让创业公司变得强大

September 20262026年9月

One of the most useful heuristics I have when doing office hours with startups is to ask: what would make this company more powerful? Asking how the company could make more money is a good heuristic too, but it tends to yield incremental improvements. Whereas thinking about how to make it more powerful will sometimes make it orders of magnitude more valuable.在与创业公司进行办公时间交流时，我最有用的启发式方法之一是问：什么能让这家公司变得更强大？问公司如何能赚更多钱也是一个好的启发式方法，但它往往只会带来渐进式的改进。而思考如何让它变得更强大，有时会让它的价值提升几个数量级。

There are a lot of different variants of this question, depending on the type of company. Is there a way to transform the company from a mere component supplier into the one that owns the relationship with the customer? Or the related question: is there a way to make the money flow through it? It's always good when money flows through you. [1]根据公司的类型，这个问题有很多不同的变体。有没有办法把公司从一个单纯的组件供应商转变为拥有客户关系的一方？或者相关的问题：有没有办法让钱流经它？当钱流经你时，总是好事。[1]

Is there a way to create something akin to an app store, where other companies can build upon your product? Then all their efforts to create valuable things make you more valuable too. Ideally this is combined with owning the customer relationship and making the money flow through you. [2]有没有办法创造出类似应用商店的东西，让其他公司可以在你的产品之上进行构建？这样，他们为创造有价值的东西所做的一切努力，也会让你变得更有价值。理想情况下，这与拥有客户关系以及让钱流经你相结合。[2]

Network effects make companies more powerful, and I almost always think about how to introduce them. I treat it as a kind of challenge to see if there's a way to get network effects even in things you wouldn't expect to have them. [3] It's surprising how often it can be done. And when it can, this sometimes transforms the idea completely; what had been a service is now a marketplace. The deluxe version is full app store, but if there's no more direct way to do it, you can often induce network effects by letting your users share something. For example, if you opt in, we'll tell you how you're doing compared to other users. The obvious AI variant is to let your users opt in to training your model on their interactions with it. Many will resist that, but if some don't, the model they get to use will outperform the vanilla one used by the others.网络效应让公司变得更强大，而我几乎总是会思考如何引入网络效应。我把它当作一种挑战，看看有没有办法在你原本不会预期有网络效应的东西里也获得网络效应。[3] 能做到这一点的情况多得令人惊讶。而当能做到时，这有时会彻底改变这个想法；原本是一项服务的东西，现在变成了一个市场。豪华版就是完整的应用商店，但如果没有更直接的方式做到这一点，你通常可以通过让用户分享某些东西来诱发网络效应。例如，如果你选择加入，我们就会告诉你，与其他用户相比你表现如何。明显的AI变体是让你的用户选择加入，用他们与模型的交互来训练你的模型。很多人会抗拒这一点，但如果有些人不抗拒，那么他们能使用的模型就会胜过其他人使用的普通版本。

Often you can induce network effects by generalizing the idea, which makes the startup doubly more powerful. For example, if I were talking to a startup building a way for agents to pay for things, the first question I'd ask is whether the agents could also pay one another. If they can do that, you become a marketplace. And being a marketplace is so valuable that if it wasn't immediately obvious what agents could pay one another for, it would be worth spending a lot of time trying to think of something. If you could, it might be worth tilting the whole company toward that, and if necessary even becoming a market maker to get it rolling.你常常可以通过推广这个想法来诱发网络效应，而这会让创业公司强大一倍。例如，如果我在和一家正在构建代理支付方式的创业公司交谈，我首先会问的问题是，这些代理是否也能互相支付。如果它们能做到这一点，你就变成了一个市场。而成为一个市场是如此有价值，以至于如果代理之间能为什么互相支付并不是立刻显而易见，那也值得花大量时间去努力想出一些东西。如果你能想出来，也许值得把整个公司都向那个方向倾斜，如果有必要，甚至成为做市商来让它启动起来。

These hypothetical transformations of the original idea don't always yield anything promising. Far from it. But they're always worth considering; if nothing else, trying to transform an idea helps you understand it better.这些对原始想法的假设性转变，并不总是能产生任何有前景的结果。远非如此。但它们总是值得考虑；至少，尝试转变一个想法能帮助你更好地理解它。

There are certain kinds of thinking where ideas start to seem almost physical. Most programmers have probably experienced it. Manipulating startup ideas feels this way too. One transformation that feels especially physical is the strategy of going full stack: instead of selling your technology to companies doing x, you use the technology yourself to do x in competition with them. You can almost see the idea stretch as it engulfs what had been the customer. And now that you own the outside surface, the shape of that probably changes too.有些类型的思考会让想法开始显得几乎是物理性的。大多数程序员大概都体验过。操纵创业想法感觉也是如此。一种感觉特别具有物理性的转变，是走全栈路线：与其把你的技术卖给做x的公司，不如你自己使用这项技术来做x，与它们竞争。你几乎能看到这个想法在吞噬原本的客户时被拉伸。而既然你现在拥有了外表面，那个外表面的形状大概也会改变。

There's a variant of going full stack where you eat your way gradually through the customer by doing all their hardest work for them. In the limit case, you're doing all the brainwork and they're just running errands for you. At which point, as in the full stack case, the real customer is their customer; the initial customer is now just a kind of hand puppet.走全栈路线有一种变体，就是你通过替客户做所有最难的工作，逐渐把他们吃掉。在极限情况下，你在做所有的脑力工作，而他们只是在替你跑腿。到那时，就像在全栈情况下一样，真正的客户是他们的客户；最初的客户现在只是一种傀儡。

Another thing I'm always looking for is tails that could wag the dog. The history of startups is full of these. Paypal started out doing security for hand-held devices. They created Paypal as a demo of their security software. But then eBay sellers started using it to take payments, and after a couple months the founders acknowledged that this was the business they were now in, even though they hadn't meant to be. So whenever founders build something peripheral to the main product I always ask: could this be the real product?我一直还在寻找的另一件事，是可能反客为主的“尾巴”。创业公司的历史充满了这种例子。Paypal最初是做手持设备安全性的。他们创建Paypal，是作为其安全软件的演示。但后来eBay卖家开始用它来收款，几个月后，创始人们承认这就是他们现在所在的业务，尽管他们原本并不打算如此。所以每当创始人构建了主产品的某种外围东西时，我总是会问：这会不会才是真正的产品？

It's exciting when you notice users "misusing" your product to do something you hadn't intended. This means there's something they want so desperately that they'll not only use any solution you offer, but even use things that aren't meant to be solutions. When you see something like that, don't be annoyed that your users are using your product wrong; listen for the message they're sending, because it could be valuable.当你注意到用户“误用”你的产品去做你未曾打算的事情时，这很令人兴奋。这意味着他们如此迫切地想要某样东西，以至于他们不仅会使用你提供的任何解决方案，甚至会使用那些本来并非解决方案的东西。当你看到这样的情况时，不要因为用户错误地使用你的产品而恼火；倾听他们发出的信息，因为它可能很有价值。

The reason Paypal grew so fast was that it helped users make money. Few things make you more powerful than that. When you help users make money, they're (a) quick to adopt your product and (b) will pay a lot for it. So your revenues grow doubly fast. Most of the most successful companies we've funded help their users to make money, as does YC itself.Paypal增长如此之快的原因，是它帮助用户赚钱。很少有东西能比这更让你强大。当你帮助用户赚钱时，他们（a）会很快采用你的产品，（b）会为此支付很多钱。所以你的收入会以双倍速度增长。我们投资过的大多数最成功的公司都帮助用户赚钱，YC本身也是如此。

Playing the long game makes you more powerful, because most other people you encounter won't be. Most startups competing with you will be run by opportunists hoping to be acquired. Most big companies you deal with will be run by executives who don't expect to be there for more than a few years and are only thinking about this quarter's numbers. So tradeoffs that only pay off in 10 years will usually be underpriced. The classic one is to offer great terms in order to acquire users. I generally advise startups to sell as cheaply as they need to at first; get all the users, then worry about your margins. But there are usually also deeper, structural ways to play the long game. [4]玩长期游戏会让你更强大，因为你遇到的大多数其他人不会这么做。与你竞争的大多数创业公司，都会由那些希望被收购的机会主义者经营。与你打交道的大多数大公司，都会由那些不指望在那里待超过几年、只想着这个季度数字的高管经营。所以那些只有在10年后才会有回报的取舍，通常会被低估。经典做法是提供极好的条件来获取用户。我一般建议创业公司一开始尽可能便宜地卖；先拿到所有用户，然后再担心利润率。但通常也有更深层的、结构性的方式来玩长期游戏。[4]

Being generous makes you more powerful. As Tim O'Reilly said, you should create more value than you capture. Many hard-headed business types would write this off as idealistic hippy stuff, but in fact this is the route to becoming really rich. Squeezing every last penny out of customers is a distraction. It gets you 2x returns at most. Whereas discovering some new thing you could make for them could easily get you 10x or 100x returns. They're two different ways of looking at the world, and the O'Reilly way makes more, for those who can do it. [5]慷慨会让你更强大。正如Tim O'Reilly所说，你应该创造比你获取的更多的价值。许多精明的商业人士会把这当作理想主义的嬉皮士东西而不屑一顾，但事实上，这是变得真正富有的途径。从客户身上榨取最后一分钱是一种分心。它最多给你带来2倍的回报。而发现某种你可以为他们制造的新东西，则很容易给你带来10倍或100倍的回报。这是两种不同的世界观，而对那些能做到的人来说，O'Reilly的方式赚得更多。[5]

The classic example of generosity leading to power is when companies open source their software. They literally give away the product, but by giving it away they both make it a standard and make users trust it more. That makes it spread, and in the end they end up with a small piece of a much, much bigger pie.慷慨通向权力的经典例子，是公司将自己的软件开源。他们实际上把产品送出去，但通过送出去，他们既让它成为标准，也让用户更信任它。这使它传播开来，最终他们得到的是一块大得多的馅饼中的一小块。

If you don't want to go full open source, you can get some of the benefits by making your product extensible. One end of that continuum is the app store, but if you don't restrict or charge for extensions you may ultimately end up with a bigger ecosystem. [6]如果你不想走完全开源路线，你可以通过让你的产品可扩展来获得其中一些好处。这个连续谱的一端是应用商店，但如果你不限制扩展或不对扩展收费，你最终可能会得到一个更大的生态系统。[6]

The ultimate in extensibility is to let your product be called via an API. Many companies shrink from that because they dislike the loss of control that comes with it. They want their software to be used only in the way they intended. There may be some specialized domains where you want to be rigid about this, but I suspect it's usually a mistake. Especially now that agents are replacing human users. Who knows what they'll want to do? So err on the side of having APIs. Especially when you're a larval startup and have nothing to lose.可扩展性的终极形式，是让你的产品可以通过API被调用。许多公司对此退缩，因为他们不喜欢随之而来的控制权丧失。他们希望自己的软件只按他们 intended 的方式被使用。也许在某些专业领域，你会想对此保持严格，但我怀疑这通常是个错误。尤其是现在代理正在取代人类用户。谁知道它们会想做什么？所以要偏向于拥有API。尤其是当你还是一家幼虫期的创业公司、没有什么可失去的时候。

Strangely enough, selling to earlier stage companies makes you more powerful. Founders are often surprised by this, because the earlier you sell to startups, the less money they have. But if you get them as customers at the very beginning and you charge based on usage, your revenues will grow at startup rates.说来奇怪，向更早期阶段的公司销售会让你更强大。创始人常常对此感到惊讶，因为你越早向创业公司销售，它们的钱就越少。但如果你在最初就把它们变成客户，并且按使用量收费，你的收入就会以创业公司的速度增长。

This is why Stripe makes a point of signing up companies at the first possible moment. With payments infrastructure, if it ain't broke, you don't fix it, and since Stripe ain't broke, companies that install it never churn. And selling to early stage startups is so straightforward. The founders are sophisticated and decide quickly. If you have the best product, you win. Whereas if you're making something you can't sell to companies till they have 500 people, you're in a much weaker position. Now you're doing enterprise sales, which takes forever and is notoriously not a domain where the best product wins.这就是为什么Stripe特别强调要在尽可能早的时刻签下公司。对于支付基础设施，如果它没坏，你就不会去修，而既然Stripe没坏，安装了它的公司就永远不会流失。而向早期创业公司销售是如此直接。创始人们很老练，决策很快。如果你有最好的产品，你就赢。而如果你做的东西，直到公司有500人时你才能卖给他们，那你的处境就弱得多。现在你在做企业销售，那要花很长时间，而且众所周知，这不是一个最好的产品会赢的领域。

When I ask a startup how big a customer has to be before they'll buy their product, I always hope the number will be low. And if it's not, I always ask if there's some way to tweak the product so they can sell it earlier. The ideal is something they can Collison-install right now for their batchmates. [7]当我问一家创业公司，客户必须多大才会购买他们的产品时，我总是希望这个数字会很低。如果不是，我总是会问，有没有办法调整产品，让他们能更早地卖出去。理想的情况是，他们现在就能为他们的同批同学进行“Collison式安装”。[7]

More generally, having customers who decide fast makes you powerful. Not just because of the speed, but because customers who decide fast tend to decide based on how good you are. Startups usually make the best stuff (if you were both small and mediocre, how could you even survive?) and when customers decide fast, making the best stuff leads straight to making the most money. Whereas selling to customers like hospitals and school districts is like walking through mud. Whenever I meet a startup selling to customers like that, I ask if there's some way they could at least start by selling to a subset of the market that decides faster. [8]更一般地说，拥有决策快速的客户会让你更强大。不仅仅是因为速度，还因为决策快速的客户往往会根据你有多好来决策。创业公司通常做出最好的东西（如果你既小又平庸，你怎么可能生存？），而当客户决策很快时，做出最好的东西会直接通向赚最多的钱。而向医院和学区这样的客户销售，就像在泥里走路。每当我遇到向这类客户销售的创业公司时，我都会问，有没有办法至少先从市场中决策更快的一个子集开始销售。[8]

There's a strategy similar to getting the customers early, which is to get the data early. Rippling used this technique brilliantly. Their goal was always to be both the operating system for applications dealing with employee data and most of the applications running on it. They didn't know exactly what this would look like; it would have to evolve; so they started by writing onboarding software, because that's where the life of employee data begins. And since there was more at stake for them than just the onboarding software market, their onboarding software was way better than it needed to be, and spread rapidly.有一种策略类似于尽早获取客户，那就是尽早获取数据。Rippling出色地使用了这种技术。他们的目标始终是既成为处理员工数据的应用程序的操作系统，又成为运行在其上的大多数应用程序。他们并不确切知道这会是什么样子；它必须演化；所以他们从写入职软件开始，因为那是员工数据生命开始的地方。而且由于对他们来说，利害关系不只是入职软件市场，他们的入职软件远比它需要的更好，并迅速传播开来。

Upstream is almost always good, whether it's with money or user relationship or customer stage or data.上游几乎总是好的，无论是就金钱、用户关系、客户阶段还是数据而言。

Since startups make the best stuff, they're strongest on level playing fields. They're weakest in markets dominated by companies you'd describe as mafia. Record labels are mafia. PBMs are mafia. In these worlds you don't win by having the best product. Indeed you may only even exist for as long as the mafia chooses to allow you to. Which is not to say they can't be defeated. They probably can be, but you'd have to do it by coming in from the side — by somehow making them irrelevant, rather than by frontal attack. Then you wouldn't depend on beating them to succeed; it would be an ancillary benefit of winning in another dimension. [9]由于创业公司做出最好的东西，它们在公平竞争的环境中最强。它们在由你所谓黑手党式公司主导的市场中最弱。唱片公司是黑手党。PBM是黑手党。在这些世界里，你不是靠拥有最好的产品来赢。事实上，你可能只有在黑手党选择允许你存在的时间内才能存在。这并不是说它们无法被击败。它们大概可以被击败，但你必须从侧面切入——通过以某种方式让它们变得无关紧要，而不是正面攻击。这样，你就不必依赖击败它们来成功；那会成为在另一个维度获胜的附带好处。[9]

Often what you're doing when you explore ways to make a company more powerful is finding ways not to be held back by other companies. If you're a component supplier and have to live in a (sometimes literal) box created by another company, you can escape that if you can find a way to own the customer relationship. If you're selling to companies that are big and bureaucratic, you can escape that by selling to them when they're smaller and decide faster, or by using your technology yourselves to compete with them. This pattern is so common that you can use it as a heuristic for generating ways to make an idea bigger. In what ways is the current idea being held back by other companies?当你在探索让公司变得更强大的方法时，你经常在做的事情，是找到不被其他公司拖累的方法。如果你是一个组件供应商，必须生活在另一家公司创造的（有时是字面意义上的）盒子里，那么如果你能找到一种拥有客户关系的方法，你就能摆脱它。如果你在向庞大而官僚的公司销售，你可以通过在他们更小、决策更快时向他们销售，或者通过自己使用你的技术与他们竞争，来摆脱这一点。这种模式如此常见，以至于你可以把它当作一种启发式方法，用来生成让想法变得更大的方式。当前这个想法在哪些方面被其他公司拖累了？

Often as not, though, startups are being held back by themselves. A surprising percentage of the advice I give to startups has the word "just" in it. You don't need to x. Just y. One way startups' ideas get twisted into knots is by evolving from something else; there's now a part they don't need, and they don't realize it yet. But with very early stage startups especially, the reason the idea is complicated is often fear. The company is unconsciously cowering by doing something less ambitious than they could. Just y is often, in effect, just stand up straight. And when they do they're much taller.不过，创业公司常常是被它们自己拖累的。我给出的创业建议中，有令人惊讶的高比例都带有“只要”这个词。你不需要x。只要y。创业公司的想法被扭成一团的一种方式，是从别的东西演化而来；现在有一部分是他们不需要的，而他们还没有意识到。但尤其是对于非常早期阶段的创业公司来说，想法之所以复杂，原因往往是恐惧。公司正在无意识地退缩，做着比它们所能做的更缺乏雄心的事情。“只要y”常常实际上就是“只要站直”。而当它们站直时，它们就高得多。

But all these strategies for making startups more powerful have one thing in common — or more precisely, have to obey one constraint. They all have to make things better for the customer. You can't add network effects or make the money flow through you or go full stack just because you'd like to. You can only do these things when the result is better for the customer. Otherwise you won't have any uptake.但所有这些让创业公司变得更强大的策略都有一个共同点——或者更准确地说，都必须遵守一个约束。它们都必须让客户的情况变得更好。你不能仅仅因为你想要，就添加网络效应，或让钱流经你，或走全栈路线。只有当结果对客户更好时，你才能做这些事情。否则你不会有任何采纳。

These are strategies for making startups powerful in the long term, but startups that execute them don't usually have any power at the point when they do. And indeed this initial weakness of startups is why, on the whole, they're good for the world. Newly founded startups are too weak to force anything on anyone. The only way they can become powerful is to make customers' lives better. In fact this constraint is so rigid that you can run it backwards to generate ideas. What would the perfect world look like, from the customer's point of view? If there's a component of that world that the startup could transform itself into, it probably should.这些是让创业公司在长期变得强大的策略，但执行这些策略的创业公司在执行时通常还没有任何权力。而事实上，创业公司这种最初的弱小，正是它们整体上对世界有益的原因。新成立的创业公司太弱，无法把任何东西强加给任何人。它们变得强大的唯一方式，是让客户的生活变得更好。事实上，这个约束如此严格，以至于你可以把它反过来运行来生成想法。从客户的角度看，完美的世界会是什么样子？如果那个世界有一个组成部分是创业公司可以把自己转变而成的，那它大概就应该这么做。

Notes注释

[1] You can also make tokens flow through you, and this usually means that money flows through you too, since the tokens have to be paid for. In theory this puts you in a powerful position. You own the customer relationship, and the model companies are in effect component suppliers. The question is how easy it would be for them to engulf you, or even your customers.[1] 你也可以让token流经你，而这通常意味着钱也流经你，因为token必须被付费。理论上，这让你处于一个强大的位置。你拥有客户关系，而模型公司在实际上是组件供应商。问题是，它们要吞噬你，甚至吞噬你的客户，会有多容易。

[2] If you can't create an app store, can you at least define the standard for how different companies' products interact? In a new field there's often no standard yet. But don't worry that you're too small to propose one. If you're one of the first in the field, you presumably have as good ideas as anyone about what such a standard should look like. And everyone is so hungry for standards that the first to be proposed tends to win, no matter who proposed it.[2] 如果你不能创建一个应用商店，你至少能不能定义不同公司产品如何交互的标准？在一个新领域里，通常还没有标准。但不要担心你太小，不能提出一个标准。如果你是这一领域最早的人之一，你大概对这样的标准应该是什么样子有着和任何人一样好的想法。而每个人都如此渴望标准，以至于第一个被提出的标准往往会赢，不管是谁提出的。

[3] YC itself is an instance of network effects in something you wouldn't expect to have them. We didn't intend it to be, but we realized very quickly that that was what we'd stumbled upon.[3] YC本身就是网络效应出现在你原本不会预期有网络效应的东西中的一个实例。我们并不是有意如此，但我们很快意识到，这正是我们偶然发现的东西。

[4] There are even times when it's worthwhile to sell at a loss. But be careful when you do this, because if you give away too much, you lose the signal that customers send by paying you. If your product is a $10 bill that you sell for $5, your growth rate isn't telling you anything useful.[4] 甚至有些时候，亏本销售是值得的。但这样做时要小心，因为如果你送出去太多，你就会失去客户通过付钱给你所发送的信号。如果你的产品是一张10美元的钞票，而你以5美元出售，那么你的增长率就没有告诉你任何有用的东西。

[5] The O'Reilly way of looking at the world is more common among founders. When companies switch from making new products to squeezing more profit out of existing ones, it's often because control has passed from the founders to hired managers.[5] O'Reilly式的世界观在创始人中更常见。当公司从制造新产品转向从现有产品中榨取更多利润时，往往是因为控制权已从创始人转移到雇来的管理者手中。

Partly this is because only founders tend to have the inclination or the ability to create new things. But it's also because founders have experienced weakness. Hired CEOs take the power of the companies they run for granted, whereas founders remember the days when the company was so weak that it had to delight users to survive.部分原因是，只有创始人往往有创造新事物的倾向或能力。但也是因为创始人经历过弱小。受雇的CEO把他们所经营公司的权力视为理所当然，而创始人记得公司曾经弱到必须取悦用户才能生存的日子。

[6] Perhaps flexibility in this department will be the key to finally displacing Apple. One thing you can be sure of is that they'll be restrictive about hardware and software that integrates with theirs.[6] 也许在这个方面的灵活性，将成为最终取代Apple的关键。有一点你可以确定，那就是对于与他们的硬件和软件集成的硬件和软件，他们会很限制。

[7] If you switch from asking "What size customers should we target?" to "At what point in their life should we acquire customers?" it becomes clear that targeting bigger companies is just targeting a given company later. As long as you're confident that customers won't churn, why not lock them in early? Why do slow, brittle enterprise sales when you could just sell to early stage startups and then grow with them? This kind of situation is exactly why YC emphasizes growth rate rather than absolute numbers. If your growth rate is good enough, the absolute numbers will take care of themselves. And the way to get the fastest growth rate is to sell to the customers who grow the fastest and decide the fastest.[7] 如果你从问“我们应该瞄准多大尺寸的客户？”转向问“我们应该在客户生命周期的哪个时点获取他们？”，就会很清楚，瞄准更大的公司，其实只是更晚地瞄准某个既定公司。只要你确信客户不会流失，为什么不早早锁定他们？既然你可以直接卖给早期创业公司，然后与它们一起成长，为什么还要做缓慢而脆弱的企业销售？这种情况正是YC强调增长率而非绝对数字的原因。如果你的增长率足够好，绝对数字自会照顾自己。而获得最快增长率的方法，就是卖给那些增长最快、决策最快的客户。

This way of looking at the world comes naturally when you're playing the long game. If you think in quarters, it seems like potential customers have fixed sizes. If you think in decades, you can see they have trajectories.当你玩长期游戏时，这种世界观会自然产生。如果你按季度思考，潜在客户看起来就像是有固定尺寸。如果你按十年思考，你就能看到它们有轨迹。

[8] In a normal industry, customers who are slow to adopt new technology represent an opportunity for startups. The slower they are, the more likely you can win by going full stack. What makes selling to hospitals and school districts so grim is that you generally can't — though there are some opportunities to go around schools and go directly to serving students.[8] 在一个正常行业里，采用新技术缓慢的客户对创业公司来说代表着一个机会。它们越慢，你就越有可能通过走全栈路线获胜。向医院和学区销售之所以如此令人沮丧，是因为你通常不能这么做——尽管有一些机会可以绕过学校，直接服务学生。

[9] Apparently one thing record labels and PBMs have in common is that they're full of lawyers. So this is presumably a way to recognize such companies.[9] 显然，唱片公司和PBM有一个共同点，那就是它们都充满了律师。所以这大概是识别这类公司的一种方式。

Thanks to Sam Altman, Patrick Collison, Aaron Epstein, Ankit Gupta, Diana Hu, Pete Koomen, Jessica Livingston, and Harj Taggar for reading drafts of this, and to Diana for reminding me about token flow.

感谢Sam Altman、Patrick Collison、Aaron Epstein、Ankit Gupta、Diana Hu、Pete Koomen、Jessica Livingston和Harj Taggar阅读本文草稿，并感谢Diana提醒我关于token流的问题。