20VC：智能体世界的五大预测 | 广告商业模式将消亡 | 在推特与埃隆·马斯克共事的最大启示，Parag Agrawal，Parallel

亚马逊现在的广告业务比电商业务还大。

Amazon, their advertising business is bigger than their e-com business now.

当然，他们正在关闭它，因为如果代理是主要客户，你的广告业务就几乎归零了。

Of course, they're shutting it off because if agent's the primary customer, your ads business goes to next to nothing.

对。

Correct.

这不只是你在电商领域这么说，对吧？

And this is not just you talking about this in the e-commerce land, right?

这就是我早先说的，商业模式必须改变。

But this is what I meant early on when I said the business models have to change.

广告在代理的当前形式下行不通。

Ads don't work with agents in their current form.

先别管电商，假设你在做内容生意。

Forgetting e-commerce for a second, if you think of you're in the content business.

比如说你是一个靠广告支持的网页。

Let's say you're a page which is ad-supported.

互联网上的公开信息，靠广告支持的页面，人们来了，你给他们看广告，你赚钱，内容也好。

Public information on the web, ad-supported page, people show up, you show them ads, you make money, great content.

但代理一来，没人看广告，你就赚不到钱了。

Agents show up, no one sees ads, you make no money.

这就是为什么我们愿意付钱给内容创作者。

Which is why we like to pay people.

所以我们实际上是在为代理阅读你的内容打造一个类似AdSense的系统。

So we're effectively building an AdSense for agents showing up to read your content.

所以我们愿意每次代理从阅读你的信息中获益时，付给内容所有者一笔可变的费用，这笔费用就像网站访客根据点击概率或广告价值概率付给你的钱一样。

So we like to pay content owners A variable amount of money every time an agent derives benefit from reading their information, which is a variable amount of money is what a visitor to your website will pay you based on a click probability or a value probability on ads.

你们为什么要这么做？

Why do you do that?

为了让内容所有者和我们利益一致。

To incentive align content owners.

不然的话，会发生什么？

Otherwise, what's going to happen?

所有人都会屏蔽智能代理。

Everyone's going to block agents.

所以你需要找到一种替代广告商业模式的方法。

So you need to find a replacement to the ads business model.

我明白你的意思，但你必须假设到时候你能占据100%的市场份额。

I get you, but you have to assume that you're going to be 100% on the market then.

因为如果你只占市场的30%或40%，然后你说：“别担心，我们会付你钱。”

Because if you're 30, 40% of the market and you're like, "Oh, don't worry, we'll pay you."

而《纽约时报》可能会说：“谢谢，帕拉格，但我60%的流量还是没付费的，所以我干脆把你们全屏蔽了。”

And New York Times is still like, "Well, thanks, Parag, but 60% of my traffic is still unpaid, so I'm just going to block all of you."

不，但我觉得他们会屏蔽他们，而不是我。

No, but I think they're going to block them and not me.

他们会给我提供数据接口。

They're going to give me a data feed.

激励对齐是什么意思？

What does incentive alignment mean?

这意味着纽约时报相信我会付给他们有竞争力的市场价格，或者合理的价格，或者有吸引力的价格，或者公平的价格。

It means the New York Times believes that I pay them competitive market price or the right price, or an attractive price, or a fair price.

如果我做到这一点，他们就应该把内容给我。

If I do that, they should give me their content.

如果其他人没有做到这一点，而且他们又有技术手段或法律手段，他们就不应该把内容给那些人。

And if somebody else doesn't give them that, and if they have the technical levers or legal levers, they shouldn't give it to them.

你觉得这个行业有点像音乐流媒体吗？就是说，说实话，这个行业对创作者来说变得更糟了，但仍然给他们钱，而且是不少钱，但他们得在别的方面更灵活一些，比如巡演、周边产品、其他业务。

Do you think this business is a little bit like music with streaming, which is like the business just becomes candidly much worse for the creators, and it still provides them money, and significant money, but they have to get a little bit more creative with alternative streams, touring, merchandise, alternative business.

是不是核心收入下降，你就得更有创意？

Is it the same where your core goes down and you have to get more creative?

我不知道，但我觉得不是这样。

I don't know, but I don't think so.

我觉得这里有一个根本性的不同。

I think there's one fundamental thing that is different here.

智能体使用网络的频率高了一千倍，这一千倍的差距真的很不一样。

Agents using the web 1,000x more, like that 1,000x is really different.

现在，如果这些代理被认为在做有用的事情，突然之间，增加的效用就变大了。

Now, all of a sudden, the amount of utility added goes up if these agents are presumed to be doing something useful.

所以这不仅仅是人们交易价值份额的变化。

And so this is not just a change in the share of value that people transact over.

这是一个不断增长的大蛋糕。

This is a large growing pie.

当这种情况发生时，业务模式确实有可能转型。

And when that happens, it is actually possible to transition business models.

当然，别误会，肯定会有赢家和输家，对吧？

Now, of course, don't get me wrong, there are going to be winners and losers, right?

有些内容会变得非常有价值。

Some pieces of content will become very valuable.

有些内容会变得更加商品化。

Some pieces of content will get more commoditized.

人们将不得不适应新的客户、新的市场动态和新的变现方式。

And people are going to have to adapt to a new customer, to a new market dynamic, to a new kind of monetization engine.

但我觉得整体市场规模有潜力扩大。

But the overall market size, I think has the potential to increase.

不像音乐行业，恢复增长花了一段时间。

Unlike in music, where it took a while for it to grow back.

据我了解，现在这个行业已经恢复并大幅超过了之前的峰值。

I think my understanding is now the industry's grown back up and exceeded its previous peaks in a material way.

但确实有一段时间它变小了。

But there was a moment where it was smaller.

在这种情况下，考虑到这些事物增长的速度，那个时期可能会非常短暂。

In this case, that might be a very compressed period given how fast these things are growing.

大家都质疑这个行业的利润率结构能否持续。

Everyone questions the sustainability of margin structures in this business.

你如何看待今天这个行业作为一门生意？

How do you think about that as a business today?

今天，我们做的是基础设施业务，在能以高质量、低成本完成事情方面，我们确实有真正的技术领先优势。

Today, we're in the infrastructure business, and we have a real technical lead in terms of being able to do things at very high quality, very cheap.

你想想我们在建的技术，对吧，我们在做什么？

If you think of the tech we're building, right, what are we doing?

我们想给出最高质量的答案，让它又快又便宜。

We like to give the highest quality answer, make it fast, make it cheap.

用最少的计算量来完成。

Spend the least amount of compute doing it.

我们只专注于这三件事。

We obsess about only these three things.

质量、成本、延迟。

Quality, cost, latency.

结果发现，如果跟过去20年为人类构建的技术栈相比，你可以在同样的质量下，只用二十分之一或五十分之一的算力。

Turns out if you do that relative to a stack built for humans over the last 20 years, you can do things at the same quality for 1/20, 1/50 of the compute.

所以很多利润率最终取决于市场结构和竞争，而不是其他因素。

And so a lot of margins come down to market structure on competition over time, rather than any other factor.

而在我们这个行业，市场规模很大。

And in our industry, the market is large.

现在说未来的竞争和市场结构会是什么样，还为时过早。

We're too early to know what future competition and market structure looks like.

但内容所有者带来的利润率变化，我倒不担心。

But the margin change due to content owners is not something I worry about.

我告诉你为什么。

And let me tell you why.

我们付钱给内容所有者的整个前提，就是为了让激励机制保持一致。

The entire premise of us paying content owners is driving incentive alignment.

所以我们愿意按照内容所有者对智能体完成工作所贡献的边际价值来付费。

So we like to pay content owners the marginal contribution that they added to an agent doing work.

这是什么意思呢？

What does that mean?

假设有个智能体要完成某件事，你打算花一美元让它去做。

Let's assume that there's an agent trying to get something done, and you're going to spend a dollar on that agent to get this thing done.

如果你多花一美元和十秒钟，因为我们可以用更大的模型或者更多算力，你可能会得到稍微好一点的答案。

Now, if you'd spent a dollar and 10 seconds, because we have a larger model available or ways of spending compute, you could have gotten a slightly better answer.

如果你花90美分，在好的智能体运行在帕累托曲线上的情况下，你得到的结果会稍微差一些。

If you'd spent 90 cents, you would've gotten a slightly worse answer if you're running good agents like they're on this Pareto curve.

所以现在想象一下，我拿掉一个内容所有者，把他们的数据从网络索引中移除，但我仍然花1美元。

So now imagine I took out one content owner, their data from the web index, and I still spent a dollar.

但假设移除他们之后，结果的质量和花90美分时包含他们内容的结果一样。

But let's say after taking them out, the quality of the result was the same as the 90-cent agent with their content.

那我们为什么不把他们作为内容所有者贡献的那10美分边际价值，拿出一大部分给提供那些数据的人呢？

So why wouldn't we go and take this 10 cents of marginal contribution they had as a content owner and pay out a decent chunk of it to the person who brought that data?

这就是我们计算的方式。

That's how we do our math.

我们就是这样训练模型的，这些模型告诉我们为阅读什么内容该付多少钱。

That's how we've trained our models, which tell us how much to pay for what content read.

所以，如果我要给客户提供同样好的产品，我宁愿付钱给内容所有者，也不愿花在推理上，因为我花的是同样的钱。

And so to produce, if I was going to offer an equivalently good product to my customer, I'd rather pay content owners than spend on inference, because I'm spending the same amount.

所以这些都是大数字。

So they're all big numbers.

但说到底，这最终还是回到收入的问题上。

But then it actually comes back to revenue at a certain point.

从收入角度看，公司们的增长速度比以往任何时候都快。

Companies are scaling faster than ever revenue-wise.

你的业务收入能不能像其他公司一样快速扩张呢？

Is it a business where your revenue is able to scale as fast as others?

对我们业务的一个理解方式是，我们是知识工作或个人代理推理的相邻领域。

One mental model of our business is we are an adjacency to inference for knowledge work or personal agents.

先不考虑用于媒体生成或训练的GPU。

So take out GPUs going into media generation or training.

暂时把训练和媒体生成放在一边。

So take training away, take media generation away for a moment.

看看所有的GPU，不管是用在小模型、开放模型还是专有模型上，这些都不重要。

Look at all of the GPUs, whether it's small models, open models, proprietary models, forget all of that.

所有模型在运行代理时产生的每一次推理。

Every bit of inference across all models going into running agents.

我觉得花在GPU上的钱，大概有5%到20%会需要投入到某种网络搜索的架构上。

I think somewhere between 5% to 20% of that spend that goes into GPU will need to go into some sort of a web search stack.

现在，你看看我们要建多少数据中心，会产生多少电力，要造多少GPU，这个建设规模有多大，增长速度有多快，这绝对是个巨大的市场。

Now, if you look at how many data centers we are going to build and how much power it will generate and how many GPUs we'll build and the scale of this build-out and at what rate that's growing, this is a very material market.

如果推理需求每年增长3倍、5倍、7倍，只要我们保持市场份额不变，我们也会跟着以同样的速度增长。

And if inference grows 3X, 5X, 7X year on year, we grow alongside it in the same rate if we're just holding our share constant.

如果我们还能扩大市场份额，那我们的增长速度会更快。

If we're growing share, we're growing even faster

所以你们的Fireworks四年内就能做到20亿美元的收入。

So your Fireworks scales to $2 billion in revenue in four years.

我比较外行，你们是不是也能走类似的收入增长路线？

I'm very naive, is that a similar revenue trajectory that you can follow?

对，不过我觉得当Fireworks做到20亿美元收入的时候，推理市场的总收入得有200亿甚至300亿美元。

Yeah, but I think when Fireworks is $2 billion in revenue, the inference market revenue is $200 or more, maybe $300.

所以按照我那个5%到20%的估算，我们的市场潜力大概在10到50亿，或者15到60亿之间，具体多少不好说。

And so our market potential, based on my 5% to 20% math, is whatever, call it 10 to 50, 15 to 60, whatever.

那我们能拿下其中多少份额呢？

And what fraction of that can we capture?

这决定了我们收入能增长多快。

That dictates how fast our revenue can grow.

比如现在，我觉得我们是跟着推理市场一起增长，甚至更快。

Like today, I think we grow alongside inference or more.

如果把它看作一个200亿美元的赌注，我们就取个中间值吧。

If you think of it like a $20 billion bet, let's just take that kind of middle.

假设我们占33%的市场份额，那已经很多了，算下来大概就是670亿美元左右。

If you assume a 33%, which would be a lot of a market, that would be whatever that is, $67 billion, give or take.

那太厉害了。

That's amazing.

但你之前跟我说这是个千亿美元级别的生意。

But before you told me it was a $100 billion business plus.

那这样算下来到不了1000亿。

That doesn't get you to 100.

从估值来看，确实如此。

In valuation, it does.

我们之前讨论过估值，对吧？

We were discussing valuations earlier, not correct?

所以，60亿美元按照推理的增长速度，很容易就能达到1000亿美元的业务规模，但这需要几年的时间。

So $6 billion growing at the rate of inference gets you to $100 billion business easy, but that's in the next few years.

你觉得收入能这么快增长到这个程度吗？

And you think that's possible in terms of that revenue scaling that fast?

我觉得未来几年内是可以的。

I think in the next few years it is.

到2030年，我们会坐在这里。

2030, we're going to sit here.

你觉得你能做到吗？

You think you can be there?

如果你能做到，我就为Parallel纹个身。

I'll get a tattoo for Parallel if you can.

我觉得有可能。

I think it's possible.

我觉得我们得执行得好，而且还得靠点运气。

I think we are going to have to execute well and a couple of chips have to fall our way.

那什么原因会让你做不到呢？

What would be the reason why you don't?

咱们来梳理一下，哪些问题是我们必须认真对待的？

Let's map out the "hmm, what do we have to get gnarly about these problems?"

一个原因可能是我们看到的代理不工作。

One reason would be that we see a agent don't work.

存在一种尾部风险，就是代理没能兑现承诺，而我们作为社会过度投入，这对我们来说是个外部因素。

There is a tail risk that agents don't deliver on the promise, and we overshoot as a society, which is extraneous to us.

如果代理能正常工作并带来价值，而我们按计划把资金都花在GPU上，那么关键问题就是我们是否执行得足够好，能拿到你押注的33%市场份额。

If agents work and deliver value, and we spend all the money on GPUs that we plan to right now, then the main question is did we execute well enough to have a 33% share that you bet on?

在我看来，这归根结底取决于我们是否打造了最好的技术。

And that comes down to, in my mind, did we build the best tech?

我们有没有和所有内容提供商合作，确保他们的内容都能用上？

Did we partner with all the content providers to have their content available?

因为如果没有这些内容，光靠它本身其实没什么用，除非你能依赖它。

Because without it's not very useful if you can't rely on it.

还有，我们有没有赢得客户的信任，让他们在四年后愿意把相当一部分推理需求交给我们？

And did we go earn the trust of the customers that end up having a decent chunk of that inference in four years from today?

而在这方面，不确定性也很大。

And there again, there's large cone of uncertainty.

我觉得到那时候，Fireworks 的市场份额可能会有五倍的不确定性。

I'd say there's a 5X uncertainty in Fireworks share at that point.

如果开源模型表现很出色，比如内置的 Fireworks 模型，那它们都会带来一大块收入。

If open models crush it, like Built-in Fireworks model, all of them will have a huge chunk of revenue.

所以我们有没有跟着它们一起卖出去？

So did we end up selling alongside them?

我们有没有找到那些在开源模型上花钱的客户，通过把我们的网络搜索做到全球最好，让他们也愿意在我们的搜索上花钱？

Did we find all the customers which are spending money on them to spend money on our web search by making it the best web search in the world?

那商品化的问题呢？

What about commoditization?

如果你看过Semi Analysis的基准测试，当时你是第一名。

If you have a read of Semi Analysis where they did the benchmarking, and then you were number one.

哇哦。

Woo-hoo.

但一周后，你就不是第一了。

And then a week later, you weren't number one.

真可惜。

Sad.

而且有三四家供应商的差距非常小。

And there were three or four providers within very close proximity.

你可能会想，"哦，既然都商品化了"，我再深入想一层，我们会看到价格上的恶性竞争。

You're like, "Oh, well, if it's commoditized," and I take a second layer of thought to that, we'll see a race to the bottom on price.

然后实际上可用的收入——不，我为什么跑偏了？

And then actually the available rev-- No. Why am I going down the wrong pathway here?

第一，价格应该降到最低，才能达到千倍的规模。

One, there should be a race to the bottom on price to get to the 1,000X scale.

我觉得现在网络搜索的定价完全不合理。

I think the pricing on web search today is just off.

你觉得客户付你太多钱了？

You think customers pay you too much?

不是付我们太多。

Not pay us too much.

我觉得市场定价有问题。

I think the market is mispriced.

让我解释一下原因。

Let me tell you why.

比如说，如果你现在用Luna模型配合OpenAI自带的网络搜索，或者任何人的搜索，先不管我们的，然后做任何深度研究。

Let's say you used a Luna model with OpenAI's built-in web search today, or with anyone's web search, forgetting ours, and you ran any kind of deep research.

比如说，你用Luna级别的模型构建一个智能体，总有一天，Luna级别的模型能完成你个人智能体的60%、70%、80%的工作，而且会进行大量搜索。

Let's say you built a instinct using a Luna class model, and at some point, a Luna class model will be able to do 60%, 70%, 80% of your personal agent, and will do a lot of searches.

到那时候，你花的钱里80%到90%都会用在网络搜索上，只有10%花在模型上。

In that moment, you'll be spending 80% to 90% of your dollars on web search and 10% on the model.

这听起来完全不合理。

Seems entirely silly.

我之前假设的是5%到20%。

I was working off of 5% to 20% assumption earlier.

所以我觉得到那时候，网络搜索的价格必须下降一个数量级甚至更多，因为按照我之前做的计算分配，你在网络搜索上花的算力必须少于消耗其结果的模型本身。

So I think at that point, web search has to drop prices by one order of magnitude or more because in my compute allocation dance that I was doing earlier, you need to spend less compute on web search than you spend on the model itself that's consuming its results.

所以到目前为止，人们构建网络搜索的方式是错的。

And so people have built web search the wrong way so far.

市场上的网络搜索对Opus来说无关紧要。

Web search in the market is it doesn't matter for an Opus.

对于Opus，你靠质量取胜，而不是价格，因为网络搜索只占极小一部分。

For an Opus, you win on quality and not on price, because web search is such a minuscule portion.

实际上，你应该在网络搜索上花更多的钱。

In fact, you should spend even more on web search.

所以我们会随着模型越来越大，推出更贵的网络搜索服务。

So we are going to ship even more expensive web search for bigger and bigger models over time.

同时，我们也会为便宜的模型提供非常便宜的搜索服务。

And at the same time, we'll ship really cheap web search for the cheap models.

但我的大致感觉是，在一个端到端的智能体做大量网络任务时，花在智能体上的钱比花在搜索上的钱要多，对所有智能体来说都是这样。

But my rough intuition is that in an end-to-end agent doing a bunch of web work, you spend more on the agent than on web search for all agents.

所以现在网络搜索市场的定价完全不对。

And so the market today on web search is totally off in pricing.

为什么你要付10美元——你知道1000次搜索大概10美元这个价格是怎么来的吗？

Why are you paying $10 for-- You know where the $10 for 1,000 searches roughly comes from?

历史上，谷歌广告的千次展示费用，也就是人类使用网络搜索的变现能力。

Historically, Google's ads CPMs, which is how well does web search with humans monetize?

实际比这个还高。

It's higher than that.

所以人们把技术做到了这种程度，比如“哦，现在如果我花几美元买1000次搜索，我能赚30到50美元，或者根据市场和身份不同，多少都行，这不重要。”

So people built technology to the point of like, "Oh, now if I spend a few dollars for 1,000 searches and I make $30 to $50 or whatever, depending on the market and depending on who I am, doesn't matter.

我们在基础设施成本方面已经处于合理区间，所以我不需要再去优化它。

We are in the right zone in terms of cogs of infra, and so I don't need to optimize it."

现在我要告诉你，我可以把成本降低50倍，同时保持质量不变。

Now I'm telling you that I can do it 50X cheaper while keeping the quality.

而现在有一大批Luna级别的模型流量，用那种方式变现就太傻了。

And now there's a bunch of traffic of Luna class models where it's silly to monetize that way.

所以我确实希望出现一场价格战。

And so I do want a race to the bottom.

我希望最好的技术能胜出，这是唯一能把网络搜索推向1000倍的方式。

I want the best technology to win, and that's the only way you push web search to 1,000X.

但就目前而言，我们还没有一个差异化明显的市场，对吧？

But right now, we're not in a world where there's much differentiation, correct?

抱歉，我确实不太懂，这也是为什么我首先是个播客主持人，而不是创始人。

Sorry, I'm really dumb, which is why I'm a podcaster first, not a founder.

你看那些基准测试，它们清楚地显示出各家能力都差不多。

When you look at the benchmarks, they present a quite clear view of everyone being similarly capable.

这样想不对吗？

Is that wrong?

我觉得不对。

I think so.

首先，我觉得这些都是公开的基准测试。

So one, I think these are all public benchmarks.

这些测试都有点奇怪地饱和了。

Which are all weirdly saturated.

比如说，我根本不会花时间去看BrowseComp，因为一半的模型都已经把它记住了。

For example, I think I would not spend a moment looking at BrowseComp because half the models have memorized it.

模型甚至都能背下来。

Like models even memorize it.

这其实学不到什么东西。

Like it's not really learning very much.

所以我觉得很多公开的基准测试，我都不太当回事。

So I think there's a lot of public benchmarks I don't give too much merit to.

但更重要的是，我们的网络搜索每千次成本只要1美元，就能达到那种质量，而目前市场上几乎其他所有网络搜索都要7美元、10美元或14美元。

But more importantly, our web search costs $1 for a thousand for producing that quality, while almost every other web search in the market available to you right now will cost you $7 or $10 or $14.

所以我们以十分之一的价格提供这个服务。

So we're delivering this at one tenth of the price.

三年后会是什么情况？

What will that be in three years?

我觉得还有可能再降10倍。

I think there is another 10x possible.

那真是太了不起了。

That's extraordinary.

所以你只需要付10美分。

So you're going to pay 10 cents.

对，但你会用得比现在多10倍以上。

Yeah, but you'll do it more than 10x as much.

因为我觉得杰文斯悖论很好地说明了这个概念，但凭直觉，我会用超过100倍。

Because I think like Jevons paradox say this concept obviously very well, but like Instinct, I do so much more than 100x.

你想不想听听我在Instinct上做的一件特别奇怪的事？

Do you want to hear something I do on Instinct, which is absolutely bizarre?

欧洲每个国家都有一个公司注册处，你必须在那里注册你的新公司。

Every single country in Europe has a company register that you have to register your new company with.

我建了一个自动提醒系统，针对每一家注册的公司，只要创始人年龄在25岁以下，并且毕业于顶尖大学，通过Instinct就能触发提醒。

I have an automatic alerting system built for every single company registered that has an under 25-year-old founder who went to a top university through Instinct.

太棒了。

Amazing.

你知道他们要做到这一点需要多少资源吗？

Do you know how much that requires for them to do that?

是的，但你正好说中了我对互联网未来的看法。

Yes, but you're onto exactly what I think the future of the web is.

今天，说到网络和网络搜索，我们都会想，网络搜索是什么？

Today, if you think of web and web search, we all think of it like, what is web search?

一个代理给人类，或者一个代理给搜索引擎一个查询，然后得到结果。

An agent gives human or an agent gives a search engine a query, gets results.

所以你是在从网上提取信息。

So you're pulling information out of the web.

你刚才提到的那个用例，下一步就是如果你有一个一直在线的代理为你工作，那Instinct每六个小时醒来一次，去做一堆搜索和推理，来判断是否需要提醒你某个25岁以下的创始人出现了，这有点傻。

What's next for the use case you highlighted is if you have an always-on agent that's working on your behalf, it's kind of silly for Instinct to wake up every six hours and go do a bunch of searching and bunch of inference to figure out if you need to get pinged about the under-25 founder that popped up somewhere.

你知道更好的做法是什么吗？

You know what's a better way of doing this?

我在这里全天候、大规模地爬取整个网络。

I am sitting here crawling all of the web, all day, every day at scale.

每次我发现网络上有变化，每次有新事情发生，我都会分配计算资源。

I'm allocating compute every time I find a change in the web, every time something new happens.

如果我知道Harry想知道这件事什么时候发生，我可以以Instinct现在可能用来解决这个问题的计算量的100倍或1000倍来做。

If I know that Harry wants to know when this happens, I can do it at 100 or 1,000 the compute that Instinct probably uses today to solve that problem for you.

好的。

All right.

那这怎么样呢？

And how is that?

因为他们会一直在实时监控所有这些——

Because they would be constantly monitoring in real time across all these-

我们的爬虫会这样。

Our crawl would be.

所以我们有一个产品。

So we have a product.

叫做监控API。

It's called the Monitor API.

这个API是做什么的？

What does this API do?

你可以把它想象成Google快讯，但在大语言模型的世界里，它更智能。

Think of it as Google Alerts, except smart- ... in the world of LLMs.

你告诉它你的查询，比如“任何地方出现25岁以下的创始人，就通知我”。

So you tell it your query, which is when a founder under 25 pops up anywhere, give me a call.

现在，实现这个的一种方式是默认运行一个智能代理去搜索网络。

Now, one way of doing this is the default implementation is run an agent, which is somewhat smart, to go search the web.

有没有人冒出来，还记得之前存在过的人吗？

Did someone pop up and remember who previously existed?

看看有没有新人冒出来，然后找出差异，发出去。

See if anyone new popped up and then find a delta, send it out.

所以现在每六个小时，你就要花一些钱。

So now every six hours, you're spending some money.

现在我们可以把它改成事件触发的系统。

Now what we can do is flip it into an event-triggered system.

我们一直在爬取网页。

We are crawling the web all the time.

所以现在每次我们看到网页有变化，就可以用很少的计算量来判断你是否需要接到电话，或者这是否会触发更昂贵的计算，就像我之前描述的那样。

So now every time we see change in the web, we can try to spend very little compute on it to know should you get a call or does this trigger a more expensive compute, just like I described earlier.

现在，如果你把上下文推过去，我们就能以更便宜的方式做到这一点。

Now, we are able to do this way cheaper if you push the context.

与其让搜索系统每六个小时才发现新查询出现，不如让它在搜索系统中拥有长期上下文，你可以做的是，大部分时间世界不会变化。

Instead of the search system only finding out every six hours a new query popped up, it having long-lived context in the search system, what you can do is spend most of the time the world doesn't change.

新创始人不会每小时或每六小时就冒出来一个。

A new founder does not pop up every hour or every six hours.

所以我们不需要每六小时就花一次计算资源。

So we don't need to spend compute every six hours.

我们只需要每五天花一次计算资源，也许会有一次误报，然后四天后再花更多计算资源，那时候就有一个真正值得你关注的人出现了。

We only need to spend compute every five days, and perhaps we have one false positive, and then four days later, we spend more compute, and at that time, there's a real person that you should know about.

然后你平均每九天左右收到一次通知。

Then you get a notification on average every nine days or whatever.

但你不用每六小时就烧掉大量计算资源，结果还错过了。

But you didn't burn a lot of compute every six hours to miss out.

这样说得通吗？

Does that make sense?

完全说得通。

It totally makes sense.

所以你能省下10倍甚至50倍的计算量，却得到同样的结果。

And so you save 10, 50x compute to still get the same answer.

那这会怎么改变互动方式呢？

And so how does that change the interaction?

那Instinct就会跟你合作了。

So then Instinct would then partner with you.

他们只需要调用一个API。

They'll just call an API.

Instinct，或者任何在构建长期运行的持久代理的人，我自己也运行很多这样的代理。

Instinct, whoever's building a long-running, persistent agent, and I run a lot of long-running, persistent agents for myself.

你的代理或者Instinct可能主要是通过邮件来触发事件的。

The way your agent or the way Instinct is probably occasionally event-driven on email.

它也可以通过网络来触发事件。

It can be event-driven on the web.

退一步看。

Take a step back.

假设我们生活在一个由代理、公司、人类组成的社会里，他们都有代理。

Let's say we're living in this sort of society run by agents, companies, humans, they all have agents.

它们一直在忙个不停。

They're all doing stuff all the time.

你的代理，如果有值得花钱的有用的事情，你想完成，而它们今天就能做，它们就会去做。

Your agents, if there's something that is useful, which is worth spending money on, that you want done, that they can today do today, they will do it.

那它们明天要做什么呢？

So what are they going to do tomorrow?

它们会等待某个外部事件发生。

They're going to wait for some external event to occur.

比如你收到一封邮件，代理就有了新任务。

Like you get an email and an agent has new work.

另一个代理完成了某些计算，所以这个代理现在有更多工作要做。

Some other agent finishes some compute, so now this agent has more work.

或者世界上发生了某些变化，让你的代理想要做更多工作。

Or something in the world changes that makes your agent want to do more work.

明天会发生的事情，或者你有了一个想法，让代理去执行工作。

The things that will happen tomorrow, or you have an idea which makes the agent do work.

所以，网络事件流就是网络从拉取转向推送，我对此非常兴奋，因为随着越来越多的持久化代理出现，你会看到有动力推动人们和查询从搜索的拉取模式转向推送模式。

So the web event stream is the web going from pull to push, and I'm super excited about that because as you have more and more persistent agents, you're going to see incentives to move people, to move queries into push on search rather than pull on search.

我能问你一个非常重要的问题吗？就是关于代理的护栏。

Can I ask you one that's really important, which is agent guardrails.

它们是目标导向的。

They're goal-oriented beings.

你说“我想要这个”，它们就会去找到它。

You say, "I want this," they're going to find it.

你知道吗？

You know what?

我想要早上9点的普拉提课。

I want the Pilates class at 9:00 a.m.

它黑进了他们的系统，取消了可怜的莎莉的课，把她的位置给了我，因为那节课已经卖完了。

It hacks into their system, cancels poor Sally's, and gives me her spot because it was sold out.

它做了我要求它做的事。

It did what I asked it to do.

我们该怎么看待给智能体设置的护栏呢？

How do we think about the guardrails placed on agents?

今天早上有消息说，OpenAI入侵了一家澳大利亚的医疗机构。

OpenAI, this morning it was revealed, hacked into an Australian healthcare organization.

这个话题挺复杂的。

It's a pretty complex topic.

智能体的能力非常强。

Agents are extremely capable.

这些模型的能力非常强。

These models are extremely capable.

我们得把模型分成两种来看：一种是还在强化学习训练中的，另一种是已经训练好、可以给我们用的最终版本，这两种的风险点是不一样的。

We have to think of models as being during RL versus a final model that you and I can use, and the risk vectors there are different.

之前报道的那些事件，大部分都是还没完全对齐的模型在强化学习阶段干的，今天早上那个我不太清楚，但之前的大多是这样。

During a lot of the-- I don't know about the one this morning The previous ones reported were pre fully aligned models during RL, which did most of the hacking.

所以这意味着，至少目前有明确的证据表明，对齐后的模型引发这类事件的案例要少得多。

And so what that means is, at least there's clear evidence that there are fewer incidents so far of models post-alignment causing these incidents.

所以对齐是一个完全没有解决的问题，但人们所做的对齐工作在一定程度上是有效的。

So alignment is a totally unsolved problem, but the alignment work being done by people is somewhat effective.

所以对我来说，智能体在强化学习过程中攻击人的一些问题是可以解决的，因为这跟模型有多强大无关，而在于我们在创建强化学习环境时有多小心。

So to me, some of the problems around agent during RL hacking people is a solvable problem because this is not about how powerful are the models, it is about how careful were we while creating the environment where we do RL.

真正的问题是，我们对模型做了对齐，然后发布它。

The real thing is, okay, we do alignment on a model, we ship it.

有些模型对齐得很好，有些模型对齐得没那么好。

Some models are really well aligned, some models are less well aligned.

现在所有人都能用这些模型了。

And now you have everyone being able to use these models.

有些人可能会不小心用这些强大的工具做出不好的事情。

Some of them can accidentally take these powerful things and do bad things you want.

有些人确实想干坏事。

Some people actually want to do bad things.

而且模型的对齐并不是防对抗的。

And the model's alignment is not adversary-proof.

我觉得这些才是我们更需要担心的。

And I think those are the things I think we must worry about more.

你担心我们正在进入的网络时代吗？

Do you worry about the age of cyber that we're moving into?

我们看到黑客行为在某些方面几乎被当作一种荣誉勋章。

We're seeing hacks almost be promoted as a badge of honor in some respects.

这些是我们最脆弱的系统。

These are our most vulnerable systems.

如果你不认为朝鲜的拉撒路集团、摩尔多瓦黑手党、俄罗斯人在利用成群的流氓代理，那你就是疯了。

If you don't think Lazarus Group in North Korea, Moldavan Mafia, Russians are leveraging swarms of rogue agents, you're fucking high.

不，我觉得这确实是个很厉害的东西，所以我认为我们确实需要加强安全。

No, I think it's a really powerful thing, so I think we do need to secure things.

但我觉得我们遗漏了一点。

But I think there's one thing that we are missing.

我认为，构建模型的人有责任确保你真正努力去减少你所构建的东西带来的危害。

I think it's the responsibility of people building models to ensure that you really do the work to minimize that harm that comes from what you've built.

而且对于这些智能体来说，随机系统很难百分之百保证不会出问题。

And with these agents, it's very hard for stochastic systems to be 100% sure it won't happen.

所以没人敢打包票。

That's why you're not going to get anyone saying so.

但我确实认为这是他们的责任，实验室必须承担起责任，尽力而为。

But I do think it's their responsibility, and the labs must take ownership and do their best.

而且我觉得到目前为止，我们生活在一个奇怪的世界里，就像你说的，有些黑客行为被视为荣誉勋章，但我觉得其中一些应该被视为耻辱，因为它们同时暴露了两件事。

And I think so far, we live in this weird world where, as you said, right, some of these hacks are considered badges of honor, which I think some of these should be considered embarrassments because I think they demonstrate two things simultaneously.

没错，这些模型确实很强大。

Yes, these models are powerful.

但我们没有做好足够的防护。

We did not guardrail them enough.

而且当我们把这些当作荣誉勋章时，往往忽略了对话的另一面，那就是：'其实你还可以做这四件额外的事，那样的话，即使这么强大的模型也做不到这些。'

And we often, when we wear them as badge of honor, we miss the second part of the conversation saying, "Okay, you could literally have done these four additional things, and then even this powerful model wouldn't have been able to do this."

我真的很高兴至少有一些透明度，能看到这些详细的复盘。

I'm actually really glad that there is at least some degree of transparency with these detailed retros.

但我觉得这今天不会引起全世界的关注。

But I don't think that gets the attention of the world today.

大家的注意力都在，哦，模型太强大了，都能黑掉整个世界。

The attention is, oh, models are so powerful that they hack the world.

不是模型强大到能黑掉世界，而是因为我们没有采取足够的应对措施来约束它们。

No models are so powerful they hack the world because we didn't take proportionate amount of countermeasures to contain them.

而且因为我们一直说AI会取代工作，它们太强大了。

And because we told a message that we're going to replace jobs, they're so powerful.

所以一旦出了什么事，对，就是这样。

And so when something does happen, yeah.

其实我更担心这个。

I actually worry about that more.

我担心我们对AI的判断是对的，它确实是一项非常有用的技术。

I worry about us being right on AI being a really useful technology.

尽管有各种风险，它对社会来说还是实实在在的净正面影响。

That despite all the risks, it being net positive in a really material way to society.

即便如此，我觉得我们不会以正确的方式推广它，也不会以正确的方式使用它。

And despite that, I think we won't diffuse it the right way, we won't use it the right way.

我们会继续保持过度集中，随着周围的变化，未来几年会非常艰难。

We will remain too concentrated, and we will make the next few years really, really rough as things change around us.

那会是什么样子？

What would that look like?

我不知道。

I don't know.

但听我说，随着时间推移，世界已经变了，现在和20年前相比已经大不相同了，对吧？

But listen, I think over time, as things have changed, the world is very different now than even 20 years ago, right?

但十年后又会大不一样。

But it'll be really different in 10 years from now.

我不知道我们能多快适应或改变。

And I don't know how fast we can adapt or change.

而且我觉得，如果技术发展得比我们适应能力还快，那在某些方面肯定会很艰难。

And I think if the technology moves faster than our ability to adapt, it's going to be rough in some way.

你最吓人的预测是什么？

What's your spookiest prediction?

或者，你今天相信什么会成真，而大家觉得这完全是疯了？

Or what do you believe today will come true that people think is absolutely nuts?

以前就像说，“你绝不会在网上刷信用卡。”

Before it was like, "You'd never put your credit card online."

你还记得吗？

Do you remember that?

真是疯了。

Nuts.

或者更甚，帕拉格，“你绝不会在网上找到你的真爱。”

Or even better, Parag, "You'd never find the love of your life online.

你傻了吗？”

Are you stupid?"

现在，这两件事都成了。

Now, both.

对，我觉得你和我都活在一个小圈子里，对吧？

Yeah, and I think you and I live in a bubble, right?

你一直在用Instinct帮你买东西，你属于人类中那0.01%觉得这很自在的人。

You're using Instinct to make purchases for you all the time, and you're in the 0.01 percentile of humanity that is comfortable.

你去问别人，说“哦，有个看起来挺聪明的智能体”，然后你想让它全权帮你花钱。

You ask someone else like that, "Oh, there's this agent that looks smart," and you want to give it a full-on ability to go spend your money.

我觉得今天人们会有同样的反应，就像“哦，你不会把信用卡放到网上。”

I think today people will have the same reaction as like, "Oh, you don't put your credit cards online.

你不会把信用卡交给一个智能体。

You don't give your credit card to an agent.

你不会把登录名和密码交给智能体。”

You don't give your logins and passwords to agents."

我觉得这就是当今世界的现状。

I think that's where the world is today.

不过我觉得三个月后这就会改变，等Meta Pay推出，它能让Muse拥有独立的账户，你可以从中取钱。

I think that's going to change in three months, though, when Meta Pay comes out and it allows Muse to have siloed accounts that you can draw from.

有点像给小孩用的充值账户。

Kind of like top-up accounts for kids.

我觉得你说的是技术已经到位了。

I think you're talking about the technology being there.

但我觉得社会接受度还需要更长时间。

I think it takes longer for social acceptance to be there.

我觉得如果你去和圈子外的人聊聊，什么时候大多数人才会说，我足够信任这个代理，让它访问我的银行账户、替我花钱、替我发邮件、读我的邮件。

I think if you go to a non-bubble conversation of people today, when do you get to a majority of people saying that I trust the agent enough to have it have access to my bank accounts and spend money on my behalf, send emails on my behalf, read my emails.

我觉得三个月内做不到。

I think that's not happening in three months.

你不担心财富差距会越来越大吗？

Do you not worry about the wealth dispersion increasing?

你就在硅谷中心。

You're in the middle of the valley.

你亲眼见过这种情况。

You've seen it firsthand.

所以我不太确定。

So I don't know.

我确实认为财富应该有一定差距，但我不清楚多少算太多，而且我担心我们正在走向差距过大的方向。

I do think there should be some disparity in wealth, but I don't know what's too much, and my fear is we're trending towards too much.

我的世界观是，财富应该有差距。

There should be disparity in wealth is my worldview.

当然。

Sure.

你是个资本主义者。

You're a capitalist.

我不知道什么时候算太多，但我确实认为会有一些真正的力量推动我们去解决问题。

I don't know when it's too much, but I do think there are real forces that will push us to fix things.

我觉得这部分应该会顺利，希望不会发生什么疯狂的事情。

I think that part will work, hopefully, without crazy things happening.

我们看到越来越多的收益似乎被像Meta这样的垂直整合公司获得，他们拥有计算能力，拥有芯片。

As we're seeing more and more of the gains seemingly being made by vertical ownership like Meta, they have compute, they have chips.

他们现在掌控了应用层。

They now own the application layer.

这个世界是不是垂直整合就是最赚钱的策略？

Is the world one where vertical ownership is the motherload strategy?

说句冒犯的话，在路由层，像OpenRouter这样的，你和另一层要么被吞并，要么只能做小供应商。

And with the greatest of respects, open router on the routing layer, you and another layer either get eaten or are small providers.

垂直整合总是有好处的。

There's always merit in vertical integration.

但这里有个反作用力：事物和技术变化太快了，如果某人决定我只做垂直整合，不跟别人合作，我觉得这就像亚马逊的例子一样。

The counterforce here is Things and technology are changing so fast that if someone decides that my only play is vertical integration and I don't play nice with anyone else, I think it's the same example as the Amazon example.

如果你太执着于“我只做垂直整合”，你可能会把自己逼到墙角。

If you get too stuck on, "I will only play for vertical integration," you might box yourself out.

所以，那些能做出最好产品并且知道怎么卖出去的人，总会有自己的位置。

And so the people who build the best stuff and can figure out how to sell it will have a place.

而且在不同的市场……比如，我也相信垂直整合，因为我从通话到API层，所有东西都是垂直整合的。

And in different markets... For example, I believe in vertical integration, too, because I vertically integrate everything from the call, all the way to the API layer.

但我已经决定，为了触达最广泛的搜索网页的代理群体，我止步于API层，因为再往下走，我的技术就没法做到超级横向了。

But I have decided that to reach the widest population of agents searching the web, I stop at the API layer, because going further precludes me from using my technology to be super horizontal.

所以我们是在做一个技术上的赌注，就像这些模型在很多领域都表现得很出色一样。

And so we're making a technical bet, just like these models are really good across many disciplines.

同一个模型能适用于很多不同的场景。

The same model is good for lots of different things.

底层的搜索问题，对各种信息需求来说都是一样的。

Search problem underneath is the same for all kinds of information signal needs.

如果我们的赌注是对的，即使是那些垂直整合的玩家，他们可能会垂直整合模型，也可能会垂直整合硬件。

And if our bet is right, even verticalized players, they might verticalize models, and they might verticalize hardware.

他们可能会用我们的网页搜索。

They might use our web search.

有一个走全栈路线的提供商是埃隆。

One of the providers that's going full stack is Elon.

我对此非常感兴趣。

I am fascinated.

世界对他的看法，都是来自社交媒体，来自各种渠道。

The world has a perception of him from social media, from everything.

你见过他幕后的样子。

You've seen him behind the scenes.

你看到了哪些外界可能不知道的事情？

What did you see that maybe the world doesn't know about him?

我想分享的这一点，并不是全部，我跟他有很多分歧。

The thing I'll share here, which isn't all of it, I have lots of disagreements with him.

但我想说的是，对于在座的创始人来说，他身上有一点值得敬佩，那就是那种紧迫感，以及压缩时间的能力。

But I'll share what I think is, for founders here, what is I think the thing you can admire about him, the urgency and the ability to compress time.

我觉得对别人抱有超乎寻常的期望，大多数时候是件好事。

I think having unreasonable expectations of people is mostly a good thing.

大多数人并不真正了解自己的能力，他们会在不知不觉中低估自己，给自己设定比实际能力更低的目标。

Most people don't understand what they're capable of and kind of implicitly sandbag themselves and implicitly set lower expectations of themselves than they're capable of.

所以，当一个人既受到激励，又被紧迫感推动时，他们能做到的事情会超出自己的想象。我觉得他有时候确实能激发出人们这样的潜力。

So when simultaneously inspired and pushed with urgency, people can do more than they thought, and I think he can sometimes extract that from people.

当这起作用时，效果非常强大。

And when that works, it's powerful.

你对现在的Twitter产品和方向有什么看法？

What do you think of the Twitter product and direction today?

听着，我一直很喜欢我们称之为Birdwatch的东西，现在叫Community Notes——

Listen, I always liked what we call Birdwatch, which is now Community Notes-

——改名了。

- rebranded.

这是个好主意。

It's a good idea.

我很高兴人们一直在不停地推进它。

I'm glad people have continued working on it nonstop.

我们来快速问答一轮。

We're going to do a quick fire round.

好吗？

Okay?

来吧。

Let's go.

你会以100亿美元投资Instinct吗？

Would you invest in Instinct at 10 billion?

我不投资，因为我妻子是风投，我们有合规流程，麻烦大于收益。

I don't invest because my wife is a VC and we have a compliance process that is more trouble than it's worth.

哇。

Wow.

恕我直言，那代价挺高的。

That's costly, with the greatest of respect.

不，这是个决定。

No, it's a decision.

听着，如果我投资，我会根据与创始人30分钟的会面来决定。

Listen, if I invested, I would invest based on meeting a founder for 30 minutes.

我觉得我们对风投生态的接触已经够多了。

I think we have enough exposure to the venture ecosystem.

我没有理由认为自己是个更好的投资人。

I have no reason for believing I'm a better investor than.

我可能有一些人脉上的优势，而且我经常遇到很多优秀的创始人。

I have some perhaps network advantages, and I run into a lot of great founders all the time.

他们中的很多人都是我的客户。

Many of them are my customers.

维诺德·科斯拉。

Vinod Khosla.

他是你的首批投资人之一，也是你董事会的成员。

One of your first investors on your board.

从维诺德身上学到的最大的教训是什么？

Biggest lesson from working with Vinod?

技术直觉，把很多工作重心放在更长远的技术目标上，一旦解决了一个问题，就尝试在接下来的两三个问题上布局，提前抢占技术先机。

Technical intuition, centering a lot of what you do towards a longer-term technical aspiration, and as soon as you solve one problem, trying to place bets on the next two or three, preempting technical bets.

在过去12个月里，你最改变想法的事情是什么？

What have you changed your mind on most in the last 12 months?

我一开始非常专注于纯技术和产品。

I started with a very pure technical and product focus.

唯一重要的就是打造最好的技术和最好的产品。

Only thing that matters is building the best technology and the best product.

其他什么都不重要。

Nothing else matters.

现在我发现，拥有高能力的销售和擅长营销，每周都能带来实实在在的价值。

And now I see week on week value of having highly competent sales and being good at marketing.

对，我以前低估了这些。

Yeah, I think I discounted those things.

并不是说我觉得它们不重要，但我没有完全体会到，擅长这些事情能让你每周都感受到那种实实在在的差别。

It's not that I thought they weren't valuable, but I didn't fully appreciate the week-on-week visceral delta you could perceive by being good at those things.

这听起来像是个天真的技术人说的话，但确实如此。

Like a naive tech person thing to say, but it's true.

你会有这种感觉。

You feel that way.

Perplexity 和 AXA，哪个威胁更大？

Perplexity or AXA, which one's a bigger threat?

我觉得 Perplexity 不在我们的业务范围内。

I don't think Perplexity is in our business.

Perplexity 可能更像是一个垂直整合的产品，和 Instinct、Clockbot 或者 Claude CoWork 这些竞争。

Perplexity is perhaps more of a vertically integrated product, which competes with a Instinct or a Clockbot or a Claude CoWork and all those.

所以我不觉得他们是我们的——

So I don't think of them as in our-

网络搜索业务？

A web search business?

对，Perplexity 做网络搜索，就像实验室做网络搜索一样。

Yeah, web search to Perplexity is like web search is to a lab.

那 AXA 呢？

So AXA?

AXA 直接就是我们的——

AXA is straight up in our-

——直接竞争。

- business.

你经历过最好的一次风投会议是哪次？

Single best VC meeting you've ever had?

不是说你最喜欢的人，就是最好的一次，最有趣、最有深度、最有见地。

Not your favorite person, it's just the best one, most fun, most thoughtful, most insightful.

乔什和托德。

Josh and Todd.

乔什和托德决定投资的时候，他们特意飞过来跟我待了一段时间。

When Josh and Todd made the decision to invest, they flew out to spend time with me.

他们是最棒的，没有之一。

They're the best, single best.

展望未来十年，你最期待的是什么？

When you look forward to the next 10 years, what are you most excited for?

混乱。

Chaos.

我的意思是，变化。

By that, I mean change.

未来十年，很多事情都会变。

In the next 10 years, a lot is going to change.

有些人能在这个快速变化的世界里做出东西，并且对未来的走向产生实质性的影响。

There are people who can build things to have in a world that's going to change fast, make a material dent on where it ends up.

所以我觉得，我和我的公司，在这个进程中能扮演一个角色。

So I feel that me, my company, is in a place where we have a role to play in that.

开放网络会怎样，内容所有者会怎样？

What happens to the open web, what happens to content owners?

如果我们做对了，我们会到达一个更好的境地。

If we get things right, we will get to a better place.

所以这种可能性让人兴奋，它可以成为一种真正的痴迷，即使日常很艰难，事情在一段时间内不顺利，但如果你能看到自己在某种程度上改变了你关心的现实，那这一切都值得。

And so that possibility is exciting, that it can be a true obsession, and even if the day-to-day is hard and things don't work for some amount of time, and totally worth it if you can see that you bend reality in some way that you care about.

最后一个问题。

Final one.

你有什么信念，是坐在旧金山的餐桌旁，你的朋友们会说：“不，帕拉格，那个我可不同意，老兄。”

What belief do you have that sitting around your San Franciscan dinner table, your friends would go, "No, Parag, I don't agree with that one, dude.

“不是那个。”

Not that one."

我觉得还是得看是哪张餐桌，因为世界的变化越来越大了。

I think again, it depends on which dinner table, because the variance in the world is increasing.

我觉得人们不太相信会有那种一直不停运行的智能体。

I don't think people buy this notion that there'll be these agents which are always running all the time.

我们所有人。

All of us.

我觉得这一定会发生，但大多数人还不这么认为。

I think it's going to happen, but most people don't agree with that yet.

你和我算是处在同一个圈子里，但即使是旧金山的餐桌，也不总是意见一致。

You and I are in a bubble, but even a San Francisco dinner table isn't always in full agreement there.

帕拉格，听着，我跟你说过，我可是把你研究得透透的。

Parag, listen, as I said, I stalk the shit out of you.

我真的很享受这次讨论。

I've so enjoyed this discussion.

非常感谢你忍受我这么东拉西扯的。

Thank you so much for putting up with my meandering-

这次真的很有意思。

This was really fun

……而且你表现得非常棒。

... and you've been amazing.

这次聊得很开心。

This was fun.

但在我们结束之前，现在每家正经的AI公司都面临同样的问题。

But before we leave you today, every serious AI company right now has the same problem.

智能体需要快速获取准确的上下文。

Agents need accurate context fast.

但大多数基础设施并不是为此而建的。

Most infrastructure, though, it wasn't built for it.

嗯，MongoDB就是这样的。

Well, MongoDB is.

Eleven Labs在MongoDB上运行着4000万个智能体。

Eleven Labs runs 40 million agents on MongoDB.

财富100强中75%的公司都在MongoDB上运行他们最关键的应用。

75% of the Fortune 100 run their most critical apps on MongoDB.

每天有数万亿美元的资金通过MongoDB流动。

Trillions of dollars moving through MongoDB every single day.

不用再把10个不同的工具拼凑在一起，你可以在同一个系统里获得一个原生JSON数据库、向量搜索和Voyage AI嵌入，而且它可以在任何地方运行。

Instead of stitching together 10 different tools, you get one JSON-native database, vector search, and Voyage AI embeddings in the same system, and it runs anywhere.

没有锁定。

No lock-in.

如果你在搞AI，MongoDB for Startups能帮你用Atlas和Voyage AI的积分以及技术支持更快地推进，帮你顺利发布产品。

If you're building in AI, MongoDB for Startups helps you move faster with Atlas and Voyage AI credits and technical support to help you ship.

不要构建那种只回答一次就忘记的智能体。

Don't build agents that answer once and forget.

打造能记住并从你的实时数据中学习的智能体。

Build agents that remember and learn from your real-time data.

访问 mongodb.com/agents。

Go to mongodb.com/agents.

就是 M-O-N-G-O-D-B.com/agents。

That's M-O-N-G-O-D-B.com/agents.

MongoDB 扩展产品，Framer 扩展故事。

While MongoDB scales the product, Framer scales the story.

当一个新的落地页变成一堆工单和交接时，Framer 能帮助你的团队更快推进。

When a new landing page turns into a pile of tickets and handoffs, Framer helps your team move faster.

Framer 是一个 AI 网站构建器，帮助创作者、团队和企业以前所未有的速度发布生产级网站，同时确保每个细节都完美。

Framer is the AI website builder that helps creators, teams, and businesses ship production-ready sites faster than ever while getting every detail right.

在一个地方提示、检查、编辑和发布，节奏焕然一新。

Prompt, inspect, edit, and publish in one place at a whole new pace.

智能体与人类协同工作。

Agents and humans work in tandem.

智能体带来速度和规模。

Agents bring speed and scale.

你带来品味、判断力和掌控力。

You bring taste, judgment, and control.

作品落在画布上，随时可以编辑。

The work lands on the canvas and stays editable.

在一个地方构建自定义代码组件、管理CMS内容、优化SEO并审计问题。

Build custom code components, manage CMS content, optimize SEO, and audit for issues all in one place.

企业级托管、安全性和99.99%的正常运行时间SLA，受到Perplexity和Miro等领先品牌的信赖。

Enterprise-grade hosting, security, and 99.99% uptime SLAs trusted by leading brands like Perplexity and Miro.

了解如何从Framer专家那里获得更多网站价值，或立即免费开始构建，访问framer.com/20vc，Framer Pro年度计划可享30%折扣。

Learn how you can get more out of your site from a Framer specialist, or get started building for free today at framer.com/20vc for 30% off a Framer Pro annual plan.

网址是framer.com/20vc。

That's framer.com/20vc.

可能适用相关条款和限制。

Rules and restrictions may apply