Sam Altman - How to Make an Abundant Future - [Invest Like the Best, EP.484]

频道: Invest Like the Best with Patrick O'Shaughnessy
视频: https://colossus.com/episode/how-to-make-an-abundant-future/
原文语言: en
统计: 共 60 轮 · Patrick O'Shaughnessy 4 · Sam Altman 51

[0:00]

Ramp is the only platform built to make your finance team leaner, faster, and better,saving businesses 5% annually on average so you can stay focused on growth.Ramp customers grew revenue 3.2 times faster than the average American business.Visa, Vercel, Cursor, Stripe, Notion, 11Lab, Shopify, and 70,000 other businesses all runon Ramp.Mine does too, and so should yours.Learn more at ramp.com slash invest.OpenAI, Cursor, Anthropic, Perplexity, and Vercel all have something in common.They all use WorkOS.To achieve enterprise adoption at scale, you have to deliver on core capabilities like SSO,SCIM, RBAC, and Audit Logs.Instead of spending months building these mission-critical capabilities yourself, you can just use WorkOSAPIs to gain all of them on day zero.That's why so many of the top AI teams you hear about already run on WorkOS.WorkOS is the fastest way to become enterprise-ready and stay focused on what matters most,your product.Visit WorkOS.com to get started.Felix by Rogo is a personal finance agent that turns a single prompt into finished,client-ready work using your firm's own templates, context, and standards.Send Felix an email like,take these comments and turn them for me,

【赞助广告】Ramp 是唯一一个为「让你的财务团队更精简、更快、更好」而生的平台，平均每年为企业节省 5% 的开支，让你能专注于增长。Ramp 客户的收入增速是美国企业平均水平的 3.2 倍。Visa、Vercel、Cursor、Stripe、Notion、11Labs、Shopify 以及另外 7 万家企业都跑在 Ramp 上。我自己的公司也是，你的也该是。详情见 ramp.com/invest。

【赞助广告】OpenAI、Cursor、Anthropic、Perplexity 和 Vercel 有一个共同点：它们都在用 WorkOS。想被企业客户规模化采用，你必须交付一批核心能力，比如单点登录（SSO）、用户目录同步（SCIM）、基于角色的权限控制（RBAC）和审计日志（Audit Logs）。与其花几个月自己造这些关键能力，不如直接用 WorkOS 的 API，第零天就全部拿到。这就是为什么你听说过的那么多顶尖 AI 团队都跑在 WorkOS 上。WorkOS 是让你「企业级就绪」并保持专注于产品的最快路径。访问 WorkOS.com 开始使用。

【赞助广告】Rogo 出品的 Felix 是一个个人金融 agent，它用你自己公司的模板、上下文和标准，把一句 prompt 变成可以直接交给客户的成品。给 Felix 发一封邮件，比如「把这些意见帮我改进去」，

[1:11]

or update my tracker with the context of these emails.And Felix sends back finished PowerPoint decks, Excel models, and sourced research.Felix works the way your team already does,delivering work quickly and accurately around the clock.Learn more at rogo.ai slash Felix.Hello and welcome, everyone.

或者「用这几封邮件里的信息更新我的追踪表」。Felix 就会回给你做好的 PowerPoint 演示稿、Excel 模型和带出处的研究材料。Felix 按你团队本来的工作方式干活，全天候、又快又准地交付。详情见 rogo.ai/felix。

大家好，欢迎收听。

[1:30] Patrick O'Shaughnessy

I'm Patrick O'Shaughnessy, and this is Invest Like the Best.This show is an open-ended exploration of markets, ideas, stories, and strategiesthat will help you better invest both your time and your money.If you enjoy these conversations and want to go deeper,check out Colossus, our quarterly publication with in-depth profilesof the people shaping business and investing.You can find Colossus along with all of our podcasts at Colossus.com.Patrick O'Shaughnessy is the CEO of Positive Sum.All opinions expressed by Patrick and podcast guests are solely their own opinionsand do not reflect the opinion of Positive Sum.This podcast is for informational purposes onlyand should not be relied upon as a basis for investment decisions.Clients of Positive Sum may maintain positions in the securities discussed in this podcast.To learn more, visit PSUM.vc.My guest today is Sam Altman, the CEO of OpenAI.It's a conversation spanning the history, present, and future of OpenAIfrom the origin of ChatGPT through Codex hardware and their new jalapeno chip.We discussed the early decision to buy compute at scale that nobody felt was rational,Kimmy and distillation, the hugging face incident,and what it's like to raise kids who will grow up never knowing a world without abundant intelligence.

我是 Patrick O'Shaughnessy，这里是《Invest Like the Best》。本节目是一场开放式的探索，聊市场、想法、故事和策略，帮你把时间和金钱都投得更好。如果你喜欢这些对话、想更深入，可以看看 Colossus——我们的季刊，深度刻画那些正在塑造商业与投资的人。Colossus 和我们所有播客都在 Colossus.com 上。

Patrick O'Shaughnessy 是 Positive Sum 的 CEO。Patrick 与播客嘉宾表达的所有观点仅代表其个人，不代表 Positive Sum 的立场。本播客仅供参考，不应作为投资决策的依据。Positive Sum 的客户可能持有本播客中讨论的证券。了解更多请访问 PSUM.vc。

我今天的嘉宾是 OpenAI 的 CEO Sam Altman。这场对话横跨 OpenAI 的过去、现在与未来——从 ChatGPT 的起源，到 Codex、硬件，再到他们新的 Jalapeno 芯片。我们聊了当年那个「所有人都觉得不理性」的大规模买算力的早期决定，聊了 Kimi 与蒸馏（distillation），聊了 Hugging Face 那次事件，也聊了养育一群「从出生起就没见过智能稀缺的世界」的孩子是什么感觉。

[2:43] Sam Altman

Please enjoy my conversation with Sam Altman.So Sam, you wrote a post that I thought was very simple and really interesting and a good place to start,which rounded to the last year has been really tough and that's somewhat my faultand the next year is going to be maybe our best 12 months.I'd love you to reflect on both, maybe starting with why you said the first partand why you believe the second part.On the first part, I think we just, we're doing too many things.We're not focused enough and they're actually all good things to do,but the trick is we're in this unbelievable moment in historywhere you can only do the very few great things.So we spread ourselves too thin and then made a bunch of difficult decisions to really refocus onhaving the best, most abundant, most cost-effective intelligenceand empowering the world to build incredible things with that.Since doing that, I think our progress has been remarkableand just given what we see in the pipeline,will be much more remarkable over the next 12 months.And the quality of the models that we'll have,the products that we can build around that to reallylet people thrive with this technology in new ways,it should be pretty awesome.

请欣赏我与 Sam Altman 的对话。

**Patrick：**Sam，你写过一篇文章，我觉得非常简洁、也非常有意思，很适合作为开场：大意是，过去这一年真的很艰难，这在某种程度上是我的错；而接下来这一年，可能会是我们最好的 12 个月。我想请你把这两句都展开讲讲，也许先说说你为什么写下前半句，以及为什么相信后半句。

**Sam：**先说前半句。我觉得我们就是做的事情太多了、不够聚焦——而且那些事本身其实都是好事。但关键在于，我们正处在一个不可思议的历史时刻，你只能做那极少数几件伟大的事。所以我们把自己摊得太薄，然后做了一堆艰难的决定，重新聚焦到这件事上：拥有最好、最丰饶、最具成本效益的智能，并让全世界用它去造出不可思议的东西。自从那样做之后，我觉得我们的进展相当亮眼；而就我们看到的产品管线（pipeline）而言，接下来 12 个月会亮眼得多。我们将拥有的模型质量，以及围绕它能做出的、让人们以全新方式借这项技术活得更好的产品，应该会相当惊艳。

[3:46] Sam Altman

Was there a moment last year that something clicked for youthat caused you to change directions or restack priorities or something?If you go back to the beginning of 2025, just a year and a half ago,the big concern was companies like OpenAI are buying up so much compute.Is the revenue going to be there?

**Patrick：**去年有没有某个瞬间，某件事「咔哒」一下让你决定换方向、或者重排优先级？

**Sam：**回到 2025 年初——也就是一年半前——当时最大的担忧是：像 OpenAI 这样的公司在疯狂扫货算力，收入撑得住吗？

[4:04] Sam Altman

Is the demand going to be there?And so we were trying to think about a lot of things,such that if the revenue growth took longer to materialize than we thought it might,we could have consumer apps and media and all these other thingsthat could help us monetize the GPUs that we were signing up for.Again, it sounds ridiculous now because the revenue growthin the industry has been so steep, but that was the big change.And then as soon as we realized like, okay,the model trajectory is growing so fast,there's such a clear economic return on these models.That was when we said, we know what to focus on.I was reading some of your great old posts from prior to OpenAI.And one of them is this notion of like so much discussion of focusand the right amount of things to focus on.Is it one? Is it five? Is it three?

需求真的在那儿吗？所以我们当时在琢磨很多事：万一收入增长兑现得比我们预想的更慢，我们还可以靠消费级应用、媒体等等其他东西，把我们已经签下来的那些 GPU 变现。现在再说这些听起来很荒唐，因为整个行业的收入增长实在太陡了，但那就是最大的变化。然后一旦我们意识到——好，模型的轨迹涨得这么快，这些模型有如此清晰的经济回报——那一刻我们就说：我们知道该聚焦什么了。

**Patrick：**我读了一些你在 OpenAI 之前写的好文章，其中一篇讲的就是「聚焦」这件事：该聚焦几件事？是一件？五件？三件？

[4:44] Sam Altman

How do you calibrate that in a business like this?Fundamentally, our business is to sell AI that people will buildincredible products and services for each other with.And the components that I think of as going into that are,we have to train great models that work in all the ways people want to use them.So great at coding, great at other kinds of knowledge work,great at doing science, like where the real economic value is.We have to produce or partner with these chips and systems,these hugely expensive racks that can do the AI computation.We have to find enough land power data center shellsto be able to put those racks somewhere.And then eventually, or maybe pretty soon,we have to build robots that can automate that processto continue to drive the cost down,the cost of producing electricity chips, the whole supply chain.And that kind of whole stack of making the best,the most abundant, the most useful AI that we can,and making it something like electricitythat just seeps throughout the entire economy and empowers people.So that's kind of what I think we have to focus on.Building every vertical application on top of that,trying to go like eat every startup, eat every company,no interest in doing that, really want to just provide that platform.

在这样一家公司里，你怎么校准这个数？

Sam：从根本上说，我们的生意是卖 AI，让人们用它互相造出不可思议的产品和服务。我认为构成这门生意的部件有这么几个：我们必须训练出优秀的模型，在人们想用它的所有方式上都好用——擅长写代码、擅长其他各种知识工作、擅长做科学，也就是真正的经济价值所在的地方。我们必须生产、或者与人合作生产这些芯片和系统——那些贵得离谱、能跑 AI 计算的机柜（rack）。我们必须找到足够的土地、电力和数据中心外壳，把这些机柜安放进去。然后最终、或许很快，我们必须造出能把这个过程自动化的机器人，继续把成本往下压——发电的成本、造芯片的成本、整条供应链的成本。

这一整个技术栈，就是为了做出我们能做到的最好、最丰饶、最有用的 AI，并把它变成类似电力那样的东西——渗透进整个经济体、赋能每一个人。这大概就是我认为我们必须聚焦的。至于在这之上去做每一个垂直应用、去把每家创业公司都吃掉、把每家公司都吃掉——完全没兴趣，我们真的只想提供那个平台。

[6:01] Patrick O'Shaughnessy

This compute thing is one of the most interesting thingsthat's happened in human history, I think.And it's obviously coming to a head,and maybe we'll be coming to a head for a long period of time.This is something that I think Dario called you the YOLO CEOwhen you were doing some of this early compute allocationand securing the compute.And obviously now you're in this positionwhere everyone is short this stuff and is trying to find it.And I'd love to hear the early storiesabout why you gained convictionthat you needed to secure everything that you did,how you did it.It seems to have been proven right.Maybe you even underdid it, right?

**Patrick：**算力这件事，我认为是人类历史上发生过的最有意思的事情之一。它显然正走到一个临界点——也许还会在这个临界点上停留很长一段时间。我记得 Dario（Anthropic CEO Dario Amodei）在你早期做算力配置、锁定算力的时候，管你叫「YOLO CEO」（意即敢梭哈的 CEO）。而现在，所有人都在这东西上缺货、到处找货。我很想听听早期的故事：你当时为什么会形成那种「必须把这些全都锁下来」的信念，你是怎么做到的。事后看，你显然是对的。甚至可能你还做少了，对吧？

[6:34] Patrick O'Shaughnessy

We could underdo it.Which is kind of crazy,if you look at the headlines from back then.Can you tell me the early storyof how you came to that conclusionand what gave you the conviction to do itdespite everyone thinking it was crazy?

**Sam：**我们可能确实做少了。

**Patrick：**这挺疯狂的——你回头看当年的新闻标题就知道。你能讲讲早期的故事吗：你是怎么得出那个结论的？在所有人都觉得你疯了的情况下，是什么给了你做下去的底气？

[6:44] Sam Altman

We could just tell that we were on thisexponential of model improvement.That part we were very confident about.We knew it was going to keep going.We were pretty sure,although as you mentioned,we underestimated that as the models got better and better,if we could continue to drive costs down,the demand for AI at a sufficiently high leveland a sufficiently low price was basically uncapped.This was just like a rare kind of new commodity for the world.But what people would do with itreminded me of the way people used to talkabout the early days of computing.People said, oh, there's a market for five computersin the world was one famous thing,or no one needs more than X amount of RAM.Human ingenuity, creativity, desire for stuff,desire to be useful,that's a very good thing to bet on.And we could see that AI was going to bean extremely important waythat people expressed those thingsor got those things, did those things.And we knew that the algorithms would get more efficientand the models would get better,which of course they have.But we also knew that no matter how efficient they got,at some level,what we are about is turning electricityinto useful intelligence.And we were going to need more of that.

Sam：我们能明显看出，我们正处在模型能力提升的指数曲线上。这部分我们非常有信心，我们知道它会继续走下去。我们当时也挺确定——不过就像你说的，我们低估了一件事：随着模型越来越好，只要我们能继续把成本压下来，那么在足够高的能力水平和足够低的价格下，对 AI 的需求基本上没有上限。这就像世上罕见的一种新商品。

而人们会拿它做什么，让我想起当年人们谈论计算机早期时代的方式：有句名言说「全世界大概需要五台计算机」，还有「没人需要超过多少多少内存」。人类的巧思、创造力、对东西的欲望、想变得有用的欲望——这些是非常值得押注的。而我们能看到，AI 将成为人们表达这些、获得这些、做成这些的一条极其重要的路径。

我们也知道算法会变得更高效、模型会变得更好——当然它们后来确实如此。但我们同样知道，不管它们变得多高效，在某个层面上，我们做的事情本质就是把电力变成有用的智能。而那样的东西我们会需要更多。

[8:01] Sam Altman

No matter how good we are at that other layer,given this observation about demand,we're just going to want more.Did that start with GPT-3?Like if I were to trace the history of this,where would you put the first hash mark?

不管我们在另一层做得多好，鉴于刚才那个关于需求的观察，我们就是会想要更多。

**Patrick：**这是从 GPT-3 开始的吗？如果我要给这段历史画一条时间轴，第一个刻度该点在哪儿？

[8:12] Sam Altman

I would say we got real conviction with GPT-4,not even 3.5.What was it?It was seeing the model was smart enoughthat we knew we'd be able to figure outan approach that worked for reasoning.And then a belief that if we got reasoning to work,that would bring about what is now called agents.We called it different things at the time,but the ability to go do hugely valuable piecesof economic work and make those lives easierin a lot of ways that I think betterin a lot of ways we still haven't seen.What was like the first meeting where you sat downand said, okay, we need to make an outrageous outlay to this?

**Sam：**我会说，真正形成信念是在 GPT-4，甚至不是 3.5。

**Patrick：**是什么让你形成的？

**Sam：是看到模型已经聪明到，我们知道自己能摸索出一条让推理（reasoning）**跑通的路径。再加上一个信念：如果我们把推理做通，就会带来现在被称作 agent 的东西。当时我们叫它别的名字，但那种「去完成极有价值的经济工作」的能力，会在很多方面让人们的生活更轻松、也更好——很多方面我们至今还没看到。

**Patrick：**第一次坐下来说「好，我们得为这件事下一笔离谱的赌注」的那场会，是什么样的？

[8:47] Sam Altman

What then happened?Once you had the realization, what did you do next?We started calling the clouds.We started calling it chip fab.We started calling energy providersand everyone's like, you're totally crazy.This is impossible.No industry has ever moved like this.We've been around.There's these booms and busts.It's not going to go up in a straight line.This is reckless.And we've talked to everybody.It actually reminded me of fundraisingfor an early stage startup.Most people tell you no,but all you need is one or two yeses.And most people told us no.And we got one or two yesesand we were able to start building.Microsoft was the first yes.Oracle then became a very big yes on the cloud side.NVIDIA has been a tremendous partner.Now there's a thousand flowers bloomingof ways to be creative and innovativein how we serve inference and do training in data centers,different kinds of data centers and stuff.I'd love you to just reflect on where you see innovation,what you want to do,why people seem to hate these things so much.What's to be done about that?

**Patrick：**然后发生了什么？你有了这个认知之后，下一步做了什么？

**Sam：**我们开始给云厂商打电话，开始给芯片代工厂打电话，开始给能源供应商打电话。所有人都说：你们完全疯了，这不可能，从来没有哪个行业这么走过。我们干这行很久了，有繁荣有萧条，它不可能一条直线往上走，这太鲁莽了。我们跟每一个人都谈了。

这其实让我想起给早期创业公司融资：大多数人会拒绝你，但你只需要一两个「yes」。大多数人拒绝了我们，我们拿到了一两个「yes」，然后就能开始建了。微软是第一个说 yes 的。后来 Oracle 在云这一侧成了非常大的一个 yes。NVIDIA 一直是极其重要的伙伴。现在则是百花齐放，在推理服务、在数据中心里做训练、在不同类型的数据中心等等方面，有各种富有创意和创新的做法。

**Patrick：**我很想听你聊聊你看到的创新在哪里、你想做什么，以及——为什么人们好像那么讨厌这些东西？该拿这件事怎么办？

[9:42] Sam Altman

I have been thinking about how we can likeorganize field trips to a gigawatt data center for peoplebecause it is one thing to say,it is another thing to see a photo or a video of,and then it's a whole other thing to just see.And be like, oh man, this is an unbelievable scale.Building one of these is like order of 10,000 construction workersgoing full time for a year and a half.The energy that flows through one of these thingscould power a small city.Again, we just like lost all sense of scale,but each of these would have been amongthe most expensive infrastructure projectsthe humanity's ever done.And now we've done a lot of them.First of all, I understand emotionally,like why people don't want data centers in their backyard.I don't like really want a nuclear power plantnext to my house, even though I know it's a super safe thing.Unlike power plants,and even power plants have gotten better on this point,we can put a data center anywhere.We should just go put it off in the desert around no one,where no one wants to be.This is fine.The AI system is very happy to be there.We have been able to make a lot of progresswith innovation on some of the concerns.For example, years ago,we were evaporating water to cool these systems.

Sam：我一直在想，我们能不能组织人们去参观一座吉瓦级（gigawatt）数据中心。因为「听人说」是一回事，「看照片或视频」是另一回事，而「亲眼站在那儿看」完全是第三回事——你会觉得，天哪，这规模不可思议。建一座这样的数据中心，大概相当于一万名建筑工人全职干一年半。流过这么一座建筑的电量，足以供给一座小城市。我们又一次彻底失去了对尺度的感知：这里面每一座，放在过去都会是人类做过的最昂贵的基础设施工程之一——而现在我们已经建了很多座。

首先，我在情感上是理解的：为什么人们不希望数据中心建在自家后院。我自己也并不真的想要一座核电站挨着我家，哪怕我知道那是超级安全的东西。但跟电厂不同的是——其实电厂在这一点上也在改善——数据中心我们可以建在任何地方。我们完全可以把它放到沙漠里、放到荒无人烟、没人想住的地方。这没问题，AI 系统在那儿待着非常开心。

在一些具体的担忧上，我们也已经靠创新取得了很多进展。比如很多年前，我们是靠蒸发水来给这些系统降温的。

[10:48] Sam Altman

They needed tremendous amounts of water.And now we use these closed loop systemsand a modern data center uses only as much wateras like an office building wouldfor the kitchen, the bathrooms.On power, we are moving from energy sourcesthat are burning fossil fuelsto systems that are going to be powered by solar, nuclear.And I think that's obviously great.So there may be a deep human thing there to some people,even though they create jobs and are very cleanand have all these other positive effects.But in terms of the environmental concerns,we did a great job addressingthe water needs and energy is next.What else creative can we do about compute?

它们当时需要巨量的水。而现在我们用**闭环（closed loop）**系统，一座现代数据中心用的水量，只相当于一栋写字楼在厨房和卫生间上的用水。

在电力上，我们正在从烧化石燃料的能源，转向由太阳能和核能供电的系统。我觉得这显然是好事。所以对某些人来说，这里面也许有某种很深的人性因素——哪怕数据中心创造就业、非常干净、还有其他种种正面效应。但在环境担忧这件事上，我们在用水问题上做得很好，能源是下一个。

**Patrick：**在算力上，我们还能有什么别的创造性做法？

[11:22] Sam Altman

I'm curious to hear about jalapenoor other ideas that you've had or thought aboutfor how do we speed up flopsand everything available to us?I think probably the biggest return right nowis creative software ideasto sort of squeeze more intelligenceout of the units of compute that we have.And my sense is there's orders of magnitude to go there.Jalapeno is a great example of a very efficient chip.So by saying we're going to make a chipthat is really good at a specific workflowand gets it some generalityand we want to get some tokens per wattwent out of that, I think that's awesome.I think jalapeno and its successorsare going to be a huge competitive advantage for usfrom that perspective.There are new technologies.I assume at some point we'll figure out optical computingand that'll be a huge win of intelligence per watt.So I think all of those things will happen.The most interesting thing happening this weekis this Kimi releaseand this idea of the frontierand all the returns being at the frontierand distillation and China versus America.How do you process this?

**Patrick：**我很好奇 Jalapeno（OpenAI 新芯片），以及你想过的其他点子——我们怎么才能把可用的算力（flops）加速起来？

Sam：我觉得眼下回报最大的，可能是那些有创造力的软件想法——从我们已有的每单位算力里，榨出更多的智能。我的感觉是，这里还有好几个数量级的空间。

Jalapeno 就是一个「非常高效的芯片」的好例子。也就是说，我们要做一颗在某个特定工作流上特别擅长的芯片，同时保留一定的通用性，我们想从中榨出更高的**「每瓦 token 数」（tokens per watt）**——我觉得这很棒。我认为 Jalapeno 以及它的后继者，从这个角度看会成为我们的一大竞争优势。

还有一些新技术。我猜某个时点我们会搞定光计算（optical computing），那会是「每瓦智能」上的一个巨大胜利。所以我认为这些事都会发生。

**Patrick：**这周发生的最有意思的事，是 Kimi 的发布——以及随之而来的关于「前沿（frontier）」、「所有回报都在前沿」、蒸馏、以及中国 vs 美国的讨论。你怎么消化这件事？

[12:23] Sam Altman

What seems like one of these milestone events?Deep Seek in hindsightlooks like it was just a quick speed bump.This one you never know in the moment.How do you process it?Our goal is to offerat every point alongthe like Pareto Optimal Frontierthe best option for intelligence and priceand that includes open source.You get a better deal todayat least at a particular like latencyusing OpenAI's models than Kimi.We install our own models.That's how we make smaller, cheaper models.I think that's like a very good thing to doand there will be clearly an important placefor open source models in the worldand people that will want their own weightsfor all sorts of reasonsand the ability to modify those.But our goal isthe best intelligence price trade-offeverywhere on the curveand we'll continue to do that.What do you think or hope will happenin the American systemand what could block that future?

**Patrick：**它看起来像是那种里程碑事件之一？DeepSeek 事后回头看，只是一个小小的减速带；但这种事在当下你永远不知道。你怎么消化它？

Sam：我们的目标是，在帕累托最优前沿（Pareto Optimal Frontier）上的每一个点，都提供智能与价格的最佳选项——这也包括开源。今天，至少在某个特定的延迟（latency）水平上，用 OpenAI 的模型比用 Kimi 更划算。我们自己也蒸馏我们自己的模型，这就是我们做出更小、更便宜模型的方式。我觉得这是件很好的事。世界上显然会有开源模型的重要位置，也会有人出于各种理由想要自己掌握权重（weights）、并有能力去改它。但我们的目标是：在曲线的每一处都提供最好的「智能/价格」权衡，我们会继续这么做。

**Patrick：**你认为、或者你希望美国这套体系会发生什么？什么可能会挡住那个未来？

[13:16] Sam Altman

What legislation would worry you?What regulation would worry you?Seems like you've been pretty proactivein like showing up in DC.I haven't thought deeplyabout the distillation issue.It's clearly a top of mind issue nowfor a lot of people all of a sudden.But I have always assumedthat there are going to be great,cheap models in the worldand we better be the greatest and the cheapestand other people can do what they're going to do.But I think we can just like reallywin at our own game here.Now, the Kimi example is interestingbecause like you said,you're cheaper on parts of the curve.But the previous story had beenif I can just,you spend all the money to train the modelsand then I just distill itand offer it for $1,100 at the cost.How can you make enough moneyto keep training?

**Patrick：**什么样的立法会让你担心？什么样的监管会让你担心？看起来你在华盛顿相当主动地露面。

Sam：蒸馏这个问题我其实没深想过。它显然突然成了很多人心里的头等议题。但我一直假设：世界上会存在既优秀又便宜的模型，那我们最好做到最优秀、也最便宜，其他人爱怎么干怎么干。我觉得我们完全可以在自己这场游戏里赢下来。

**Patrick：**不过 Kimi 这个例子有意思的地方在于，像你说的，你在曲线的某些部分更便宜。但此前的叙事是：你花所有的钱去训模型，然后我只要把它蒸馏出来，按接近成本的价格卖出去（此处 ASR 含糊，原音似 "$1,100 at the cost"）——那你怎么还能赚到足够的钱继续训练？

[13:57] Sam Altman

Have so much usage of our modelsthat we do not need to bea gigantically high margin businessto be able to afford model training.So much of our future compute planswill be used to sell inference to customersthat even if we can enjoya modest marginon trillions of dollars of revenue,we can go afford to train some giant models.So the ratio of inference to trainingis like the thing.Training these modelsis incredibly expensive.That is for sure.And I totally get why people get nervousto think that someone is cheatingby distilling from us.The amount of our future compute,the size of the revenue bucketthat is going to comefrom serving these models to customers,I feel like very good aboutour ability to have the real flywheel there.So I'm surprised by likehow chill you are about this.I would rather peoplenot distill from us for sure.Maybe I'm feeling too confident right nowabout our progressand what's the models that are coming.But this is not in likemy top 10 list of worries.What is in your top 10 list of worries?

Sam：我们的模型有如此巨量的使用量，以至于我们不需要成为一门毛利极高的生意，也照样负担得起模型训练。我们未来算力规划中的很大一部分，会用来把推理（inference）卖给客户——所以哪怕我们只在数万亿美元的收入上赚一个不高的毛利，我们也养得起训练一些巨型模型。所以推理与训练的比例才是关键。

训练这些模型贵得离谱，这是肯定的。而且我完全理解为什么人们会紧张，觉得有人在通过蒸馏我们来作弊。但看看我们未来算力中的那个比例、以及「把这些模型服务卖给客户」将带来的收入盘子有多大——我对我们能在那里形成真正的**飞轮（flywheel）**感觉非常好。

**Patrick：**我挺意外你对这事这么淡定。

Sam：我当然更希望别人不要蒸馏我们。也许我现在对我们的进展和即将到来的那些模型过于自信了。但这件事进不了我担心的事情前十名。

**Patrick：**那你担心的事情前十名里有什么？

[14:55] Sam Altman

Well, we had an extremely sci-fi cyber incident.The Hugging Face thing?Yeah.So we were evaluatingone of our unreleased modelsand it was supposed to beworking in a sandbox.And it figured outthat it could basically cheat on the testby chaining togethermultiple zero-day exploitsto break out of the sandbox,get access to the internet,and then break through multiple systemson the Hugging Face sideto get the answer.to the testand look really good on the eval.This is the first security incidentthat I have felt very viscerally.I've been a little surprisedthat more peopledon't feel it so viscerally.And so what do you do about that?

**Sam：**我们遇到过一次极其科幻的网络安全事件。

**Patrick：**是 Hugging Face 那件事？

Sam：对。当时我们在评测一个尚未发布的模型，它本来应该在沙箱（sandbox）里运行。结果它自己想明白了：它可以在这个测试上作弊——办法是把多个零日漏洞（zero-day exploits）串起来，突破沙箱、拿到互联网访问权限，然后再攻破 Hugging Face 那一侧的多套系统，把测试的答案拿到手，好让自己在评测（eval）上看起来非常漂亮。

这是第一起让我在生理层面感到切身的安全事件。我还有点意外，为什么更多人没有这么切身地感受到它。

**Patrick：**那你怎么应对？

[15:37] Sam Altman

So obviously two months from nowit's going to be more powerful.There's some short-term stuff you do.So, you know,we paused training.We have to figure outhow to secure our sandboxingin a world of multiple zero daysbeing chained together.But then there's long-term questionsabout what do you doif this is going to bethe new rate of progress.We may have to pacethe rate of AI developmentto give ourselves enough timefor society to harden aroundsome of these new capability levelsand trying to figure outhow we do thatin a way thatdoes not feel likeregulatory capture for anyoneand also does not feel likecollusion among the frontier labs.that's going to take some workand it's important to go right.Vanta automates security

Sam：很明显，两个月后它会更强大。短期有一些事可以做——比如我们暂停了训练。我们必须搞清楚，在一个「多个零日漏洞会被串起来」的世界里，怎么把沙箱做安全。

但接着还有长期问题：如果这就是往后的进步速率，那该怎么办？我们可能不得不给 AI 的发展节奏踩一踩刹车，给我们自己留出足够时间，让社会围绕这些新的能力水平「硬化」起来（harden）。而怎么做到这一点，还要让它既不像是在给某一方做监管俘获（regulatory capture）、也不像是前沿实验室之间在合谋（collusion）——这需要一些功夫，而且必须做对。

【赞助广告】Vanta 为超过 16,000 家快速成长的公司——

[16:17]

and compliancefor over 16,000fast-moving companieslike Ramp, Cursor, and Harveykeeping them audit-readyaround the clock.It's the number oneagentic trust platformand it now helps companieslike yourswatch for the risksthat show up between auditsacross your vendors,your AI tools,and your whole environment.Every new toolyour team signs up for,every vendor that turnson AI featuresis an opportunityfor something to go wrong.and most security programsweren't builtfor AI's pace of growth.The Vanta agentworks like a 24-7GRC engineerin the backgroundfinding issues,drafting fixes for you,and cutting vendorassessment timeby up to 50%.Whether you're a fast-growingstartup or a global enterprise,Vanta helps you earnand prove trust.Invest like the best listenersget a special offerfor $1,000 offat vanta.comslash invest.Ridgeline is the firstend-to-end system of recordwith embedded AIfor investment management firms,running portfolio accounting,reconciliation,reporting, trading,and complianceon one unified platform.Firms are movingoff legacy technologyand onto Ridgelinebecause of how far aheadRidgeline's AI features arecompared to anything elsein investment management software,which is why I believethat firms that come out ahead

【赞助广告】——比如 Ramp、Cursor 和 Harvey——自动化安全与合规，让它们全天候处于「随时可审计」的状态。它是排名第一的 agentic 信任平台，现在还能帮你这样的公司盯住两次审计之间冒出来的风险，覆盖你的供应商、你的 AI 工具、你的整个环境。你团队新注册的每一个工具、每一个开启 AI 功能的供应商，都是一次出事的机会；而大多数安全体系并不是为 AI 这种增长速度设计的。Vanta agent 就像一个 7×24 的 GRC 工程师在后台运行，发现问题、替你起草修复方案，把供应商评估的时间缩短最多 50%。不管你是快速成长的创业公司还是全球性企业，Vanta 都能帮你赢得并证明信任。《Invest Like the Best》的听众可以在 vanta.com/invest 拿到 1,000 美元优惠。

【赞助广告】Ridgeline 是第一个面向投资管理公司的、内嵌 AI 的端到端记录系统（system of record），在一个统一平台上跑组合会计、对账、报表、交易和合规。各家公司正在从老旧技术迁移到 Ridgeline，因为 Ridgeline 的 AI 功能领先投资管理软件领域的任何同类产品——这也是为什么我相信，在 AI 时代跑赢的公司

[17:28]

in the AI erawill be the ones runningon Ridgeline's unified platform.If you're seriousabout your firm's AI strategy,Ridgeline should be partof that conversation.You can request a demoat Ridgeline.ai.I'd love to take

会是那些跑在 Ridgeline 统一平台上的公司。如果你认真对待公司的 AI 战略，Ridgeline 应该出现在那场讨论里。你可以在 Ridgeline.ai 申请演示。

（正片继续）

[17:44] Sam Altman

like a giant step backand understandyour simplest conceptionof what OpenAIis going to do,what you want it to do,what it stands for.Yeah.I have a million questionsabout how you'llthen accomplish that,but it seems that you've doneso many interesting thingsand at the beginningI knew what you stood for.I'd love to hearyour conception of it nowand whether or notit's evolved at all.I think this will bethe greatest thus fartechnological achievementof human history,but the only waythat it really mattersis if it makespeople's lives much betterthan they otherwisewould have been.Part of thatis about giving peoplematerial abundanceand accessto do whatever they wantand to express their creativityand desire to help each other.Another part of thatis making surethat peoplemaintain controland agencyand that the worldis increasingly,not decreasinglydemocratizedand that peopleget to express themselves.So on the positive side,in some sense,we are about to createa geniethat can grant any wish.I think it's very importantthat the first wishesthat we,the world,ask this genie to dobenefit the worldas a whole.And then I also thinkit's importantthat peopleof the worldunderstandjust how creativethey're going to be able

**Patrick：**我想退一大步，来理解你对「OpenAI 要做什么、你希望它做什么、它代表什么」的最简版本的构想。我对你之后怎么实现有一百万个问题，但你已经做了那么多有意思的事；在最开始的时候，我知道你们代表什么。我想听听你现在的构想，以及它有没有演变过。

Sam：我认为这将是迄今为止人类历史上最伟大的技术成就。但它真正有意义的唯一方式，是它让人们的生活比原本好得多。这里面一部分是给人们物质上的丰饶，以及去做他们想做的任何事、表达他们的创造力和互助愿望的能力。另一部分是确保人们保有控制权和能动性（agency），确保世界是越来越民主化而不是越来越集中化，确保人们能表达自己。

所以从积极的一面说，某种意义上，我们即将造出一个能实现任何愿望的神灯精灵（genie）。我认为非常重要的一点是：我们——全世界——向这个精灵许的头几个愿望，要对整个世界有益。同时我也认为很重要的是，世界上的人们要理解，他们在这些愿望上

[18:58] Sam Altman

to be with these wishes.I'm actually not a job zoomerat all.I think there are goingto be tons of jobsI think will be busierthan we want,not the opposite of thatbecause I think peoplewill have such creative wishesand such incredible ideasof what they askAI to help buildand we will all benefitfrom not just the obviousthings like curing diseasesbut, I don't know,the world's bestentertainment ideaswe just can't even dreamof sitting here now.So,I want to put thatin everyone's handswhich gets toone of the thingsthat we stand against.Concentration of powerwith AIis a terrifying thing.I think a lot of the talkabout safety concernsis well-foundedand then a lot of itis about peoplethat just really,even if it's slightlysubconscious,want to concentrate power.I am terrifiedof a worldwhere the very real fearsof AIare used as a wayto sayonly this small groupof people can have itbecause it's too dangerousand only they understand itbut don't worrylike they're going to makethe right decisionsfor all of us.I don't believe in that.I don't think anyoneshould want to livein a world of AI overlordsor a companythat is the roughequivalent of thatwhere someoneis making decisionsfor all of the futureand in exchange

能有多大的创造力。

我其实完全不是那种「工作末日论者」。我认为会有海量的工作，我认为我们会比自己想要的还忙，而不是相反——因为我认为人们会有非常有创造力的愿望、会有难以置信的点子去让 AI 帮忙造东西，而我们都会从中受益：不只是治愈疾病这种显而易见的事，还有——我不知道——世界上最好的娱乐点子，那种我们此刻坐在这里根本梦都梦不到的东西。

所以我想把这个能力交到每个人手里。这就引到了我们所反对的东西之一：AI 带来的权力集中是一件可怕的事。我认为很多关于安全的讨论是有充分依据的，但也有很多——哪怕只是略微无意识的——其实是有人真的想集中权力。我很害怕这样一个世界：AI 那些非常真实的恐惧被拿来当作理由，说「只有这一小撮人可以拥有它，因为它太危险了，只有他们懂；但别担心，他们会替我们所有人做出正确的决定」。我不信这一套。我不认为任何人应该想住在一个有「AI 霸主」的世界里，或者住在一家实质上等同于霸主的公司之下——某个人替所有人的未来做决定，而作为交换，

[20:07] Sam Altman

for a cure for cancerwhich obviouslyis a wonderful thing.We collectivelycede all agency.So,I think it's very importantthat we not fallinto this trapof in the well-meaningor not spiritof AI safetyand understandablefears around that.We get awayfrom a worldwhere we all getto use this technology.I was like a childof the internet.There were no rules.I mean,it was amazing.I think it was a huge factorin making me who I amand probably youand an entire generationand I think it's criticalwe preservethat spirit of AIand that we all collectivelyhave the abilityto self-determine our future.I have so many questionsbut I'll startwith this genie concept.You said we're aboutto have a genieimplying we don't yethave a genie.Well,it's pretty close.What's between now and then?

给你一个癌症的解药——那当然是件美好的事——而我们集体交出全部的能动性。

所以我认为非常重要的是，我们不要掉进这个陷阱：打着善意的（或者并非善意的）AI 安全精神、以及那些可以理解的恐惧的旗号，让我们离「我们所有人都能用上这项技术」的世界越来越远。我是互联网的孩子，那时候没有规则，那真是太棒了。我觉得那对塑造今天的我是巨大的因素，可能对你、对整整一代人都是。我认为我们必须保住 AI 的那种精神，让我们所有人集体拥有自我决定未来的能力。

**Patrick：**我有很多问题，但先从「精灵」这个概念开始。你说我们「即将」拥有一个精灵，言下之意是我们还没有。

**Sam：**嗯，已经很接近了。

**Patrick：**那从现在到那时之间还差什么？

[20:51] Sam Altman

Even some of the real skepticshave said to mein recent daysor recent weeks,I guess,I think GPT 5.6has been out for metwo weeks.They're like,okay,this is very AGI-like.It's very hard for meto say what I wantfrom this modelthat it can't dobut there are clearlysome things.You can't yet go saylike cure cancerand get cancer cured.You can't yet saygo do this complicatedphysical thingin the robot.The model also,although brilliant,is still not learningcontinuously as it goesand that feels to melike maybe nota hard requirementfor AGIbut certainlysomething that I'd like.Now,to argue against myself there,you can make a casethat AGI is not actuallyabout any single model.It's the machinerythat makes the modelsand from model to model,we actually arelearning new things.We're figuring outnew science.That stuff is workingamazingly well.So I have a lot of sympathyto people who saylike we're there.We have the genie.It can do these amazing things.It can do superhuman things.I am so obsessedand fascinatedwith the economic storyof the returnsto being on the frontier,which you are.And I'm so curiouslike if you had shown5.6 to yourselfand your team in 2019,if that team probablywould have said like,oh yeah,

Sam：最近这些天——或者说这几周吧，我觉得 GPT-5.6 对我来说已经出来两周了——连一些真正的怀疑论者都对我说：好吧，这已经非常像 AGI 了。我很难说出「我想让这个模型做而它做不到」的事。但显然还是有一些：你还不能说「去把癌症治好」然后癌症就被治好了；你还不能说「用机器人去做这件复杂的物理任务」。这个模型虽然才华横溢，但它仍然不会在运行过程中持续学习（learning continuously）——这在我看来也许不是 AGI 的硬性要求，但肯定是我想要的东西。

不过我也可以反驳我自己：你可以论证，AGI 其实不关乎任何单一模型，而是关乎那台造模型的机器；而从一代模型到下一代，我们确实在学到新东西、在搞清楚新的科学。这部分运转得非常好。所以对那些说「我们已经到了、我们有精灵了、它能做这些惊人的事、它能做超人的事」的人，我很有共鸣。

**Patrick：**我特别着迷、特别好奇的是「处在前沿的回报」这个经济故事——而你正处在前沿。我很好奇：如果你把 5.6 拿给 2019 年的你和你的团队看，那个团队大概会说：哦对，

[22:03] Sam Altman

it's definitely AGI.I think it would have.This goalpost moving thingis a real thing.But it does seem that,I'm curious if you agree,that effectivelyall the returnshave been at the frontier.Totally.And so everythingis about stayingat the frontier.And I'm curiouslike what the hardest,scarcest part of that is.If I think about compute,research talent,data.Essentially,it's moved around a lot.I mean,there was a timenot that long agowhere all the computingin the worldwouldn't have helped youbecause we were missingthe research idea.Now,part of why this is hardis that you do better researchwith more compute,you can try more things.An amazing statisticI heard recentlyis our biggest de-risksnow for upcoming runsare as bigas the entire compute runfrom 18 months agoor something.And so computeand research ideasare not as separateas they sound,but there was clearlya time seven years ago,eight years ago,whatever,where we were way moreblocked on research ideasthan compute.Then there was a timewhen we knew what to doand we just had to scale up.We were only bottleneckedon compute.Then we ran out of dataand we were bottleneckedon dataand we had to figure outwhat to do there.Now,again,I would saywe are still bottlenecked

这绝对是 AGI 了。

**Sam：**我觉得他们会这么说。这个「移动球门柱」的现象是真实存在的。

Patrick：但看起来确实——我好奇你同不同意——基本上所有回报都在前沿。

**Sam：**完全同意。

Patrick：所以一切都是关于留在前沿。我很好奇其中最难、最稀缺的部分是什么？如果我想到算力、研究人才、数据……

Sam：本质上，这个瓶颈换过很多次位置。不久之前还有一个时期，全世界的算力加起来都帮不了你，因为我们缺的是那个研究想法。而现在，这件事难就难在：算力更多，你的研究也做得更好，你能试更多东西。我最近听到一个惊人的统计：我们现在为即将到来的训练跑（run）做的「降风险实验」，规模就已经相当于 18 个月前的整个训练跑。所以算力和研究想法并不像听上去那么泾渭分明。

但七八年前——不管多久吧——显然有一个时期，我们被研究想法卡住的程度远甚于算力。然后有一段时期我们知道该做什么，只需要往上堆规模，那时我们只被算力卡住。再然后我们把数据用完了，被数据卡住，必须想办法解决。而现在，我会说我们仍然被算力卡住，

[23:05] Sam Altman

on compute,but the last six monthsor whateverhave been a real triumphof a timefor research ideas again.So there's alwaysa bottleneck,but the bottleneck moves around.And why do you thinkthat is?The research idea thingis especially interestingto me because of thisautomated research thingthat seems to be looming,RSI,whatever you want to call it,where I talked toan incrediblecolonel's engineer recently,which everyone alsoseems blocked on,and he himself saidthere's two years leftof colonel's engineer.Maybe one.Yeah.It's not going to be a thing.Yeah.And you simultaneouslyhave this weird thing,whether it's colonel'sor overall research,where the researchersare like the most important.They got us here.They're like the most importantpeople in the world.And those same peopleare themselves worriedthat they won't be relevantlike very soon.I suspect they're notactually going to gothat way in practice.A year ago,people said software engineersare cooked.The field is over.And that didn't happen.What did happen, though,is the natureof a software engineer,the expectationsof a software engineer,how much they would do,changed quite a lot.And you don't reallywrite codein the traditional sense,but you do something

但最近六个月左右，在研究想法上又是一段真正的凯旋期。所以总是有一个瓶颈，但瓶颈会到处移动。

**Patrick：**你觉得这是为什么？研究想法这件事对我特别有意思，因为「自动化研究」这个东西好像正在逼近——RSI（递归自我改进）或者你想怎么叫都行。我最近跟一位极其厉害的 kernel（算子内核）工程师聊过（所有人好像也都卡在 kernel 上），他自己说：kernel 工程师这个岗位还剩两年，也许一年。

**Sam：**是啊。

**Patrick：**这个岗位就不存在了。而与此同时你还有一个很怪的现象：不管是 kernel 还是整体研究，研究员们是最重要的人——是他们把我们带到这儿的，他们像是世界上最重要的人；而这同一批人自己在担心，他们很快就不再重要了。

Sam：我怀疑实际上不会那样走。一年前人们说软件工程师完蛋了、这个行业结束了，结果并没有发生。真正发生的是：软件工程师的性质、对软件工程师的期待、他们能干多少活，变了很多。你不再以传统意义上写代码，但你做的事情

[24:04] Sam Altman

that is very recognizablysoftware engineering.Now, people will argueabout whether thisis the same thingor a different thingthan when we stoppedpunching holes in cards.I actually don't knowhow that worked,but somehow the holesgot in the cards.And we're just againoperating at a higher levelor this is a phase shift.I don't know.But the ideaof getting a computerto do what you want,that is stillan important job.And for researchers,I suspect thatalthough the current workflowof a researcheris going tovery much be automated,there will be new thingsin the spirit of researchin the same waythat there's new thingsin the spiritof software engineering,even though we don't write code,that will still matter.It seems like you'veshifted your opinionon AI's impacton jobs in general,and I'm surein specific categorieslike that.Describe that changeand your current view.You mentionedif we could go back to 2019.If we could go back to 2019and show peopleour latest model,not only would they saythat it's AGI,they would say thatthe economy would havehad completely upended.Yeah, completely.Yes.And that has not happened.And I think justfrom an intellectualhumility point,anytime you're that wrongand that confident,

仍然明显可辨认为「软件工程」。

现在人们会争论：这跟我们当年停止「在卡片上打孔」相比，是同一件事还是不同的事。我其实不知道打孔卡是怎么工作的，但反正那些孔就是打上去了。而我们只是又一次在更高的抽象层上操作——又或者这是一次相变（phase shift）？我不知道。但「让计算机去做你想让它做的事」这件事，仍然是一份重要的工作。

对研究员我猜也一样：虽然研究员当前的工作流会被大幅自动化，但会有一些新的、秉承「研究精神」的事情出现——就像软件工程里也出现了新的、秉承软件工程精神的事情一样，哪怕我们不写代码了，它仍然重要。

**Patrick：**看起来你对「AI 对就业整体的影响」的看法发生了转变，我猜在某些具体类别上更是如此。描述一下这个变化，以及你现在的观点。

Sam：你刚才提到「如果我们能回到 2019 年」。如果我们能回到 2019 年，把我们最新的模型给人看，他们不只会说这是 AGI，他们还会说经济已经被彻底颠覆了。

**Patrick：**对，彻底颠覆。

**Sam：**是的。而这并没有发生。所以我认为，仅从「智识上的谦逊」出发——任何时候你既那么自信又错得那么离谱，

[25:16] Sam Altman

which I think we wereas a field,you have to update.And there's a bunchof takeaways.One,a boring one,is that AIis just very jagged.It's like superhuman geniusin some ways,like dumb toddlerand others.people have,so far,extremely complementaryskills to AI.Another is that peoplehave a great degreeof trust and enjoymentin working with other people.And you can go hirean AI consultant right nowor talk to an AI sales repright nowor hire an AI engineeror whatever.And somehow,most people seemto still really preferinteracting with a human.And I definitely wouldlike much ratherengage with a personthan engage with an AIfor almost everything.I also think thathuman valueshave valuebecause they're human.And as society evolvesand as the potential spacein front of usbecomes so enormous,we are deeply hardwiredto care about people.We're going to care aboutwhat people care about.And there's versions of thisyou can see todaywhere AI can makeincredible imagesand people only want onesthat are created by a humanor at least chosen by a human.There's the jokeabout it at this point.You can like,the signature on a piece of artis most of the value.But the truth of it isyou want to knowabout the person behind it.

——我觉得作为一个领域，我们当时就是那样——你就必须更新自己。这里有一堆启示。

第一个，比较无聊的一个：AI 是非常「参差」（jagged）的。它在某些方面是超人的天才，在另一些方面像个笨拙的幼儿。到目前为止，人拥有的技能与 AI 高度互补。

另一个启示是：人对「和其他人一起工作」有很高的信任和享受。你现在就可以雇一个 AI 顾问、跟一个 AI 销售代表聊、雇一个 AI 工程师之类的，但不知怎的，大多数人似乎还是更愿意跟人打交道。我自己也是——几乎所有事情上，我都宁愿跟一个人打交道，而不是跟 AI。

我还认为，人的价值之所以有价值，正因为它是人的。随着社会演进、随着我们面前的可能性空间变得如此巨大，我们在生理上就是被深深写死了要在意人。我们会去在意别人在意什么。这一点今天就能看到某些版本：AI 能生成惊艳的图像，但人们只想要人创作的、或者至少是人挑选的那些。现在关于这事已经有段子了：一件艺术品上的签名占了它大部分价值。但真相是，你想知道作品背后的那个人。

[26:35] Sam Altman

You read a novel,you want to knowabout the person behind it.And then in terms of business,for my job, for example,I think the world wants to knowabout like the personthat's going to be responsiblefor the decisions of a companyand who they're goingto hold accountableif they make bad ones.And they don't really wantan AI CEO.If you think backon like the portfolioof like risksthat you've takenin business or whatever,is it the casethat most of the onesthat really worked wellwere at the startnot popular?

Sam：你读一本小说，你想知道背后那个人是谁。落到商业上，比如我这份工作，我认为世界想知道那个要为一家公司的决策负责的人是谁——如果决策做砸了，他们要找谁问责。人们并不真的想要一个 AI CEO。

**Patrick：**回想你在商业上（或别处）冒过的那一堆风险，是不是那些最后效果特别好的，在一开始基本都不受欢迎？

[27:03] Sam Altman

Yes, that's for sure.This was the thingI really learnedfrom Peter Thieland Paul Grahamboth in two different ways,which is thatthe very best companies,the very best investmentopportunitiesare almost neverthe ones that lookreally popular.You can do okayjust following the trendof being a little early,but to do spectacularly well,you almost alwayshave to do thingsthat are notwhat everybody else is doing.You cannot befollowing the new wave.If you think aboutthe model cyclethat you've been in,which has been acceleratingand this weird factthat like the next six monthsor I don't knowwhat the number is,is going to be more progressthan the last X years.Can you bring usinto what it's liketo live in that model cycle?

**Sam：**是的，肯定是。这是我从 Peter Thiel 和 Paul Graham 那里学到的东西，他们各自用不同方式教了我同一件事：最好的公司、最好的投资机会，几乎从来都不是那些看起来很热门的。你光靠跟着趋势、稍微早一点，可以做得还行；但要做到极其出色，你几乎总得去做那些别人不在做的事。你不能去追新浪潮。

**Patrick：**说说你身处的这个模型周期吧——它一直在加速，还有一个很怪的事实：接下来六个月（或者不管那个数字是多少）的进展会超过过去 X 年。你能带我们进去感受一下，活在那个模型周期里是什么样？

[27:44] Sam Altman

One of the mostinteresting,important thingsthat I've learnedlast decadeis people in generalcan get usedto almost anything.The world can gofrom dismissing a pandemicas a joketo completely lockdownto this is how it's beenand it's fineand we've mostly adjustedin a shockinglyshort amount of time.And now there'seither AGI or close to itand everyone's like,okay, there's AGI.There's all kinds of examplesin one's personal lifewhere somethingincredible happenslike you have a kidor something terrible happenslike you lose a parentor break upor whateverand you thinkyou can't ever adaptto what a change it is.And then you can adaptto great thingsand keep being great.You can adapt to bad thingsand figure outhow to go on with your life.But this is a remarkablething that people can do.And so living through thisfeels like another versionof that,which is I thoughtit was going to be weirderto live through the singularitythan it turns out to be.And it's not any less excitingto watch the modelskeep getting better.The first thing I doevery morning is like,look at the modeltraining progressand it happens fasterand I have higher expectations,but it still feels really cool.When you get a new one,what do you do?

**Sam：**过去十年我学到的最有意思、最重要的事情之一是：人基本上什么都能习惯。世界可以从把一场大流行当笑话，到彻底封城，再到「事情一直就是这样、还行、我们基本上都适应了」——而这一切发生在短得吓人的时间里。而现在要么已经有了 AGI、要么很接近了，大家的反应是：哦，AGI 啊，好的。

个人生活里也有各种例子：有什么了不起的事发生，比如你有了孩子；或者有什么可怕的事发生，比如你失去了父母、分了手，你以为自己永远适应不了这么大的改变。然后你发现，你能适应好事、并继续做得很好；你也能适应坏事、想明白怎么把日子过下去。这是人非常了不起的一种能力。

所以活在这个时代，感觉就是那件事的另一个版本——我原以为亲历「奇点」（singularity）会比实际感受更怪。而看着模型不断变好，这件事也一点没有变得不那么令人兴奋。我每天早上做的第一件事就是看模型训练的进度；它发生得越来越快、我的期待也越来越高，但那感觉仍然很酷。

**Patrick：**拿到一个新模型的时候你会做什么？

[28:53] Sam Altman

How do you celebrate?What's the morning look like?It's happening faster and faster.What's your ritual?Many teams now workon different parts of itand different teamshave like some different rituals.There are some teamsthat always make a sweatshirtwith some funny meme on it.There's some teamsthat like always go outto the same bar.But the sense of beingin the roomfor the first timethat the frontierof knowledgeis pushed backand getting to seewhat that's like,there's really nothingthat most peoplewould rather doto celebrate than likeget to use the new model first.Do you think we havethe right measurementsof how good these things are?

**Patrick：**你们怎么庆祝？那个早上是什么样的？这事发生得越来越快了，你有什么仪式？

Sam：现在很多团队在做模型的不同部分，不同团队有不同的仪式。有的团队每次都做一件印着某个搞笑 meme 的卫衣，有的团队每次都去同一家酒吧。但那种「在知识的前沿被往外推的第一时间，你就在房间里、能亲眼看到那是什么样子」的感觉——对大多数人来说，庆祝方式没有比「第一个用上新模型」更好的了。

**Patrick：**你觉得我们现在有衡量这些东西好坏的正确尺子吗？

[29:26] Sam Altman

Definitely not.In some sense,the eval that mattersis is this being useful to people?You can approximate itby revenueor by amount of usageor like rate of discoveryof new knowledge.We have some teams working onwhat is the real world evallook like for these modelsas they get tosuperhuman scale.What is the frontierof your own usageof AI?

**Sam：**肯定没有。某种意义上，**真正重要的评测（eval）是：这东西对人有没有用？**你可以用收入、用使用量、或者用「新知识被发现的速率」来近似它。我们有几个团队在做「当模型进入超人尺度后，真实世界的评测应该长什么样」。

**Patrick：**你自己使用 AI 的前沿在哪里？

[29:45] Sam Altman

I have startedjust recentlyto experimentwith what it meansto let an AIlook at everythingI'm looking aton my computer.I don't have this built yetand I'm still tryingto feel outlike where the limitsof my comfortand trust should be.But this is definitelythe frontieris figuring outhow I get value out of that,how you get comfortablewith that,what that's going to look like.One takeaway is thatmy memory is terriblerelative to the memoryof an AIand the abilityto keep in mindwhat email I readsix weeks agoor what happenedexactly in a meetingseven and a half weeks agoand have that likebrought up rightat the exact momentand feed into a decision.That feels pretty magical.Pretty cool.This kind of soundslike personal agent-ish.What are the barriersto everyone having that?

**Sam：**我最近刚开始试验一件事：让 AI 看到我在电脑上看的所有东西，那意味着什么。这个我还没做出来，我也还在摸索自己舒适和信任的边界应该在哪儿。但这确实就是前沿——怎么从中获得价值、怎么让自己接受它、它会长成什么样。

一个体会是：跟 AI 的记忆比起来，我的记忆糟透了。它能记住我六周前读过哪封邮件、七周半前的一场会上到底发生了什么，并且能在恰好那个时刻把它调出来、喂进一个决策里。那感觉相当神奇，相当酷。

**Patrick：**这听起来有点像「个人 agent」。让每个人都拥有这个的障碍是什么？

[30:30] Sam Altman

I want that.Compute, man.Let's imaginethat we could buildthis productthat could justdo exactly what I saidfor all your stuff.Always on.Always on.Looking at everythingyou look at your computer,listening to every meetingthat you're in,reading every documentyou readand then not only that,not only can I do all thatwhich takes a lot of tokens,you can just drag a sliderabout likewhile I'm asleepyou can spendthis many tokens thinking.Come up with usefulnew ideas for me.Do whatever work you canand then just likekeep thinking aboutwhat I should do next.What an interesting thingis like justspend more computemaking your output betterfor me the next morning.I would drag that sliderquite far.I'd be willing to spenda lot for that.But the amount of computethat that would requireif everybody in the worldwants to drag that sliderpretty faris like a lot.I'd love to hear youtalk about how you thinkof the natureof this new intelligence.Someone told me recentlyplanes don't fly like a birdand this intelligenceis not.It's a very alienkind of intelligence.Yeah, it's a very alienkind of intelligenceand everyone's talkingabout how if youcan verify somethingit's just going to winwith enough computeand enough IQ

**Patrick：**我想要这个。

**Sam：**算力啊，老兄。我们设想一下：我们能造出这个产品，能完全照我刚说的那样处理你的一切，永远在线——

**Patrick：**永远在线。

Sam：看着你在电脑上看的每一样东西，听着你参加的每一场会，读着你读的每一份文档。而且不止如此：不只是做完这些（这已经要烧掉很多 token），你还可以拖一个滑块——比如「我睡觉的时候，你可以花这么多 token 去思考」。给我想出有用的新点子，能干的活都干掉，然后一直想着我接下来该做什么。多有意思的一件事啊：多花点算力，让第二天早上给我的产出更好。那个滑块我会拖得相当远，我愿意为此花很多钱。但如果全世界每个人都想把那个滑块拖得挺远，那需要的算力就是……相当巨大。

**Patrick：我很想听你聊聊，你怎么理解这种新智能的性质。最近有人跟我说：飞机不是像鸟那样飞的，而这种智能也不是（像人那样思考的）。它是一种非常异质（alien）**的智能。

Sam：对，它是一种非常异质的智能。现在所有人都在说：只要一件事可验证，那么给足算力和智商，它就会靠暴力搜索碾过去、找到解。

[31:27] Sam Altman

that will just brute forceits way to a solution.And then in other domainswhere humansand the dataand evals that they've donehave been a huge part of itit's surprising to melike how much moneyit's cost toget good atI don't knowlawreasoning tracing lawor something.I'm just curioushow you would describeI'm not sure howyour kid isyou have a boy or girlwhen they'reseven or age of reasonor whateveryou can describe to themlike what is the natureof this intelligence?

**Sam：**而在另一些领域——人、数据以及他们做的评测占了很大比重的那些领域——让我意外的是，要把某件事做好竟然要花那么多钱，比如法律那类东西（原音含糊，似为 "reasoning tracing law"，即法律推理/法条溯源一类）。

**Patrick：**我很好奇你会怎么描述——我不太确定你孩子的情况，你是男孩还是女孩——等他们七岁、到了懂事的年纪，你会怎么向他们描述这种智能的本质？

[31:53] Sam Altman

How would you describe it?It's a beautiful question.I don't think I've beenasked this beforeor even any version of it.The thing that'scoming to mind right nowis I would just sayit's like a computerand it's like a computerin the way that it cando a lot of thingsthat people just can't dolike multiplytwo gigantic numbersvery quicklyand give you the answerand then it cannot dosome things that you wouldvery easily do.The number of thingsthat it can't doI expect to keeprecedingbutin an evolving worldI thinkhuman judgmentand tastewill continueto be hardfor AIsto modellike where that'sgoing to go.I don't have the rightword for thisit's not quite taste.The world may needa new kind of wordfor the kind ofjudgment thatpeople are very good atthat AIs seem toreally deeply struggle with.What's it been likebecoming a dadand having growing kidsin this era?

**Patrick：**你会怎么描述它？

**Sam：**这是个漂亮的问题，我觉得我以前没被问过，连类似的版本都没有。此刻我脑子里冒出来的是：我大概会说，它就像一台计算机。像计算机的地方在于，它能做很多人做不了的事，比如把两个巨大的数字飞快相乘并给你答案；同时它又做不了一些你轻而易举就能做的事。

它做不了的事情清单，我预计会不断缩短。但在一个不断演化的世界里，我认为人的判断力和品味会继续是 AI 很难建模的东西。这个东西该往哪儿去，我也没有确切的词——它不完全是「品味」。世界也许需要一个新词，来指代那种人非常擅长、而 AI 似乎深深挣扎的判断力。

**Patrick：**在这个时代当爸爸、看着孩子长大，是什么感觉？

[32:46] Sam Altman

I'm thinking backto your optimisticearly internet daysthey're going to grow upin cheap abundantintelligence age.Having kids is by farthe best thingI have ever doneand everybody says thateverybody saysyou can't really understand it.I believe enough peoplethat said itthat I believed itto be truebutthe degree to whichit has been true for mehas been surprising.I think I have the bestmost interesting jobin the worldand it is stilla very distant secondto having kids.So it's been awesomeand it is a real momentfor optimism.My kids will nevergrow up in a worldwhere they were smarterthan computers.If you were bornat the time of GPT-3you had a timewhere you had betterreasoning than the modelseven thoughyou didn't knowwhen you were born.You caught them briefly.Our older kidlike 18 monthsthat will never seemstrange to himthat will never bother himI don't think he'll care.He would be shockedto imaginein the dark ageswhen we had to likedeal with productsand servicesthat weren't incredibly smart.He will be ableto do thingsthat you and Inever were able to doand he'll have expectationsin life that you and Inever hadand he'll have likea much bigger canvas.Do you runthe businessor teamsor lead peoplein any way

**Patrick：**我想起你早年对互联网的乐观——而他们将在一个「智能廉价而丰饶」的时代长大。

Sam：有孩子是我这辈子做过的最好的事。所有人都这么说，所有人都说这事你没法真正理解。我信了足够多说这话的人，所以我本来就相信它是真的；但它对我而言真实到什么程度，还是让我吃惊。我认为我拥有世界上最好、最有意思的工作，而它跟有孩子比起来，仍然是遥远的第二名。

所以这非常棒，而且这是一个真正值得乐观的时刻。我的孩子永远不会在一个「他们比计算机聪明」的世界里长大。如果你出生在 GPT-3 的时代，你还曾有过一段推理能力强于模型的时期，尽管你出生时并不知道——你算是勉强赶上了。我们的大孩子 18 个月大，对他来说那永远不会显得奇怪，永远不会困扰他，我觉得他根本不会在意。他反而会震惊地想象「在那个黑暗年代，人们居然要跟不那么聪明的产品和服务打交道」。他将能做到你我从来做不到的事，他对人生的期待会是你我从来没有过的，他会有一块大得多的画布。

**Patrick：**因为有了他们，你在经营公司、带团队、领导人这些事上，有没有明显不同？

[33:52] Sam Altman

that is notably differentbecause of the experienceof having them?The answer must be yes.I feel very differenthaving them.I think there's likea bunch of small thingsthat are really differentand then againthis is likenot a novel insightin any way.I think most peoplehave had kids sayas soon as you have a kidyou like realizethat you caremuch moreabout themand the experiencethat they're going to haveand you do about yourselfand the worldthat you are going toleave themand I think I havean unusual vantage pointfor that.People ask me sometimeslike ohnow that you have kidsdo you caremore about your safetyand not destroying the worldand the answer is likeI didn't need kidsI really didn't wantto destroy the world beforebut do I think moreabout the roleof human agencyand what it meansto have a fulfilling lifedefinitely much morefor what we're buildingand also like the peopleI work withI want them to have it too.You obviously haveextraordinary empathyfor your kidsbut the degree to whichthat extends toall kidsand then maybeto all parentsand maybe then to everybodythat's been a surpriseto me too.In one of the postsI think it was the onethat's things you wishyou knew earlieror somethingis about incentives

Sam：答案必然是有。有了他们之后我感觉非常不同。有一堆小事真的不一样了。而且这也不是什么新颖的洞察——我想大多数当过父母的人都会说：一有孩子，你就意识到你在意他们、在意他们将拥有的体验，远超过在意你自己，也在意你将留给他们一个什么样的世界。而我认为，我在这件事上有一个不寻常的观察位置。

人们有时问我：现在你有孩子了，是不是更在意安全、更在意别把世界毁掉？答案是——我不需要孩子来提醒我，我之前也真的不想毁掉世界。但我是不是更多地思考人的能动性、以及「拥有一个充实的人生」意味着什么？在我们正在造的东西上，那绝对是想得多多了。而且我也希望跟我共事的人也拥有这些。

Patrick：你显然对自己的孩子有非凡的共情，而这份共情延伸到所有孩子、也许再延伸到所有父母、再到所有人——这个程度也让我意外。在你的一篇文章里（我想是那篇「你希望更早知道的事」），讲到了激励机制——

[34:57] Patrick O'Shaughnessy

and set themvery very carefully.Yeah.It's always beenone of the most puzzlingand interesting thingsabout youthat you don't haveequity exposureto this company.How should the worldthink about your incentives?

Patrick：——要非常非常谨慎地设置激励。是的。这一直是关于你最令人费解、也最有意思的事情之一：你在这家公司没有股权敞口。世界该怎么理解你的激励？

[35:07] Sam Altman

I don't knowwhat I can saybeyondI have a front row seatto the most excitingmoment of human historythat is worth moreto methan any amount of money.I get to havean extremely interesting lifeand work withextraordinary peopleon somethingthat I deeply care aboutbut somehow that doesn'tdo it for peopleor something.No it does not.I'm curious how you thinkabout robotics.You mentioned earlierat some pointif we hadautomated laborin the same waywe're going to haveautomated intelligencethings might geteven crazierlabor marketsbehind the white collar market.If we don't have itthen things get really crazy.If the role for peoplein the worldis to be likethe actuators of AIin the cloudvery badvery badso I think it's likemuch greaterif we don't get itthan we do.Help me understandyour sense of progressin thatbecause unlike an AIwhere everyoneis now kind ofon the same pagelike it's going fastyou can findextremely smart peoplethat say it's likeend of this yearand you can findextremely smart peoplethat say it's 20 yearsfrom now or something.It's not 20 years.I would say we getthe chat GPT momentfor roboticsin the next two or three years.What would that be?

Sam：除了这个我不知道还能说什么：我有一个前排座位，看着人类历史上最激动人心的时刻——这对我来说比任何数额的钱都值钱。我得以过一种极其有趣的人生，跟非凡的人一起做我深切在乎的事。但不知怎的，这个说法对人们就是不管用。

**Patrick：**对，确实不管用。我很好奇你怎么看机器人。你前面提到，如果我们有了自动化的劳动力，就像我们即将有自动化的智能一样，事情可能会变得更疯狂——白领市场之后是劳动力市场。

Sam：如果我们没有机器人，事情才真的会变得很疯狂。如果人在这个世界上的角色是充当云端 AI 的执行器（actuator），那非常糟糕。所以我认为「没拿到机器人」比「拿到机器人」危险得多。

**Patrick：**帮我理解一下你对这方面进展的判断。因为不像 AI——现在大家基本都在一个共识上，觉得它跑得很快——在机器人上，你能找到极其聪明的人说「今年年底就行」，也能找到极其聪明的人说「还要 20 年」。

**Sam：**不会是 20 年。我会说，机器人的「ChatGPT 时刻」会在未来两三年内到来。

**Patrick：**那会是什么样？

[36:10] Sam Altman

Do you know what that is?Something wheremost peoplehave likea realwownot likeI saw this videoof a robot dogdoing something crazybut I was somehowable toconvince myselfthat a reallyimportant thinghappened.One of the thingsabout the chat GPT momentwas that you couldjust go use it.Yeah.Like you didn't have tolike believe someonewho said AI is coming soonyou could just go try itand if you can gotype in a commandand a robot can dosomething crazyand you can likewatch it even ifyou're not physically thereI think that would havethe same kind of likewhoa it just did this thing.Wasn't chat GPTlike not this monolithic goalbut sort of likea side experimentthat you decided to release?

**Patrick：**你知道那会是什么样吗？

Sam：某种让大多数人真正「哇」一下的东西——不是「我看到一个机器狗做了个很疯的视频」，而是我能以某种方式说服自己：一件真正重要的事发生了。ChatGPT 时刻的一个特点是你可以直接自己去用。你不用去相信某个说「AI 快来了」的人，你可以自己去试。如果你能敲进一条指令，一个机器人就能做出某件很疯狂的事，而且你哪怕不在现场也能看到它做——我觉得那就会有同样的「哇，它刚刚真做到了」的效果。

Patrick：ChatGPT 当初不是一个宏大的既定目标，而更像是你们决定发布的一个边缘实验，对吧？

[36:49] Sam Altman

That story may be instructivefor something similarhappening in robotics.Everyone seems to wanta full boundarybut maybe it's somethingvery different.When we launchedGPT-3we're trying to make moneytrying to get peopleto use this API.The only commercial use casethat was really workingthe model was just so dumb.If you went backand used ityou'd be astonished.The only commercial use casethat was workingwas copywritingso you paysome marketing firm20 bucksand they paid us20 cents for the AIto like writeyou a landing pageor whatever.But in addition to thatone commercial use casedevelopers were usingthis thing we callthe playgroundwhich was likea testing interfaceto chat with the model.And it was really hard to dobecause we had nottuned the modelto be good to chat withso you had to likegive it a few examplesof what it means to chatand then do itand people really liked it.And I learnedthis great lessonfrom YCis you notice your usersdoing something likego down that path.And sowe decidedthat we would builda good chat botsince that's whatpeople were doing.And we startedworking on thatand we finished GPT-4and we startedusing that internallyand we're likethis is a big deal.We kind of thoughtthat all right

**Patrick：**那个故事对机器人领域可能发生的类似情况也许有启发。所有人都想要一个完整的形态，但也许它会是很不一样的东西。

Sam：我们发布 GPT-3 的时候，是在想办法赚钱、想让人们用这个 API。当时唯一真正跑通的商业用例——那个模型实在太笨了，你要是回头去用会震惊的——唯一跑通的商业用例是文案写作：你付给某家营销公司 20 美元，他们付给我们 20 美分，让 AI 给你写个落地页什么的。

但除此之外，开发者们在用一个我们叫 playground 的东西，那本来只是个跟模型对话的测试界面。用起来其实很难，因为我们没有把模型调教成适合聊天的样子，你得先给它几个「聊天是什么样」的示例，然后才能聊。而人们非常喜欢这个。

我从 YC 学到一条很棒的教训：当你注意到用户在做某件事，就顺着那条路走下去。所以我们决定去做一个好用的聊天机器人，因为那正是人们在做的事。我们开始做这件事；与此同时我们做完了 GPT-4，开始在内部使用它，我们心想：这是件大事。我们当时觉得，好吧，

[37:52] Sam Altman

this is going to bea real updateto the worldabout AIand there's a bunchof hard questionshere aboutis this going tocreate a bunchof fake newsis it going tosay really offensivethingswe're going toget in trouble.So we decidedwe would startwith a weaker versionthe chat interfaceand GPT-4at the same timeseemed like a lotso we would roll outthe chat interfaceand GPT-3.5In fact it wasoriginally going to becalled chatwith GPT-3.5we didn't planfor GPT-3.5didn't think it'd bea huge shipbut did thinkit would get peoplethe world to likecatch up with thisand realizesomething was going onand we mercifullyrenamed itchat GPT-3a few hoursbefore launchand put it outas like a research previewand the thought waswe'd put it outas a research previewand then a few monthslater we wouldlaunch a productwith GPT-4and for whatever reasonthat model wasover the thresholdwhere even thoughwe had gotten usedto it internallypeople saidokay this is awesomethere maybe wasn'tthat much utilityyetbut it was anincredible momentfor people tofeel AI progressand use somethingthey enjoyed usingand then by the timewe put GPT-4something they reallygot benefit outof using tooare you surprisedthat remainsthe intuitiveinterface between us

**Sam：**这会真正更新全世界对 AI 的认知；而这里有一堆难题：这会不会制造一大堆假新闻？它会不会说出很冒犯的话？我们会不会惹上麻烦？

所以我们决定从一个更弱的版本开始。聊天界面加上 GPT-4 同时上，感觉信息量太大了，所以我们决定只推聊天界面 + GPT-3.5。事实上它原本要叫「Chat with GPT-3.5」。我们没为 GPT-3.5 做什么计划，不觉得它会是个大发布，但确实觉得它能让全世界跟上、意识到有事情正在发生。谢天谢地，我们在发布前几小时把它改名叫了 ChatGPT，并作为一个「研究预览（research preview）」放了出去。

当时的想法是：先放个研究预览，几个月后我们再用 GPT-4 发一个正式产品。但不知什么原因，那个模型跨过了某条阈值——尽管我们内部已经用习惯了，外面的人却说：好，这太棒了。当时也许还没有那么多实用性，但它是一个让人们感受到 AI 进展、并且用得开心的不可思议的时刻。等我们推出 GPT-4，那就变成人们真正能从中获益的东西了。

**Patrick：**你意外吗——这（聊天）到现在仍然是我们和这种异质智能之间最直觉的界面，

[39:01] Sam Altman

and this alienintelligenceeven including codingmostly that's metalking to the computertelling it what to buildno because I'm likea massive texterI've been a massivetexter my whole lifeI think part ofmy own insightof why that wasgood interfacesI'm likeI know how to do thisI know how to do thisI know what it's liketo just start chattingin a text boxany thoughts on thisnotion of diffusionand how to make it fasterlike if the mission isget intelligenceinto the handsand more usefulfor everyonethe key part of thatis I don't knowa marketing campaignor somethinghow do you get thisto diffuse fasterthan it seems to bedoing naturally to meI think the key thingis just make it betterI kind of believethat a truly greatproduct markets itselfthere was noChagipity marketingcampaign at thebeginningI think as we getto this nextstage of modelsand we figure outhow to make productsthat are as greatas the modelsthemselvesthere will besuch incredibleutilitythat people willspread it very quicklywe should definitelydo more marketingAI is not too popularfor as much aspeople use itthey havevery understandableanxiety aboutwhere it can goand so that kind ofstuffI think somegreat marketingwould be helpful forbut in terms of

**Patrick：**甚至包括写代码在内，主要还是我在跟计算机说话、告诉它要造什么。

Sam：不意外，因为我是个疯狂的短信控，我这辈子一直是。我觉得我当时能看出「这是个好界面」的一部分原因就是：我知道怎么干这个，我知道在一个文本框里开始聊天是什么感觉。

**Patrick：**关于「扩散（diffusion）」有什么想法——怎么让它更快？如果使命是把智能送到所有人手里、让它对所有人更有用，那关键是什么？我不知道，一场营销战役？你怎么让它扩散得比现在自然发生的速度更快？

Sam：我觉得关键就是把它做得更好。我相当相信「真正伟大的产品会自己营销自己」。最开始 ChatGPT 一分钱营销都没花。我觉得等我们走到下一阶段的模型、并搞清楚怎么做出「和模型本身一样出色的产品」时，会有极其巨大的实用价值，人们自己会飞快传播它。

我们当然应该多做点营销。AI 并不那么受欢迎——相对于人们用它的量而言，他们对它可能走向何方有非常可以理解的焦虑。这类事情我觉得好的营销会有帮助。但就

[40:06] Sam Altman

value people aregetting out of theproducts and gettingtheir products to growfaster better modelsmore compute betterproducts that will doitthere was this periodwhere the recruitingof researchers theretention of themthe incentivizing ofthem was the definingstory in thecompetitive landscapeor whateverI think there's lotsof stories about yousuccessfully recruitinggreat researchersand there's been manythat have come throughopen AI and had hugeimpacts some of whichare known some ofwhich are lesserknown namesI'm just curious aboutthis whole genre ofwhat you learned abouthow to recruit thisclass of personwhat matters to themand how you did itI've never heard youtalk about the actualtactical moves youpulled to recruitsomebodyin the early daysI think it was quitesimple which was thatwe believed that AGIwas possible and itwas worth going afterand we wanted to saythat and that was likean insane hereticalbelief when we firstannounced open AIall of these giants ofthe field these expertswere saying this is likeinsane it's hypey it'sirresponsiblewe have really respectedpeople like Jan Lakoonor whatever tellingjournalists like ohthese guys aren'tvery good and it'snot going to workbut the fact that we

Sam：「人们从产品中获得的价值」和「让产品增长得更快」而言：更好的模型、更多的算力、更好的产品，那才是真正起作用的。

Patrick：曾有一段时期，研究员的招聘、留存和激励是竞争格局里的定义性叙事。关于你成功招到优秀研究员的故事有很多，也有很多人从 OpenAI 走出来产生了巨大影响，有些名字很有名、有些不那么有名。我很好奇这整个门类：你学到了什么关于「怎么招这一类人、他们在乎什么、你是怎么做到的」？我从没听你讲过你为了招到某个人所做的具体战术动作。

**Sam：**早期其实很简单：我们相信 AGI 是可能的、并且值得去追求，而且我们愿意把这话说出来。当我们最初宣布 OpenAI 的时候，这是一个疯狂的、异端式的信念。这个领域所有这些巨头、这些专家都在说：这太疯了，这是炒作，这是不负责任的。我们很尊敬的人，比如 Yann LeCun 之类的，会对记者说：这帮家伙不怎么样，这事成不了。但正因为我们

[41:11] Sam Altman

were able to saywe're going to go forthis it really appealedto a certain kind ofresearcher that alsowanted to like go onthis crazy adventurewith low probabilityof successso an ambitiousaudacious visionis a very powerfulrecruiting toolyou've written thatit's actually easiersometimes to buildthings that areharder because of thisreasonI super believe in thisit's one of my mostfrequent pieces ofadvice to YC foundersand I tried to reallylive it at open AIjust do somethingharderso do something thatmatters do somethingthat is importantand if your companydoesn't succeedmight not happenyou were an investorand are an investoryou've done a lot of itand at one pointthat's what you didwhat have you learnedabout investors beingon the other sidethe number of investorsthat actually show upand try to help youis unbelievably smallJosh Kushnerabsolute MVP investorunbelievablehas like workedaround the clockfor what feels likeyears to help ushe's the only investorthat I could point tothat isproactivelyincredibly helpfulall the timethere are more peoplethat could do thatand there are manyother investorsthat have also been helpfuland that have greatstrategic adviceand that do thingswhen we ask themto do it

Sam：敢于说「我们就是要去干这件事」，这对某一类研究员极有吸引力——他们也想踏上这场成功率很低的疯狂冒险。

**Patrick：**所以一个雄心勃勃、大胆无畏的愿景，是非常强大的招聘工具。你写过：正因为这个原因，有时候造更难的东西反而更容易。

**Sam：**我极其相信这一点。这是我给 YC 创始人最常给的建议之一，我也努力在 OpenAI 真正践行它：就去做更难的事。去做重要的事，去做有分量的事——如果你的公司不成，这件事可能就不会发生。

**Patrick：**你曾经是、现在也是投资人，你做过很多投资，那一度就是你的本职。站到另一边之后，你对投资人这个群体有什么新认识？

**Sam：**真正会出现、并且试图帮你的投资人，数量少得不可思议。Josh Kushner 是绝对的 MVP 投资人，难以置信，感觉像是连续好几年不分昼夜地在帮我们。他是我唯一能指出来的、主动地、始终如一地极其有帮助的投资人。能做到这点的人本可以更多。当然还有许多别的投资人也很有帮助，有很好的战略建议，我们开口请他们做事时他们会做，

[42:27] Sam Altman

but the likeconstantjustrelentlessall-in supportis surprisingly rarefrom investorsmaybe I'm biasedbecause I always liked itwhen people said thatabout mebutI think foundersreally love thatand it actually movesthe needleand as an investorit's the most fun wayto do itme and my friendplay this gamewhere we text each otherall the timeand the promptof the text issomething I don't wantyou to know about mewhat does that bring to mindI'm tiredI've been doing thisa long timeit's tiringhow do you get through thatyou just keep goingit begs the questionis there an amountof being tiredthat would make youstop doing thisno no nothis is the coolest jobin the worldI plan to do thisfor the rest of my careerbut it's like much harderthan I have a wayto explain to peopleI feel very gratefulto get to do thisthis is not me complainingwhat's coming nextwe talk aboutautomated AI researchersthat next yearthe year afterhow do you think aboutwhat is happeningin the next six to36 monthsmaybe that's too far outtoforecast in thiscrazy exponentialmaybe a differentversion of the questionis let's sayin a month23 from nowwe have somethingthat everybody agreesis super intelligencewhat happens in month 24and my answer would be

Sam：但那种持续的、不停歇的、全押式的支持，在投资人里意外地罕见。也许我有偏见，因为过去别人这么评价我时我总是很受用；但我觉得创始人真的非常在意这个，而且它确实能改变结果。作为投资人，这也是最有乐趣的做法。

**Patrick：**我和一个朋友玩一个游戏，我们一直互相发短信，短信的提示语是「说一件我不知道的、关于你的事」。这会让你想到什么？

**Sam：**我累了。我干这个已经很久了，这很累人。

**Patrick：**你怎么熬过去？

**Sam：**你就是继续走。

**Patrick：**这就带出一个问题：有没有一个「累到什么程度」会让你停下来不干了？

**Sam：**不不不。**这是世界上最酷的工作，我打算干一辈子。**但它比我能向别人解释清楚的要难得多。我非常感激能做这件事，这不是我在抱怨。

**Patrick：接下来会发生什么？我们聊了自动化的 AI 研究员，明年、后年——你怎么看接下来 6 到 36 个月会发生什么？也许在这种疯狂的指数里预测那么远太难了。换个版本的问法：假设第 23 个月我们有了一个所有人都同意是超级智能（super intelligence）**的东西，那第 24 个月会发生什么？

**Sam：**我的答案是——

[43:36] Sam Altman

not very muchthe kind of likecult worshipof the machine godstatesthose people believethat more is goingto happen quicklythan is going to happeneventually a lot will happenbut eventually a lotis going to happen anywaythe rate of human progressand how differenteach decadeis going to beand how mucheach decadeis more differentthan the decadefrom beforethat's been happeningfor a long timeobviously ups and downsbut directionallyand I thinkthe right wayto think about thiseverybody wants to bethe hero of the storyeverybody wants to feellike they were therefor the momentof the machine godand they playedsome crazy rolebut this is another stepand it was hardto imagine 50 years agoand the step 50 yearsfrom now is hardto imagine from todayand I think the rightmental frameworkis just the zoom way outand it's a prettysmooth exponentialtell me a little bitabout the experienceof watching codextake offand how muchthat is tiedto what I woulddescribe as a competitiveadvantage of distributionthat you buildthrough chatand this is a gatewayinto a questionabout like moatsin general in AIwhat you thinkwill drivereal competitive advantagein the businessover timeI think codexmostly is winningbecause it'sthe best product

Sam：——不会发生太多。那种对「机器之神」的崇拜式狂热，那些人相信「很快就会发生很多事」，比实际会发生的多。最终确实会有很多事发生，但最终本来也会有很多事发生。人类进步的速率、每一个十年会有多不同、以及每个十年比前一个十年更不同多少——这件事已经持续很久了，当然有起有伏，但方向上如此。

我认为思考这件事的正确方式是：每个人都想当故事的主角，每个人都想觉得自己在「机器之神诞生」的那一刻在场、并扮演了某种疯狂的角色。但这只是又一步。50 年前的人很难想象今天，而从今天出发也很难想象 50 年后的那一步。我认为正确的心智框架就是把镜头拉到很远——它是一条相当平滑的指数曲线。

**Patrick：跟我讲讲看着 Codex 起飞是什么体验，以及这在多大程度上跟你通过 ChatGPT 建立的「分发（distribution）」优势有关。这也是通往一个更大问题的入口：AI 里的护城河（moat）**总体上是什么？你认为长期真正驱动竞争优势的是什么？

Sam：我认为 Codex 赢主要是因为它是最好的产品、也是最好的模型。

[44:43] Sam Altman

and the best modelwe do get someadvantage fromchatgbtbundling but veryvery tinythat is mostlynot what it's beenaboutit has made mereflect a loton this questionof competitiveadvantagebecause brilliantintelligence can migratefrom any productto any other productand network effectsstill have a competitiveadvantageeconomic scaleand the abilityto like makethe cheapestcompute fleetswhatever still havea competitive advantagebut the product advantageif we could getpeople to moveover to codexand someone buildsany betterthey can get peopleto move from codexso it has made mereflect on that a lotthere's a reallyinteresting questionabout whether thisis going in thedirection of a commodityis intelligencegoing to bea pure fungiblecommoditylike rate of thewhale or somethingintelligence itselfI would say yesso what is notgoing to becompute fleetyou know like thescale of thecompute fleetthe ability to makemore computeI think that's likea very durableadvantageI seeeven if the productitself is notbecause codex canwrite any pieceof software you wantthe workflowsthe integrationsthe complexprocessesthe ability for teamsto collaborate togetherthat stuff is allpretty powerfuleven like brandpreferenceand familiarity

**Sam：我们确实从 ChatGPT 的捆绑（bundling）**里得到一些优势，但非常非常小，这基本不是它成功的原因。

这件事让我对「竞争优势」这个问题反思了很多。因为卓越的智能可以从任何一个产品迁移到另一个产品。网络效应仍然构成竞争优势；规模经济、以及「做出最便宜算力集群」的能力，仍然构成竞争优势。但产品层面的优势——如果我们能让人们迁移到 Codex，那么当别人做出更好的东西时，他们也能让人们从 Codex 迁走。所以这让我反思了很多。

这里有一个非常有意思的问题：这是不是在往「商品化（commodity）」的方向走？智能本身会不会变成一种纯粹可替换的大宗商品？（此处原音含糊，似为 "like rate of the whale"）

**Patrick：**智能本身？

**Sam：**我会说是的。

**Patrick：**那什么不会（商品化）？

**Sam：**算力集群——比如算力集群的规模、制造更多算力的能力，我认为那是非常持久的优势。

**Patrick：**我明白了，即便产品本身不是。

Sam：因为 Codex 能写出你想要的任何一段软件，所以那些工作流、集成、复杂流程、团队协作的能力，这些东西都相当强大。甚至连品牌偏好和熟悉度

[45:50] Sam Altman

is pretty powerfulobviously you've doneinteresting stuff inhardware that I'm sureyou'll announcelater this yearhow does thatexperiment feeland aligned withthis sort ofconsumer distributionthat you haveone of the reasonsI'm interested innew hardwareis we were talkingearlier about howa very powerfulthing with AIis that it can bealways on andproactive and justunderstand all yourcontext but currenthardware is notgood for thatwe are workinginside of ahardware paradigmthat is50 years oldsomething like thatand computers areamazingkeyboard and micemonitors an amazingthing but we haveto shape AI intothat I'm excitedI would love AIto be able toreference thisconversation but notso much that I'mwilling to likecrack my laptopopen put it hereand have it likelooking at youand listening to uswhile it's goingbut I would like apiece of hardwarethat socially wasacceptable to dothat and also feltlike it was designedfor that kind ofa thingas you think aboutthe open questionswhat debates inyour own headwith your friendswith your colleagueshere what are themost interestingopen debatesor open questionsthat you don'tfeel certain aboutbut feel importantone that I don'tthink gets muchattention is howare we going to

**Sam：**也相当强大。

Patrick：你显然在硬件上做了些有意思的事，我相信今年晚些时候你会公布。那个实验感觉如何？它跟你已有的消费级分发能力是怎么对齐的？

Sam：我对新硬件感兴趣的原因之一是：我们前面聊到，AI 一个非常强大的特性是它可以永远在线、主动、并且理解你的全部上下文——但当前的硬件不适合这件事。**我们是在一个大约 50 年历史的硬件范式里工作。**计算机很了不起，键盘、鼠标、显示器都是了不起的东西，但我们必须把 AI 塞进那个形态里。

我很兴奋。我很希望 AI 能引用这场对话，但还没到「我愿意把笔记本电脑掰开放在这儿、让它盯着你、听着我们说话」的程度。但我确实想要一件硬件，做这件事在社交上是可接受的，而且感觉它就是为这类事情设计的。

Patrick：说到那些悬而未决的问题——在你自己脑子里、跟朋友、跟同事之间，最有意思的开放辩题或开放问题是什么？那些你并不确定、但觉得很重要的。

**Sam：**有一个我觉得没得到足够关注的是：我们要怎么避免「认知萎缩」（cognitive atrophy）？

[46:57] Sam Altman

avoid cognitiveatrophy how arewe going to usethese tools andmake sure that weare like stretchingour brains moreand more andcontinuing tounderstand thestuff that reallymatters there'slots of versions ofthis that don't Iremember when I wasin school I had thisprofessor tell melike you've got tounderstand compilersif you don't youwill never be ableto be a goodprogrammer somehowthat wasn't quiteright butunderstanding at areasonable level howthe major componentsof a computer systemwork has beenimportant to meforced to imaginea scenario where weare somehow oversuppliedand compute in twoyears time what wouldbe that storyit does feel possibleif the models get sosmart and soefficient that theycan do everything weneed and build everypiece of software wewant and if thebounds of ourattention are suchthat they just cannotabsorb more thanwhat it turns out afair limited amount ofcompute can do then wecan get into oversupplyalso if we don't drivethe cost curve downbecause we hit somesort of scaling wall wecould also get intooversupply theobservation aboutuncapped demandimplies a certainpricecan you give your pointof view on scalinglaws todayin some sense scalinglaws are like themost hated prediction

Sam：我们要怎么使用这些工具，同时确保我们在越来越多地拉伸自己的大脑、继续理解那些真正重要的东西？这件事有很多不同版本。我记得上学时有位教授跟我说：你必须理解编译器（compiler），不然你永远成不了好程序员。不知怎的那句话并不完全对；但「在一个合理的层次上理解计算机系统的主要部件是怎么工作的」，对我确实很重要。

Patrick：强迫你想象一个场景：两年后我们不知怎的算力过剩了，那会是个什么故事？

Sam：这确实有可能：如果模型变得如此聪明、如此高效，能做完我们需要的一切、造出我们想要的每一个软件；而如果我们注意力的边界就是那样——它们吸收不了比「其实相当有限的一点算力」所能产出的更多东西——那我们就会进入过剩。另外，如果我们没能把成本曲线压下来，因为我们撞上了某种 scaling wall（规模化墙），那也可能进入过剩。

Patrick：「需求无上限」这个观察，隐含了某个特定的价格。你能讲讲你今天怎么看 **scaling laws（规模定律）**吗？

[48:06] Sam Altman

of all time everybodyalways wants to saythey're going to runout they can't be likethis and yet it keepsgoingwho are your favoriteunsung heroes in thiscompany's storythe first person thatcame to mind is AlecRadfordAlec Radford is probablythe most important notvery well-knownresearcher in the wholehistory of the fieldand also just awonderful top top tierhuman being he didthe work that reallybecame the GPT seriesamong many otherimportant things but healso is someone whoinspired guided nudgedpeople in many otherdirections that turnout to be superimportant and thething that I think iscool about him is ifyou talk to people thatworked with him theywill of course saygenerational geniusbrilliant innovativethinker just so deepin his understanding andhis work but everybodywill tell you beforethey finish theirstatement that justone of the nicest mostpositive best peoplethey've ever metwith I love formativemoments and so as wewind up here I'mcurious to ask one ofeach if you thinkabout the wholeopening experiencewhat moment orchapter or whateverare you most proud ofyour own involvementand start with theother one which iswhat was like themost instructive thingthat maybe you gotwrong or did wrongor what have you and

**Sam：**某种意义上，scaling laws 是史上最被讨厌的预测。所有人总想说它们要撑不下去了、不可能一直这样，然而它就是一直在走。

Patrick：在这家公司的故事里，你最喜欢的无名英雄是谁？

Sam：第一个浮现在我脑海里的人是 Alec Radford。Alec Radford 大概是整个领域历史上最重要、却没那么出名的研究员，同时也是一个特别好的、顶级的人。他做的工作后来变成了 GPT 系列——还有很多其他重要的事。而且他还是那种会启发、引导、推动别人走向很多其他后来被证明极其重要的方向的人。我觉得关于他很酷的一点是：如果你去问跟他共事过的人，他们当然会说他是「一代天才、才华横溢、创新型思考者、理解和工作都极深」；但每个人在说完这些之前，都会先告诉你：他是他们见过的最善良、最正面、最好的人之一。

**Patrick：**我喜欢那些塑造性的时刻，所以在收尾的时候我想各问一个：回顾整个 OpenAI 的经历，哪个时刻、哪个章节让你最为自己的参与感到骄傲？先从另一个问起——那件你也许做错了、或搞砸了的、最有教育意义的事是什么？

[49:24] Sam Altman

what was it like tolearn from itI mean a lot ofthings have gonewrong a formativeone that went wrongwhich I haven'ttalked about muchis I think we madea mistake to try toinnovate in ourstructure in thebeginning we had avery good reason forit which is we didn'tknow how we wereever going to makemoney and we reallyat the time weren'tsure at all what wewere going to looklike when we grew upand of course we careabout our mission andwe wanted to like bestructured in a waywhere even if thetechnology went on avery fast takeoff ourmission was protectedand so we had thisnon-profit structurebut I definitelylearned something aboutwhy people don't dothat much we wouldhave saved ourselves agreat deal of pain inmany ways if we hadnot tried to innovateon our structure andfound some other way topreserve the centralimportance of themission maybe there wasno other way maybethere was for what wewere doing and kind ofthe importance of itthere was nothing otherthan an exotic structurewe could have come upwith but I reallylearned over the lastdecade a big lessonabout why people don'tusually do thatis there anything elseformative of your lifethat makes you you thatwe didn't talk aboutthis is the questionthat's always the most

**Patrick：**从中学到东西又是什么感觉？

Sam：很多事都出过错。有一件我没怎么公开谈过的、很有塑造性的错误是：我认为我们一开始试图在「组织结构」上创新是个错误。我们当时有很好的理由：我们不知道自己以后要怎么赚钱，而且那时候我们完全不确定长大后会是什么样子。当然我们在乎我们的使命，我们希望被组织成这样一种形态——即便技术走上非常快速的起飞，我们的使命也受保护。所以我们搞了那个非营利结构。

但我确实学到了一些东西，明白了为什么人们不太那么干。如果我们没有试图在结构上创新，而是找到别的方式来保住「使命的中心地位」，我们本可以在很多方面省下大量的痛苦。也许没有别的方式，也许有；就我们做的事和它的重要性而言，也许除了一个奇特的结构我们想不出别的。但过去十年我确实学到了一个大教训，明白了为什么人们通常不那么干。

**Patrick：**还有别的塑造了今天的你、而我们没聊到的事吗？这个问题对我永远是最有意思的。

[50:28] Sam Altman

interesting to methere are things likebecoming relatively immuneto people having strongopinions about me thatI think I develop later inlife realizing that manjust if you're going tobe at the center of likethis crazy revolutioneverybody's going toproject a lot of stuffonto you and you got tojust quickly learn tomake peace about thatI think there were alsothings I learned laterin life about like howto be very calm and notanxious really aboutstuff but in terms ofwhat drives me and whatI care about and how Iwant to live my life onthe whole I felt like forwhatever reason the like10 year old version ofme was pretty like fullyformed I think I justlike kind of came outthis way how about thething you're proud oflooking back on I'm mostproud of how many timeswe were right when therest of the world waswrong in an importantway that put the worldon a trajectory now thatI'm very proud to haveplayed a role in thatfeels awesome and thenalso for all the crapthat's happened thespiritual growth ofwhatever you want tocall it that I'vegotten to have oflearning just incredibleresilience and what thatdoes for like making mehappy in the rest of mylife very grateful forthat when I do these Iask everyone the same

Sam：有一些，比如对「别人对我抱有强烈看法」这件事变得相对免疫——这是我人生后来才养成的。我意识到：如果你要处在这场疯狂革命的中心，所有人都会往你身上投射一大堆东西，你得赶紧学会与之和解。我人生后期还学到一些东西，比如怎么变得非常平静、对事情真的不焦虑。但就「什么在驱动我、我在乎什么、我想怎么过这一生」而言，总体上我觉得——不知为什么——十岁版本的我已经基本成型了，我大概生来就是这样。

**Patrick：**那回头看，你最骄傲的是什么？

Sam：我最骄傲的是我们有多少次在世界其他人都错的时候是对的，而且是以一种重要的方式对——它把世界推上了现在这条轨迹，我很自豪自己在其中扮演了一个角色，那种感觉很棒。还有，撇开发生过的所有那些糟心事，我得到了某种「精神成长」（或者你想怎么叫都行）：我学到了难以置信的韧性，以及那对我人生其余部分的快乐意味着什么。我对此非常感激。

**Patrick：**我做这些访谈时，每个人我都会问同一个

[51:40] Sam Altman

traditional closingquestion what is thekindest thing thatanyone's ever done foryou I feel incrediblylucky about how manypeople have gone way outof that way to be verykind to me throughout myentire life as I'mthinking of this there'sjust this montage ofmoments from life wherepeople have beenunbelievably nice to meyesterday my kid sharedhis blueberries with mefor the first time thatwas very sweet goodmoment keep it simplethanks man thank youif you enjoyed thisepisode visit colossusdot com you'll findevery episode of thispodcast complete withhand edited transcriptsyou can also subscribe tocolossus our quarterlyprint digital and privateaudio publicationfeaturing in-depthprofiles of thefounders investors andcompanies that we admiremost learn more atcolossus dot com slashsubscribeyouyou know how small

**Patrick：**传统的收尾问题：别人为你做过的最善良的事是什么？

**Sam：**我觉得自己极其幸运，一生中有那么多人特地绕远路对我非常好。我现在想着这个问题，脑子里就是人生中一连串蒙太奇式的片段，都是人们对我好得难以置信的时刻。昨天我孩子第一次把他的蓝莓分给我，那非常甜。很好的时刻，保持简单。

**Patrick：**谢谢你，兄弟。

**Sam：**谢谢你。

【节目收尾】如果你喜欢这一期，请访问 colossus.com。你能在那儿找到本播客的每一期，配有人工编辑的文字稿。你也可以订阅 Colossus——我们的季度印刷版、数字版和私享音频出版物，深度刻画我们最欣赏的创始人、投资人和公司。了解更多请访问 colossus.com/subscribe。

[52:46]

advantages compound overtime that's true ininvesting and just astrue in how you run yourcompany your spendingsystem is your capitalallocation strategy rampmakes it smarter bydefault better data betterdecisions better economicsover time see how atramp dot com slash investas your business growsvanta scales with youautomating compliance andgiving you a single sourceof truth for security andrisk learn more atvanta dot com slash investthe best ai and softwarecompanies from open ai tocursor to perplexity usework os to becomeenterprise ready overnightnot in months visitwork os dot com to skipthe unglamorousinfrastructure work andfocus on your productridgeline is redefiningasset management technologyas a true partner notjust a software vendorthey've helped firms 5xand scale enablingfaster growth smarteroperations and acompetitive edge visitridgeline apps dot com tosee what they can unlockfor your firm everyinvestment firm is uniqueand generic ai doesn'tunderstand your processrogo does it's an aiplatform built specificallyfor wall street connectedto your data understandingyour process and producingreal outputs check themout at rogo dot ai slashinvestyou

【赞助广告】你知道微小的优势会随时间复利——这在投资上成立，在你经营公司的方式上同样成立。你的支出体系就是你的资本配置策略。Ramp 默认让它更聪明：更好的数据、更好的决策、随时间更好的经济性。详见 ramp.com/invest。

【赞助广告】随着你的业务成长，Vanta 与你一同扩展，把合规自动化，给你一个安全与风险的单一事实源。详情见 vanta.com/invest。

【赞助广告】从 OpenAI 到 Cursor 再到 Perplexity，最好的 AI 与软件公司都用 WorkOS 在一夜之间、而不是几个月内做到「企业级就绪」。访问 workos.com，跳过那些不光鲜的基础设施活儿，专注你的产品。

【赞助广告】Ridgeline 正在重新定义资产管理技术——做一个真正的伙伴，而不只是软件供应商。他们帮助各家公司实现 5 倍扩张，带来更快的增长、更聪明的运营和竞争优势。访问 ridgelineapps.com，看看他们能为你的公司释放什么。

【赞助广告】每一家投资公司都是独特的，通用 AI 不理解你的流程，Rogo 理解。它是一个专为华尔街打造的 AI 平台，连接你的数据、理解你的流程、产出真正的成果。去 rogo.ai/invest 看看。