we have always believed that.

没错。

Yep.

这看起来确实如此。

It seems like very true.

每个月都显得更真实了。

It seems even more true every month.

所以是的。

So yeah.

关于这一点，LLM领域和训练方面其实也有不少争论，你觉得呢？

What what do you think and there's some debate on this on the LLM side on the on the training side too.

但你觉得现在的瓶颈是什么？和三年后相比呢？

But what what do you think is the bottleneck today versus, let's say, three years from now?

我觉得五年后的事太难预测了。

I feel like five years is too far to predict.

是啊。

Is it yeah.

我记得你以前说过，现在阻碍搜索的已经不再是智能问题了。

I think you've said in the past it's no longer intelligence in terms of bottlenecking search.

对。

Mhmm.

是数据的问题吗？还是可获取的数据？

Like, is it data, accessible data?

你觉得这方面会怎么发展？

How do you think about how that evolves?

是的。

Yeah.

我的意思是，一开始的瓶颈其实是基础设施，这挺有意思的。

I mean, well, initially, the bottleneck is gonna be actually the infrastructure, which is kinda interesting.

没人意识到这一点。

No one realizes.

但如果你的搜索量是谷歌的十倍、百倍、甚至千倍，那支撑这种规模的基础设施简直大得惊人。

But, like, if you if you get, for example, 10 x, a 100 x, a thousand x more searches than Google, you you the infrastructure to handle that is insanely large.

但还没人把它建起来。

It just hasn't been built yet.

所以我们特别期待去探索各种新型的向量数据库，比如那种吞吐量特别高的。

So we're really excited to explore all sorts of cool new vector databases that have, like, super high throughput, for example.

对。

Mhmm.

类似这样的东西。

Things like that.

是的。

Yeah.

所以这其实是个挺有意思的瓶颈。

So that that's, like, an interesting bottleneck.

就像计算资源有瓶颈一样，基础设施也会遇到瓶颈。

Like, in the same way there's compute bottlenecks, there'll be, like, infrastructure bottlenecks.

我的意思是，迟早会解决的。

I mean, it'll it'll be solved at some point.

没错。

Yes.

还有其他的瓶颈，比如数据瓶颈。

And and then, like, other bottlenecks are, like, data bottlenecks.

智能体越来越需要询问关于世界的问题，但这些数据可能不在网上，甚至根本没被记录过。

Like, so agents are increasingly gonna wanna ask questions about the world, and that data might not be on the web, it might not even be recorded anywhere.

所以，未来会越来越倾向于如何收集全世界的数据，真正把那些被埋没的信息挖掘出来。

And so, like, you know, there will be this trend towards how can we accumulate all the world's data, like, literally unearth the world's data.

我对此特别兴奋，因为世界上充满了信息，但并不是所有信息都被记录下来了。

I'm I'm really excited about this because the world is filled with information, and it's not all recorded.

对。

Mhmm.

人类的历史，确实如此。

You know, the history of humanity Totally.

人类的历史就是世界不断被记录的过程——当初我们还是狩猎采集者的时候，就有大量信息，后来开始写下东西，世界上的信息量就飞速增长，但至今仍有大量信息未被记录。

The history of humanity is the world's been know, when we're we're we're hunters and gatherers, there's all sorts of information, and then they started writing things down, and, like, the amount of information in the world has been, like, skyrocketing, there's still so much information that's not recorded.

你知道吗？从最早的陶泥板，甚至更早的洞穴壁画，人类第一次把信息记录下来，之后是陶泥板、报纸，再到如今的数字时代，但你脑子里的信息——比如卫星图像——远不止这些，简直多得惊人。

You know, you go from all the way from a clay the first clay tablet or really the first, like, paintings on on on caves would be arguably the first time things were written down, and then, like, and then, you know, clay tablets, and, you know, now I've and newspapers, and then, obviously, now we have, like, the digital age, but there's so much information in your head, like, satellite images, like, they're not just, like Totally.

我们要吃掉整个世界的信息汤，把所有能搜索到的内容都整合起来。

Eating the world's soup of information that we could search over.

是的。

Yeah.

要真正理解这个世界，比如农作物产量如何影响一家公司，或者人们在想些什么，又或者我们该如何让世界更紧密地联系在一起，都需要更深入地理解世界。

And, like, to to fully understand the world, to fully understand how crops crop yields are gonna, you know, affect, you know, some company, you know, like, or, like, what are people thinking about the world, or how do we unite the world?

这要求我们对世界有更深层次的把握。

Like, that requires, like, understanding the world at a deeper level.

所以我认为，在很多方面，瓶颈还是数据。

So I think the bottleneck will be data in a lot of ways.

而一旦你拥有了这些数据，新的瓶颈就会出现——那就是信息的检索。

And then once you have the problem is once you have all this data, now the bottleneck will be retrieval.

没错。

Like, Right.

没错。

Right.

这想法简直疯狂。

Imagine, it's just a crazy idea.

想象一下，随着人类不断扩张，我们已经开始思考非常遥远的未来。

Like, imagine, you know, we're as we expand as a species, like, we're thinking, like, very far far in the future.

想想看，我们已经进入太阳系了，世界上发生的事情多得数不清。

Like, think about, like, we're in the solar system, like, there's so many things going on in the world.

每天都在积累海量的数据。

There's so much data being accumulated.

要从这些数据中检索信息，成本非常高。

The retrieval over that is very expensive.

所以，这里其实有一些非常根本、非常有趣的问题。

And so, they're actually really fundamental there are interesting fundamental questions here.

现在互联网上，大概有万亿个网页是真正重要的。

Like, right now the web is like, let's call it a trillion pages that matter.

如果网页规模扩大一千倍，变成一千万亿个页面，会发生什么？

What happens if the web were a thousand times bigger, like a quadrillion pages?

意思是，每个人都开始上传数据。

Meaning, like, everyone started uploading data.

那样的话，任何能在一万亿页面上运行的搜索算法，到了一千万亿页面可能就撑不住了，成本可能会高出一千倍，这根本不现实——毕竟我们希望搜索越来越便宜，而不是越来越贵。

Well, then, like, any search algorithm that works over a trillion pages no longer like, it might work over a quadrillion pages, but it might be a thousand times more expensive, and that's not practical because if anything, we want search to get cheaper, not more expensive.

那什么样的搜索算法才能应对一千万亿页面？这些问题太有意思了，我真不记得以前听过谁讨论过。

So what kind of search algorithms to work over a quadrillion These are fascinating questions that, like, I don't know if I've ever ever heard anyone else talk about.

其实，说到我们未来生活在火星之类的情况，倒是有位创始人确实深入思考过这个问题，就是埃隆。

Well well, actually, was gonna say the at least the thinking about when we're living in, you know, on Mars or whatnot, there's probably one other founder that has thought about that extensively, Elon.

我去年听了你上《潜空间》播客的那期，是我最喜欢的播客之一。

And, you know, I listened to your when you you went on the latent space podcast last year, one of my favorite podcasts.

你当时提到自己曾在SpaceX工作过。

And you talked about actually working at SpaceX.

那段经历怎么样？

How was that experience?

那埃隆的领导风格里，有没有哪些东西你后来也用在了自己公司的管理上？

And, like, have are there elements of Elon's leadership style that you've taken as, you know, CEO of your own company?

是的。

Yeah.

首先，那段实习经历简直太神奇了。

Well, first of the internship was magical.

比如说，我亲眼见证了第一次火箭在驳船上着陆。

So, like, for example, I saw the first landing on the barge.

太震撼了。

Wow.

对啊。

Yeah.

就在任务控制中心外面。

And like, like, just outside mission control.

现在回想起来，我都还起鸡皮疙瘩。

Like, I'm getting tingles just thinking about it.

但说实话，看到大家齐心协力做一件这么了不起的事，真的特别鼓舞我，对我影响很大。

But, yeah, like, like, just just seeing that, like, people coming together to do something magical for like, that was very inspiring and and made a big impact on me.

至于我的领导风格，我觉得自己确实吸收了一些我认为是马斯克最出色的地方。

And, yeah, I mean, in terms of my leadership style, yeah, I like to think that I've incorporated some of what I think are the best aspects of Elon.

比如说，他特别注重细节。

So and, you know, for example, like, he's very detail oriented.

他连最琐碎的事情都要过问。

Like, he gets into details of everything.

不管好坏吧，我在Exa也是这么做的。

And, like, for for better or worse, like, I do that too at Exa.

从算法，比如向量数据库的算法，到办公室的布置，我都希望每个角落都能让人感到振奋和激动。

Like, everything from like the algorithms, like the VectorDB algorithm to like the office space and making sure like every part of the office is like like really just like inspires and like excites.

是的。

Yeah.

这也体现了我们的热情。

And like like shows our passion.

这种态度贯穿了方方面面，从市场推广到工程开发，再到产品上市，真的很让人兴奋。

And that that goes across everything, our marketing to the engineering to to go to market, and and so that that that's exciting.

很明显，这种方法没法持续扩大规模，所以随着公司成长，我一直在学习：哪些细节值得我亲自去关注？

Obviously, doesn't scale, so one thing I've been learning as we scale the company is like, what details can I choose to go into?

我常把自己比作一个在公司上空飞行的自我，一旦看到某个细节觉得有必要修正，就立刻俯冲下去处理，处理完再飞回来。

And so it's I like the ego metaphor of like, you know, I'm an ego flying above the company, and then when I see some detail that I think it should be important to to fix, like, I go dive down and go into it and then come back up.

我觉得埃隆身上确实有这些特质。

So I think Elon has some of those properties.

显然，埃隆每天都是全身心投入，这一点毋庸置疑。

Obviously, Elon is, all in every day for Mhmm.

算起来，都几十年了吧？

What is it, like, decades?

是的。

Yeah.

我才做了五年，但我打算一直做下去，做上几十年。

I've I've only been doing it for five years, but I intend to do it for decades.

他另一项特别擅长的事，就是特别会用那些容易传播的名称，用富有感染力的方式激发大家。

One last thing that he does really well, which is, like, he's very good at memetic, like, names and, like, inspiring through, like, memetic things.

比如你想想，SpaceX的使命根本不是什么改进火箭、飞得更远之类的。

Like, you know, like, you realize that SpaceX's mission is not, like, improve rockets to get to or yeah.

也不是说要把火箭做得多好。

Like like, oh, make make rocket travel really fa like, good.

而是让人类成为多星球物种。

It's like it's like make humanity interplanetary.

这种说法才真正有传播力。

Like, that's a really good memetic thing.

所以我经常思考项目命名的事，每周一我都会在全公司面前开站会。

So I I think a lot about the names of, like, projects and, like, when I when I I, you know, I do a team stand up every Monday in front of the whole company.

现在我们有个双层办公区，大家都在顶层围绕着工作。

Now we have, like, this, like, double floor office where people now are surrounding on the Top Floor.

这感觉真的很棒。

It's really cool.

就像个体育场一样。

It's like a stadium.

太棒了。

Love that.

我会发表一段讲话，喜欢把我们做的事简化成最核心的部分。

And I give, a speech, and I and I like to, like, simplify what we're doing into like, what is the core?

什么是真正核心的传播点？

What is the memetic core?

然后想出一个酷炫的名字，这能激励我。

And come up with a cool name, and that inspires me.

顺便说一句，名字特别重要，因为公司发展到一定规模后，人们总在你不知道的场合里交流，而名字能真正锚定项目的使命。

By way, a name is really important because it's when you have a company of a certain size, people are constantly communicating in ways you're not part of those conversations, and the name really, like, grounds the the mission of that project.

是的。

Yeah.

当然。

Absolutely.

特别重要。

Really important.

我可能一整天都在琢磨一个名字。

Like, I I I could think of probably a day of just just about a name.

话题有点跳，但说到名字，我特别喜欢 GOAT。

So I just totally divergent, but on the topic of names, one, I love GOAT.

能聊聊这个名称背后的象征意义吗？

Talk a little bit about the symbolism of that.

除了‘史上最强’这个意思，还有 Exa 这个名字。

And then I mean, besides, you know, greatest of all time, obviously, but, and then also Exa.

你之前提过 Exa 的含义，再详细说说为什么给公司取名叫 Exa 吧。

You shared before what Exa means, but talk more about why you named the company Exa.

是的。

Yeah.

没错。

Yeah.

当然。

Sure.

好的。

Okay.

其实那个GOAT的事是这样的，隔壁开了一家咖啡店，好多人都在发帖，连旧金山市长都经常晒。

So the GOAT thing is basically, there was a coffee shop that opened up next door, and so many people so, like like, the mayor of SF kept posting about it.

很多人都说那家咖啡店特别棒，叫Hedge Coffee。

And so a lot of people go it's a really cool coffee shop, hedge coffee.

路过的人特别多。

And and so so many people were walking by.

我们就想，得想办法让这些人停下来看看Exa是做什么的。

We're like, we gotta, like, have them stop and and look at what Exa is.

我们在办公室入口贴了一些Exa的海报，但还是在想，怎么才能让他们真正停下来？

So we have some posters about Exa on the entrance to our office, but we were like, do we get them to stop?

于是我灵机一动，干脆贴个山羊吧。

And so I was like, just just put a goat.

我不清楚自己怎么想的，但就是决定放一只山羊，其实我心里还真的想过弄只真山羊。

Don't know why I thought this, but just put a goat And I actually some part of me wanted a real goat.

我也不知道该怎么养这只山羊。

I don't know how we'd maintain the goat.

我觉得就算这样，回报率也还是会很可观。

I think the ROI would still be valuable.

但最后我们买了一只做得挺精致的假山羊。

But then instead we bought, like, a a nice, like, fake goat.

哦，对啊。

Oh, yeah.

我特别喜欢那只山羊。

I love that goat.

我们还给它披上了公司周边的装备。

And we put, like, the swag on him.

没错。

Yeah.

所以它看起来真的很漂亮。

So it's like it's a beautiful yeah.

而且经常有人停下来观看。

And and people stop by.

真的有很多人会特意过来。

So like we're actually like tons of people stop by.

哇，真不错。

Like Wow.

太美了。

It's so beautiful.

就连周六周日，咖啡店最忙的时候，也有人专门过来，我会过去跟他们打招呼。

Even on Saturday and Sunday, which is like when the coffee shop is very popular, like people come by and I I go and I say hi to them.

我会跟他们聊几句。

I talk to them.

跟他们聊聊Exa的事。

I tell them about Exa.

其实特别漂亮。

It's actually very beautiful.

我们这儿其实是孩子们上学路上必经的一个点。

We're actually like a a important stop for for kids on their walk to school.

你知道吗，小时候我们都有那么几个固定打卡的地方，比如冰激凌店。

Like, you know, when you're a kid, you had, like, those stops, so you, like, move to the ice ice cream store.

GOAT 就是这样一个地方，孩子们总爱在这儿拍照。

Like, the GOAT is a stop, and they're always taking pictures.

非常

Very

可爱。

cute.

早早地就开始吸引潜在用户了。

Starting the recruiting funnel early.

这挺好。

That's good.

是的

Yeah.

但这些小细节居然这么重要，真有意思。

But it's interesting how much those things matter.

是的

Yeah.

没错。

Absolutely.

我们确实招过一些人，就是因为那个GOAT。

We've hired we've hired people because of the GOAT.

不管怎样，Exa 这个名字真不错。

Anyway, Exa, I think, is a great name.

我很喜欢 Exa。

I love Exa.

Exa 的一个优势是，它是个特别棒的前缀。

One value of Exa is, like, it's a it's a great prefix.

所以呢，Exa加任何东西，Exa数据、Exa这个、Exa那个。

So, like, Exa anything, Exa data, Exa this, Exa that.

但Exa其实代表的是10的18次方，跟Google的10的100次方正好相反。

But but Exa means tenth to the eighteenth, which is in contrast to Google, is 10 to the hundreds.

对。

Mhmm.

当时的想法是，我们其实已经被信息淹没了。

And the idea there was like, you know, we're kind of overwhelmed with information.

虽然你在谷歌上技术上能搜到海量内容，但那并不等于信息有条理，或者你能拿到最优质的知识。

Even though you could technically find lots of things on the web on Google, that doesn't mean it's like organized and you get the highest quality knowledge.

在这里，10的18次方的意思是，你提取的是最关键的信息。

Like, in here, it's like tenth to the, like, know, tenth to the eighteenth is like, you're you're extracting the most important information.

信息量当然还是很大，10的18次方，但不会让人觉得压得喘不过气。

It's still very large amount of information, tenth to eighteenth, but it's not, like, overwhelming.

对。

Mhmm.

所以，这正是其中一个想法。

So that was one of the ideas.

这名字真棒。

Great name.

我们也特别喜欢这个名字。

We love the name as well.

还有那些山羊。

And the goats.

我孩子特别喜欢那些山羊。

My kids love the goats.

这个话题其实已经被聊得很多了，但最近新闻里又频繁提到，所以我还是想提一下——你有没有看过威尔·梅尼塔斯最近写的一篇文章，叫《苦干糊糊》？

So I I wanted to this this topic feels a little bit overtalked about, but it's been in the news more recently, so I'm gonna bring it up, which is I don't know if you saw Will Menitas wrote an article recently called Grind Slop.

这文章其实带点反潮流的味道，我特意摘了一段。

And it's almost this backlash of and in fact, I there's a I pulled a quote from it.

他说，我们正见证一种看似自律的现象，实则可能是人类历史上最奢侈的经济资源浪费。

He said something about we're witnessing a phenomenon that masquerades as discipline, but represents perhaps the most extravagant squandering of economic surplus in our civilization's history.

这话也太重了吧。

Extreme words.

对吧？

Right?

他还提到好多文章都在吹捧这种拼命工作的方式，其实这些话题更多该讨论的是工作与生活的平衡之类的。

And he he talked about sort of all of these articles that are glorifying the grind over what they're doing, you know, all of these things that, you know, maybe are more relevant to work life balance, etcetera.

我先说一件让我印象特别深、特别兴奋的事，那次我去Exa，大概是晚上八点吧，办公室里人特别多。

So I I guess I'll start with saying one of the things that I was actually most impressed and excited by when I visited Exa at, god, I think it was, like, 8PM, was that there were a ton of people in the office.

不只是因为要装样子才待在那儿，毕竟你和Jeff根本不在乎这个。

And it wasn't just that they were there for FaceTime because obviously that's not what you and Jeff care about.

而是他们真的对自己手头的工作充满热情。

But they were excited about what they were working on.

更重要的是，他们和一起做事的人也特别投缘。

And I think really importantly, they were excited by who they were doing it with.

所以你能再多聊聊吗？你们是怎么看待这种全情投入的文化的？就是你们刚才说的那份使命。

And so just say more about, like, how you think about that culture of going all in and going after what you guys are, you know, sort of the the mission that you just talked about?

你觉得在文化中，哪些方面特别需要平衡？

And, like, what do you think the important things to balance are in culture?

是的。

Yeah.

对。

Yeah.

这对我来说至关重要，任何想做出出色成绩的公司都该重视这一点。

I mean, this is critical to me, and and it it should be critical to any company that wants to do great things.

员工得对自己做的事充满热情，还得玩得开心。

Like, you people need to be super excited about what they're doing and have a ton of fun.

对吧？

Right?

没错。

So, like Yeah.

你可能也听说了，在APM那边，大家都是笑着工作的。

You might have also heard at APM, like, are just laughing.