Sam Altman：创办 OpenAI 与押注不可能

David Senra

我刚提到 Tobi Lütke，以及我之前录过他。你为什么说他是现在最有意思的 CEO 之一？

最打动我的一点是：在 AI 很早的时候，以及曲线上的每一个节点，他都是最往前倾的 CEO。他自己在写软件，自己在试，还给我们极其详细的产品反馈、模型能力反馈。别人还没这么说的时候，他就说：我们不是 NPC 公司，所以我们会采用智能体，否则就完了，我们要自己做。每次跟他聊，他都在任何人——CEO 与否——的最边上。他自己做、自己懂，手感很深，总是比别的 CEO 早六到八个月。

你还记得吗，大概一年半前，也许是 2024 年，他写过那封信，说你要做的第一件事是看 AI 能不能解决你的问题。那时候大家觉得很荒唐。这就是我的意思：他一直领先，一直是对的，而且很 no bullshit。没有炒作，就是：它现在真正能做什么，我觉得很快能做什么，我要怎么把公司往这儿推。对你这种位置的人来说，有一个大型公司 CEO 能给这么准、这么前沿的产品反馈，好处一定很大。很多人会给反馈。他是唯一站在「大公司 CEO」和「极其准确、细节、卡在前沿的反馈」交叉点上的人。

他跟我说——不确定是在那期节目里还是之后——我们回头看 2026，会是每家公司都可能被抢走的一年。会有人做出 AI 原生版的 Shopify，他说那个人会是我。所以晚上他好像在自己重做：如果从零开始、用现在的技术，你会怎么做。他真的自己上手。大多数到那个层级的 CEO，都是团队管团队去落地，把毛边磨平来让你高兴。你自己不干，很难有手感。据我所知他整晚都在亲手做。

我不确定 2026 会不会是每家公司都觉得自己可能被抢走的一年。在这一点上我可能和他有一点不同。但我懂那个精神，也感觉那件事正在发生。

你觉得这甚至可能吗？不管是 2026 还是 2046？

显然不是字面意义上的每家公司。有些东西非常反 AI。AI 越好，一些跟 AI 完全无关的生意反而更难被竞争掉，因为我们会更想要那种真实的、非科技的体验，或者更在意球队之类的。所以不是所有东西。但会有很多软件生意非常「谁都可以来抢」。

那你是不同意时间表？

对，我不同意时间表。我觉得会再久一点。我热爱创业公司，觉得那是经济里最酷的东西，职业生涯都在试图真正理解它们。我以为到了 GPT——大概是 2023 年——很快就会有比实际多得多的软件生意被颠覆、被抢走。速度上我错了几件事，其中一件是：经济的惯性太大了。人们继续做同样的事，继续向同一家公司买，继续想用工具的同样方式。很多方面这其实是件好事，会让眼前这场大转型走得更平滑、更慢。我为此感恩。但也意味着，即便技术这么惊人，我们所有人在时间表上都太野心勃勃了。AI 是人类发明过的最不可思议的技术之一。社会和会更慢地适应。

有意思。开录前我们聊过这些历史平行。你刚才说的时候，我甚至没在想 OpenAI，而是在想 80 年代 Larry Ellison 的传记。他说：伙计们，这不是软件问题，是人的问题。我们得说服他们。软件能装上，他们不用。得改他们的行为。Netflix 开始寄 DVD、还没开始流媒体的时候，人们还去 Blockbuster，这让我非常吃惊。习惯的力量、人们做事的方式，比技术宅意识到的难改得多。

I just brought up Tobi Lütke and the fact that I recorded with him previously. Why did you say that you think he’s one of the most interesting CEOs right now?

One of the things that struck me the most about Tobi is in the very early days of AI, and then at every moment along the curve as it’s developed, he has been the most forward-leaning CEO. He’s in there writing the software himself. He is experimenting with it. He sends us extremely detailed feedback on the product offering, on the capabilities of the models. Before anybody else was saying this, he was like, we are not an NPC company, and thus we are going to adopt agents. Otherwise we’re totally screwed. We’re going to build it ourselves. Every time I talk to him, he is at the edge of what anyone, CEO or not, is doing. He builds himself, he understands, he has a great deep feel, and he is always six, eight months ahead of any other CEO.

Do you remember when he wrote, it was probably like a year and a half ago, maybe 2024, that letter saying that the first thing you have to do is see if AI can solve your problem. Even back then people went crazy. They thought it was ridiculous. This is my point. He’s just consistently been ahead. He has been correct. He’s leaned in. He’s very no-bullshit. There’s no hype. Nothing other than: here’s what it can really do right now. Here’s what I think it’ll be able to do soon. Here’s how I’m going to push the company. I never even thought about how much of a benefit it is for somebody in your position to have somebody like that giving you intense, very direct, clear product feedback. A lot of people send product feedback. He is the only person at the intersection of CEO of a large company and extremely accurate, detailed, on-the-cutting-edge feedback.

He told me — I don’t know if it was on the episode or after — he was very adamant. He’s like, we’re going to look back on 2026 as a year that every business was up for grabs. Somebody was going to build the AI-native version of Shopify, and he said it’s going to be me. At night he is apparently trying to rebuild: if you started from scratch, what would you do with the current technology. He does it himself. He’s using these tools himself. He’s writing software himself. He’s trying the models himself. He’s trying to reimagine his workflows himself. Most CEOs, when you get to that level, have teams of people managing teams of people trying to implement the thing and make you happy and smooth the rough edges. I think it’s very hard to get the feel if you’re not actually doing the thing. And he does it so hands-on all night long as far as I can tell.

I’m not sure if 2026 will be the year that every business feels up for grabs. I might disagree with him a little bit there. But I get the spirit of that, and I do understand it. It feels like that’s happening.

Do you think that’s even possible? Like whether it’s 2026 or 2046?

Obviously not literally every business. I think there are some things that are very anti-AI. The better AI gets, the more some businesses that have nothing to do with AI will be harder to compete with, because we’ll really want these authentic, non-technological experiences, or we’ll care more about sports teams or whatever. So no, not everything. But I think there will be many software businesses that are very up for grabs.

Would you disagree on the timeline then?

Yeah, I disagree on the timeline. I think it’s going to take a little bit longer. I love startups. I think startups are the coolest thing in the economy. And I’ve spent my career trying to really understand startups. I thought when we got to GPT, which was back in 2023 I think, that very quickly after that there was going to be much more disruption, software businesses being up for grabs right away, than it turned out to be. The thing that I think I was wrong about, in terms of the speed: one of them is the economy just has so much inertia. People keep doing the same things they’re doing. They keep buying from the same company. They keep wanting to use their tools in the same way. I think it’s actually a positive in many ways, and it’s going to make this big transition in front of us go smoother and slower. I’m grateful for it. But I think it means we’ve all been too ambitious on timelines, even with this incredible technology. I think AI is one of the most incredible technologies humanity’s ever invented. Society and the economy will adapt more slowly.

It’s funny. We were talking before we started recording that there are all these parallels to history. When you were just talking I was thinking of reading this biography of Larry Ellison in the ’80s. He was just like, guys, this isn’t a software problem, it’s a people problem. We have to convince them. We can install software. They’re not using it. We have to change their behavior. My own example was after Netflix came out and started shipping DVDs, even before they started streaming. It was amazing to me that people still went to Blockbuster. Force of habit and the way people do things — changing behavior is just much harder than the tech nerds realize.

David Senra

回到 Tobi 那封在公司里催采用的信。你部分是在发明这些东西，所以采用得比谁都快。有没有你自己都吃惊的行为：我知道有更好的办法，我甚至在做那个可以更好的产品，可我还是过不了这个习惯？

百分之百。从来没人问过我这个。我等这个问题等很久了。我觉得自己心理上最不一致的一点是：我用电脑的方式，二十年来都一样。我现在有一个叫 Codex 的魔法东西，所有人也都有。这意味着我应该完全换一种方式用电脑。我不该点来点去，把一个通讯软件粘到另一个；不该漫无目的地滚邮件，找哪封最不痛苦；不该还用老办法管待办、做这些粗活。可我脑子里好像编码了一件事：干这种活才叫工作、才叫生产力。你要是问我，我绝不会说喜欢那样干，事实上我会说相反的话，而且我觉得我是认真的。但用显示偏好看：我现在有更好的办法，更快，可以让 Codex 处理更多日常——过邮件堆、过待办——我还是那样干。这说不通，除非我其实偷偷喜欢，或者那样会让我感觉良好。

你觉得要怎样才能更深入地采用自己的产品？

我不太确定。它在逐渐发生。也许正确答案就是这些事必须逐渐发生，彻底改掉根深蒂固的习惯和工作流很难。我觉得我们还能用这项技术做出更好的产品，让过渡更无缝。但现在感觉大家都骑在两个世界之间：电脑还能按老办法用，Codex 又能用这种惊人的新方式用我们的电脑，我们不确定什么时候用哪个。我觉得这主要是产品失败。现在这个阶段让我想起 iPhone 之前的智能手机。我是早期用户，大概 2003、2004 年有过 Palm Treo。缺的不只是多点触控，更缺的是让 iPhone 成为 iPhone 的那些产品想法。我们现在有了所有技术零件，但还没有那个完全改变人与技术接口的 iPhone 时刻。

If we go back to this uproar of Tobi writing that letter to the people inside of his company — you’re adopting this faster than anybody else because you’re partially inventing them. Is there something where you’re actually shocked at your own behavior: I know there’s a better way to do this, I’m even creating the product that could be better, and yet I still can’t get over this force of habit?

A hundred percent. No one has ever asked me this before. I have been waiting for this question. The thing to me that feels most psychologically inconsistent about myself is that I have for 20 years been using computers the same way. I now have a magic thing called Codex. So does everybody. That means I should completely be using my computer in a different way. I should not be clicking around, pasting from one messaging app to another. I should not be scrolling mindlessly through my emails and trying to figure out which one is least painful for me to open and respond to when I don’t want to be dealing with it. I should not be keeping a to-do list and doing these rote computer tasks in the same way that I have for so long. And yet there’s something in my mind that is encoded that doing this kind of stuff is what it means to work and what it means to be productive. And if you asked me, I would never say I like doing it that way. In fact I would say the opposite, and I think I would mean it. But by revealed preference, I have a better way to do it now. I can do it faster. I can be using Codex for more of just my day-to-day — got to get through the stack of emails, got to do the stuff on my to-do list — and I still do it that way. And it makes no sense other than I must secretly like it or feel good about it.

What do you think has to change for you to actually adopt your own product in a more deep way?

I don’t really know. It’s happening gradually. And this might be the right answer, which is these things have to happen gradually, and totally changing someone’s ingrained habits and workflows is difficult. I think there are better products we can build with this technology. They will make it more seamless to do that. But right now it feels like we’re all kind of straddling these two worlds: we still have a computer we can use the old way, and we have Codex that can use our computer in this amazing new way, and we’re not sure which to use when for what. And I think this is mostly a product failure. The phase that we’re in now reminds me of smartphones before the iPhone. I was an early adopter. I had a Palm Treo in 2003 or ’04 or whatever. A lot of the technology was there. It was missing multi-touch, but mostly it was missing the product ideas that made the iPhone the iPhone. And I feel like we are now in a world where we have all of the technological pieces, but we have not had the iPhone moment of completely changing how someone interfaces with technology.

David Senra

我们有共同朋友 Josh Kushner。他说 Steve Jobs 想问题和办公司的方式，和他认为你做的方式很像。Jobs 不是自己写代码、做硬件，但他说：我是零号病人，我做自己想用的产品。苹果那套本质上就是他想要的。有个故事：团队为新 MacBook 准备了很紧张的一小时会，Steve 走进来，开关机，开盖有延迟，说：让这个变成那样，然后走了。你怎么改进产品？是只通过自己的需求吗？

我现在大部分精力在研究和算力上。我很希望能在产品上花更多时间。我们有很棒的人在想产品，但我们能做的最重要的事，是造出聪明的模型，并能高效、充裕地为很多人运行它们。这件事做对了，我相信其他都会跟上。哲学上我很倾向去找那个高杠杆、很难、能让指数继续的问题。对我们来说就是模型和算力。我也觉得那些问题天然适合我。

为什么天然适合你？

按我们这种方式扩展算力，需要复杂的供应链，很多有意思的合作要谈，我喜欢做这些。也有有意思的财务挑战：如何给可能已经是、或正在迅速变成史上最贵的基础设施项目融资。从自己设计芯片，到晶圆厂和机架供应链，再到电力系统——我一直对能源感兴趣——这些会碰到一起。围绕这种规模的算力，技术、商业、政策、供应链、物流上有很多有意思的问题。

我以前做创业投资。职业生涯里我发现和创业投资最接近的，就是管一个研究项目。它们也有很多不同。平均研究员和平均创始人表面上看不一样。但怎么找非共识押注、怎么决定在哪里有信念、怎么理解指数增长、怎么管理并识别离群人才，有很多相似。我们就坐在这栋研究楼里。

多说一点，你从创业投资里学到的，和做研究有什么平行。

很大一块是幂律。投资里大家老说这个：你得给大脑重新编程，因为我们天生不太这样想——你最好的一笔投资会超过其余所有加起来，第二好的又会超过剩下所有加起来。至少 AI 研究也是这样。我们起步的时候，人们觉得 AGI 完全不可能。那是 2015 年。我们因为要做 AGI，被领域里所有知识巨人锤。后来真正聚焦大语言模型，又被锤，说这完全荒唐。我至少从创业背景里懂：高风险押注没关系，只要做成了会超级值钱。研究也长这样。能成大事的研究者，是那种非共识、新鲜路子、能量高、非标准的人。

非标准再具体一点。他们是尖的吗？

你不会想投那种对上一千个人刚讲过的同一个想法只有一点点不同、却说服自己完全新颖的创始人。他们多半在跟羊群挤同一条赛道，做「该做的事」——创业。他们听 Peter Thiel 说得够多，知道你该有点不同，于是去模仿，但不是真的。什么时候有人就是想得跟大多数人不一样，并且愿意守住非常不受欢迎的信念，对我来说非常清楚。也许是错的。但如果对，他们会非常对。对比那种薄薄一层包装、其实大家都有的同一个想法。2015 年底我们创办 OpenAI 时，全世界没几个 AGI 项目，DeepMind，再一两个。那是非常非共识的事。同一年可能有成千上万创始人在做照片分享应用。今天很多人想办 AI 实验室。有那么一小撮人，两三个，在做完全新的、AI 好到现在才可能、但看起来还不像好主意的事。作为投资人我一直想投的就是那个，而且多半成了。研究里也一样：很多人追上一个成了的东西，少数人高信念走向一个还没人要的想法。我觉得我们曾经是、现在也是那些人最好的研究实验室。

几个月前跟 Demis 聊，他说就你们这种追法，钱和资本的要求，不会再有第四个更大的玩家，但有一个小数——我编个 10%——某个僧侣式的研究者会从一个我们从没考虑过的角度切入。

完全是。我不知道怎么给它一个数字，但确实有某种可能。我喜欢这个。这就是事情保持兴奋的原因。

We have a mutual friend in Josh Kushner. He says there’s a big comparison to be made between the way Steve Jobs thought and ran his company and the way he thinks that you do. He wasn’t writing the code or building the hardware, but he’s like, I am patient zero. I am making products that I myself want to use. Essentially everything we saw with Apple is just basically what he wanted. There’s a story where the team prepares a huge presentation on a new MacBook. He walks in, on, off, tries to open it, there’s a delay, goes, make this like that, and walks out. How do you approach it? How do you improve the product? Are you just doing it through your own needs?

Most of my effort right now is on research and compute. I would love to be able to spend more time on product. We have great people thinking about the product here, but the most important thing that we can do is to create smart models and to be able to run them efficiently and abundantly for a lot of people. If we can get that right, I believe that everything else will follow. Philosophically, I’m very inclined to say, try to find the high-leverage, difficult problem that will continue the exponential. And for us, this is models and compute. I also just think those are problems that naturally suit me.

Why do they naturally suit you?

To scale compute in the way that we’re doing, this requires a complex supply chain. There’s a lot of interesting partnerships to figure out, which I like doing. There’s interesting financial challenges of how you’re going to finance what is probably already, or at least rapidly becoming, the most expensive infrastructure project in history. The technology questions that go into building out compute at this scale — from design your own chip, to the supply chain of fabs and people that make racks, to the power systems for these things. I’ve always been interested in energy. It all comes together. So there are a lot of problems that are interesting across technology, business, policy, supply chain, logistics altogether around building compute at this kind of scale.

I used to be a startup investor. And the thing in my career that I have found closest to startup investing is managing a research program. There are all these ways in which they’re really different too. The average researcher and the average founder on the surface look different for obvious reasons. But there’s a lot of similarities about how you find the non-consensus bets, how you decide where to have conviction, how you understand what exponential growth looks like, how you manage outlier talent and how you identify it even more. This is the research building that we’re in and it’s where I sit.

Say more about why the parallels between what you learned in startup investing and doing research.

One big one is the power law. People talk about this all the time in investing, which is you have to reprogram your brain because we don’t seem naturally wired to think this way, where your best investment will outperform all of your other investments put together. Your second best will outperform everything else put together after that. And AI research, at least, is like that as well. When we started, people thought it was totally unlikely or almost impossible that AGI was possible. What year is this? 2015. We just got hammered by all of the intellectual giants of the field for saying that we were going after AGI. And then when we started really focusing on large language models, I got hammered again saying this is completely ridiculous. I understood at least from my startup background this point: high-risk bets are okay as long as you take the ones where if they work, it’s super valuable. And research looks this way. The kind of people that make great researchers are sort of non-consensus, fresh approach, high energy, non-standard people.

You’ve got to say more about non-standard. Can you be more specific? Are they spiky?

You don’t want to fund a founder who has a very slightly different take on the same idea as the last thousand people you talked to, who has tried to convince you and maybe themselves that somehow they’re completely different. But it’s mostly trying to fit in with the herd and be on the same track as everybody else and do what they’re supposed to do, which is start a startup. And they’ve heard Peter Thiel say enough times that there’s something you’re supposed to be doing different, that they kind of try to emulate that, but they don’t really mean it. It’s very clear to me when you have someone who just thinks differently than most other people and is willing to stand by convictions that are very unpopular. It may well be wrong. But if right, at least they’re going to be really right, versus someone who is a thin veneer on the same idea that everybody else has. In late 2015, we were starting OpenAI. There were very few AGI efforts in the world. There was DeepMind, one or two others. It was a very non-consensus thing to do. In that same year there were probably many, many thousands of founders starting photo-sharing apps. Today a lot of people want to start AI labs. There are some handful of people, two, three, whatever, doing something completely new that actually wasn’t possible until the AI got this good, but doesn’t seem like a good idea yet. And that is the thing that as a startup funder I always wanted to fund, and the thing that mostly worked for me. There’s a similar thing for researchers. There were a lot of researchers that would chase whatever the last thing was that worked, and there were a small number of researchers that had high conviction toward a new idea. And I think we were and are the best research lab for those people.

I remember talking to Demis about this a few months ago and he thought, in terms of chasing after it the way that you guys are, the money, the capital required, you’re not going to have a fourth bigger player, but there’s this 10% chance — I’ll just make up the number — that there’s just some monk researcher that’s going to approach it in a way that’s an angle we’ve never even considered.

Totally. I don’t know how to put a number on it, but there is some chance. I love that. I think that’s why stuff stays exciting.

David Senra

让我困惑的是，你从创始人变成投资人，再变回创始人。为什么是 2015 年？什么让你对人工智能感兴趣到这种程度：这事这么非共识，人们觉得我疯了，我还是要做？

我一辈子都对 AI 感兴趣。我是很宅的小孩，周五晚上玩电脑、看科幻、读科幻。我一直觉得 AI 会是最惊人、最疯狂的东西。从没想过真能去做，但一直爱它。上大学某种程度上就是为了学它。大一到大二那个夏天在 AI 实验室干活，什么都不成。很记得一位教授跟我说：这些方向你都可以试，我们唯一知道不行的是深度学习。试了很久，那是最保证职业生涯糟糕的路。我当时是好骗的大学新生，就当那是真的，去干别的了。大概 2005 年，很清楚 AI 当时不成。我碰巧进了创业，然后非常爱上它。我甚至不叫它职业弯路，因为回头看超级有用，成为创业投资人也很好。硅谷常见路径是创始人然后半退休当投资人。

我听你说过你会把这件事做完整个职业生涯。希望你能来很多次。我们不需要更多退休去投资的创始人。投资人已经太多了。

对。我不想变成那样。我想说的是，我能走反方向：先当投资人，再过来。这非常少见。我非常感恩，因为如果你真的作为投资人去研究和观察公司，会得到一套不可思议的学习和模式匹配，对管 OpenAI 超级有用。但这是相反的正常方向，非常罕见，我强烈推荐。

为什么有用？

如果你在管公司，过去五年十年只面对过有限几次类似的关键决策，看过什么成、什么不成。要做高风险战略转向，或以很乱的方式开掉一个高管，你只有自己那点有限经验。但作为投资人，你几乎整天都在看那些关键时刻。你得不到天天管公司的操盘练习，但你看过很多大的关键时刻。那个数据集的丰富程度太棒了。所以做决定时会在脑子里过：这家公司遇到类似的事发生了什么，这个创始人犯了这个错，那个创始人做对了。

What is confusing to me: you went from founder to investor back to founder. Why in 2015? What got you interested in artificial intelligence to begin with, that you’re saying this is such a non-consensus thing, people think I’m crazy, I’m going to do it anyway?

I had been interested in AI my whole life. I was a very nerdy kid. I was the kind of kid that spent Friday nights playing on my computer and watching sci-fi, reading sci-fi. And I always thought that AI would be the most amazing, kind of craziest thing. I never thought I’d actually get to work on it, but I always loved it. I even came to college sort of to study it. I worked in the AI lab the summer between my freshman and sophomore year, and nothing was working. Very memorably, a professor told me you could try all of these things, all these directions. The one thing we know doesn’t work is deep learning. We tried that for a long time. It’s the most guaranteed way to have a bad career. I was an impressionable freshman in college. I assumed that was true. So I pursued these other things. It was clear to me at the time — this is kind of 2005 — that AI was not working. And I happened to accidentally get into startups, but then very much fell in love with it. I wouldn’t even call it a career detour because it was super helpful looking back, and becoming a startup investor was great. The sort of normal career path in Silicon Valley is you’re a founder and then you sort of semi-retire, you become an investor.

I heard you say that you’re going to work on this for the rest of your career. Hopefully you’re going to come on the show multiple times. We don’t need more founders that retire and invest. We have too many investors.

Yeah. I don’t want to be that. What I was going to say is, the fact that I got to go in the other direction — I was an investor first, and then I came. It’s pretty unusual. Very unusual. And I’m super grateful for it because you get this unbelievable set of learnings and pattern matching if you really study and watch companies as an investor, that has been super helpful to me running OpenAI. But it’s the opposite of the normal direction. So it’s just a very rare thing and I strongly recommend it.

Why is it helpful?

If you’re running a company, you have faced some number of similar decisions in your past, some number of crux decisions, and you’ve seen what works and what doesn’t. If you have to make a high-stakes strategy shift or fire an executive in a really messy way, you have whatever your own limited previous experience was over the last five or ten years. But as an investor, you kind of watch all the crux moments. You don’t get the operating practice that you do just day in and day out running a company. But you’ve seen a lot of the big crux moments a lot. Kind of all day long you see those. So just the wealth of the data set that I have — that was awesome. So that plays in your head when you have a decision: this is what happened when this company had a similar thing, or I saw this founder make this mistake, or this founder got it really right.

David Senra

你研究过工业革命，我们也聊过都读过的传记。我朋友说我像一个温度被调高的、训练在历史上最伟大创业者上的 LLM。对你来说，那是一个大洞见，还是每个场景里这些人做过什么会自己拼起来？我读你博客很多年，你写得很短，编号列表。你从硅谷最好的投资人之一，不去过轻松的日子，去干人们觉得不可能、还会被嘲笑的最难的事。你当时住在圣路易斯。AI 为什么会吸引一个小孩？

我觉得它吸引每一个电脑宅。我不觉得这点我有多特殊，它只是感觉不可能。大多数人会说：那当然是最酷的东西，但完全不可能。我奇怪的地方是：好，那试试。所有人都会觉得它超棒、值得去追。

所以那是你小时候的性格：别人说你不能做，第一反应是抵抗？

不是抵抗，而是：你确定吗？为什么不？试试，看看会怎样。也许我可以。也许我们可以。我是很乐观的小孩。一件事看起来越不可能，我越感兴趣。能发明一种技术，让我们做其他所有事、以前所未有的方式赋能人——没有别的单一技术能这样——这对我来说天生就极其吸引。我想要那个东西，想能做其他所有事。我能记得的性格里还有：给人们更多力量、更多能力，很有意思。某种意义上这是技术的整条弧线，我肯定一直是技术宅。但 AI 是我能想象的那个最强版本。

小时候你觉得它能让你做什么？有什么是没有 AI 就做不成的？

很难分清多少是当时真想的，多少是现在的工作给记忆上了色。小时候我很迷机器人。学校有机器人社。当时的机器人可笑得很。夏令营有一只能用电脑在地板或桌上控制的小乌龟，我觉得酷极了。电脑控制的物理东西在动，我一直觉得惊人。现在我极度关心 AI 能如何推进科学发现。成年后的记忆里，我觉得小时候也觉得这很酷，但那不太可信，我猜这就是记忆被染色的例子。可我们现在能让 AI 去发现新物理、治病，它已经在为数学做的那些——我觉得这会是最重要的领域之一，甚至比自动化其他任务更重要，只是帮我们理解更多东西。我们前面聊过《无穷的开始》，以今天的位置重读，我会想：天，AI 真的会帮我们做这件事，理解一切，或尽可能多。我肯定也对那种星际迷航式的巨大繁荣和丰裕、以及 AI 能如何驱动它感兴趣。也许对科学感兴趣的记忆更真。我也爱科学：因为我们聪明，就能想办法理解世界、做预测、做没有这种深理解就做不到的事。这天生就很棒。

有意思的是，人类想从 AI 得到的东西一直很稳定。我刚重读 Claude Shannon 传记第二次。忘了他和 Alan Turing 1940 年代在贝尔实验室天天喝咖啡聊 AI。他们觉得不可避免，觉得 15 年后就会有比人聪明的计算机。你想让计算机做什么？解数学、写诗、治病。你一遍遍听到这些。

我读过那些人当时写的不少东西。他们不在这儿看到这些，我非常难过，因为他们对一切都那么对。我们终于到了这个时刻：AI 在解新颖的数学问题，在发现别的东西，它在写诗——写得好不好可以争，我会说还不太好——但它在这儿。我想他们会说：好，你们做成了，就是这个。那会非常酷。

奇怪的是所有人都说它永远做不了 X。音乐行业的人说它永远做不出伟大音乐。那它会做播客吗？当然会。它至少会做我们能做的一切，甚至现在就比我们做得更好。人类心理上有个很深的缺陷：它永远超不过我恰好活着的这个当下。但这里有个更有意思的问题：假如它真做出很棒的播客，两个 AI 的对话比你我更有意思，人们会在意吗？还是因为我们都对人着迷、那不是真人，他们会想要真人的那一版？

对我来说更有意思的是：这两个我可能喜欢或不喜欢的人，在进行一场对我有意思的对话。像我另一个播客那种只是「我在这本书里读到一个有意思的想法」，也许会被冲击。尤其对我们这些这事发生前出生的人。也许对你儿子不同。但我觉得人类永远会被人类吸引。我真心这么信。很多别的工作会面临显著转型，但关于人、关于人和人的连接、关于人喜欢别人的那些事，在后 AI 世界里会更值钱，而不是更不值钱。我可能不是谈这些的合适人选，因为尽管我的工作是数字的、向全世界广播，我深深渴望更模拟的生活。我读纸书。我不想开 Zoom。我喜欢实体的东西。

我也那样。我不读电子书。我不喜欢 Zoom 会。我喜欢在真实世界里和人在一起。肯定有一撮怪人，大概不少住在城市里，不喜欢人类、只想和电脑交流。但我觉得那是人类的极小一部分。这也是为什么我觉得整体上世界不会那么不同。即便有超级智能，人们仍然会在根本上被接线成在意别人、想待在别人身边、和别人互动。会有一些人迷上模型，觉得人类碍事，或者是需要对付的危险。对大多数人来说，人就是全部意义。我觉得这非常重要。当我们发现那样的人，要把他们点出来，确保他们不掌握权力。我当然同意。也许我最担心的 AI 两大风险，有一点互相拉扯。一个是失控：AI 以某种我们无法保证想要的控制的方式变得太强。另一个是权力太集中：一家公司、一个模型或一个人权力太大。这两件事的根本都是：无论哪件发生，都是非常反人类的立场。正确做法是说：我们要人深深掌控未来，要人被深深赋能。人是这一切的全部意义。我们不会坐在这儿因为不信任或不喜欢人，就把控制逐渐交给一个 AI 模型。把所有信任和世界上的决策权都交给这个模型，是非常厌世的说法。但我觉得世界上有些人认为那才是正确结局。

We were talking about that you studied the Industrial Revolution. We were talking about some great biographies. My friend Daniel says this about me: you’re like an LLM trained on history’s greatest entrepreneurs, but with the temperature turned up. Do you find that you have one big insight of all that, or for any given scenario you have what all these people did? I’ve read your blog for years. You’re a great writer, very succinct. I love numbered lists. You’re one of the best investors of all time in Silicon Valley. You could just be rich and not really have to work. Then you’re like, I’m going to do the hardest thing ever, the thing people think is impossible, I think you’re made fun of. You were living in St. Louis. Why would AI appeal to you back then?

I think it appealed to every kind of computer nerd. I don’t think that’s that unusual about me. It just felt impossible. I think most people would say, of course that’d be the coolest thing ever, but it’s totally impossible. I think the weird thing about me was, okay, let’s try. I think everybody thought it. Everybody would think it’s awesome and something to go for.

So wait, that was a personality trait of yours as a kid, that people told you that you couldn’t do something, like your initial response was resistance?

Not resistance, but like, are you sure? Why not? Let’s try. Let’s see what happens. Maybe I can. Maybe we can. I was a very optimistic kid. And also the more something seemed impossible, the more intrigued I was. The idea that we could invent a technology that would let us do everything else, that would just empower people in this way that no other single technology could — that always seemed innately, incredibly appealing to me. It’s like, I want that thing. I want to be able to do everything else. Another personality trait as long as I remember is: it is interesting to really give people a lot more power, a lot more ability. In some sense this is the whole arc of technology, and I was for sure always a technology nerd. But AI is the strongest version of that I can imagine.

What did you think that it would enable back then? When you were a kid, this seems like a cool technology, I want to do X, I can’t do X unless AI is invented.

It’s always hard to remember how much of this is the stuff that I actually thought at the time versus how much my current work has colored my memories of it. As a kid, I was very into robots. We had a robots club in my school. And the robots at the time were laughably bad. I even remember at summer camp we had this little turtle that you could control with a computer on the floor or on the table and thought that was just the coolest thing. There’s something about physical stuff moving controlled by a computer that I always thought was amazing. Now I am extremely interested in what AI can do to advance scientific discovery. In my memory as an adult, I think I thought that was cool as a kid too, but it feels just implausible, and I assume that’s an example of where the memories have gotten more colored. But now the fact that we can have AI go discover new physics and cure diseases, and what it’s already doing for math — I think this will be one of the most important areas, even more important than automation of other tasks that AI can do, just to help us understand more things. We were talking earlier about this book, The Beginning of Infinity, and rereading that book from today’s vantage point, I’m like, man, AI is really going to help us do this important thing of understanding everything, or as much as we can. I was definitely interested in the Star Trek version of huge prosperity and abundance and what AI could do to drive that. Maybe the memory of being interested in science is more real. I also loved science and just this idea that because we were smart, we could figure out how to understand the world and make predictions and do things that we couldn’t without this deep understanding. That seems innately awesome.

It’s interesting how consistent over time what humans want from AI is. Something you’re describing is very similar. I just reread the biography of Claude Shannon for the second time. I had forgotten that he and Alan Turing used to meet every day for coffee when they were both at Bell Labs, like 1940s, and they would just talk about AI. They thought it was inevitable, and they thought it was going to happen like 15 years from then. What would you want the computer to do? Solve math problems, write poetry, cure diseases. You hear this over and over again.

I have read a bunch of things that those guys wrote at the time, and I am so sad they are not here to see it, because they were so right about everything. We’re finally at the moment where AI is solving novel math problems, it is discovering other stuff, it is writing poetry. You can argue about how good or not — I would say not very good — but it is writing poetry. It’s here. I think they would have said, all right, you’ve done it, this is it, we’ve got it. And that would have been so cool.

This is the weird thing where everybody’s just like, oh it’ll never do X. I talk to people in the music industry: it’s never going to make great music. Then they’re like, well do you think it’s going to make a podcast? Of course it’s going to. It’s going to do everything that we can do, at least, I would say better than what we can do even right now. There’s a deep human psychological flaw there. But here is a more interesting question. Let’s say it does make a great podcast. Two AIs are having a more interesting conversation than you and I are. Do you think people will care? Or will they want the one with the real people because we’re all obsessed with people and the fact that it’s not real people?

For this, it’s more interesting. It’s like, oh, these two people that I maybe am predisposed to like or dislike are having a conversation that’s interesting to me. For strict reference, maybe my other podcast where I’m just saying, hey, this is an interesting idea I read in this book, that could maybe get disrupted. But especially for people that were born before this happened — maybe it’s different for your son. For me, I think humans are going to always be drawn to humans. I really deeply believe that. I think there’s a lot of other jobs that could face significant transition, but stuff that’s about people, stuff that’s about people’s connection to people and people liking other people, that stuff feels like it gets more valuable in the post-AI world, not less. I may be actually the wrong person to talk about this stuff because I kind of deeply desire, even though my entire work is a digital broadcast all over the world, I deeply desire more of an analog life. I like reading physical books. I didn’t want to get on Zoom. I like physical shit.

I’m like that too. I don’t read e-books. I don’t like Zoom meetings. I like to be with people in the real world. I definitely think there’s a subset of weirdos, and there’s probably a lot of them that live in the city, that don’t like humans and only want to communicate with computers. But I think that’s a tiny percentage of humanity. This is why I think the world is on the whole not going to be that different. Even with superintelligence, people are still going to be very fundamentally wired to care about other people, to want to be around other people, to interact with other people. There will be some people who just get obsessed with the models and just think humans are in the way, or a danger to be contended with. And for most people, it’ll be the whole point. I think it’s very important. And when we do find people like that, they should be called out and make sure they don’t acquire power. I certainly agree with that. Maybe the two big risks that I’m most worried about with AI, which are a little bit in tension. One is a loss of control, where AI somehow just becomes too powerful in a way that we can’t guarantee the control we want. And the other is power gets too centralized, where you have one company or model or person with too much power. And in both of these, the fundamental thing is: I think it’s a very anti-human position for either of these things to happen. The right approach is to say we want people deeply in control of the future. We want people deeply empowered. People are the whole point of this all. We are not going to sit here and gradually hand over control to an AI model because we don’t trust or like people. It’s a very misanthropic thing to say we’re going to just put all of our trust in this model and let it have all the power on decision-making over the world. But I think there are some people in the world who think that’s the right outcome.

Sam Altman

还有另一个版本：因为我们不信任人，所以必须限制谁能拿到这项技术、他们能怎么用，所有这些可怕的事都可能发生。出于对这些的恐惧，我们要把权力集中到少数公司手里，不让别人用。但我们会给他们一些好处。我对此的漫画是：AI 领域有些人实际上在说，我们会给世界治好所有病，会让东西变得很便宜，交换条件是人们放弃对自己未来的自主和影响力，以及权力；以安全为名，还有绝对猖獗的不平等。有人能拿到巨大的财富和权力，其他人只得到「还不错的一切」。这是糟糕的销售话术，非常反人类的销售话术，可不知怎么有人愿意这么说。

你觉得他们为什么愿意这么说？

我觉得是恐惧和权力。人们谈 AI 风险时，有很多人被那些风险的量级吓住，觉得必须保护世界，于是说：我们应该在这里用很多自由换安全，因为这和其他我们见过的风险不一样。但那也最终变成一种为很多追求权力的行为辩护的方式。

There’s like another version of this, which is because we don’t trust people, we have to limit who gets access to this technology and how they can use it and all of these terrible things could happen. And out of fear of those, we are going to concentrate power in the hands of a few companies and they’re going to, you know, we’re not going to let other people use this. But we’ll give them some benefits. My caricature of this is, I think there are some people in the AI field who effectively say, we’re going to give the world a cure to all disease and we’re going to make stuff really cheap in exchange for people giving up their autonomy and impact over the future and power, and also in the name of safety, and also just absolutely rampant inequality. There will be people that have access to huge amounts of wealth and power and other people just get a pretty good everything. And this is a terrible sales pitch. This is a very anti-human sales pitch that somehow people feel willing to make.

Why do you think they feel willing to make that?

I think it’s fear and power. I think when people talk about the risks of AI, there are a lot of people who are so nervous about the magnitude of those risks and get so taken by that and feel a need to protect the world from that, that they’re like, we should trade off a lot of liberty for safety here because this is unlike other risks we’ve seen. But then I think that also ends up kind of a way to justify a lot of power-seeking behavior.

David Senra

我读到 Claude Shannon 或 Alan Turing 说的，至少在我读过的书里，更多是乐观的：我们会发明让生活更好、能为我们做事的东西。

你完全对。回到 Shannon、Turing 那个时代，他们谈 AGI 会多么美妙、它会做的所有事。我们起步时，受到末日派很大压力。我同意末日派的部分是：这是强大的技术，每一级都该偏向安全、谨慎行事。我不同意的部分是：这是不可解的问题。回到 OpenAI 刚开始，我觉得会有两个被广泛持有的看法。第一，我们绝对造不出很像 AGI 的东西，当然也不是十年内。第二，即便造出来了，我们也肯定没法让它安全。如果你有一个在很多方面比很多最聪明的人更聪明的 AI，末日派会说世界那时肯定已经被毁了，对齐一定失败了。这些关于十年后会发生什么的立场被说得很满。我们造出了当时大多数人会说「很像 AGI」的东西，发生了很多好事，那些疯狂的坏预测没有发生。所以人们该更新对未来的预测。前面还有更高赌注的挑战要解决。但我们的方法——这也是我从创业学到的——做事的方式是把东西放到世界里，从真实客户拿反馈，看哪里坏、哪里不坏。那是做出好产品的方式，也是做出安全产品的方式。我们在 AI 安全上的进展，比起步时大多数人以为的多得多。

为什么？因为有十亿人每周在用你们的产品？

每次我们拿到新一级模型，就放到世界里。看什么成、什么不成，哪里需要我们放松护栏因为人们有好事想用，哪里有对齐失败、安全系统失败。ChatGPT 出来还不到四年。十亿人在用，用在敏感、重要的事上。能在这么短时间、这么强的技术上交付一个被广泛认为安全的东西——当然有问题——我觉得象牙塔里绝对做不到。我相信这就是你建造好的、安全的、鲁棒的、有用的技术和产品的方式。我觉得这是 Y Combinator 的伟大功课。对当时大多数 AI 安全的人来说，走到这一步还保有我们现在这种安全保证，会显得完全不可能。从这里会更难，但我不觉得你断开和现实的连接就能解决。

为什么从这里更难？因为世界上最聪明的人和最聪明的模型差不多一样聪明，而这个方向马上要翻转？

方向上我觉得对。模型能力强得不可思议，轨迹又这么陡，未知的未知也许相对不难，但从绝对角度看更难。我们得做一堆困难决定：什么时候推迟开发，什么时候说好，现在该接触现实了，或者再等久一点好好研究。我最近跟人聊，印在脑子里的是：FAA 让飞行变得极其安全。飞行表面上看极度危险，你登机却不怎么想。飞机在人类历史长河里并不老，一开始绝不是这样。他们有极其扎实的事故报告，极其清醒，从不试图挥手糊弄，想尽可能抽取信息。某种意义上，对任何新技术，那种方法都非常有效，也常常被低估。所以我们开始部署模型、说要把 ChatGPT 放到世界里时，我们知道模型不完美，会幻觉，还会做别的事。但我们也知道世界必须体验这项技术。我们必须学会如何让它安全。我们必须把力量交到人手上。我们不能只用这个来强加世界观，也不能坐在实验室里试图想遍所有影响——那反正不成，因为社会和模型会共同演化。我们必须作为这个共同产品一起做。然后做非常好的事故记录。出了问题就研究。发非常清楚的复盘。尽可能学。不仅改进自己的技术和产品，也试着把这些功课分享给其他做 AI 的人。到目前为止效果出奇地好。那是技术史上的好例子，也是创业的好例子。

When I read what Claude Shannon was saying or Alan Turing, at least in the books that I’ve read, it’s more of an optimistic, like, we’re going to invent things that make our lives better and can do things for us.

You are totally right that if you go back to the Claude Shannon, Alan Turing era, they talked about how wonderful AGI would be and all the things that it would do. And when we started, we really had a lot of pressure from the doomers. The part of the doomers that I agree with is that this is a powerful technology and we should err on the side of safety and we should act with caution at each level of technology. The part of the doomers that I don’t agree with is that it’s an unsolvable problem. If you go back to the beginning of OpenAI, I think there would have been two widely held opinions. Number one, not at all, and certainly not in 10 years, were we going to build something that was very AGI-like. And then, conditioned on if we did, we certainly were not going to be able to make it safe. If you had an AI that was smarter in many ways than a lot of the smartest people, most of the smartest people, then the doomers would say surely at that point the world would have been destroyed, the alignment thing would have failed. And there were just these very confidently held positions about what would have happened a decade on. We have built something that I think most people would say at the time would have said it’s very AGI-like, and a lot of good things have happened, and the kind of crazy bad predictions of the world have not happened. So I think that should update people’s predictions about the future. There are still higher-stakes challenges in front of us to solve, but our approach — this is another thing I learned from startups — of the way you do things is to put things out into the world, get feedback from real customers, see where they break, see where they don’t break. That is the way you make a good product. That is also the way you make a safe product. And we have made way more progress on AI safety than I think most people thought we would when we started.

Why? Because so many people, there’s a billion people using your products on a weekly basis?

Each time we get a new level of model, we put it out in the world. And we see what works, what doesn’t work, where people need us to relax the guardrails because they have good things they want to use it for. Where we have alignment failures, where we have safety systems failures. ChatGPT has only been out less than four years. A billion people use it, and sensitive, important stuff. And the fact that we can deliver something that is broadly considered safe — of course there’s issues with it — in that short of a time frame with such a powerful technology, I think there is no way we could have done that in an ivory tower. And this is how I believe you build good, safe, robust, useful technology and products. And I think it’s a great learning of Y Combinator. And it would have seemed to most of the AI safety people totally impossible to get to this stage and still have the level of safety guarantees we have. Now, I do think it gets harder from here, but I don’t think you’re going to solve it by disconnecting yourself from reality.

Why does it get harder from here? Because we’re about as smart as the smartest people in the world are about as smart as the smartest models in the world, and that’s going to flip right now?

Yeah, directionally I think that’s right. I think that the models are just so incredibly capable and improving on such a steep trajectory that the unknown unknowns, maybe they don’t get harder relatively, but from an absolute perspective they seem harder. And I think we’ll have to make a bunch of difficult decisions about when we delay development, when we sort of say, okay, you know what, let’s have contact with reality now, or let’s wait longer to really study this more. I was talking to someone recently and something that stuck in my mind is that the FAA has helped make flying incredibly safe. Flying on the surface seems like this extremely dangerous thing. And you probably get on an airplane without giving it much thought. Airplanes are not that old in the long trajectory of human history. And this was certainly not the case at the beginning of airplanes. They have extremely robust accident reporting, extremely clear-eyed. They never try to hand-wave over something. They want to extract as much information as possible. And in some sense, I think with any new technology, an approach like that works very well and is often underappreciated. So when we started deploying our models, when we said we’re going to put ChatGPT out in the world, we know the model’s imperfect. We know it hallucinates. We know it can do these other things. But we also know that the world has got to experience this technology. We’ve got to learn how to make it safe. And we’ve got to put the power in people’s hands. We cannot just use this to impose our worldview. We cannot use this to go sit in a lab and try to think through all the impacts, which won’t work anyway, because society and the models are going to co-evolve. We have to all do this together as this joint product. And then we’ll do very good accident recording. We will study when something goes wrong. We will put out a very clear post-mortem. We will learn as much as we can. We will not only improve our own technology and products, we’ll try to share those learnings with other people building AI. I think that’s worked surprisingly well so far. And that was good examples from history of technology, good examples from startups.

David Senra

一件让人晕的事：人人都在用 AI，人人又好像恨 AI。到底怎么回事？

人们总是害怕快速的社会经济变化。工业革命也不是人人都暖。社会对急剧变化有一点内建惯性和怀疑，大概是好事，能在动荡时帮一把，也是我不想硬刚的人类生物学。另一面是：很多造 AI 的人会说有 25% 概率毁灭世界，但我们还是要往前冲，否则坏人先做成；或者说明年 50% 的工作会没。这个领域没有把好处讲清楚，也没有讲清楚下行怎么缓解。即便有人提 UBI、工作变成可选项，也很少有人谈：为什么人必须在世界上有更多权力和个人自由，而不是更少。影响自己的未来、集体设计社会往哪走、随之而来的自主，对大多数人极重要。AI 领域的人自己感觉得到，却不怎么承认这对别人同样重要。

回到前面那套销售话术的漫画：亲爱的草民，我们赐你们治癌、物质财富和好娱乐，你们别抱怨，未来的决定交给我们这些仁慈独裁者。不好。作为创业的学生，我最信的就是赋能人去做新东西、坚持信念、有自由去办公司、发明技术、追想法。即便多数人永远不想办大公司，他们也懂这件事有多重要。一听到有人暗示 AI 之后这种空间会变少、少数人会有权力、但他们会做伟大决定并保证大家安全，人们会觉得非常可怕。

即便多数人不想办特别大的公司，很多人想办小公司。那一直很难，需要不少特权、运气和资源。我们马上会看到有史以来最大的一波人去办小生意。AI 正在赋能这件事。不知为什么这个领域——包括我们——谈得不够，也没做出足够加速它的产品。但会看到更多。

Tobi 跟我说：你我是同一行，都在制造更多创业者。他做基础设施，我做教育和激励。AI 作为行业，对外讲清楚这件事做得特别差。《Intel Trinity》里 Bob Noyce、Andy Grove、Gordon Moore 意识到集成电路会吓到客户，三个人停下手头的事，去给客户、投资人、整个国家上课，一度开的课比当地社区学院整个课表还多。为什么 AI 没人这么干？

没有借口，我们应该做得更多。我们试过一些版本，还没做对。

One thing that has to be disorienting: everybody uses AI, everybody hates AI. What the hell is going on there?

People are always afraid of rapid socioeconomic change. People did not have universally warm and fuzzy feelings toward the Industrial Revolution. Some built-in inertia and skepticism of rapid change is probably a good feature of human society, and a feature of human biology I believe in never trying to fight too hard. A lot of people building AI have been off saying there’s a 25% chance we’re going to destroy the world and yet we’re going to race ahead because otherwise those bad guys will do it first, or 50% of the jobs are going to go away next year. We have not, as a field, done a very good job of explaining the benefits and how the downsides can be mitigated. Even when people have answers like universal basic income or work will be optional, there has been very little discussion of how and why it’s important that people have more power and personal freedom, not less. The ability to influence their own future and collectively design where society goes, and the autonomy that comes with that, matters a lot. People in the AI field feel it for themselves, but they don’t spend much time acknowledging how important that is to other people.

To go back to that caricature of the sales pitch: dear peasants, we will bequeath upon you a cure for cancer and material wealth and great entertainment. You stop complaining and we’ll make all the decisions about the future and just trust us, we’ll be benevolent dictators. Not good. As a student of what made this economic miracle work, really empowering people to do new stuff, push on what they believe in, and have the freedom to create companies and invent technology — there is nothing I believe in more strongly. Even if people never want to start a big company, they understand how important that is. When you hear people saying there is going to be less of that with AI because a small number of people are going to have the power, but they’re going to make great decisions and keep everybody safe, I think that’s very scary to them.

Even if maybe most people don’t want to start really big companies, a lot of people want to start smaller companies. That has required a fair amount of privilege, luck, and resources. We are about to see the greatest boom in people starting smaller businesses that we have ever seen. AI is empowering that. For some reason the field, including us, has not talked about that enough, even though we see the signs, and we have not built enough products to accelerate that. But I think we’re going to see a lot more of that.

Tobi said you and me are in the same business: we’re both trying to create more entrepreneurs. He’s building infrastructure; I’m building educational and inspirational podcasts. Out of any new industry, AI is doing the worst job I’ve probably ever seen at getting out and talking. In The Intel Trinity, Bob Noyce, Andy Grove, and Gordon Moore realized the technology would scare customers, so those three stopped what they were doing and educated potential customers, investors, the entire country. At one time they were putting on more classes than the local community college had in their catalog. Why isn’t anyone in AI doing that?

No excuse that we should be doing more. I think we’ve tried versions of this. We haven’t gotten it quite right.

David Senra

我听你在别的播客说：你甚至在考虑该让 AI 看到电脑上多少东西。

最新一代模型，我不想说已经够聪明——我们永远该追求更聪明——但已经相当聪明。我现在更受限于 AI 对我有多少有用上下文。我想让它尽可能多地知道，好帮我。我不打算读完内部 Slack 的每一条，也不打算读完客户讲 ChatGPT 哪里成、哪里败的每一个故事。研究论文我大概也能多读一些，但很耗心力。我想要一个不断试图帮我的智能体，能看、能理解我没时间、没精力覆盖的上下文，在我做大决定时给出好建议。我们一直正确地聚焦模型智力，产品侧还没想够：给模型比任何人都多的上下文，帮人做大决定，意味着什么。我们正站在一种非常不同的共事方式的门口。聪明人很多，但没人能在几秒内准确用上几万页上下文。AI 能，这将是全新的、不可思议的补充。

你读过这么多传记，总会有模模糊糊记得的东西，如果能精确调出某段轶事，就能在那一刻帮到某个创业者。我从 2018 年起把每本书的笔记和高亮都放进一个库，后来又把 Founders 全部逐字稿加进去。我每天用它做节目。Claude Shannon 那期问：Bob Noyce 怎么说的，Rockefeller 怎么做的。书是我读的、笔记是我记的，七年前的我不记得了。

太棒了。这就是我的意思。

I heard you on another podcast saying you’re even considering how much you should let AI see every single thing that’s on your computer.

With the latest generation of models, I don’t want to say they feel smart enough because we should always aspire for them to get smarter, but they’re pretty smart. I feel more limited at this point by the amount of useful context the AI has on me. I want the AI to know as much as it can to help me. I’m not going to read every post on our internal Slack. I’m not going to read every story a customer has about where ChatGPT works or fails them. I probably could read more research papers than I do, but it takes a lot of mental energy. I would love an AI agent that is constantly trying to be helpful, that can look at and understand more context than I have time or energy for, and can bring that to bear and give me good advice when I have to make a decision. We’ve focused correctly so much on model intelligence that on the product side we have not yet thought enough about what it means to give a model more context than any person could have and help advise that person on their big decisions. We are just on the precipice of a very different way of working with AI. There are plenty of very smart people, but no one that can read tens of thousands of pages of context in some small number of seconds and use that accurately. AI can. That’s going to be very new and an incredible supplement.

You’ve read all these biographies. There are times you vaguely remember something that, if you could recall a specific anecdote, would really help an entrepreneur in that moment. Since 2018 I’ve kept every note and highlight from every book in a database, then added all the transcripts of Founders. I use it every day to make every episode. Working on Claude Shannon, I ask what Bob Noyce said about this, or what Rockefeller did. I read the book. I took the note. I don’t remember it because it was seven years ago.

That’s awesome. This is what I mean. That is so cool.

David Senra

你的主业是算力和研究。你们必须自己做产品吗？Codex 从哪来？收入从哪来？你现在有多少产品？

我觉得我们更该是平台公司，而不是产品公司。当然会做产品。我们刚把 ChatGPT 和 Codex 合在一起。以前是 ChatGPT、Codex 和 API。Codex 这名字起得不幸，它不只是编程，几乎能做任何工作，把人搞糊涂了。大多数人要的是一个通向个人或公司 AGI 的单一界面，能帮他们做需要的事；再加上一套 API，想在上面建什么就建什么。这就是我们该给世界的平台。在成本—性能曲线的每一个点上，我们都会卖最好的 AI。你要高端去发现科学，可以；你要便宜的去做大量不需要天才智力的活，我们也有。人们想用很多 AI，要低成本、要快、要好用、要有上下文、要顺。一个直接界面，一套 API，两者还会越来越合。我不觉得我们该去建每一个产品品类，不该跟所有客户竞争，不该试图吞掉整个经济。我希望一亿家新公司和八十亿人用它，用出各种新方式。

学到这一点，你杀掉了哪些好点子？

杀掉好点子、把它们牺牲给伟大点子，大概是创业者最难学的课。再怎么以为自己会做，谁都做得很糟。我自己也很糟。去年我们砍了 Sora：产品不错、好玩、酷，但吃很多算力，不如把算力放进 Codex 重要。我们砍了网页浏览器 Atlas：我觉得它是最好的浏览器，但不如把人才放别处重要。算力、人、资源都有限。我们想清楚：面向知识工作、最终面向科学的通用智能，是我们能做的最重要的事。为了生成智能而上游的所有事——自己的芯片、自己的数据中心、基础设施软件、训练模型——都很重要。然后把 AI 当服务提供出去，让人用在工作、科学发现、个人效率上。灵活的通用平台，别的少做。

Your main focus is getting more compute and then research. Do you have to build your own products? You built Codex. Where’s all the revenue coming from? How many products do you have now?

I think we should be more of a platform company than a product company. We will build products, of course. We just merged ChatGPT and Codex together. We used to have ChatGPT, Codex, and the API. Codex was unfortunately named: it was not just coding; it could do any kind of work, which confused people. What most people want is a single interface to their own personal or their company’s AGI that can help them with whatever they need, and then an API to build anything they want on top of it. That is the platform we should offer. At every point on the cost-performance curve we will sell great AI. You want really high-end AI to discover science. You want inexpensive AI to do a massive volume of work that doesn’t require genius-level intelligence. People want to use a lot of AI, at low cost, fast, working well, with their context, smooth. One direct interface to the product, one API, and those eventually come more and more together. I don’t think we should go build every product category, compete with all our customers, or subsume the entire economy. I think we should offer this platform and try to have 100 million new businesses and 8 billion people use it in all kinds of new ways.

What mistakes did you make to have to learn that? You had to kill some good ideas to go after the great with full intensity.

Killing good ideas to go after the great ones is kind of the hardest lesson for any entrepreneur. It sucks, and no matter how much you think you’re going to do it, everyone seems to do terrible at this. I’m terrible at this. Last year we killed Sora, which was a good product and fun and cool, but used a lot of compute and was not as important as Codex where we put the compute. We killed our web browser Atlas, which I think was a great product. I think it was the best web browser, but not as important as somewhere else we could put that talent. In a world of limited compute, limited people, limited resources, we said the general intelligence for knowledge work and eventually for science is the most important thing we can do. Anything upstream of generating the intelligence — building our own chip, building our own data centers, writing good infrastructure software, training models — that’s all really important. Then let’s just offer this AI as a service and get people to use it for work, for scientific discovery, to be more productive. A flexible general platform, and not a lot of other things.

David Senra

做复杂工作的人都需要有人帮他整理思路。Charlie Munger 有猩猩理论：一个相对聪明的人可以对着猩猩把问题讲完，人离开时已经更好了。Munger 对 Buffett 就是这个角色。谁在你生活里扮演这个角色？

大概三类。一是一直在这儿的很多研究员，大家一起走过，有一套共享语言、直觉和标准。技术会往哪走、现在是什么形状，公司外我复制不出这个。商业和世界的问题，很长一段时间里 Paul Graham 和 Peter Thiel 是我学得最多的两个人，现在卡住一个非常非显然的问题时，我还是找他们。找了很久，没人有同样那种超级非线性的思考。如果 LLM 是预测下一个词，这两个人是我最预测不了下一个词的人。那是超有价值的技能。你卡住了，他们会说：那些选项都不行，这儿有一件现在看起来完全显然、你却没想到的事。

这更像是给你自己思考的提示，还是明确的「去做 X」？Peter 有什么能分享的例子？你们砍 Atlas、算力受限、必须聚焦，听起来很像他会说的。

经常是具体的事。ChatGPT 刚上线时很怪：大家不知道用来干什么，涨得很快，但感觉不稳定，像低价值增长，人只是好奇地跟它聊。公司里很多人说得另找方向，这不可持续。大概上线后两个月，我们在谈另外五六件可以转去做的事。他说：除了它在涨——这既罕见又好——去做别的是显然的错。它的力量就是 Google 那个文本框的力量：一个你可以打任何东西、它就会做对的文本框。它对不上当时硅谷那套——要信息流、要网络效应、要锁定；那时还没有记忆，所以大家慌。人们追了二十年 Google 的商业模式，这是第一件冒出来的。空文本框对 Google 有效，为什么不加倍押这个？它在涨，非常灵活，除了不合当下硅谷智慧，所有迹象都在。我们就死磕 ChatGPT，很好。那是简单的天才。

Paul Graham 呢？

很多 YC 创始人的梗：你去 office hours，他会晃着手指说 you know what you should do。后面那句话有时很棒，有时很糟，重要的是那种创造力和开放的风景，以及试试很多东西。我们聊过迭代部署的精神：把 V1 早早发出去，因为客户反馈会让它好得多。上线 ChatGPT 前我甚至没问他该不该发，但我知道他会说什么。我知道它还早、还让人尴尬，正确的事就是拿出去放到人面前。关于「产品让你尴尬时就该发」这件事，我知道他会说什么。那不一定是最有价值的战术建议，但排得上。回头看这么多 YC 创始人，行动快、能迭代，和成功的相关程度让我吃惊。

Anybody engaged in complicated work needs somebody to help organize their thoughts. Charlie Munger has this orangutan theory: a relatively smart human could sit down with an orangutan, tell it all his problems, and leave better off. Munger played this role for Buffett. Who plays this role in your life?

Kind of three categories. One, a lot of the researchers that have been here forever. We’ve all been through it together, and we’ve developed this shared language, intuition, standards. I have not been able to replicate that with anybody outside the company when it comes to the shape of what’s happening and where the technology is likely to go. In terms of business and the world, for a long time Paul Graham and Peter Thiel have been two of the people I have learned the most from, and they are still the two I go to if I really have a very non-obvious problem I’m stuck on. I have not found anyone else after a lot of looking with the same ability to think in a super nonlinear way. If what LLMs do is predict what word comes next, those are two of the people I can predict the least. That is a super valuable skill. You go with, I feel really stuck, and someone can tell you none of those options are good. Here’s this thing that now seems totally obvious and correct that you didn’t think of.

Is this more a prompt for your own thinking as opposed to explicit advice, do X? What would be an example from Peter? Killing these good ideas, compute constrained, we have to focus — that sounds like him.

It’s often a specific thing. After we launched ChatGPT it was this weird thing: people didn’t really know what to use it for, it was growing super fast, but it felt unstable, almost like low-value growth. People were using it because they were interested in talking to it. A lot of people in the company were like, we got to figure out something else, this is not sustainable value. Maybe two months after ChatGPT launched, we were talking about a list of five or six other things we could focus on instead. He was like, it’s an obvious mistake to do anything besides the fact that it’s growing, which is rare and great. The power of this is the power of the Google text box: a text box you can type anything into and it does the right thing. It doesn’t match the current Silicon Valley system of feeds and network effects and lock-in — we had none of those, and that’s why everyone was worried; this was before we had memory. People have been chasing the Google business model for 20 years and this is the first thing that’s come up. Clearly the empty text box worked for Google, so why don’t you just double down on that? It’s growing, it’s very flexible, and it has all of the signs other than it doesn’t fit the current Silicon Valley wisdom. So we went super hard on ChatGPT, and it was great. That was an example of very important simple genius.

What about some advice from Paul Graham?

This is a meme for many YC founders: you’d go see him for office hours and he would shake his finger: you know what you should do. Sometimes the thing that came after that was great, sometimes terrible. The important thing was a kind of creativity and open landscape, let’s try a lot of things. We talked about the spirit of iterative deployment: ship a V1 early. It could be much better later because of feedback from customers. I don’t even think I asked him before we launched ChatGPT if he thought we should launch it, but I knew what he would say. I knew it was still early, still embarrassing, and the right thing was to get it out in front of people. In terms of launch when you’re embarrassed of the product, I know what he’s going to say. I won’t say the most valuable piece of tactical YC advice, but it’s been up there. I’m astonished looking back at all my data points of YC founders how much moving fast and being iterative correlates with success.

David Senra

这对话里 YC 出现太多次了。你经历过、管过、一直有关联。能展开讲讲它为什么这么深？

有一支乐队卖的专辑不多，却影响了后来所有音乐人。也许是 Velvet Underground。用这个比喻对 YC 不公平，因为按传统指标、创造的市值，YC 会是最有价值的科技公司之一。但 YC 对过去二十年科技、创业的影响程度，我觉得只被理解了一部分。OpenAI 就是例子：不只是产品怎么发出去，也包括研究实验室怎么管。即便没走过 YC 的这一代大科技公司负责人，很多人会讲类似的故事。

那是一套操作系统——做这五件事——还是一种办公司的哲学？

两件大事。有一点操作系统，但更重要的是办公司的哲学：迭代部署、技术人掌权、愿意在公司各层押有能量和野心、经验可能更少的年轻人。再就是它对整个生态的改变。如果把时钟拨回 2004，只把技术推到 2016，创业生态、当创始人意味着什么、资本怎么流、公司怎么管都不改，我不觉得 OpenAI 会可能。YC 给创始人更多杠杆、年轻技术创始人能融到很多钱、没有漂亮履历也能做野心勃勃的事——没有这些，OpenAI 不会出现。

You’ve mentioned YC way too many times in this conversation. How impactful going through YC and then running YC, being affiliated with them, has clearly been on your life. Can you expound on this?

There is some band that wasn’t that successful, they didn’t sell that many albums, but they influenced all of the musicians that came after. I think it might be the Velvet Underground. It’s not fair to talk about YC this way because measured by traditional metrics and market cap created, YC would be one of the handful of most valuable tech companies. But the degree to which YC totally influenced everything that has happened in the last 20 years of the tech industry and startups, I think is only sort of understood. OpenAI is an example of that. Not just how we have shipped our products, but the philosophy of how we run our research lab. If you go talk to many of the other people running this generation of large tech companies, they would tell you similar stories, even if they didn’t go through YC.

Is it an operating system that YC is giving you, do these five things, or is it more a philosophy of building companies?

Two major things. There is some of the operating system of what to do. But I think it was the philosophy of how to run companies: iterative deployment, technical people in charge, and being willing to bet on young people with a lot of energy and ambition but maybe less experience throughout all levels of a company. And then the related change to the whole ecosystem. If we ran the clock back to 2004 and projected technology forward to 2016, but not anything else about the shape of the startup ecosystem, what it meant to be an entrepreneur, how capital flowed, how you run companies — I do not think OpenAI would have been possible. The changes YC induced: more leverage going to founders, young technical founders having the ability to raise lots of capital, the ability to work on ambitious things without a very proven resume. I don’t think OpenAI would have been possible.

David Senra

这和你从创始人变成投资人再变回创始人有关吗？

那些好处也有。我不会说自己真的从创始人到投资人再到创始人，因为第一次当创始人并没有那么成功。你还是会从失败里学到一些，但我觉得从成功里学到的多得多。

停。这个要展开。

有一部伟大的俄国小说，我很尴尬叫不出名字。大概是《安娜·卡列尼娜》。开头是：不幸的家庭各有各的不幸，幸福的家庭都一样。回头看自己失败的功课，我学到的是关于 grit、决心、以及不该做什么的泛泛东西。大多数事都不成，不成的原因很多，正确的因果更难拼。真做成了——Y Combinator 哪部分真有效、OpenAI 哪部分真有效——把那些功课往前用，比用「没做成的反面功课」有用得多。每种数据点当然都该学。但按我自己的经验，从成功里学到的课非常好，我该更多地用它们；从失败里学到的要么相当泛、我其实已经知道，要么会挡到别的。我觉得对很多人一般也是这样。

我们知道该做什么，缺的是不断提醒。有人说我那个 Founders 播客是创业者的教堂：同一类人格在历史上反复出现，同一本书、同一个故事要一遍遍回去。你写过 YC 结束时还在重复同一件事。他们离开教堂就不做了。不是课本身，是不断提醒这件事重要。

我极度同意。但被提醒的最好是正面的那些：多跟用户聊、更早发产品、拿更多反馈、招人标准更高、招得更快。

你有一条很棒的推：跳过会议、跳过饭局，就是做产品、卖产品。你没在做、没在卖，其实就这些。Doug Leone 跟我讲过 Nubank 创始人，从 VC associate 出去做成最成功的公司之一，他一辈子只见过这一个。你从成功里学，是因为那十年半你看过也许一万家公司？

完全是。很多人会去找那些真做成的课。就像你说的，同一件事在不同行业里一遍遍发生。但你必须被提醒很多次，而且它不风光。Peter Thiel 说：这已经在成，你为什么还做别的。一定有过这种时刻：这建议我给过创始人一百万次，然后发现自己没在用。

Is this all tied to the fact that you think there was a benefit in you going from founder to investor for a long period of time back to founder?

There are all those benefits too. I wouldn’t say I really went from founder to investor to founder because the first time I was a founder didn’t really work out that well. You still start a company. You learn some lessons from failure, but I think you learn way more from success.

Hold on. We’re not moving on from that. You got to say more about that.

There’s some great Russian novel. I’m very embarrassed not to know this. I think it’s Anna Karenina. It starts with all unhappy families are unhappy in their own way, all happy families are the same. When I look at the lessons of where I have failed, I learned something generic about grit and determination and something not to do. Most things don’t work. There are a lot of reasons why things don’t work, and it’s harder to put together the correct causation. When I’ve had something really work — what parts of Y Combinator really worked, what parts of OpenAI really worked — trying to apply those lessons going forward has been much more helpful than trying to apply the anti-lessons of what didn’t work. You should of course learn from every data point. But in my own experience, the lessons I learned from success were very good, and I should have applied those more. The lessons I learned from failure were either fairly generic and I kind of already knew them, or got in the way of something else. I think this is generally true for a lot of people.

We already kind of know what we should do. It’s the constant reminder. The best description of my other podcast Founders is: it’s church for entrepreneurs. The same personality type has appeared throughout history. We go back to the same books, the same stories over and over again. You even said when YC ended, you’re repeating the same thing, telling it to them all the time. Then they leave the church and they stop doing the same stuff. It’s not even the lessons. It’s the constant reminder that this is important.

I extremely strongly agree with that. But I think it is better to be reminded of the thing: talk to your users more, ship products earlier, get more feedback, hold a higher bar for who you recruit and who you hire, and more quickly.

You have one of the greatest tweets: skip the conferences, the dinners, everything else, just make the product and sell the product. If you’re not making it and you’re not selling it, that’s all you actually have to do. I just talked to Doug Leone, and he talked about the founder of Nubank, who was an associate VC and then found one of the most successful companies. He dedicated his life to this. That’s the only one. Are you taking their successes as instructive because you were exposed to maybe 10,000 different companies in that decade or decade and a half?

Totally. People often try to go look for those lessons of the things that really worked. As you’ve said, it’s kind of the same thing over and over again, done in different industries. But you have to be reminded of it a lot, and it’s unglamorous. Peter Thiel saying no, this is working, why are you doing anything else but the thing that is working. There’s got to be examples where you’ve given this advice to other founders a million times, and then you catch yourself not even applying your own advice.

Sam Altman

完全有。很多例子。也该问：新东西是什么，哪件事你以前没有建议可套。OpenAI 和我以前能模式匹配的任何东西真正不同的是：从创办到第一个产品，四年半。这和 YC 的建议完全相反。管研究团队在很多方面像挑选和辅导创始人，但在没有客户外部信号的世界里怎么管——四年半不该有产品，对创业来说通常是灾难性建议——非常难。我们试过各种办法，用「研究是否真的在成」去替换「客户是否真的喜欢产品」。有一件成了：Dota 2 那些日子，我们用强化学习打电子游戏，挂了一块排行榜，不同想法的表现大家看得见，客观、真实，人们想往上爬。我们得发明各种办法去模拟终端用户。那是我完全没有模式可套的新问题。

你怎么走过来的？

我们问过一批待过历史上伟大研究实验室的人。OpenAI 起步时，硅谷人人都把研究实验室当虚荣项目，包括我。贝尔实验室和 Xerox PARC 鼎盛时期的书非常流行。讨论很多，活着还记得怎么做的人不多。我们跟 Alan Kay 聊了很多，也跟另外几个人聊。有些建议很好，有些不太能翻译到当下。

你们也没有贝尔实验室那种巨型印钞垄断。本田创始人也觉得研发必须独立出去、分开持股。

我们没有。回想早期，主导记忆是我在试图融钱、又融不到。那么多力气，那么沮丧。真希望当时有那种现金机器。对 OpenAI 最清楚的记忆之一：我 2015 年进来，第一天是 2016 年新年后。大约 11、12 个人出现在 Greg Brockman 的公寓，大概周一或周二早上 9 点、9 点半。就算是 1 月 4 日。大家走进来，很多兴奋，像开学第一天。很快有人环顾房间：那我们现在干什么？有人说该弄块白板。Greg 派人去找。白板来了。再环顾：现在该干什么。能量塌下来。没人知道该做什么。这不是产品创业那种「做这个产品、去跟客户聊」。我们说要做 AI。也许该写论文。好，写论文。也许该想想想法。好，想想想法。每个人都有「我完全不知道自己在干什么」的时刻。那是我的一次：我们刚把这事启动了，没人知道下一步。

于是我们做会做的事。很多东西不成，后来找到一种做研究押注、再评估的节奏。远谈不上完美，但找到了一条可以往前走的梯度。我们搞清楚怎么给非常聪明的人他们需要的资源，怎么确保不完全在荒野里迷路。若干年里，多半是混乱的跌撞，我们最终做出了大部分重大发现。从无监督情感神经元那篇论文变成 GPT-1，再变成后来的 GPT。缩放定律的工作给了我们信心：不只是去买算力，还有怎么把模型做大。还有很多别的。这个过程里我们学到排行榜有用，也学到给研究者很想打动的重要人物做外部演示有不可思议的力量。也学到一堆不成的，比如假截止日期。

从公寓里 12 个人、连白板都没有，到十年后十亿人在用。一定非常迷失方向。

非常奇怪的经历。

Totally. I’ll give many examples of that. It’s also instructive: what was the new thing, what didn’t you have the advice for. The thing that was really different about OpenAI than anything I had pattern matching for is it was four and a half years from when we started the company to when we launched our first product. The opposite of YC advice. There were all these ways in which managing a research team was similar to selecting and advising founders. Learning how you manage through this part of the world where you don’t have the external signal from customers — and you’re doing what would normally be catastrophic startup advice, not shipping a product for four and a half years — that was very difficult. We tried all of these things about how we replaced the signal of do customers actually like the product with is our research actually working. One of the things that worked is during the Dota 2 days, when we were trying to use RL to beat people at a video game, we put up a leaderboard and people could just see how different ideas were performing. That was objective and real. People wanted to go up that. We had to try all of these things to basically simulate end users. That was a totally interesting new problem I had no pattern matching for.

How did you work your way through that? What was your thinking?

We asked a bunch of people who had been at great research labs of the past. OpenAI started at a time when everybody in Silicon Valley, as their vanity project, including me, wanted to start a research lab. There were all these books about the heyday of Bell Labs or Xerox PARC. They were very popular. There was a huge amount of discussion, but not a ton of people that had, in living memory, how to actually do it. We talked a lot to Alan Kay. We talked to a handful of other people, and we got some advice about what made a really good research lab. Some of it was really good. Some of it didn’t translate as well to the current moment.

You also didn’t have this giant monopolistic profit-printing machine. Bell Labs was spun out independently. The founder of Honda thought research and development had to actually be separate, spun out of the company with separate ownership, just like Bell Labs.

We did not have that. When I think back to those early days, I mostly feel like I was trying and failing to raise money. That’s my dominant memory of the early days of OpenAI. So much effort. So frustrating. I wish we had some sort of cash machine like that. One of my clearest memories of all of OpenAI: I came in 2015, but the first day was right after New Year’s in 2016. Twelve of us or 11 of us showed up at Greg Brockman’s apartment, like 9, 9:30 on a Monday or Tuesday morning. Let’s say it’s January 4th. Everybody walks in. A lot of excitement. It feels like the first day of school. Then very quickly people look around the room: well, what do we do now? Someone says, okay, we should get a whiteboard. Greg gets someone to go off and find a whiteboard. Whiteboard comes. Look around again. What are we supposed to do now. You just feel the energy in the room collapse. None of us know what to do. It was not like building a product startup. It was not, let’s build this product and let’s talk to customers. We said we want to make AI. Maybe we should write some papers. Okay, let’s write some papers. Maybe we should think about some ideas. Okay, let’s think about some ideas. Everybody’s got their moments of I have no idea what I’m doing. That was one of mine. We have just launched this thing. None of us have any idea what we’re going to do.

So we did what we know how to do. Eventually we figured out a lot of things didn’t work. Eventually we figured out a kind of rhythm for making and then evaluating research bets. Far from perfect, but we did find a gradient that we could progress along. We figured out how to get the resources that very smart people needed and how to make sure we were not completely getting lost in the wilderness. Over some number of years, mostly chaotic stumbling, we eventually made most of the big discoveries. What started as the unsupervised sentiment neuron paper turned into GPT-1 and then eventually GPT whatever. The scaling laws work that gave us the confidence not only to buy the compute, but the understanding about how to scale up our models, came together. And many other things too. Through this process we learned things like that idea of leaderboards that worked. We also learned the incredible power of external demos for an eminent person that the researchers really wanted to impress. And then we learned a bunch of things that didn’t work, like fake deadlines.

That has to be so disorienting. You had 12 people in an apartment, don’t even have a whiteboard, don’t know what to do. Fast forward a decade. You have a billion people using it.

Very strange experience.

David Senra

你写日记吗？

第一个孩子出生后，我下班回家摇他睡觉，得找话讲，就把一天、我们在挣扎什么、我担心什么讲给他听。挺有意思，有一天对他也会有意思。于是开始每个周日给他写信：先讲，再写下来。大概只写了八封左右。

几个孩子？继续写那些信。创始人往往 70 岁才写自传，回望时希望能再来一遍，时间已经冲走太多。他们都重复：真希望当时写了日记。即便不为现在，Michael Moritz 的《The Little Kingdom》写的是苹果头六年，书结束时 Steve 还没被踢出苹果。你会想要这个。写给孩子的心态没法藏：我真的在意孩子会怎么看我，这件事这周感觉不好，下周得换做法。

两个。我没读过那本书。也许我会再找到一种办法做这件事。

你会做的。Sam，谢谢你抽出时间。太棒了。

谢谢。也很感激。

Do you keep a journal?

When my kid was born, my first kid, I would get home at the end of the day and be rocking him to sleep and just talk. I needed to come up with stuff to talk about, so I would just tell him about my day and what we were struggling with and what I was worried about and what was happening. It was kind of fun for me, and someday it’ll be interesting for him to have this. So I started writing him, like every Sunday I would write him a letter. I would talk just to talk, and then I would write it down. I only ever did like eight of them or something.

How many kids do you have? Keep writing the letters. Most of the time founders don’t write autobiographies when they’re 40. They write them when they’re 70, looking back, wishing they could do it again, and so much has been lost. They all repeat this: I wish I journaled. Even if you don’t, have a book written, even if it’s internal. Have you ever read The Little Kingdom by Michael Moritz? First six-year history of Apple. That book ends, Steve hasn’t even been kicked out of Apple yet. You’re going to want this. The mindset of writing to your kid: you really can’t hide behind anything. I really care what my kid’s going to think about me. This thing happened, didn’t feel great about it, better do it differently next week.

Two. I’d never read it. Maybe I’ll find some way to do it again.

You’re going to do it. Sam, thanks for taking the time. This is awesome.

Thank you. Appreciate it.