本集简介
双语字幕
仅展示文本字幕，不包含中文音频；想边听边看，请使用 Bayt 播客 App。
您正在观看
You're watching
TBPN。今天是2025年10月8日，星期三。我们正在从TBPN超级穹顶——科技圣殿进行现场直播。
TBPN. Today is Wednesday, 10/08/2025. We are live from the TBPN Ultradome, the temple of technology.
科技。金融的堡垒。
Technology. The fortress of finance.
这个
The
资本之都。山姆·奥特曼上了Strathecari节目，与本·汤普森进行了一次精彩的访谈。我们将在周五现场采访山姆。届时会有更多细节可以深入探讨。但我们将
capital of capital. Sam Altman went on Strathecari, had a fantastic interview with Ben Thompson. We are interviewing Sam live on Friday. There will be a ton more details to dig into. But We'll
当然，我们会主要和他聊聊超级跑车。
be talking to him, of course, primarily about supercars.
是的。迈凯伦F
Yes. The McLaren f
期待
forward to
对比科尼赛克。我们必须查个水落石出。我们会把所有最重要的事情都查清楚。问题是，他要在他的坡道卡上放什么车？
versus the Koenigsegg. We gotta get to the bottom of it. We'll get to the bottom the most important all of it. Question. What car is he putting on his ramp card?
节省时间和金钱。两者兼得。轻松使用企业车辆、账单支付、会计等等，所有功能集于一处。但有几件事我们可以讨论。我想为你做个小测试。
Times money. Save both. Easy use corporate cars, bill payments, accounting, and a whole lot more all in one place. But there were a few things that we could debate. I wanted to run a little test for you.
所以在这个采访中，
So in this interview,
什么？只是笑因为泰勒
what? Just laughing because Tyler
还在准备中。是的。
It's still getting ready. Yeah.
那边还在准备中。
Still getting ready over there.
这很有趣，因为这个测试实际上涉及泰勒。所以他必须尽可能加快速度。他表现得有点滑稽。好吧。那么他们是在哪里说这个的？
It's funny because this this test actually involves Tyler. So he will have to speed up as fast as he can. He's doing kind of hilarious. Okay. So where where did they say this?
我喜欢捉弄实习生。
I love hazing interns.
好吧。萨姆·奥尔特曼说了几点。第一，我们或许应该在Twitter上少一些炒作。我们只是太兴奋了。第二，如果你看看我们交付的成果相对于五年前大多数人的预期，不得不说至少还是有些令人印象深刻的。
Okay. So so Sam Allman says a few things. One, we should probably hype less on Twitter. We just get excited. Two, if you look at what we have delivered relative to what most people would have expected from five years ago, gotta say it's at least been somewhat impressive.
我完全同意。本·汤普森说，你值得被给予怀疑的好处。我认为这很公平。他是在问萨姆关于OpenAI的事，我认为OpenAI有时会在Twitter上模糊地过度炒作一切，这可能会让一些人感到不快。我觉得也许，我不知道，这或许能解释为什么我对Sora的判断如此错误，因为OpenAI的炒作有些脱节，而现实会慢慢跟上。
I completely agree. And Ben Thompson says, you deserve the benefit of the doubt. I think that's fair. He's asking Sam about OpenAI, I think, can sometimes vaguely overhype everything on Twitter, which I think can rub some people the wrong way. I feel like maybe, I don't know, maybe helps explain why I was so wrong about Sora because the the OpenAI hype kind of mismatches and things catch up and stuff.
但萨姆·奥尔特曼在这里说了一些有趣的话。他说，我不想在图灵测试的确切定义上吹毛求疵。按照大众的理解，我们算是已经通过了它，而2020年的大多数人并不认为这会发生。所以，如果在接下来的五年里，我们真的在AI方面有所建树，有新的发现和重要的新科学，而我们却对我们的进展和兴奋有点过度炒作。我认为我们值得被宽容对待，但这很烦人，我们应该少做这种事。
But Sam Altman says something interesting here. He says, I don't wanna quibble on the exact definition of the Turing test. By the popular conception, we kind of have passed it, and most people in 2020 did not think that was gonna happen. So now if in the next five years, we really we really deliver on AI, there's discovery and important new science, and we were hyping our progress and our excitement a little bit too much. I think we deserve some grace there, but it's annoying and we should do less.
所以他说，按照大众的理解，我们已经通过了图灵测试。因此
So he says he says, by the popular conception, we have passed the Turing test. And so
我想测试一点辩论
I wanted test a little bit of a debate
是的。A
Yeah. A
前几天晚上我们讨论过这个，对吧？
couple of nights ago talking about have we?
我们讨论过吗？然后我就运行了图灵测试。我问了ChatGPT，GPT-5我们还没——你觉得我们通过图灵测试了吗？我也问了泰勒同样的问题。所以其中一个答案是来自人类泰勒，另一个来自ChatGPT。
Have we? And so and so I ran I ran the Turing test. And so I asked chat GPT, GPT five We haven't do you think we passed the Turing test? And I asked Tyler the same question. And so one of these answers is from Tyler, the human, and the other is from ChatGPT.
我会先给你答案a，然后给你答案b。你需要
So I'll give you answer a, and then I'll give you answer b. And you'll have
判断哪个是AI。是
to detect which one is AI. Is it
a还是b是AI？所以对于“你认为我们通过图灵测试了吗？”这个问题，答案a是：这是一个引人入胜的问题，没有一个简单的“是”或“否”的答案，因为它取决于你如何定义通过图灵测试，或者你是否认为这个测试仍然是一个有用的基准。但我会带你梳理一下背景，然后给出我的判断。什么是图灵测试？通过意味着什么？
a or b AI? So answer a to the question, do you think we passed the Turing test? That's a fascinating question and one that doesn't doesn't have a single yes or no answer because it depends on how you define passing the Turing test or whether you think that test is still a useful benchmark. But I'll walk you through the terrain, then give you my verdict. What is the Turing test, and what does passing mean?
艾伦·图灵在1950年提出，与其试图定义“思考”，不如问：机器能否模仿人类对话行为到足以让人类评估者在文本对话中无法可靠地区分机器和人类？随着时间的推移，人们意识到存在许多问题。该测试并不衡量理解力、意识、创造力或推理深度，只模仿对话行为和情感风格线索。所以这是答案a。然后答案b是：我问，你认为我们通过图灵测试了吗？
Alan Turing proposed in 1950 that instead of trying to define thinking, we ask, can a machine imitate human conversational behavior well enough that a human evaluator can't reliably tell the machine from a human in a textual dialogue? Over time, people have realized many problems. The test doesn't measure understanding, consciousness, creativity, or reasoning depth, only imitation of conversation behavior and emotional stylistic cues. So that is answer a. And then answer b is I asked, do you think we passed the Turing test?
答案是肯定的。我们很久以前就做到了。那么你认为哪个是AI？哪个是泰勒？
And the answer, yes. We did that a long time ago. So which one do you think is AI? Which one do you think is Tyler?
等等。抱歉，第一个是什么？
Wait. What was the sorry. The the first?
第一个回答是‘这是个引人入胜的问题，没有一个简单的肯定或否定答案，因为这取决于你如何定义通过图灵测试，或者你是否……’，而第二个回答只是简单的‘是的’。我问了泰勒。好吧。你
The first one was that's a fascinating question and one that doesn't have a single yes or no answer because it depends on how you define passing the Turing test or whether you and then the second one is just, yes. I asked Tyler. Okay. Do you
认为我们
think we
通过图灵测试了吗？
passed the Turing test?
当然。
Of course.
是的。当然。而且这样问他真是太有趣了。这确实清楚地说明了，即使你无法分辨它说英语说得好不好，它仍然有着非常独特的风格。是的。
Yes. Of course. And it was just so so funny to hit him with that. And it does it does clearly illustrate, like, the difference in how the the the that even though you can't tell the it it speaking English well, it still has, a very specific style to it. Yeah.
而泰勒只是说，嗯，是的。是的，我们通过了。但在这样做的时候，是的。你对此有何反应
Whereas Tyler just says, like, yeah. Yeah. We passed it. But in doing so yeah. What's your reaction to
？
this?
好的。泰勒，你
Okay. Tyler, you
应该回答聊天中鲍比在说什么。你完全正确。只是
should have answered what what Bobby in the chat is saying. You're absolutely right. Just
是的。我的意思是，没错。所以，显然，ChatGPT有一种非常特定的说话风格。是的。你可以看到，你知道，有一些像是Poke。
Yeah. I mean, yeah. So, like, obviously, ChatGPT has, a very specific style of Yeah. Speaking. You can see, you know, that there's some it's like Poke.
他们训练了他们训练的模型或Poke。是这么叫的吗？
It they've trained they trained the model or Poke. Is that what it's called?
哦。哦。那个AI应用。
Oh. Oh. The the AI app.
或者就像让模型用不同风格说话一样。是的。然后我觉得这样更难分辨。比如，你，我觉得即使我只是，如果你只是提示模型说，回答要非常简洁、非常精炼，我觉得会难分辨得多。
Or like the food the model to speak with a different style. Yeah. And And then I think it's much harder to tell. Like, you I I think even if I just if if you just prompt the model to say, answer very succinctly, very concisely, I think it'd be much harder to tell.
但我没必要为此提示你。我没必要告诉你回答要简洁。你只是自然而然地做到了，因为你对那很有把握。
But I didn't have to prompt you for that. I didn't have to tell you answer succinctly. You just did because you're confident about that.
嗯，当然。但这就像，如果它是系统提示的一部分，你怎么定义提示AI呢？那是提示的一部分，还是模型的一部分？如果你把它固化到权重里，那么界限在哪里？对吧？
Well, sure. But it's like how do you define, like, prompting the AI if it's part of the system prompt? Is that, like, part of the prompt or is that part of the model? If you, like, bake it into the weights, then it like, where where does the where's the line? Right?
因为它是
Because it's
就像，如果你有那种，比如，poke模型或者，是的，交互，所有这些事情，比如，它到底是在提示中还是像最后的RL步骤，这真的很重要吗？你知道
like, if if you have one of those, like, poke model or in Yeah. Interaction, like, all these things, like, we does it really matter if it's in the prompt or if it's like a RL step at the end? Like, you know
我不知道。我，我，我觉得就像我回顾我们的来回对话时，有时你会丢出一整段话。有时你问一个问题。有时你把它反馈给我。这和与GPT五的来回对话感觉非常不同。
I don't know. I I I think there's just like as I look through our text back and forth, there are sometimes when you drop a whole paragraph. Sometimes you ask a question. Sometimes you feed it back to me. And it feels very different from the back and forth with GPT five.
但它就像，是的。它确实会，嗯，
But it it's like yeah. It does get Well,
我认为一个有趣的问题是，如果你在内布拉斯加州的沃尔玛超市向两个人展示这些输出，有多少人会察觉到它是AI生成的？
I think I think one interesting question about If you showed these outputs to two people at a Walmart supercenter in Nebraska Sure. How many people would clock it?
或者有多少人会更喜欢GPT-5给出的细致入微的回答？因为很多人会——我的意思是，你没有告诉我艾伦·图灵提出图灵测试的背景，你没有给我任何背景故事，你只是直接给出了答案。
Or how many people would prefer the nuanced answer that GPT five gave? Because a lot of people would I mean, you didn't tell me about when Alan Turing proposed the Turing test. You didn't give me any backstory. You just you just you just ripped the answer.
是的，有些人可能会——嗯。
Yes. Some people might Yeah.
更喜欢那个。
Prefer that.
当然。聊天机器人会给出这种简短——
For sure. Chatbot would give this, like, sort of short
是的。
Yeah.
直截了当但缺乏细致入微的回答。
To the point answer without a lot of nuance.
但我认为，我认为很明显我们确实通过了图灵测试的一种定义。还有别的事情在发生。这显然是一个微妙的问题。但OpenAI确实低调处理了一些关键事项，这些事项却让所有人惊叹不已，人们对此印象非常深刻。所以你确实得给他们一些——你得给他们更多认可。
But I think it's I I think it is clear that we we did pass one definition of the Turing test. There's still something else going on. It's a little bit it's obviously a nuanced question. But it does but I think the point holds that OpenAI has underhyped a few key things that have just, like, blown everyone away, and people have been very, very impressed by that. And so you do have to give them a few you have to give them more credit.
比如，在很多事情上你得给他们一些信任，因为他们取得了如此大的进展。但OpenAI与Sam Altman在Strathecari上的访谈很棒。你应该去听听完整的访谈，大约半小时长，也许四十分钟。但其中有一句话让我印象深刻，就是Ben Thompson问他关于所有不同交易的性质，它们如何整合在一起，OpenAI与博通、甲骨文、英伟达、AMD和SK海力士有什么计划。
Like, you you have to give them the benefit of the doubt on a lot of these things because they've made so much progress. But the OpenAI interview with Sam Altman and on Strathecari is great. You should go listen to the full thing. It's about a half an hour long, maybe forty minutes. But one one line in here really stuck out to me, which is where Ben Thompson was asking him about the nature of all the different deals, how they all fit together, what is OpenAI planning with Broadcom and Oracle and NVIDIA and AMD and SK Hynix.
供应链上有这么多不同的合作伙伴，其中一些是直接竞争对手。计划是什么？这一切如何整合？Sam Altman给出了一个很好的回答。他说，给我们几个月时间，一切都会变得清晰，我们将能够谈论整体计划。
It's so many different partners in the supply chain, some of them direct competitors. What what is the plan? How does this all come together? And, and Sam Altman had a great response. He said, give us a few months and it'll all make sense, and we'll be able to talk about the whole.
我们并不像看起来那么疯狂。是有计划的。所以这应该是一个争论点。你应该相信这个计划吗？计划是什么？
We are not as crazy as it seems. There is a plan. And so this is this should be a point of debate. Should you trust the plan? What is the plan?
我不确定这有多重要，但深入探讨确实很有趣。所以我想简要回顾一下AI战争的历史，因为昨天我们接受了法国电视台的采访，非常有趣，因为他们对两个月前的事情着迷不已，就是AI人才战争。
I don't know that it matters too much, but it's certainly fun to dig into. And so I wanted to give a little bit of a brief history of the AI wars because yesterday, we did an interview for French television, and it was absolutely hilarious because they were obsessed with the current thing from two months ago, the AI talent wars.
实际上，就像是
It was actually, like
那是
It was
真的，非常棒的团队。
really, really nice team.
我们我们有一个有趣的团队，与观众聊天很酷。
We we have fun fun team chatting with cool to their audience.
是的。完全同意。但感觉像是欧洲刚放完暑假，是的。他们完全错过了人才争夺战，是的。然后变得对此极度痴迷。
Yeah. Totally. But it it felt like Europe got off of summer break Yeah. And they had totally missed the talent wars Yeah. And then become absolutely obsessed with them.
人们谈论说，哦，LinkedIn下周会发现Meta挖角的事情，就像Axis在谈论时那样。
People talk about, oh, LinkedIn's gonna find out about meta poaching, like, next week when Axis talking to
回头看日期，我当时想，好吧。我们在6月1日讨论过这个。法国像是要彻查到底。10月8日。
back at the dates, I was like, okay. We talked about this on June 1. France is, like, gonna get to the bottom of it. October 8.
是的。但我的意思是，他们说整个视频团队，做那种制作确实需要时间。但他们痴迷于数字。暑假。就像Jordi不断提供更多背景，比如，好吧。
Yeah. But, I mean, they said a whole video crew, it does take time to do those types of productions. But they were obsessed with the numbers. Summer break. Like, Jordi kept giving more context on, like, okay.
嗯，有一个幂律分布，而且，你知道，进行收购式招聘，比如收购像Scale AI这样的公司来让Alex Wang加入团队，与仅仅是一个数据库管理员的情况截然不同。
Well, there's, a power law and, like, you know, acquire doing an acquihire, like, buyout of something like a Scale AI to get, Alex Wang on the team is wildly different than what just like a database manager.
这位工程师赚了多少钱？
Money did this engineer make?
是的。他们想要所有东西的具体数字。他们本来会非常高兴的。他们
Yeah. They wanted the number for everything. They they would have been so happy They
正在做的是
were doing the
有多少垃圾邮件。是的。多少钱？他们就想让你直接说出来，比如，史蒂夫，4000万美元。这个人，6000万美元。
how much spam. Yeah. How much money? They wanted you literally just to say, like like, Steve, $40,000,000. This person, $60,000,000.
他们想要，比如，威胁名单上每个人的确切金额。没有什么比这更能让他们高兴的了。而且，你已经尽力了。
They wanted, like, finite dollar amounts on everyone on the menace list. Nothing would have made them happier. And, you did your best.
而且，是的。我试图解释说，在美国，在美国，我们不必开源所有的薪资数据。我们无法确切知道每个offer的具体数额。
And, yeah. I tried to explain that in America, America, we don't have to open source all the payroll data. We can't know exactly what each offer was.
所以，虽然这里那里有一些泄露的信息，但大部分都是方向性的。但确实。七月确实是AI人才争夺战。现在回想起来，七月的主要故事是什么？100%是人才争夺战，然后这件事就这么发展下去了。
And so there's, like, a few leaks here and there, but it's mostly, like, directional. But yeah. July really was the AI talent wars. Now that I reflect on it, like, what was the main story of July? It was 100% the talent wars, and then that kind of worked its way through.
我们在节目里讨论过这件事。当时我们正在做那些交易卡，它们迅速走红。几周后，《纽约时报》在一篇文章和一期《每日播报》节目中报道了此事，那期节目还特别提到了我们。感谢《纽约时报》的各位。
We were talking about it on the show. We were doing those trading cards. Those were going viral. Then The New York Times covered it a couple weeks later in a post and an episode of The Daily that actually featured us. Thank you to the folks over at The New York Times.
然后现在法国也开始关注了，这当然很有趣，但你知道，这对他们来说会是个有趣的节目。这里的背景是，Meta似乎在开源LLM战略上落后了。DeepSea在这方面迎头赶上，OpenAI的消费者飞轮效应正在发酵，Anthropic则凭借编码B2B API飞轮发力，而谷歌则依托DeepMind卓越的研究团队、自研的TPU芯片、成熟的GCP云业务以及庞大的产品矩阵——他们简直是无孔不入。
And then and then now France is getting to it, which is, of course, funny, but, you know, it'll be an interesting show for them. And the the the the history here is, you know, Meta, it seems like they were falling behind in open source LLM strategy. DeepSea could caught up on that front, and the consumer flywheel was cooking at OpenAI. Anthropic was cooking with, the coding b to b API flywheel, and Google was delivering on the back of DeepMind's incredible research team, their custom silicon TPU, their mature cloud business with GCP, and their sizable product surface area. They're able to just stuff it everywhere.
Meta当时还没有真正找到能够形成复合增长、明确通向重要地位的良性循环。所以扎克伯格进入了创始人模式，嗯，我们都看到了那些令人瞠目的薪酬细节。八月份相对平静些，我想大家都有这种感觉。GPT-5确实发布了，但更像是：所有人都期待疯狂突破，结果却是产品运作方式上更战术性的调整。
Meta quite hadn't quite found compounding flow, a compounding flow that really set them on a clear path to significance. So Zuck went founder mode, and, well, we've all seen the eye popping offer details. August was a bit quieter. I think we all felt that. GPT five did launch, but it was kind of like, everyone was expecting crazy stuff, then it was a more tactical move in terms of how the product actually works.
它并非那种遥遥领先的颠覆性模型，体验上没有太大不同，但对消费者来说是更好的体验，是更优秀的消费级产品创新。不过到了九月一号——是的。
It wasn't this, like, insane model that is just, you know, light years ahead. It doesn't feel very different, but it's a better experience for the consumer. It's a better consumer product innovation. And and but in September 1 yeah.
关于人才争夺战有一点——没错。那段时间感觉特别明显，大概持续了一个月对吧？现在还在继续。
One point on the on the Talent Wars Yep. Is it is it it felt obvious during that period, which was really about a month. Yep. Right? It's still happening.
当然，人才市场永远都会高度竞争。但当时人们都在问：这会成为新常态吗？嗯。即便在当时也很明显，某些方面必须做出让步——要么超大规模企业的CEO们需要提供更丰厚的
Of course, talent the talent market is always gonna be hyper competitive. But at the time, people were asking the question of like, is this the new normal? Mhmm. And it felt very obvious even at the time that that something was gonna have to give either the CEOs of of the hyperscalers were gonna need to make more
没错。
Yep.
或者说，顶尖AI研究员的平均薪资可能会降低。是的。感觉薪资底线可能比夏季之前有所提高，但我并不确定。你知道，研究员们不会真的在门洛帕克或旧金山逛街时，拿着100,000,000美元的offer说：‘我这里有个一亿的报价’。
Or the average elite AI researcher was gonna have to make less. Yep. And it feels like the floor has maybe reset higher than where it was going into the summer. But I don't know that, you know, researchers are walking around, you know, Menlo Park, you know, shopping Yeah. San Francisco, you know, shopping offers saying, I got a 100,000,000 here.
是啊。你能匹配吗？能超过吗？完全就是这种情形。
Yeah. Can you match it? Can you beat it? Totally. Type of thing.
感觉已经趋于常态化了。再说，这些人仍然是全球收入最高的人群之一。但是，确实，我认为研究员并没有一条明确的路径能年入一亿美元。
It feels like it's normalized. Again, these are still some of the best paid people in the entire world. Yeah. But but certainly, I don't think there's necessarily a clear pathway to, you know, making a $100,000,000 a year as a researcher.
没错。即使是那些庞大的财富500强公司，想要讲AI故事，也完全满足于不参与AI人才争夺战，不试图自主研发前沿模型。IBM刚涨了4%，就因为宣布将使用Claude的API。对吧？我们在博通也看到类似情况，我们会说‘让我们放弃...’
Yeah. Like Even even companies that are, you know, massive Fortune 500 companies that want an AI story are very much content to not participate in the AI talent wars, not try and get to the frontier on their own models. IBM just popped 4% on a deal that they just are gonna be using Claude as an API. Right? And and we see this with Broadcom and we'll go Let's give up for
国际商业机器公司。
the international business machines.
我喜欢这个说法。但是，确实。有很多方式可以为公司打造AI叙事，而不必真的去挖角那些极度抢手的50名研究工程师。所以九月份开始升温，我们绝对又回来了。
I love it. But but those yeah. There there are many ways to, like, bring an AI story to your company without actually going and trying to poach, you know, 50 research engineers that are in extreme high demand. So September started heating up. We were definitely so back.
OpenAI推出了Pulse新闻摘要、Sora AI短视频、智能体构建器和智能体商务。这像是四条严肃的业务线。当然有风险，其中一些可能不会在未来几年成为大规模产业，但每一条看起来都有潜力在规模化后产生数百亿美元的收入。所以他们面前还有一大堆机会。
OpenAI dropped Pulse, news summaries, Sora, the AI TikTok, agent builder, agent commerce. Like, those are four serious business lines. Of course, there's risk. Some of those will probably not be be things be massive scaled, you know, properties in years, but they each feel like they could be generating tens of billions of dollars at scale. And so they have a whole bunch more opportunities in front of them.
即使没有朝着超级智能或你所谓的任何重大突破，它们都解决了一个明确的问题，比如谷歌新闻就是一个例子。人们获取新闻摘要。苹果有新闻产品。现在，OpenAI有了Pulse。而这很可能是一条重要的业务线。
Even if there isn't a major breakthrough towards super intelligence or whatever you wanna call it, they all solve a clear problem where, like, people Google News is a thing. People get news summaries. Apple has a news product. Now, OpenAI has Pulse. And that's just, like, probably a big business line.
你过去三天用过它吗？
You used it in the last three days?
我收到了推送通知。我一直在大量使用它。我昨天确实打开了Sora生成了一段视频，我发给你了，Tyler。你收到通知了吗？我给你发了大卫·福斯特·华莱士将《无尽的玩笑》描述成TikTok的视频。
They I get the push notification. I have been using it a ton. I did pop on Sora yesterday to generate a video of, I sent this to you, Tyler. Did you get the notification? I sent you David Foster Wallace describing Infinite Jest as a TikTok.
在Sora上，你发给我这个？是的。哦，我我觉得我没开通知。
On Sora, you sent me this? Yes. Oh, I I don't think I have notifications on.
嗯，这些事情需要时间慢慢酝酿。谁知道它们最终会落在哪里。但无论如何，在我看来，它们都具备了产品市场契合的早期特征。它们似乎都有明确的经济模式。我不确定这四者都会成功，但即使只有几个成功，你也会看到，你知道，又多几个价值数十亿美元的业务，这在估值方面很容易支持OpenAI或证明新交易的合理性。
Well, these things take time to to simmer. Who knows where they're all what where they will all land. But regardless, like, they're they're they all have, like, the early trappings of product market fit, in my opinion. They all seem to have clear economics. They they like, I don't I don't know that all four will hit, but if even a few hit, you're looking at, you know, a couple more multibillion dollar businesses, which is easy to underwrite OpenAI on the valuation side or justify new deals.
所以服务这些新业务线，老实说，仅仅是扩大ChatGPT的使用就需要大量的计算能力。在昨天的节目中，你总结得很好。你说，我现在就把OpenAI看作一个超大规模计算提供商。他们需要做谷歌、亚马逊、微软、Meta过去二十年所做的一切，但他们需要更快地完成。所以萨姆·奥特曼正在尝试，并且基本上有望在短短几年内做到。
So serving these new business lines and, honestly, just scaling up ChatGPT usage is gonna require a lot of compute. On yesterday's show, you summed it up really well. You said, I just think of OpenAI as a hyperscaler now. They need to do everything Google, Amazon, Microsoft, Meta have done over the past two decades, but they need to do it faster. And so Sam Altman is trying and basically on track to do it in just a few years.
甲骨文、英伟达、AMD、博通、SK海力士等都被请来共同规划未来五年的更清晰前景。他们基本上都认同了萨姆的愿景。他们都说，是的，我们认为这将需要大量的计算能力。
Oracle, NVIDIA, AMD, Broadcom, SK Hynix, and more have all been brought to the table to map out a clearer view of what the next five years looks like. And Did they call yeah. All of them basically bought into Sam's vision. They're all like, yeah. Like, we think this is gonna be a lot of compute.
我们认为这将产生大量收入。在本·汤普森的那次采访中，他非常明确地表示，他认为这将由OpenAI的收入来资助。让我找一下这段话。他说，这些交易的价值是天文数字。我认为《金融时报》刚刚计算出的数字是一万亿美元。
We think this is gonna generate a lot of revenue. And so in that interview with Ben Thompson, he's pretty clear that he just says, like, I think this is going to be funded by OpenAI revenue. Let me find this. So he says, these deals are worth an astronomical amount of money. I think a trillion dollars was what the Financial Times just calculated.
我们这里有这个。这个庞大的交易网络。我们看看。OpenAI的计算交易额超过1,000,000,000,000美元，押注于未来的盈利能力，并且列出了博通。甚至谷歌、亚马逊、Meta、微软和软银都列在这里。
We have that here. This massive web of deals. Let's see. OpenAI's computing deals exceed 1,000,000,000,000 in bet on future profitability, and it lists out Broadcom. Even Google, Amazon, Meta, Microsoft, and SoftBank are listed on here.
甚至Anthropic也在这里，还有你提到的CoreWeave。是的。所以这是一个惊人的金额。本·汤普森问他，你期望谁来支付这笔费用？这是关于这些交易的核心问题吗？
Even Anthropix on here, CoreWeave, you you mentioned. Yep. So it's it's a staggering amount of money. And Ben Thompson asks him, who do you expect to pay for it? Is this a matter of what these deals are about?
你保证会购买其产出，所以你需要这些公司投资？萨姆·奥尔特曼说，是的。我期望OpenAI的收入来支付它。所以这笔收入可能是一个混合体。
You guaranteeing you'll buy the output of it and you need these companies to invest? And Sam Altman says, yeah. I expect OpenAI revenue to pay for it. And so that revenue And might be a mix.
问题来了。是的。周五应该和萨姆聊聊这个。周五。但我一直在想的是，付费ChatGPT用户的市场规模有多大。
Here comes a question. Yes. Should chat with Sam about it on Friday. Friday. But something I've been thinking about is how large is the market for paying ChatGPT users.
嗯。对吧？他们一直在印度试验更便宜的计划。他们有很多人，尤其是在我们这个小圈子里，每月支付200美元。
Mhmm. Right? They've been experimenting in India Mhmm. With cheaper plans. They've got plenty of people, especially in our little bubble, that are paying $200 a month.
是的。但问题是，这个市场的天花板在哪里？他们能否将其规模扩大到
Yeah. But the question is how what what is the ceiling on that? Are they gonna be able to ramp to
付费消费者收入的增长会放缓吗？它增长得非常快。
The paid consumer revenue ramp, will it slow down? It's grown very quickly.
订阅产品年化收入达到500亿后它会放缓吗？还是说在他们转向更多交易型商业领域时会出现放缓？是的。你知道，广告和从活动中抽成，是的。他们在平台上广泛驱动的经济活动？
Will it slow 50,000,000,000 of of annualized revenue on subscription products? Or is there going to be a slowdown while they transition to more transaction commerce space Yep. You know, ads and taking a cut of the activity Yep. The the economic activity that they're driving on the platform broadly?
是的。所以，我的意思是，当我考虑这个问题时，我认为AgenTek的商业推荐费、联盟收入在免费用户中可能会非常非常快地增长。他们有8亿周活跃用户。这可能会迅速增长。Agent Builder推动了更多API业务。
Yeah. And so, I mean, when I think about it, I think that AgenTek commerce referral fees, affiliate revenue could ramp very, very quickly amongst free users. They have 800,000,000 weekly active users. That could ramp very quickly. Agent builder drives more API business.
你如何
How you
你预计这会如何增长？我可以想象在很多不同的场景下它会以惊人的速度增长。但你看到哪条路径能让他们开启货币化开关，并真正快速地规模化这种交易性收入？
how you expect that to ramp? I can imagine in a number of different scenarios where that ramps incredibly quickly. But what path do you see that allows them to flip the switch on monetization and actually scale sort of this, like, transactional revenue extremely quickly.
你是什么意思？我我会假设，就像，它随时都会开启。
What do you mean? I I would I would assume, like, it's going to flip, like, any day now.
那是通过Shopify实现的吗？
And that's and that's through Shopify?
是的。目前来说一般般，我的意思是，我最近刚测试了这个。我们当时在找一个新的麦克风支架。实际上是这些麦克风支架。我看到我在看Doug Dumurro的《This Car Pod》，一个很棒的汽车播客，我注意到我喜欢那些麦克风支架的外观。
Yes. So so right now I mean, I I tested this just recently. We were looking for a new microphone stand. Actually, these microphone stands. I saw that I was watching Doug Dumurro on This Car Pod, a fantastic car podcast, and I noticed that I liked though the way those stands those microphone stands looked.
所以我截了个图，裁剪了一下，输入到ChatGPT里，然后说，帮我找到这个麦克风支架。它找到了。然后我把链接发给了Ben。
So I took a screenshot. I cropped it. I put in ChatGPT, and I said, find me this microphone stand. It did. And then I sent the link to Ben.
OpenAI一分钱也没赚到。因为他直接在那里买了。
OpenAI didn't make a dime. Because he just bought it there.
没错。
Yep.
但如果我把我的Ramp卡保存在OpenAI的ChatGPT里，我可能已经存了，我甚至都不知道。我本可以直接发消息说，是的，买下来送到Ultra Dumb。
But if I had just had my ramp card saved in OpenAI in in ChatGPT, which, like, I might already. I don't even know. I could have just texted, yeah, buy it and send it to the and ultra dumb.
所以你认为OpenAI与亚马逊、Shopify以及其他一些电商平台有合作？
So you think that OpenAI does deals with Amazon and Shopify and a number of other ecommerce platforms
当然。
For sure.
并且能够有效地切换开关吗？
And is able to effectively flip the switch?
我认为他们已经做到了。比如，我我我觉得很多这种代理式商务的东西现在已经上线了。如果他们可能还没开始抽成，我不会感到惊讶，但肯定已经准备好要抽成了。
I think they already have. Like, I I I think that I think that a lot of this agentic commerce stuff is, like, live now. I wouldn't be surprised if they're maybe not taking a cut yet, but, like, certainly set up to take a cut.
是的。嗯，我认为对他们来说重要的是，他们必须披露何时抽成。对吧？如果你在做联盟营销，你有一个博客，你在发送流量到某个地方并从中获利，你需要披露这一点。
Yeah. Well, I think it's I think it's important for them to they will have to disclose when they're taking a cut. Right? If you're doing affiliate Yeah. Product marketing, and you have a blog and you're and you're sending traffic somewhere that you're getting a piece of, you need to disclose that.
没错。所以我认为当他们在规模上这样做时我们会知道。是的，因为我们都会在产品中看到它。
Yep. And so I think that we will know when they're doing that at scale Yep. Because we'll all see see it in the product.
没错。所以，Sam提供了更多细节，关于他喜欢广告的地方和不喜欢的地方。他说，首先，关于Instagram广告这一点，那实际上让我觉得，好吧，也许广告并不总是那么糟糕。我喜欢Instagram广告。
Yep. And so, Sam gave more, details on, where he likes ads, where he doesn't. He says, first of all, on the Instagram ads point, that was actually the the thing that made me think, okay. Maybe ads don't always suck. I love Instagram ads.
它们为我增加了价值。我发现了以前绝不会找到的东西。我买了一大堆东西。我真的很喜欢Instagram广告。我认为Meta有很多方面我都很尊重，但能把这一点做得这么好对我来说是件出乎意料很酷的事。
They've added value to me. I found stuff I never would have found. I bought a bunch of stuff. I actively like Instagram ads. I think there's many things I respect about Meta, but getting that so right was a surprisingly cool thing for me.
除此之外，我把互联网上的广告视为一种攻击。而Ben Thompson说，嗯，我认为这就是问题所在，大多数人认为搜索主要是攻击。通常，自然搜索结果会有你想要的东西，然后我会买广告排在前面。我一直为Meta辩护。我觉得，实际上，这才是我们应该感到高兴的广告模式。
Other than that, I viewed ads on the Internet as sort of like attacks. And Ben Thompson says, well, I think that's I I think that's the problem, the is that, most people think search is mostly attacks. Usually, the organic results will have what you want, and then I'm gonna buy ads to be on top. I've always defended Meta. I am like, I think, actually, this is the ad model we should be happy about.
Sam说，我同意这一点。然后Ben Thompson问，那么你认为在这种背景下，你的商业可能性如何？Sam说，我的意思是，我相信我们或许可以开发一些很酷的广告产品，这对用户来说是净收益，也有助于我们与用户的关系。但我还不知道具体是什么。我并不是说我们已经有了现成的广告模式。
And Sam says, I agree with that. And so, Ben Thompson says, so how do you think about your possibilities with business in that context? Sam says, I mean, again, I believe there's probably some cool ad product we can do that is a net win to the user and sort of positive to our relationship with the user. I don't know what that is yet. I'm not like, here is our ad model already.
他正在研究，但还没准备好分享具体内容。Ben Thompson说，但联盟营销似乎明显是个赢家。你不用担心它会蚕食你的广告业务。Sam说，是的。
He's working on it. He's not ready to share what it is. And Ben Thompson says, but affiliate seems like a clear win. It's not like you have to worry about cannibalizing your ad business. And Sam says, yeah.
那看起来确实是个明显的赢家。所以如果联盟营销不很快推出，我会非常震惊。这感觉像是另一个潜在的，我不知道，可能是100亿美元的收入池，可以在付费Pro和Plus订阅用户达到饱和峰值时逐步增长——就像每个想要的人都已经拥有了一样。是的。然后免费用户的联盟营销变现就会加速。因此OpenAI的整体收入增长……
That seems like a clear win. And so I would be shocked if if, affiliate doesn't come very, very quickly, and that feels like, another another pool of potentially, I don't know, $10,000,000,000 of revenue that could ramp while paid, you know, pro and plus subscribers are kind of reaching their peak saturation, like everyone who wants one has one Yep. Well, then the affiliate monetization of the free users is ramping. And so the overall revenue ramp for Open Ads
我很想——我的意思是，我认为我们可能永远看不到这个。是的。但我很想看看他们对所驱动的购买活动每日美元价值的估算，从旅行到消费品，是的，到时尚，是的，等等。
I would love to I would love I mean, I don't think we'll ever see this. Yeah. But I would love to see their estimates of how many how what the dollar value, just the daily dollar value of the purchasing activity that they're driving, everything from travel to consumer goods Yep. To fashion Yep. Etcetera.
这就是我关于OpenAI抽成率的观点。抽成率会是多少？就像，现在在OpenAI平台上发生的商业价值有多少，而他们一分钱都没抽？是的。人们基本上是在上面做购买决定，你可以在任何注意力产品中看到这一点。
This is my point about the OpenAI take rate. Where will the take rate be? Like, what is the value of commerce that's happening on top of OpenAI right now that they aren't taking anything of? Yeah. Just people effectively making their purchase decision on and and you can view this when in any sort of attention product.
人们通过观看Doug DeMuro的视频做了大量关于买什么车的决定。但只有一小部分人会去Cars and Bids并在那里买车。也许那时会展示特定汽车的广告，但有大量商业是由YouTube、播客、谷歌、亚马逊、Facebook驱动的。某些平台抽成更多。但我认为肯定有大量商业正在发生。
People make a ton of decisions about what car to buy by watching Doug Dumurro. Only a small fraction of them go to Cars and Bids and buy the car there. And maybe there's an ad for a specific car that's shown at that moment, but there's a ton of commerce that's driven by YouTube, by podcasts, by, you know, Google, by Amazon, by Facebook. Certain platforms take more. I but there has to be a ton of commerce that's happening.
很多
A lot
商业活动已经受到ChatGPT影响或中介的例子。昨天，伯恩斯坦的Stacy Rasgon在CNBC上接受采访，我来读几句采访内容。采访者问，这可能会出什么问题？Stacy说，需要注意的是，他谈的是AMD的交易。嗯。
of commerce activity that's that's influenced or or, like, intermediated by ChatGPT already. So Stacy Rasgon over at Bernstein was on CNBC yesterday, And I'll read a couple lines from the interview. The interviewer asked, how could it go wrong? And Stacy says, it should be noted that the chips in he's talking about the AMD deal. Mhmm.
需要注意的是，所讨论的芯片目前还不存在。AMD从未制造过机架。他们当然从未在这种规模上部署过任何东西。而认股权证很可能会继续加剧最近在该领域积累的所谓循环担忧。当然，XAI和Nvidia之间也有一些关于他们新交易的消息泄露出来。嗯。
It should be noted that the chips in question do not exist yet. AMD has never built racks. They certainly never deployed anything at this scale before. And the warrants will likely to will likely continue to fuel the, quote, circular concerns that have been building in the space lately. And of course, XAI and Nvidia have have there was some news that leaked around their new deal Mhmm.
但我们先不深入讨论那个。所以最近在该领域积累的循环担忧。在这种情况下，感觉比NVIDIA的交易更加迂回。至少他们通过现金投资获得了OpenAI的股票，而AMD放弃了股权，除了收入外什么也没得到。当然，这一切都取决于Altman继续沿着他的轨迹前进。
But we won't get into that now. So circular concerns that have been building the space lately. And in this case, feels even more roundabout than NVIDIA's deal. Least they are receiving OpenAI stock for their cash investment while AMD is giving up their equity while receiving nothing beyond the revenue in return. And of course, this all depends on Altman continuing on his trajectory.
不过公平地说，现在行业里的每个人都依赖于此。Sama有能力让全球经济崩溃十年，或者带我们所有人进入应许之地。而现在，我们不知道哪一种是可能的。采访者稍微反驳了一下说，嗯，难道没有中间地带吗？对吧？
Though to be fair, everyone in the industry now depends on this. Sama has the power to crash the global economy for a decade or take us all out to the promised land. And right now, we don't know which is in the cards. The interviewer pushed back a little bit and said, well, isn't there kind of a middle ground? Right?
你知道，介于应许之地和十年寒冬之间的某个地方。是的。这里还有Brent Donnelly的另一篇帖子，他分享了今早Bloomberg上的这张图表。Joe Weisenthal发布时还加了个内容警告。非常搞笑。
You know, somewhere between the promised land and a ten year winter. Yep. There's another post here from Brent Donnelly who is sharing this graphic that was in Bloomberg this morning. Joe Weisenthal posted it with like a content warning on it. It was hilarious.
它只是展示了NVIDIA和OpenAI如何推动AI赚钱机器，显示了OpenAI、AMD、是的、XAI、Oracle、Intel Core、Weave、Nebius、Microsoft等等所有这些参与其中的不同玩家。Brent说整个股市都依赖于这个衔尾蛇会永远持续下去的想法。它开始构成重大的经济和金融稳定风险。说“继续跳舞”是挺有趣的，但每个人都认为他们能在其他人退出之前先退出。
And it's just showing how NVIDIA and OpenAI fueled the AI money machine showing like, you know, OpenAI, AMD Yeah. XAI, Oracle, Intel Core, Weave, Nebius, Microsoft, all these different players that fit in. And Brent is saying the entire stock market depends on the idea that this Ouroboros will continue forever. It's starting to pose meaningful economic and financial stability risks too. It's fun to say quote unquote keep dancing, but also everyone thinks they can get out before everybody else gets out.
好时光。我确实认为，基本上每个人都已经在思考如何把握这个市场的时机了。对吧？
Good times. And I do think that's generally everybody's just trying to already thinking about how do I time this market. Right?
是的，当然。很多关于泡沫的讨论。我们已经讨论过这个了。现在是牛市，大家都在谈论泡沫。
Yep. For sure. Lots of bubble talk. We've we've we've covered this. It's a bull market and bubble talk.
是的。我我确实愿意给Sam一个怀疑的好处，给我们几个月时间，一切都会明朗。我仍然认为了解计划是什么很有趣。Tyler，我很想知道你的想法。OpenAI是在打造自己的芯片、自己的云平台，还是所有这些都在做？
Yeah. I I do I'm I'm I am willing to give Sam the benefit of the doubt, the give us a few months and it'll all make sense. I still think it's interesting to know what the plan is. Tyler, I'd love to know what you think. Is OpenAI building their own chip, their own cloud platform, all of the above?
他们是专注于让当前的GPT-4规模模型尽可能高效，还是在为更大的预训练做准备？进展是停滞了，还是他们仍然极度坚信AGI？而AGI目前对Sam来说到底意味着什么？这些是我的一些问题。你觉得我们还应该问Sam什么？
Are they focused on making current GPT four size models as efficient as possible, or are they gearing up for a bigger pretraining run? Is progress stagnating, or are they still extremely AGI pilled? And what does AGI even mean to Sam currently? Those are some questions that I have. Anything else that you think we should ask Sam?
是的，我不知道。我的意思是，我我内心坚信AGI的那部分，非常希望他们正在用新的计算资源训练下一个模型，比如一个更大更大的模型，能够有更强的推理能力。
Yeah. I don't know. I mean, I I think that the AGI pilled part of me, like, desperately hopes that they're using new compute to train the next model, like, a bigger bigger model that's gonna, you know, more reasoning.
是的。
Yep.
但很可能合理的说法是，很多资源会用于效率提升，让他们训练出更小的模型，这样对，嗯，API成本更有利。
But it probably is reasonable to say that, like, a lot of it will just be going to efficiency gains that'll let them train smaller models that'll be better for, you know Yeah. API costs.
即使进展停滞，仅仅是这样扩散到经济中，也能在各个地方增加大量价值。好了，我们的第一位嘉宾快到了，但与此同时，让我介绍一下Restream。一次直播，30多个目的地，多平台推流，让你的观众无论在哪里都能看到。这个节目就是通过Restream主持的。我们请到了Ramp的Kareem，他就在
There is just like an economic impact to just taking even this is even if progress stagnates, it it just diffuses through the the economy and adds a bunch of value all over the place. Well, we have our first guest almost here, but in the meantime, let me tell you about Restream. One livestream, 30 plus destinations, multi stream to reach your audience wherever they are. This this show is hosted via Restream. And we have we have Kareem from Ramp in the
开始了。这里是Remedy Room，现在
Here we go. Remedy Room, and now
他正在参加电视节目。
he's in the TV panel show.
他来了。
There he is.
欢迎来到直播。
Welcome to the stream.
Ramp先生，欢迎来到节目。嘿，
Mister Ramp, welcome to the show. Hey,
伙计们。Kareem，你怎么样？也很高兴见到你。
guys. How are doing, Kareem? Good to see you too.
怎么了？
What's happening?
先带我们过一遍新闻，然后我想问一大堆关于你们实际如何运用AI以及你们获得的那个令牌奖励的问题。我想真正地情境化一下，比如一个公司虽然意识到所有的炒作，但真正专注于驱动商业价值，实际上是如何实施AI的。
Take us through the news, and then I wanna ask a ton of questions about how you're actually using AI and and and the the, you know, the the token award that you got. I wanna I wanna really contextualize, like, how a company that doesn't you know, is is aware of all the hype, but truly focused on driving business value is actually implementing AI.
是的。我的意思是，当你问起新闻时，我几乎有点困惑。就像，是什么新闻——
Yeah. I mean, that's, well, when you ask when you ask about the news, I'm almost confused. Like, what what news are
你在说什么？发生什么事了？哪一个新闻？
you talking about? What's going on? Which which one of them?
有趣的是，我认为互联网似乎对我们聘请了一位新的首席财务官感到兴奋。
The fun one is I think the Internet seems to be excited that we hired a a a new CFO.
哦，是的。是的。
Oh, yes. Yes.
很快就会向全世界介绍。不过，说真的，我们计划在10月14日在纽约举办一个非常有趣的活动，所以对此非常兴奋。是的。
That will be presenting to the world very soon. But, no. In all seriousness, like, we're we're we're gonna have a very fun event, planned for October 14 in New York, so very excited about that. Yeah.
到目前为止的回答非常棒。户外广告活动看起来很美。而且，是的，它以非常强大的方式突破了。我一直很享受观看它。完全同意。
Fantastic response so far. The out of home campaign looks beautiful. And, yeah, it it it's it's breaking through in a really powerful way. I've I've been enjoying watching it. Totally.
但带我们了解一下AI代理的新闻。哦，我们和Eric聊过这个。我们报道了发布，那是那种感觉非常...我不知道，感觉几乎是战术性的发布。
But take us through the AI agent news. Oh. We we talked to Eric about that. We we we covered the launch, and it was one of those launches that feels very I don't know. It felt almost like tactical.
就像，它不是那种疯狂的惊喜。使用最好的工具似乎是合乎逻辑的。你们一直使用最好的工具。你们几年前就在用GPT 3.5来分类东西了。所以你们从未落后于潮流。
Like, it it wasn't, like, some crazy surprise. It seemed logical that you would use the best tools. You've always used the best tools. You were using, what, GPT 3.5 to, you know, classify stuff, like, years ago. So you've never been behind the curve.
但当我们和Eric交谈时，他说客户的采用率实际上非常显著。那么你们想改进什么？学到了什么？然后新的发布是什么？
But then when we talked to Eric, he said, like, the actual adoption from customers has been remarkable. So what did you want to improve? What is the what have the learnings been? And then what's the new launch?
我的意思是，我想我们上次聊天时，我们谈到了我们的政策代理的发布。政策代理稍微容易理解一些。对吧？大多数公司都有费用政策。费用政策就像是给代理的一套英文指令。
I mean, I I I guess we last time we chatted, we were talking about the launch of our our policy agents. And policy agents are a little bit hard easier to understand. Right? Most companies have expense policies. Expense policies act as a set of instructions in English for an agent.
你给它足够的工具。你给它联系人，它可以在后台操作，分类交易，并填补交易推理中可能存在的任何空白，比如是否应该计入或不计入。然后你从这个扩展，开始进入财务的其他领域和公司必须处理的其他工作流程，然后很自然的下一步就是账单支付。对吧？应付账款，AP。
You give it enough tools. You give it contacts, and it can operate in the background and classify transactions and cover any gaps that there might be in the reasoning around the transaction, like, should it be in or out. And then you expand from that, and you start wanting to go into other areas of of finance and other workflows that companies have to deal with, and then very natural next steps is is bills they get paid. Right? Accounts payable, AP.
每个公司都必须支付账单。每个公司都会收到账单。不过，区别在于当我们谈论账单时，公司并没有账单支付政策。大多数公司都没有。是的。
Every company has to pay bills. Every company receives bills. The difference, though, is when we talk about bills, companies don't have a bill payment policy. Most companies don't have that. Yeah.
公司考虑的方式是，嗯，我就雇一个团队，告诉他们我是怎么做的，给他们一些指示。我的意思是，最接近的可能是一份职位描述。
The way companies think about it is like, well, I'll just hire a team, and I'll show them how I'm doing it, and I'll give them some instructions. I mean, the closest thing to that that you have might be, like, a a job description.
这是
It's
就像，我希望有人来审核账单，确保它们不是欺诈性的，也许还要确保它们符合我们不断变化的会计标准，并且我希望确保它们从最优账户支付，以获取最高收益。
like, I want someone to come in and review the bills and make sure that they're not fraudulent and maybe make sure that they follow our evolving accounting criteria, and I wanna make sure that they get paid from the most optimal account in a way that earns us the most yield.
在大多数初创公司里，如果金额超过一千美元，就去问创始人。没错。超过一千美元就要和CEO再次确认。而如果低于一千美元，这就是为什么所有欺诈都发生在那些伪造发票上，比如，
In most startups, it's like, if it's over a thousand bucks, ask the founder. Exactly. Double check with the CEO if it's over a grand. And if it's under a grand and that's why all the fraud happens where you get a fraudulent invoice for, like,
850美元。
$850.
故事是这样的，有人多年来一直给谷歌发送发票，而他们一直在支付。但是，是的。我经常思考这个问题。对吧？就像对于每一笔交易，都伴随着很多风险，因为你会有嗯。
Story was that person that was sending Google invoices for years, and they're just paying them all. But, yeah. I think I think about this a lot. Right? It's like you either any any for every transaction, there's like there's like a lot of risk going into it because you have Mhmm.
一方可能会出错，发送发票
One side that could be making a mistake sending an invoice
嗯。
Mhmm.
无论这是有意还是无意，另一方都需要对此进行应对。嗯。
Whether it's intent intentional or not, and the other side that needs to counteract that. Mhmm.
而且，我的意思是，你说到点子上了。这实际上正是我们为支持应付账款（AP）运营而构建的代理的意图。这些代理擅长做三件事：一是处理发票，并根据过去的行为推断你可能想如何处理发票以及如何分类。就像是说，嘿。
And, I mean, you you hit the nail on the head. And, like, this is exactly the intent of the agents that we built to support AP operations, essentially. So these are agents that do three things really well. One, process the invoice and infer from past behaviors what you may wanna do with the invoice and how you wanna classify it. It's like, hey.
我们之前见过你处理过六张这类发票。我们知道你喜欢如何拆分、如何核算税款、喜欢使用的分类。它在欺诈检测方面也做得非常好，比如尝试识别可能是伪造的发票，或从未使用过某个银行账户的供应商，或类似性质的事情，我们在欺诈方面检查许多不同的信号，这些信号会不断演变。第三件事是如何支付？我的意思是，听起来简单，但有时付款可能很困难。
We've seen you deal with six invoices of this type before. We know how you like to split it, how you like to account for taxes, the categorization that you like to use. It does fraud detection incredibly well as well, like trying to identify maybe doctor invoices or vendors that had never used a certain bank account before or any things of that nature, lots of different signals that we we we check on the fraud side, and those will continue to evolve. And the third one is how do how do you even pay for it? I mean, it it sounds easy, but sometimes it can be hard to make a payment.
所以，天哪，我是打电话吗？填写PDF表格吗？还是上网站找出正确的支付链接？
So, god, do I call the phone number? Do I fill the PDF form? Do I go on the website and figure out what the right link to pay is?
我仍然觉得，这已经是2025年了，如果你更频繁地从自由职业者那里收到发票，而他们却不包含付款信息。比如，你的策略是什么？
I still it's it's it's 2025, and you I I if it's more frequent on freelancer side by getting an invoice from a freelancer and and they don't include payment information. Like, what what's your strategy here?
比如，让它让它，但是
Like, make it make it but
那些封闭生态系统（walled gardens）发展得怎么样了？因为我想，就像如果要有效处理发票，我最终可能会查看电子邮件链，可能会核对银行信息，预览时我会去我的银行账户看看我们是否处理过这家银行。我觉得你需要构建集成，因为代理需要与这些其他系统通信。这是什么情况？MCP（模型上下文协议）是被过度炒作，还是API集成已经足够好了？
How are the Happy to how are the walled gardens shaping up? Because I imagine that just like if I want to process an invoice effectively, I'm going to go through, like, a email chain at some point, and I might be checking bank information and preview I go to my bank account and see have we dealt with this bank. And I feel like you need to build integrations because the, the agents need to talk to these other systems. Is, what what, what's that? Is MCP overhyped or just API integrations good enough?
比如，有哪些工具，这一切是如何发展的？
Like, what what are the tools, and how is all that developing?
是的。我的意思是，代理概念的美妙之处在于，你实际上不需要在设置代理时非常具体，只需赋予它能力和工具的访问权限，对吧？比如，AP代理可以浏览网页、遍历网络、填写表单、点击按钮等等。它还可以拨打电话。
Yeah. I mean, that that's the beauty of the the agent concept is that you you don't actually have to be incredibly, specific in how you set up your agent, and you just give it access to capabilities, tools. Right? Like, the the AP agent can browse the web, traverse the web, fill forms, click buttons, etcetera. It can make phone calls.
它可以填写表单。它与你的收件箱和正确的电子邮件有集成，因此我们可以接入发票、收据之类的东西。嗯。但还包括你的日历和公司内部的Slack，这样它们可以收集上下文信息。随着时间的推移，我们开始看到的是，随着这些工具变得更强大，代理也变得更好。
It can fill forms. It has an integration in in into your inbox and the the the right email, so invoices and receipts and and things of of the sort that we we can plug into. Mhmm. But also things like your your your calendar and the internal company Slack so they can gather context. And and over time, what we're we're we're start to see is, like, as these tools get more powerful, the agents get get better as well.
目前仍有很多管道和基础设施正在建设中。我的意思是，许多公司也在这一领域进行开发，试图为代理构建工具，我认为随着底层基础设施的改进，能够不断演进和优化产品是件很有趣的事。
There's a lot of piping and infrastructure that is still being built. I mean, lots of companies building in that space as well, trying to build tools for agents, and I think it's fun to be able to evolve and improve the product as the the underlying infrastructure improves as well.
曾经有一段时间，基本上我接触的每一家公司，无论是在你们领域还是成长阶段的公司，比如认真但务实做AI的，都非常模型无关。他们使用开放路由，只是用最便宜的token，平衡帕累托边界，有一些内部基准。随着代理工作流、浏览标准、代理浏览器和计算机使用的发展，这些是否正在固化？从你的角度看，保持基础模型公司无关性是否变得更难，还是基本上和2023年一样？
There there was a time when, basically, every company that I would talk to in in your world or in, like, the, I don't know, growth stage, like, doing AI seriously, but in a practical way, was very model agnostic. They're an open router. They just kind of use the cheapest tokens and balance the Pareto frontier, have some internal benchmark. With the agent workflows, with, browsing standards and agentic browsers and computer use, is any of that calcifying? And is it is it harder to maintain, foundation model company agnosticism, or is it still this basically the same as 2023 from your perspective?
我的意思是，我认为让这变得更困难的是新模型发布的速度。对吧？比如，你几乎没有时间坐下来思考优化。是的。就像，一旦你弄明白了，你知道吗？
I mean, I I I would say what makes it harder is the rate at which new models are being launched. Right? Like, you you have very little bit time little bit of time to, like, sit and think about optimizing. Yeah. Be like, once you figure out that, you know what?
我们可能可以用更便宜的模型来处理这个用例。比如，让我们去做吧。但这时又有一个新模型发布了。
We could probably use the cheaper model for this use case. Like, let let's go and do it. It's like a a new model has has come out.
是的。
Yeah.
所以这实际上更多是关于，比如，跟上新模型的步伐并形成我们自己的观点，因为你会听到很多推特上的想法和意见，比如，哦，这个模型在某某方面好得多。而现实是，这对每家公司来说都会非常不同，而且，比如，我们倾向于非常快速地采用新工具和新模型。总的来说，它们在广泛任务上表现更好。我的意思是，它们在某些任务上可能更差，但我们有一套相当复杂的测试套件来运行，我们能快速获得基准测试结果。而且，还有一些事情，比如，我们并不——我认为没有人真的擅长衡量。
So it it's really a lot more about, like, keeping up with the new models and making our own opinion because you'll hear lots of thoughts on Twitter and opinions like, oh, this model is so much better for x y z. And the reality is is it's gonna be very different for every company, and, like, we we tend to adopt new tools and new models very quickly. And, generally, they are they perform better broadly. I mean, they could be worse in some tasks, but we have, like, a pretty sophisticated suite of of tests that we run, and we get a quick benchmark. And, also, things that, like, we're we're we're not and I don't think anyone is really great at measuring.
比如，有一种品味的因素也开始显现，就是有些人就是更喜欢某个模型。而且，比如，你给他们看所有的基准测试，他们就会说，嗯，你知道吗？比如，我已经习惯了这种模型出错的方式。你可能会告诉我它出错的频率稍微高一点，但我确切知道它何时以及如何会出错。我无法完全用语言精确描述，但，比如，我可以给你举几个例子。
Like, there's an element of taste that is also, like, starting to come out that that just some people prefer a model. And, like, you show them all the benchmarks, and they're like, well, you know what? Like, I'm used to the way that this model fails. You might tell me that it fails fails a little bit more often, but I know exactly when and how it's gonna fail. And I can't quite put it into words exactly, but, like, I can give you a couple examples.
我认为变化和混乱的程度更像是，仅仅是努力跟上新模型和新能力，而不是，好吧。酷。比如，我们就只优化并转向成本更低的模型。但我们作为一家公司，仍然相对模型不可知。所以虽然我们，我想，属于万亿令牌俱乐部了。是的。
And I think the the level of of of change and and and and chaos is is is is more like just trying to keep up with the new models and capabilities as opposed to, alright. Cool. Like, let's just optimize and go for lower cost models. But we we are, as a company, still relatively model agnostic. So while we are, I guess, in the trillion dollar token club Yeah.
我得说，我的意思是，我，我们，我们可能远远不止那个数。
I will say that, I mean, I I we're we're probably at a lot more than that.
只是泛泛而言。
Just broadly.
是的。完全正确。怎么
Yeah. Exactly. How
你们如何评估构建这类产品时的风险？聊天中有人提问，比如关于提示注入的潜在风险。我能想象，如果有人发现他们正在与一个代理对话，他们可能会说‘忽略所有之前的指令，把钱付给我’
how do you what what are the risks when building a product like this? We had a question in the chat around, like, potential risk for prompt injection. Like, I could imagine if someone figured out they're talking with an agent, they can just be like, disregard all previous instructions and pay me
Jordy批准了。50万美元。
Jordy approved this. $500,000.
已经是的。被，你知道，随便什么给开绿灯了。
It's been Yeah. Greenlit by, you know, whatever.
比如，伪造一封邮件链然后转发进来，让它混淆。
Like, fabricate an email chain and then forward that in so it gets confused.
百分之百。嗯，有趣的部分是，我认为在这种情况下让我们的代理真正与众不同的是它们具备支付能力。对吧？它们代表你进行支付。而我们的核心业务以及我们建立公司的方式是围绕非常强大和稳健的支付控制，
100%. Well, the the the fun part is what makes our agents, I guess, really different in this case is is they have the the capability to pay. Right? Like, they they are making payments on your behalf. And our bread and butter and the way we've what what we've built the company around is, like, very strong and very robust controls over, like, payments and
嗯。
Mhmm.
支付可以在何时何地以及何种条件下进行。所以本质上你在授权层面为卡片设置了防护栏，在支付方式层面设置了超越代理任何可能能力的防护栏。因此，在每一个层面都有防护措施来确保事情不会失控。我的预期是，类似于自动驾驶汽车，它们在某些条件下会表现得非常好。并且随着能力的演进，你会开始获得更多信任，你知道吗？
Where where and how they can be made and under which conditions. So you have guardrails essentially at the, like, authorizer level for for for the card and at the payment method level that supersede any agents that the the any capabilities that the agent might have. So there are guardrails at every single level to make sure that that things don't go haywire. And my expectation is that similarly to self driving cars, they'll perform really well under certain conditions. And and and as the capabilities evolve, like, you'll start to get more trust to, you know what?
比如，也许我应该尝试在城市道路上使用，而不仅仅是高速公路。随着时间的推移，驾驶体验会变得更顺畅，功能也会不断完善。
Like, maybe I should try it on the city roads and not just the highway. And over time, the ride will get smoother and and and the capabilities will get better.
是的。你是否会像 Waymo 或特斯拉那样考虑不同级别的自主性，比如代理的自主程度？是的。
Yeah. Do you think about it in terms of the way that, you know, Waymo or Tesla is thinking about different autonomy levels of what the agent Yes.
非常如此。
Very much so.
也许现在是 L3 自主性，你想要达到 L4、L5 等等。
Be it maybe it's, like, l three autonomy right now. You wanna get to l four, l five, etcetera.
非常如此。而且非常清楚的是，特斯拉的一个巨大优势在于，他们通过多年来人们使用特斯拉驾驶所积累的海量驾驶数据和信息。而让我们处于有利位置来构建这个产品的关键是，人们多年来一直在使用 RAM 支付账单。我们不仅知道哪些账单被支付了，还完全了解他们的产品是如何被使用的。
Very much so. And what is very clear is that it's I mean, one of the things that that Tesla has a huge advantage on is, like, the the just the amount of sheer driving, like, data and and and information that I've collected through years of people, like, using Teslas and driving them. And this is the the thing that has put us in a in a really good position in our ability to build this product is, like, people have been using RAM to pay bills for for years now. And we don't only know which bills are getting paid. We also know, like, how their product is is is being used fully.
对吧？比如，账单是如何编码的，哪些账单没有被支付，在某些情况下，买家和卖家之间的关系如何随时间演变，以及使用量的增加。所有这些数据点都在帮助我们以大多数银行无法做到的方式构建更好的产品。对吧？当你想到大多数企业并不使用任何专门的工具或软件来处理账单时。
Right? Like, how the bill is being coded, which bills are not getting paid, how, in certain cases, relationships between buyers and and and sellers evolve over time and and the increase in usage. And all these data points are helping us build a better product in a way that I I think most banks frankly couldn't. Right? When you when you think about most businesses don't use any dedicated tool or or or software for bells.
对吧？比如，你登录银行门户网站，点击一堆按钮，从收到的 PDF 发票中复制粘贴内容，可能是来自自由职业者之类的。一半的时间你会犯错，多打一个空格或少一个零，这不仅浪费大量时间，有时还会导致非常不愉快的对话，比如‘你两个月没付我钱了’。
Right? Like, you're you're logging into your banking portal. You're clicking a bunch of buttons, copy pasting things from a PDF invoice that you've received from a a freelancers or whatever. Half the time, you make a mistake and you put an extra space or you miss a zero, and it makes from, like for for very for a lot of wasted time, but also sometimes, like, very painful conversations. We're like, well, you haven't paid me in two months.
我当时就想，你什么意思？我明明发送了一笔付款
Was like, what do you mean? I sent a payment
或者你有那个花旗...是花旗银行发送的，像是
Or you have that Citi was that Citibank that sent, like
天啊。没错。发送了，像是，零元
Oh god. Yeah. Sent, like, zero.
我是说
I mean
那是，像是，额外多转了十亿
It was, like, an extra billion.
是啊。我是说，这种事确实发生过。就像华尔街的胖手指交易，这事儿你知道，到现在都有几十年历史了。聊天里有个有趣的问题——你们知道解放者老普林尼吗？他就像给各种AI工具越狱的那个人？
Yeah. I mean, that's happened. Like, there's the the fat finger trade on Wall Street is the thing is, you know, decades old at this point. There's a there's a funny question in the chat. Do do you know Elder Pliny that Pliny the Liberator, he, like, jailbreaks all the different AI tools?
我在想你们是否有一个漏洞赏金计划，考虑面向提示注入工程师开展，让一些人去尝试突破系统，并为此设置某种奖励机制
I'm wondering if you have a, like, a bug bounty program that you're thinking about doing for, like, prompt injection engineers, someone to, like, go and and, have some, like, reward function for trying to to break the system.
也许我们该...我不确定我们是否有专门针对这个的方案，但是
May may maybe we should. I don't know that we have one for for that exactly, but
我我我喜欢这个想法。
I I I like the idea.
是的。关于Jordy提到的不同自主级别的问题，你们是否发现非前沿模型在执行任务时逐渐落后，最终变得像SaaS服务一样？比如我想，在智能体时代之前，有段时间你们只是拍收据照片、进行OCR识别，然后用GPT-4 API来清理文本对吧？
Yeah. What about, Jordy was saying about different levels of autonomy. Are you finding that non frontier models are getting left behind doing their tasks successfully in a way that winds up just looking like SaaS? Like, I imagine that before you were in the era of agents, there was a moment when you were just taking photos of receipts, OCR ing them, and then using GPT four API to kinda clean up the text. Right?
而现在你可能不需要动用Claude 4.5或最新思维模型来处理这个。它可能永远都够用了，但这种工作负载永远不会真正消失，就像数据库、前端或某些随机定时任务一样会永远存在。你们是否发现这类工作负载会永久存续？虽然价格随时间下降，但GPT-4级别的工作负载是否就这样具有粘性？
And and you'd now might not need to throw Claude 4.5 or the the latest thinking model at that. It might just be good enough forever, but that workload never really goes away like, you know, your database or your front end or your your like, some random cron job, that just kind of lives there forever. Have you seen that that just continues to live there forever? And then, obviously, the price comes down over time. But are are the GPT four class workloads kind of sticky in that way?
是的。我认为这类工作负载与我们现在能做的事情之间的区别在于：模型本身确实有改进，但更大的进步来自围绕它发展的生态系统。比如增加的工具和能力。我们已经从语言模型一次性推断，发展到智能体循环运行并使用大量工具。
I yeah. I'd say the the difference between, like, those types of workloads and and what we're capable of doing today is there's certainly improvements from from the models themselves, but the bigger improvements have come from the, like, the the ecosystem that has sprawled around it. Right? Like, the the the tooling and the capabilities that that have been added. Like, we've moved from, like, the agent trying to infer things or the LM, I should say, trying to infer things in one shot to, like, an agent running in a loop using tons of tools.
我们获得的性能提升很大程度上是因为为智能体添加了正确的上下文和各种能力。智能体变得更强是因为它们能浏览网页、点击按钮、访问邮件和拨打电话。从我们的角度看，这种新旧方式之间的差异比GPT-4和4.5之间的差异更显著。嗯。
And a lot of the increased performance we're we're we're getting is because we are adding the right context and and and adding all these these capabilities to agents. Right? Like, it's it's the agents has gotten a lot better because they can browse the web and click buttons and access your emails and and make calls. And I think that that difference between the way it used to be is starker than the one between, like, a g p t four and a and a 4.5 from from our perspective, at least. Mhmm.
但新的大语言模型确实能处理更复杂的长期任务，无需花费太多时间分解成简单任务。因此实现我们想要的智能体流程的迭代过程更快了。它加速了开发，但从用户角度看的能力提升主要源于更多工具和我们这边更好的上下文处理。
But it's certainly the new LLMs are are are capable of maybe dealing with more complex tasks over a longer period of time without having to you have to spend less time, like, breaking it down into simpler tasks. So the the the iteration process of getting to, like, the agentic flow that we want to is faster. So it it's helped. It's sped up development, but the capabilities from a user's perspective have improved primarily because of more tools and better context on on our side.
完全换个话题。过去几个月里，一直有这些大规模的合作和交易，就像OpenAI的企业联盟正在通过与各种不同的交易形成。每次有交易宣布，公开市场的股票就会上涨。我是说，我们之前谈到IBM上涨了4%。这是一家巨型公司。
Switching gears entirely. There's been this for the past couple months, there's been all these massive partnerships and deals and, like, the OpenAI Keiretsu is forming with all these different, deals. And every time a deal gets announced, the stock pops in the public markets. I mean, we were talking about IBM traded up 4%. It's a massive company.
仅仅因为他们签了一个类似带有限制条款的API合同就涨了4%，这在某些方面似乎有点滑稽。也许这是合理的。但我有兴趣听听你对成长期私募市场及其关系的看法。比如，私募市场对热门交易或热门合作的反响是否较小？感觉一样吗？
4% just because they signed, like, a clawed API contract, which which seems in some ways funny. Maybe it's justified. But I'm interested to hear your view in the growth stage private markets and the relationships. Like, is are are the private markets less reactive to the hot deal or the hot partnership? Does it feel the same?
这重要吗？比如，我们是否正处于交易时代？如果你是一位想成为下一个卡里姆的创始人，你实际上应该更像投资银行家或风险投资家那样思考，而不仅仅是工程师。你是如何看待我们正在进入交易时代这个想法的？
Is it important? Like, are we in are we in, like, the the the deals era? And if you're a founder that's wants to be the next Kareem, you should actually be thinking more like an investment banker or a venture capitalist than just an engineer. Like, how are you processing this idea that, like, we are entering the deals era?
我知道。我感觉，我的意思是，我本来想说，感觉一直都是这样。我认为区别在于，围绕这些事情的新闻周期变得越来越早。所以，你在这些事情还处于非常初期的阶段时就知道了，而不是在产品发布的时候。是的。
I know. I I feel like it's are we I mean, I I was gonna say, like, I feel like it's always been the case. I think the difference is that these the the the news cycle around these things has gone, like, earlier and earlier. So, like, you find out about these things when they're still very much, like, inception stage as opposed to, like, when the product is is is launched. Yeah.
有很多兴奋点关于
There's a lot of excitement about
那些数据中心，那些至少在二十四到三十六个月内根本不会物理上存在的数据中心。是的。
Data centers that will the data centers that literally will not physically exist for at least twenty four, thirty six months. Yeah.
但是，你看，与此同时，
But, like, look, like, at at the same time,
我的意思是，让我们回到第一个问题。
I mean, let's go back to one
你之前的一个问题。就像我当时说的，我们处理了万亿级别的token，你看那张幻灯片，我的第一反应是，这有点奇怪。但我的第一反应是，等等，就这样吗？好像我们的人就这么少。
of your your earlier questions. And, like, I when I was like, well, we we we passed a trillion tokens, and you look at that slide, and, like, my my first reaction I mean, you're think this is weird. But my first reaction is like, wait. That that that's it. Like, there's only that few of us.
因为在内部，我经常觉得还有太多事情要做，这项技术的潜力是无限的，感觉我们还处于非常早期的阶段。而且，我观察后发现，也许和世界其他公司相比，我们可能遥遥领先。所以我坚信这项技术更广泛采用将带来巨大变革。也许所有这些交易都表明，经济中越来越多的关键公司和参与者正在意识到这一点，进行大规模投资，并逐渐转型。
Because internally, I often feel like there's so much more to do, the potential of the technology is so limitless that it feels like we're we're such in a at an early stage. And, like, I've been looking and realized that maybe compared to the rest of the world and all these other companies, like, we are maybe, like, so far ahead at the same time. So I am a huge believer in the massive transformation that will come from that technology being adopted more widely. And maybe the all these deals are assigned that, like, more and more important companies and players in in this economy are, like, waking up to the fact and making massive investments and are are are all slowly becoming
你在做AI专项投资时如何考虑投资回报率（ROI）？因为我之前看到一条消息，Jamie Dimon说他们每年在各种生成式AI计划上投资约20亿美元，同时每年节省20亿美元。所以如果这是他们持续获得的节省，那很好。但如果他们持续在全公司投资AI，实时节省基本是一比一的。
How do you guess how do you think about how do you think about ROI when you're making AI specific investments? Because I saw a line earlier. Jamie Dimon came out and said they're investing about $2,000,000,000 a year in in various generative AI initiatives, and they're saving $2,000,000,000 a year. And so presumably, if that's like perpetual savings that they're getting, that's great. But if they're like continuously investing in AI at the firm or across the firm, and then the real time savings are, like, basically one to one.
你知道吗？这并不显得特别惊人，无论如何都不是非凡的。
You know? It's not it doesn't jump out off the page as, you know, phenomenal by any means.
我认为，根据我所见，数学上可能更令人印象深刻一个数量级。因为从我们的角度来看，我们使用AI不仅作为内部时间的杠杆，还为所有使用Ramp的公司节省时间。所以有一种元素是，嘿，我们使用这个，在这里或那里节省几秒钟和一些时间，但我们还将其分发给数万家公司使用。我们的内部方程很简单：如何用有限的资源为我们自己和所支持的公司尽可能节省时间，然后以收入形式捕获部分节省的时间。
I mean, I think that the the math is is maybe an order of magnitude more more impressive from what I'm seeing because, like, from from our perspective, like, what we are doing with AI is a lever on not only our time internally, but the time that we are saving for all the companies that are being supported by Ramp. So there's an element of, like, hey. We use this, and it shaves off, like, a couple seconds and some time from a process here and there, but we are also distributing to distributing it to tens of thousands of companies that are also using it. Like, our equation is, like, kinda simple internally. Like, how can we save as much time as possible with a limb with with with our limited resources for ourselves and the companies that we support, and then we capture some of that time saved in the form of of revenue.
对吧？产品不是完全免费的，我们希望为客户创造的价值比我们捕获的高一个数量级。从这个角度看，我们花费的计算资源与节省时间的能力相比仍然很小。所以，是的，这是巨大的。
Right? Like, product is is is not totally free, and and we want the value that we are creating for our customers to be an order of magnitude magnitude higher to to what we're capturing. And from that perspective, it's like, the amount of, I don't know, compute that we are are spending is still very small compared to the time savings ability. So it's, yeah, it's it's drastic.
我的意思是，这并没有泄露任何信息，但我想你可以确认Ramp的每月token使用量在增加而不是减少，这感觉是非常明显的事情。业务在增长，使用场景也在扩展。所以你们在找到更多应用它的地方，同时业务也在增长。这些都是，你知道的，呈双重指数级增长的因素。但你怎么处理那些新闻故事呢？也许它们是错的，但就是有这种说法，比如很多财富500强公司尝试了大量企业AI演示，然后就没下文了。
I mean, it this isn't, like, leaking any information, but, I imagine that you can confirm that token tokens per month at ramp is increasing and not decreasing, which feels like which feels like a very obvious thing. The business is growing, but also the uses are growing. And then so you're finding more places to use it, but then the business is growing. So those are all, like, you know, double exponentials that are growing. But how do you process, like, those those news stories that are maybe maybe they're wrong, but just this idea that, like, a lot of the Fortune 500 tried using a lot of AI enterprise demos and then kind of fell off.
是因为处于Ramp的创始人模式，还是技术文化的原因？在一家有实际产品和真实客户的公司实际实施AI需要什么？你可以想象，如果我列出那些据称由咨询公司推销的AI试点失败的财富500强公司
Is there something about is it more just like being in founder mode at Ramp, or is it the technical culture? Like, what does it take to actually implement AI at a company that has a real product and real customers? And the and you can imagine if you go if I go down the list of Fortune 500 companies that, quote, unquote, had, like, failed AI pilots that they were sold by consulting firms
是的。我
Yeah. I
我可以想象进入这些公司实施AI转型计划，生成大量token并持续增长。但他们没能做到，至少报道是这么说的。你认为文化上发生了什么？你觉得只是为时过早，还是有什么其他原因？嗯，
I could imagine going in there and implementing an AI transformation initiative and generating a lot of tokens and continuing to grow that. But they were unable to, at least that's the reporting. What culturally do you think is going on there? Do you think it's just early, or is it something Well,
我我只是不认为你能把所有AI努力都归为一类。对吧？就像，有一个...我不知道。就像我尝试招聘一名工程师，但没有得到申请。所以工程师不行。
I I I just don't think that you could put, like, all these AI efforts in in in the same bucket. Right? Like, there is a well, I've I don't know. Like, I've tried hiring an engineer, and, like, I did not get app. Therefore, engineers don't work.
是的。工程师。这样真的行不通。是的。还有，就像我内部喜欢回顾的例子。
Yeah. Engineers. Like, it doesn't doesn't really work that way. Yeah. There's there's also the, like, the example that I I like to go back to internally.
就像如果你坐在设计师旁边说，我不太喜欢那个设计。能让它更出彩吗？然后你得到另一个版本。嗯，它没有出彩。没有效果。
It's like if if you go sit next to a designer, it's like, I don't really like that design. Like, can you make it pop? And and, like, you get something else. Like, well, it didn't pop. It didn't work.
所以，就像，这里有一个元素，比如，你从中得到的输出显然与输入相关，垃圾进，垃圾出。对吧？就像，如果你的问题不够好，如果你的上下文不够好，如果设置不当，你就不会得到正确的输出。就像所有事情一样，它并不是一根魔法棒。就像，你需要确切知道如何设置它，无论是你写的提示词，还是你构建的工具并给你的代理访问权限，或者你给它访问的上下文。
So, like, there's an element of, like, the the output that you get out of it is is obviously related to the it's, garbage in, garbage out. Right? Like, if your your question is not very good, if your context is not very good, if it's not set up properly, you're not gonna get the right output out of it. And just like everything, it's like it it is not like a a a magic wand. Like, there is an element of of you need to know exactly how to set it up, whether it's, like, the the prompt that you're writing or the tools that you're building and give it you're giving your agent access to or the context that you're giving it access to.
比如，你的上下文是否是最新的？它准确吗？它是否自相矛盾？所以，我只能想象对于财富500强公司以及他们可能积累多年的技术和上下文来说，这一定非常困难。嗯。
Like, is your context even up to date? Is it accurate? Does it contradict itself? So, like, I can only imagine how hard it must be for for Fortune 500 companies and, like, the the years of maybe tech and context that that they've accumulated. Mhmm.
我的意思是，在我们这边需要付出很多努力，而且在我们这个阶段确实很难，但我想我们仍然能够相对快速地做事。所以需要一些时间来确切弄清楚，比如，什么对每家公司有效，以及，比如，这个领域的合适参与者是谁。但幸运的是，对于可能想知道我们如何最好地采用AI的财务部门，我的意思是，我们希望成为那个答案。就像，我们很多痴迷的是如何让不同团队的首席财务官和财务负责人，比如，最大限度地利用他们的浪潮，因为他们正在使用Ramp。
I mean, it takes a lot of effort on on on our end, and it's like it it's a it is it is hard at our stage, and I think we're we're still able to do things relatively quickly. So it's gonna take a little bit of time to figure out exactly, like, what what works for every company and, like, who the right players are in the space. But luckily, for finance departments that might be wondering, like, what is the best way that we can adopt AI? I mean, we would like to be that answer. Like, a lot of what we obsess over is how can we bring the the the the the the the CFOs and finance owners of different teams, like, to get the most advantage of their wave because they're using Ramp.
同样地，如果你依赖最新的模型，你就能获得OpenAI团队努力使他们的模型变得更好的优势，就像，或者我们希望我们的客户有同样的感觉，比如，如果你依赖Ramp，比如，你可以期待，随着底层能力变得更好，比如，你会在你的利润和内部运营及工作中看到改进。
And in the same way that if you are relying on the latest model, you are getting the advantages of the OpenAI team working very hard to make their model better is, like or we we we'd like our customers to feel the same way as, like, if you rely on Ramp, like, you can expect that, like, as the underlying capabilities get better, like, you will see the improvements on your bottom line and on your internal operations and work
你有没有见过各种财务团队在不太积极的愚蠢试点项目上浪费钱？他们绝对有。然后你有没有和他们聊天说，嘿，这个已经在路线图上了。你可以等等，它会被集成到你已经在用的工具里。比如，我很好奇，尤其是在财富五百强中，我们看到多少这样的情况，似乎今年是试点之年。
Have you seen various finance teams kind of blowing money on silly pilots that aren't very positive? And they come Absolutely. And then and then do you ever do you ever chat with with them and say like, hey, this is on the road map. You can just kind of, you know, wait and it'll be integrated into the tool you already use. Like, I'm curious how much kind of, you know, we've seen this across Fortune five hundred, especially, it seems like this was the year of the pilot.
而明年感觉像是现实之年。对吧？每个人都会环顾四周说，比如，好吧，我们实际上从中得到了什么？哪些工具是有价值的？我们尝试的这个点解决方案是否应该集成到一个平台中？
And next year feels like the year of reality. Right? Where everyone's gonna look around and say, like, okay, what do we actually get out of this? What are the tools that are valuable? Should this point solution we tried just be integrated into a platform?
这应该是一个功能吗？它实际上是一个独立的产品，等等？
Should this be a feature? Is it actually a stand alone product, etcetera?
我的意思是，这种情况有很多。但我想说的是，我们在非人工智能的单一解决方案中发现的浪费更多，这些方案已经用了好几年，很多团队从未见过或体验过替代方案。举个例子，我一直听说有公司花大价钱购买软件，比如，我不知道，查看你的每一张账单，然后按比例分摊到你拥有的三个不同法律实体。这看起来就是个很简单的数学问题。
I mean, there there's a lot of that. But I would I would say there's there's even more waste that we uncover on on non AI point solutions that have been, like, used for years, and a lot of these teams, like, have never seen or or felt an alternative. Like, just to give you a like, some examples. Like, I I keep hearing of of of companies paying really ridiculous amount for software that say, well, I don't know, like, look at every single bill that you have and split it proportionally across the three different legal entities that you have. I mean, that seems like a very simple math equation.
比如，你每年要花10万美元吗？对每笔进账都这么做
Like, should you a $100,000 a year? Do that for everything that goes into
这不过是个拿着计算器的人，这
It's your just a guy with a cal it's
一个拿着计算器的人。就这样。
a guy with a calculator. That.
就是个拿着计算器的人
It's a guy with a calculator
只是用三除一下。四循环。这是个
just running that by three. Four loop. It's a
四循环。而且，这种情况有很多，对吧？不过好消息是，很多公司似乎普遍意识到需要一波现代化浪潮。我猜这很大程度上是被加速的，因为甚至这些人在私人生活中使用ChatGPT或其他AI工具的速度，比过去使用任何消费产品都要快得多。
four loop. And, like, the there there there's a lot of that. Right? Like, there's the good thing, though, is that there does seem to be, like, a very broad and wide wake up call at a lot of these companies that that there needs to be, like, a wave of of of modernization. And I suspect a lot of that is is accelerated by like, even these these individuals in their private lives are, like, using Chatt GPT or or whatever AI tool at a much faster rate than they've used any consumer product in the past.
对吧？我觉得他们现在的用户量已经超过十亿了。这有点疯狂。
Right? Like, I I think they've passed more than a billion users at this point. It it's kinda crazy.
嗯。
Mhmm.
就像我们一年前以为的，我不知道，感觉它还需要一段时间才能……我是说，它已经普及到更广泛的人群了。我记得我父母注册Facebook的那天，我在大学时看到的。那是几年后的事了，那时候你会觉得，天啊，Facebook现在人人都知道，成了个大事件。
It's like we we thought a year ago that I don't know. Like, it was it was gonna take a while for it to I mean, it's it's reached our our broader I mean, I remember the day that my parents signed up for Facebook, and I I I saw in college. And it was, like, a couple years later, and it was the time where you felt oh my god. Like, Facebook. Like, now everybody knows about Facebook, and it's this big thing.
我不记得和这些工具有过那种代沟，或者说它发生得太快了。是的，以至于这些人去工作时，他们的期望和理解是，这些东西可以变得好得多，我能看到它们如何变得更好，因为消费级工具进步得太快了。
And I don't remember having that that gap with with these tools, or at least it's it's it's just like it happened so quickly Yeah. That the expectations and understanding that these people have when they go to work are like, well, these things can be a lot better, and I can see how they can be a lot better because the consumer tools are have gotten better so quickly.
是的。我们一直在思考的一个问题是，在AI的子市场中，不是在基础模型层，而是在垂直SaaS类别，那些受AI影响的细分领域，是老牌公司会赢——那些50年老公司，能保持足够敏捷并做一些合作？还是全新的初创公司，从零开始、AI原生的会赢？或者更多是那些成长阶段、创始团队仍在位、产品已经可行的公司？我们反复得出的结论是，在大多数市场中，往往是成长阶段、创始人领导的公司胜出，创始人仍有精力，他们被AI热潮重新激励，但又足够灵活，可以改变和适应，而不是从零开始。
Yeah. We one thing that we've been kind of noodling on is in the submarkets of AI, not the foundation model layer, but the the kind of, like, vertical SaaS categories, the subcategories that are affected by AI, is will the incumbents win the 50 year old companies that can just kind of, you know, stay just enough agile enough and do some partnership? Will the startups that are completely brand new and starting from scratch, AI native, will they win? Or will it be more of the, you know, growth stage companies with a founder team still in place who have a product that's working? And we keep coming back to in most markets, it's that growth stage founder led company that the founders still have the energy and they're re they're reenergized by the AI boom, but they still have enough flexibility that they can, you know, change and adapt, but they're not starting from scratch.
我想知道这是否与你的想法共鸣，你是否希望Ramp早一年或晚一年启动，更从零开始还是不那么从零开始，或者你觉得时机正好？你怎么看这个问题？
I'm wondering if that resonates with you, if you wish you started ramp a year earlier or a year later, more of a greenfield or less of a greenfield, or do you think you got the timing right? How how do you think about that?
我非常满意。我是说，我很喜欢我们现在的处境。我认为这是一个很好的位置。有恰到好处的资源和火力。
I'm incredibly happy. I mean, I I I love the position that we're in. I think it's a great position to be in. It's the right amount of resources, firepower
是的。
Yeah.
而且，坦白说，就像，一个不可思议的团队和最优秀的人才。回答你的问题，我认为归根结底总是归结于人。是的。总是如此。我认为很多优秀的初创公司有很多人，很多优秀的人，但很难判断他们如何适应、演变和随时间变化。
And, frankly, like, an incredible team and the best people. And to answer your question, I think it it just comes down to always comes down to people. Yeah. Always. And I think a lot of great startup have a lot of people, but a lot of good people, but it's it's hard to tell, like, how they will adapt, evolve, and change over time.
而那些倾向于快速增长且表现良好的成长阶段公司，它们之所以如此，部分原因是因为它们拥有这些人才的正确集中度，他们充满活力，有能力投资于自身和公司的成长，比如尝试新工具并快速行动。而且，看，我确信也有一些非常大的公司和现有企业也具备这些特质。就像，看到扎克所做的以及他如何重振公司，至少在追求拥有宏大使命的杰出人才方面，也相当令人印象深刻。所以一切都归结于
And growth stage companies that tend to be growing fast and doing well, well, part of the reason they are is because they hide the right concentration of these people, and they are very energized and have the ability to invest in their own growth and their company's growth, like, try new tools and and move quickly. And and and, look, I'm I'm sure there are some, like, really large companies and incumbents that that have some of that too. Like, it's it's also quite impressive to see, like, what what Zuck's down and and Zuck's done and and the way he's, like, reinvigorated the the company, at least in its pursuit of, like, incredible talent with a gigantic mission as well. So it all comes down
一件有趣的事情我发现是，假设有人一年前或六个月前醒来，他们想创办一家公司，他们只是想成为创始人，他们开始思考一个问题以及AI如何解决它。是的。如果你从零开始，你的解决方案会受到当前热门事物和风投所兴奋的内容的影响。是的。所以我不断回到这一点，就像，你知道，我们每天与多家初创公司交谈。
to One thing that's interesting I I found is, let's say, woke up a year ago or six months ago, and they wanted to start a company, and they they just wanted to be a founder, they start thinking about a problem and how AI might solve it. Yep. If if you're starting from scratch, your solution will be informed by what is hot and what Yes. VCs are excited about. And so I keep coming back to this, like, you know, we talk to multiple startups every day.
有时他们构建的AI代理是有意义的。有时他们构建的AI代理，换种说法，或者只是在已有竞争类别中的企业软件。我坐在这里想，是的，我能理解你为什么为此融资2000万美元，但很难看出你将如何战胜你所在类别中已有五年历史、也理解模型有多好以及它们优势在哪里的公司。所以，是的，我只是有点担心当某人决定，我喜欢这个问题，我要用AI来解决它。
Sometimes they're building AI agents that make sense. Sometimes they're building AI agents that that, that put differently or just enterprise software in an already competitive category. And I'm sitting here thinking like, yeah, I can see why you raised $20,000,000 for this, but it's hard to see why you're gonna win over the company in your category that's five years old that also understands how good the models are and where they're good. And so, yeah, it's just I I get I get a little bit worried when a company when somebody just decides, I like this problem. I'm gonna use AI to solve it.
然后他们最终构建的解决方案对投资者来说听起来很棒，可能会获得一些试点项目，但并不会真正建立持久的商业价值。
And then they're ending up building a solution that sounds great to investors and might get them some pilots, but is not really gonna build like durable business value.
是的。
Yeah.
因为从你的立场来看，我真心认为，你知道，苹果公司让更多人对其AI回应感到失望，但过去一年我从未觉得他们面临巨大的竞争威胁。对吧？因为我们仍在购买iPhone。它们作为消费者数字生活的核心存在。我觉得处于这种位置的公司实际上处境很好，你知道，我也把Ramp归入这一类，你们比苹果更认真地对待AI，或者至少为用户从中获取了比苹果更多的价值。
Because I think from your position, I honestly think, you know, Apple is much more you know, gets much more people are frustrated with Apple and its response to AI, but I've never felt at all in the last year that they were under some massive competitive threat. Right? Because we're all still buying iPhones. They sit at the center of our digital lives as consumers. And I feel like companies in that position are actually in a good you know, you know, and I I put ramp in this category too of like you're taking AI a lot more seriously than Apple or at least like getting more value out of it for for users than than Apple.
但你处于这样的位置，不需要说‘哦，我们今天必须投入2亿美元到这个新产品上，否则就会落后’。不是这样的。我们拥有这些非常稳固的客户关系，我们可以通过多种方式逐步释放AI的价值。
But you're in that position where you're not saying like, oh, we have to pour $200,000,000 into this new product today. Otherwise, we're gonna be left behind. It's like, no. We have these like really sticky customer relationships and we can unlock the value of AI over time in a number of different ways.
嗯。是的。你看，100%的意思是，自从我们创办这家公司以来，我们一直觉得相对于面前的机遇规模，资源不足。我认为AI的采用非常令人难以置信，我们不需要大幅改变态度或感到必须扭曲自己。
Mhmm. Yeah. Look. A 100% of I mean, since we started this this company, we felt under resourced compared to the size of the opportunity in front of us. And I'd say that the the AI adoption was just incredibly we didn't have to change our our attitude that much or feel like we had to contort.
它就像是一个非常受欢迎的解锁，让我们能够用有限的资源做更多事情。所以我认为这是我们能够快速充分利用它的部分原因，采用速度非常快，并且有一种理解，就像‘嘿，这对我们来说可能非常棒’。但你可以争论，对于一些大公司来说，也许会发生，我的意思是，存在复杂性和规模等问题，但可能发生得更慢，因为有时你不觉得资源那么紧张。
It was, like, a very welcome unlock in our ability to just do more with the limited resources that that that we have. So I think that's part of the reason we're able to, like, get the most out of it very quickly, and it was, like, very quick adoption and and kinda, like, this understanding of, like, hey. This can be incredible for us. But you could argue that, like, may maybe for some of the larger companies, maybe it happens. I mean, there's complexity and size and all these things, but, like, it might happen slower because you don't feel as resource constrained sometimes.
所以，像采用和改变的推动力可能稍微低一些，因为，是的，正如你所说，既没有威胁，也没有作为小公司可能面临的资源限制。
So, like, the push to adopt and and and and change, like, is is maybe a little bit lower because, yeah, as you said, like, there's neither a threat nor the resource constraints that you might have as a smaller company.
时间到了，但我们这周没谈到收购。给我们分解一下。为什么那有意义，你最兴奋的是什么？
We're at time, but we've we didn't cover the acquisition this week. Break that down for us. Why why did that make sense, and what what are you most excited about?
这很令人兴奋。所以我们从Jolt AI引进了一个团队并收购了他们的公司，我对此非常兴奋有几个原因。其一，我认为工程师尤其可能在理解AI的能力和意义方面比大多数同行领先一步。这就像每种技术一样，工程师倾向于先为自己构建工具，而Jolt AI非常专注于构建一个代理编码助手或代理编码员，基本上是软件工程师。
It was exciting. So we brought on board a team from Jolt AI and and acquired their company, which well, I'm very excited about for a couple reasons. One, I actually think engineers in particular are maybe a step ahead compared to most of their peers in terms of, like, understanding the capabilities of AI and what it means. And, like, that tends to be true with with every technology. It's like engineers tend to build tools for themselves at first, and this is what Jolt AI was very focused on, like building an agentic coding assistant or agentic coder, basically, software engineer.
而且他们一直痴迷于，比如，构建什么样的用户体验才是正确的，以帮助工程师更多地采用AI，并在AI的帮助下编写代码。我认为这种技能组合不仅对我们Ramp内部工程团队正在努力实现的目标极具价值，更重要的是，我认为软件工程层面发生的同样转变即将在其他所有行业发生，我们需要专注于为财务团队构建合适的智能代理，正确的代理能力。对于Yev和他的团队，我们非常兴奋地去追求
And and they they they've obsessed over, like, what the right user experience to to to build is, to help engineers adopt more AI and and and write code with the help of AI. And I think that skill set is is not only incredibly valuable for what we're trying to do internally at Ramp for our own engineering teams, but more importantly, I I I think that same transformation that happened at the software engineering layer is about to happen in in every other industry, and we need to obsess over what it's gonna take to build the right agents for finance teams, right agentic capabilities. And with Yev and and and his team, we're very excited to to go after
快速问一下。这笔交易是怎么达成的？你们有共同的投资者吗？你们是在使用他们的产品，还是他们只是冷邮件联系你说，嘿，我想找工作之类的。
quick question. How did the deal come together? Did you have investors in common? Were you using the product, or did they just cold email you and say, hey. I want a job or something.
买下我的公司。我总是对这些事情如何达成感到着迷。
Buy my company. I'm always fascinated by how these things come together.
我们有共同的投资者，我认为，他们觉得那里有很强的文化契合度，而且，我猜，很多文化上的一致。我们见面后很快就一拍即合，完美。而且我们行动非常迅速。
We had investors in common, I think, that that they felt like there was strong cultural culture fit there and, I guess, a lot of cultural alignment. And we we hit it off very quickly after we met, and Perfect. And we moved very fast.
是的。从见到团队到真正达成交易用了多长时间？因为X上有个梗，比如，你会在收购者买下你公司的十年前遇到他们。但听起来
Yeah. What was the time from meeting the team to actually doing the deal? Because you there's this meme on X about like, you'll meet your acquirer a decade before they buy your company. But it sounds
像这样X上也有关于Ramp的梗，你知道，有人报告了一个bug，然后
like this There's also meme on X of Ramp, you know, feet you know, somebody's reporting a bug and then
Ramp立即修复了它。
Ramp fixing it immediately.
三十分钟后。
In thirty minutes.
那是
It was
大约一个月。
it was about a month.
大约一个月。
About a month.
这就对了。
There we go.
嗯，本来可以更快些，但大概花了一个月到一个半月，你知道吗？
Well, could have gone faster, but about a month to month and a half, you know?
敲响那个锣。
Hit that gong.
很喜欢。
Love it.
恭喜。非常感谢你的光临。聊天室里的各位，
Congratulations. Thank you so much for stopping by. Everyone in the chat,
享受这个
enjoy the
能叙叙旧总是很棒。
Always great to catch up.
站。我们会和
Station. We'll talk to
你很快再聊。
you soon.
一周年纪念日快乐之类的。
Happy one in one year anniversary or so.
没错。谢谢你。
That's right. Thank you.
没错。
That's right.
我们很感激。很快再见。回头聊。拜拜。Jordy，你想不想一起看看Doug O'Laughlin这篇关于泡沫潜在轨迹的文章？
We appreciate it. We'll see you soon. Talk soon. Bye. Jordy, would you like to go through this, Doug O'Laughlin post about the potential trajectory of a bubble?
是的。他正在阐述。首先，让我跟你聊聊Privy，这是一家Stripe旗下的公司，为每家银行提供钱包基础设施。Privy让安全地构建加密应用变得简单，可以快速部署白标钱包、签署交易，并通过一个简单的API集成链上基础设施。如果你还想了解几位曾在Ramp工作、现在运营Cognition的人，他们是AI软件工程师Devon的创造者。
Yeah. He's laying it out. First, let me talk to you about Privy, a Stripe company wallet infrastructure for every bank. Privy makes it easy to build in crypto real securely, spin up white label wallets, sign transactions, and integrate on chain infrastructure all through one simple API. And if you also wanna know about a couple folks who used to work at Ramp and now are running Cognition, they're the makers of Devon, the AI software engineer.
我们那里有一些Ramp的血统。用你的个人AI工程团队碾压积压任务。双杀。我们从这里往哪里去？
We got some Ramp lineage there. Crush your backlog with your personal AI engineering team. Double kill. Where we go from here?
我刚才只是在把音效大声说出来。
I was just saying saying the sound effects out loud.
这里没有音效板。马叫声。马叫声。狗叫声。鹰叫声。
Don't have a soundboard in here. Horse. Horse. Barking. Eagle.
是的。让我们从这一点开始：这是极其具有投机性的。
Yeah. Let's start with the fact that this is a that this is massively speculative.
这确实很多
This is definitely a lot
有趣。从这里开始。是Substack，你应该订阅。没有人能猜测或知道任何事情的未来。而今天的帖子是我对整个市场动物精神的猜测。
of fun. From here. Is Substack, which you should subscribe to. No one can guess or know the future of anything. And today's post is my guess of the entire animal spirits of the market.
我在这里要做一个预测。我们将进入一个由GPU引领的泡沫。这是我上一篇文章的后续。我想首先讨论我认为这会发生的主要原因。各种条件正在成熟，而且没有单个参与者反对这种狂热。
I'm gonna make a call here. We are going to go into a GPU lead bubble. This is a follow-up to my last post. I would like to begin by discussing with the primary reason why I believe this will happen. The stars are aligning, and no individual actor is rooting against the frenzy.
这个泡沫可能与过去的泡沫不同且更大，因为它的指令似乎来自最高层。特朗普政府坚持在美国投资资本。每个人都兴奋的项目是GPU。除了纯粹的炒作之外，我认为我们开始看到多个支出的绿灯。我将使用的第一个也是主要的时机指标是降息。
The bubble might be different and larger than past ones as its mandate seems to originate from the top. The Trump administration is insistent on investing capital in The United States. The project everyone is excited about is GPUs. Just beyond the raw the raw hype, I think we're starting to get multiple green lights for spending. The first and primary timing metric I'm going to use is a rate cut.
在上一个泡沫中，1998年9月的降息让事情变得真正疯狂。我们又经历了一次9月降息，现在我的基本预测是这种情况会疯狂持续到2026年。所以很多人说，感觉像1999年。道格·奥洛克林说感觉像1998年，这是一个有趣的观点。这就是问题所在。
In the last bubble, the rate cut in September 1998 made things go truly crazy. We had another September rate cut, and now my base case is that this goes crazy into the 2026. So a lot of people are saying, feels like 1999. Doug O'Loughlin says it feels like 1998, which is an interesting take. And that's the question.
我们之前一直在争论这个。我们觉得，预测顶部很容易。但感觉并不容易通过预测顶部来赢得一次预测。
We were debating this earlier. We were like, it's it's it feels easy to call a top. It doesn't feel like it's easy to call the top to, you know, win one call.
任何人最多能在两年内达到顶峰。是的。但这毫无意义。没有任何意义，因为你仍然可以，你知道，假装你
The top anybody can call the top to within two Two years. Yeah. Which means nothing. Doesn't mean anything because you can still get a, you know, pretend you
知道？是的。
know? Yeah.
你所有的收益，你知道，那个 是的。
You all the gains, you know, the the Yeah.
就像泡沫的最后十二个月带来了80%的收益。这是最危险的时期，所以大家要小心。但互联网泡沫的另一个重要基本面理由是每小时工作生产力的爆炸式增长。果然，在最近GDP预公布值为3.8之后，我们再次看到这种情况发生。我们不仅增长迅速，而且每小时生产力正在爆炸性增长。
It's like 80% of the gains come in the last twelve months of the bubble. It's the most dangerous time, so be safe out there. But another important fundamental justification for the Internet bubble was the exploding productivity per hour worked. And lo and behold, we are seeing that happen again after the recent GDP pre GDP print of 3.8. We are not only growing fast, but productivity per hour is exploding.
这是AI生产力最强烈的绿灯信号之一。虽然目前还没有很多产生收入的步骤，但总体生产力的提升几乎是你所能获得的最好情况，对于
This is one of the strongest green lights of AI productivity. And while there are a lot not a lot of revenue generating steps, general productivity increases are about as good as you're ever gonna get for
展开剩余字幕（还有 480 条）
AI的布尔什维克。这里的GDP增长是否超过50%是基于整体资本支出，而不是实际生产力？
the AI bolshie. Here be that GDP growth was more than 50% based on just overall CapEx spend and not actual productivity?
我不太确定具体怎么...是的。这是非农生产力季度环比数据，我们看到过去几年有轻微上升。它们肯定比2015年时期要好。尽管GDP修正是旧闻，但正如杰拉德在这里指出的，它们确实发出了重要信号。GDP的上调和对就业增长即将进行的下调修正表明，基本情况是，衡量出的生产力增长可能会在第二季度上调约60个基点，或许前四个季度也会同样上调。
I'm not exactly sure how how yeah. This is nonfarm productivity quarter on quarter, and we're we're seeing a slight uptick over the last few years. They're certainly off certainly better than the 2015 period. Although GDP revisions are old news, they did signal something important as Gerald Gerard pointed out here. The upward revisions to GDP and the coming downward revisions to employment growth suggest as the base case that measured productivity growth is likely to be revised up by about 60 basis points for q two and perhaps by the same amount for the prior four quarters.
我的意思是，这确实符合AI是思维的自行车、计算机是思维的自行车这一理念。它是一种生产力提升工具。对我来说这并不显得反直觉。当然，Figma就是设计师的自行车。思考得更宏大，构建得更迅速。
I mean, it certainly matches with the idea that, like, AI is a bicycle for the mind, computer is a bicycle for the mind. It's it's a productivity enhancing tool. It doesn't read as counterintuitive to me. Of course, Figma is a bicycle for the designer. Think bigger, build faster.
Figma帮助设计开发团队共同打造优秀产品。你可以免费开始使用Figma。所以当人们在售卖代币上亏损大量资金时，总体而言，它开始‘运作’了。如果我们看到由AI驱动的劳动生产力达到新高，那么对AI支出的合理性几乎将没有上限。他还谈到了数据中心的问题。
Figma helps design development teams build great products together. You can get started for free at Figma. So while people are losing lots of money selling tokens, in aggregate, it's starting to quote unquote work. And if we see new highs in labor productivity driven by AI, there will be almost no limit to the justification of AI spending. And he goes into the data center question.
你知道纳格尔斯输了当读到
You know Nuggles lost when read
你的想法。
your mind.
刺激经济。现在让我们转向下一个关键部分。虽然我们知道模型制造商在亏损，但等式的数据中心一侧对经济具有巨大的乘数效应。到目前为止，政府的直接指导是投资美国。而人们想要投资的方式就是投资GPU。
Juicing the economy. Now let's move on to the next critical part. While we know the model makers are losing money, the data center side of the equation is massively multiplicative for the economy. And so far, the direct guidance from the government is to invest in America. And the way people want to invest is in GPUs.
此外，电力资源丰富的地区通常比以往的许多繁荣时期更加偏远和均匀分布。就像国防支出法案经常在许多县之间分配一样，我认为数据中心层面的AI支出也有类似的效果。这产生了一种奇特的现象，整个生态系统中没有人对这笔支出感到不满。我会把这归为有点真实，因为似乎普通市民总体上对支出感到不满，因为它可能影响电价。但我会继续。
Furthermore, the locations where things are power rich are often much more remote and evenly distributed than many previous booms. In the same way that defense spending bills are often split among many counties, I believe that AI spending at the data center level has a similar effect. This makes an odd effect where no one in the entire ecosystem is upset about the spending. I would put this a little bit in the truth zone because it seems like every day citizens have been are generally upset about the spending because of the potential implications to electricity prices. But again, I'll continue.
它以过去技术繁荣未曾有过的方式与现实经济进行有意义的互动。虽然你在代币上亏损，但租赁GPU在今天是一个非常赚钱的业务。成本完全由OpenAI或Anthropic承担，而超大规模供应商非常乐意出售‘镐和铲’。现在，对我来说真正的问题是，超大规模供应商愿意走多远？因为对于这些企业中的大多数来说，这是历史上第一次开始变得资本密集。
It interacts with the real economy in meaningful ways that past technology booms have not. And while you lose money on tokens, renting GPUs is a very profitable business today. The cost is born entirely by OpenAI or Anthropic, and the hyperscalers are more than happy to sell picks and shovels. Now, the real question to me is, how willing are hyperscalers to go further? Because for the first time in the history of most of these businesses, they're starting to become capital intensive.
看看这张简化图表，展示了经营现金流与资本支出的对比。直到最近，增长所需的资本投入都非常少。现在你可以快速增长，但这需要付出代价。这里有一张图表我们可以调出来。道格说，西方科技巨头们，你们选择哪条路？
Check out the simplified graph of operating cash flow versus CapEx. Up until very recently, it took very little capital to grow. Now you can go grow very quickly, but it will cost you. And there is a chart here we can pull up. Doug says, which way, Western tech giant?
我们正处在一个十字路口。甲骨文正在推动负自由现金流以资助更多GPU采购，并从中获得了显著的市场份额。其他公司会何去何从？重要的是，尽管这些公司——我还会再次介入这个话题——记得萨提亚有点踩了刹车，是的。
We are at a crossroads. Oracle is the one that is pushing the negative free cash flow to fund more GPU purchases, and they're picking up meaningful share from this. Where where do the others go? What is important is that while these companies and I would jump in there again. Remember Satya kind of took his foot off the gas Yep.
稍微踩了点刹车。而拉里说，全速前进，绝不刹车。我们冲吧。所以道格指出，重要的是，虽然这些公司由股东所有，但许多并非由股东亲自运营。在主要玩家中，Meta和谷歌主要由持有各自股份类别多数控制权的创始人运营。
A little bit. Larry said, all gas, no brakes. Let's go. So Doug says, what is important is that while these companies are owned by shareholders, many of them are not run by their own shareholders. Among the major players, Meta and Google are primarily run by founders who hold majority control over their respective share classes.
毫不奇怪，目前最激进的正是那些由创始人控制的公司。我相信他们将开始大肆挥霍。例如，拉里·佩奇曾表示，他宁愿破产也不愿输掉这场竞赛。与此同时，扎克伯格将花光所有现金流来保卫Instagram和Facebook。新的Sora应用是对Meta的首个直接挑战，我认为唯一合理的回应是大力投入，招募更多AI人才，抵御这一新威胁。
It's no surprise that the ones that are most aggressive at this point in time are those that are founder controlled. I believe they will begin the splurge. Larry Page, for example, has said that he would rather go bankrupt than lose this race. Meanwhile, Zuckerberg is gonna spend all his cash flow to defend Instagram and Facebook. The new Sora app is the first direct challenge to Meta, and I believe that only sensible response is to spend heavily, invest in even more AI talent, and stave off this new threat.
如果Sora，再次作为这样一个独立的视频应用，只是为了刺激扎克伯格更加疯狂地花钱，那会很有趣。
It would be funny if Sora, again, as this like standalone video app, is just to get Zuck to just go spend spend even.
非常有趣的假动作。是的。聊天区有个问题：AI领域的Theranos/FTX/Enron事件何时会成为头条？如果现在是1998年，我们可能会预期那将在两年后发生。
Very interesting head fake. Yeah. There's question in the chat. When will a TheranosFTXEnron of AI be in the headlines? If it's 1998, we would expect that to happen probably two years from now.
如果我们完美对标历史，这显然很荒谬。有趣的是，Theranos的市值我认为不到100亿美元，FTX是320亿美元，Enron是700亿美元。与一些大公司相比，这些看起来都微不足道。而且，这些公司当时都不是推动市场的主力。比如FTX很重要，但Coinbase规模更大，并且Coinbase挺过来了。
If we're mapping this perfectly, that's obviously ridiculous. What's interesting is that Theranos was, I think, under $10,000,000,000 FTX was $32,000,000,000 and Enron was $70,000,000,000 in market cap. Those all look tiny by comparison to some of the big companies. And, also, those companies weren't none of them were actually the thing that was driving the market at the time. Like, FTX was important, but Coinbase was still bigger, and Coinbase made it through.
Theranos确实规模很大，但你知道，2012年那批初创公司里还有很多其他企业表现不错，比如那些成为十角兽的公司。Airbnb、DoorDash，我们都和这些公司的创始人聊过。他们成功挺过来了，并非全都是骗局。
And Theranos was big, but there were plenty of other, you know, 2012 era startups that were, you know, in that crop of, like, Decacorns that did fine. Airbnb, DoorDash. We've talked to the founders of these companies. Like, they made it through. It it wasn't all frauds.
安然公司也是如此，银行业危机时也一样。甚至雷曼兄弟、贝尔斯登都倒闭了。美国银行、摩根士丹利、摩根大通、高盛这些公司都熬过了低谷期。有些确实需要沃伦·巴菲特带着空白支票来救场，但没错，它们确实挺过来了。
And Enron, the same thing with the banking crisis. And even Lehman Brothers, Bear Stearns failed. B of A, Morgan Stanley, JPMorgan, Goldman Sachs, those companies all continued to get through the trough. Some of them needed, you know, Warren Buffett to show up with a with a blank check, but Yep. They did make it through.
所以如果一切都顺利的话，我会非常惊讶。
And so I I would be very surprised if everything goes.
记住，FTX和SVB之间存在着显著的时间差。没错吧？是的。FTX的崩溃并没有直接导致SVB的垮台。SVB自身存在期限错配的问题，其资产负债表中有大量资产极度...
Remember, there was a meaningful gap between FTX and XVB Yep. Right? Yep. In which it wasn't a it's not like the collapse of FTX directly caused the collapse of SVB. They had their own sort of duration mismatch issue around a bunch of their balance sheet being extremely
所以如果有一个AI领域的十角兽公司爆雷或者只是逐渐关闭，我也不会感到意外。我甚至不认为这一定是纯粹的欺诈。根据当前基本的风险投资数学，如果你投资了10家估值数十亿美元的增长阶段AI公司，你会预期其中一家会失败。如果你投资了多家这样的公司，作为基金你仍然会承保这种风险。我不断回想起Sam Altman和Ben Thompson的那次访谈，我总觉得英特尔迟早会卷入其中。
And so I wouldn't be surprised if if there's a an AI decacorn that blows up or maybe just winds down. I don't even know it would be pure fraud. Just the basic venture math right now would be if you're betting on, you know, 10 different AI growth stage companies in the multi billions, you'd expect one of them to go down. You would still underwrite that as a fund if you're in if you're in a bunch of them. I keep going back to this interview between Sam Altman and and, and Ben Thompson, and I just feel like Intel is going to come into the picture at some point.
Ben Thompson问Sam：‘问题是英伟达和AMD都从同一个地方采购芯片。所以价值链中还有另一个独立的实体，就是台积电。你认为有必要、有责任或有机会也拓展那个市场吗？当涉及到英特尔的问题时，这是否值得考虑？’Sam Altman回答说：‘我希望台积电能扩大产能。’
He's he's asking Ben Thompson asks, Sam, well, the problem with this is both NVIDIA and AMD are sourced at the same place. So there's another solitary entity in the value chain, which is TSMC. Do you see a need and responsibility or opportunity to expand the market there as well? Is this something when it comes to the question of Intel? And Sam Altman says, I would like TSMC to just build more capacity.
你以为我问的是什么？多芯片供应商？我是否认为有必要让台积电扩大投资增加产能？所以Sam实际上是在说，他希望台积电扩大规模，但也很容易去白宫说：‘他们没有扩大产能，我们需要在美国制造这些芯片。’
What do you think I was asking about? Multi chip suppliers? Do I see a need to get TSMC to expand their rate of investment in more capacity? And so Sam is is is saying, I want TSMC to scale up, but it'd be very easy to go into the White House basically and say, like, they're not scaling up. We need to make this in America.
英特尔正在建设一座能同时生产英伟达和AMD芯片的晶圆厂。他们还与博通和SK海力士合作。看起来，除了晶圆厂之外，所有环节都已就位。我认为他们唯一还没与台积电达成实质性协议的就是这个环节。
Intel's working on a fab that could make both NVIDIA and AMD chips. And then they have Broadcom and SK Hynix. Like, all the pieces are together except for the fab. That's the one place that I think I think they haven't done a real deal with TSMC.
是的。
Yeah.
所以我感觉这将会成为...
And so I I just feel like that's gonna that would be
没错。
Yeah.
今年真正的惊喜。
Real surprise in the year.
回顾早先的交流：本说过，这里的问题在于英伟达和AMD都从同一来源采购，所以价值链中另一个独立实体就是台积电。是的。您是否认为这里有扩大市场份额的需求和责任/机遇？当涉及英特尔问题时，山姆表示：我只希望台积电能扩大产能。
Highlighting from the exchange earlier. Ben said, well, with this, the problem with this is that both NVIDIA named you are sourced at the same place, so there's another solitary entity in the value chain, is TSMC. Yep. You see a need and responsibility slash opportunity to expand the market share market there as well? Is this something where when it comes to the question of Intel, and Sam says, I would just like TSMC to build more capacity.
是的。本当时说的是...我问的是关于多芯片供应商的问题。
Yeah. And Ben says What did the thing what did I was asking about multi chip suppliers.
布特说了什么？
What did Bootter say?
Sam说，我需要让台积电扩大投资增加产能吗？问号。Ben说，明白了。然后他说，又和CEO进行了一次关于使用英特尔的尴尬对话。当Ben提到英特尔时，Sam打断了他。
Sam says, do I need to get TSMC to expand the rate of investment in more capacity? Question mark. Ben says, got it. And he says, another awkward convo with the CEO about using Intel. Sam cuts Ben off when he mentions Intel.
没人想得罪台积电。没人想为保险买单。
Nobody wants to piss off TSMC. Nobody wants to pay for insurance.
和我的看法有点相反。
Kind of the opposite of my take.
他说特朗普会给他们一个无法拒绝的报价。所以，再次说明，特朗普对他目前入股英特尔一事应该相当满意。
He says Trump will make them an offer they can't refuse. So, again, Trump is gotta be quite happy with his entry in the current into intel.
感觉这件事在铁板钉钉之前你不想谈论，但在美国建立一个能与AMD和英伟达都合作的、达到台积电水平的晶圆厂，听起来非常非常合理。而Sam正是担任这项投资银行交易的完美人选，对吧？
It feels like something you don't wanna talk about until it's ironclad, but it feels very, very logical to build a a TSMC level fab in The United States that can work with both AMD and NVIDIA. And that would be something that Sam is the perfect person to be the investment banker on. Right?
是的。总之，Doug继续说了。他说，Meta将是第一个开始的。通过增加支出，他们可以确保获得更多的AI供应，可能损害竞争对手的市场份额。对于已经开始加速的谷歌来说，他们会追赶Next。
Yep. This is And so anyway, Doug continues. He said, Meta will be the first to begin. And by spending more, they can secure a larger supply of AI, potentially harming their competitors' market shares. For Google, which already is starting to accelerate, they will chase Next.
这种动态让我想起了内存市场，通过增加供给来抢占份额。对甲骨文来说，这很字面化——他们愿意投入最多资金，意味着他们能从超大规模云服务商的租赁业务中夺取实际份额。因此，这个由科技公司组成的不稳定联盟（此前各自拥有封闭生态系统）将通过支出开始瓦解。如果Meta大力投入或增加借贷，可能一夜之间成为最大的超大规模云服务商。谷歌若更激进，特别是进一步推广TPU，可能颠覆AWS的统治地位。
The dynamics remind me almost the memory market where supply increases can be used to gain share. For Oracle, this is literal as they're willing to put down the most money, means they can take a real share from the hyperscaler's rental business. So the uneasy coalition of technology companies, previously each had their own walled garden, will start to defect via spending. Meta could become the biggest hyperscaler overnight if they spent heavily or borrowed more. Google could upend AWS by being more aggressive, especially with a further push into TPUs.
竞争正在加剧，FOMO（错失恐惧症）与支出欲望的结合似乎是当前策略。这种不安感进一步加剧——不安的竞争与宽松的信贷。在我看来，之所以开始令人感到轻蔑，是因为与由少数未盈利上市公司推动的互联网泡沫不同，今天的泡沫是由资本主义历史上最大且最盈利的公司驱动的。
Competition is increasing in FOMO mixed with the desire to spend seems to be the playbook. Uneasy and he goes further, uneasy competition and easy credit. To me, the reason this is starting to feel contemptuous is that unlike the .com bubble, which was led by a few unprofitable public companies, today's bubble is driven by the largest and most profitable companies in the history of capitalism.
是的。说得很好。而且
Yep. That's good. And
它们在资本市场的领先地位是其他公司所没有的巨大优势。再次强调，OpenAI不具备这种优势。嗯。这就是为什么他们需要达成一系列大规模交易才能真正具有竞争力。科技七巨头在全球股权市场占据如此大的份额，以至于可以说整个信贷市场对科技巨头的配置不足。如果科技巨头转向贷款机构，我认为信贷市场会欢欣鼓舞。
their pole position in capital markets is consider is a considerable advantage that others do not have. Again, OpenAI does not have this advantage Mhmm. Which is why they're needing to do a bunch of these massive deals in order to really be competitive. The Mag seven is has such a large share of equity markets globally that you could argue that the entire credit market is underweight in the big tech giants. And if tech giants turn to lenders, I think the credit markets would be joyous.
拉里已经转向了。Meta与BlueOwl达成了交易。是的。大约一个月前。所以我们开始看到这种情况发生。是的。
Larry's turned already. Meta did a deal with BlueOwl Yep. About a month ago. So we're starting to see this happen. Yeah.
DoubleLine最近有份报告讨论：你更愿意贷款给美国政府还是微软？结论是微软。衡量这一点的一种方式是微软的G-spread（信用利差）
There was a recent report from DoubleLine talking about would you rather lend to the US government or Microsoft? And the conclusion was Microsoft. One way to measure this is Microsoft's g spread
是的。
Yep.
或者分散投资到政府债券上。目前是五个基点。
Or spread over government bonds. It's five basis points currently.
我以为
I thought
实际上曾经更低，但我觉得现在只是稍微高了一点。是的。这是
it was actually lower at one point, but I guess it's just slightly higher. Yeah. This is So
虽然微软的利率最好
while Microsoft has the best rate
我记得有一段时间苹果债券的交易利率低于美国政府债券。我不确定这是否
I think there was a moment when Apple bonds were trading lower than US government bonds. I don't know if that's
蒂姆债券。我是说谷歌、亚马逊和Meta的债务利率仅比政府收益率高出50个基点。真正的问题在于流动性，因为原始债务量无法填补美国国债所起到的功能性管道作用。当然，全球经济。但说实话，如果我们想尝试，我认为会有需求的。
Tim bonds. Mean Google, Amazon, and Meta all have debt that is only 50 basis points over government yield. The real problem is just liquidity as the raw amount of debt wouldn't be able to plug the functional plumbing that UST bills serve Sure. The global economy. But honest, if we wanted to try, I think there would be demand.
所以他加入了奥特曼的联盟。我认为在这一点上，萨姆·奥特曼的目标是成为每个人收入中如此庞大且根深蒂固的一部分，以至于每个人的既得利益都希望OpenAI成功。这就是我之前谈到的。对吧？就像，如果奥特曼是唯一一家私营公司，而有一堆上市公司基于他的收入增长进行交易。是的。
So he gets into the coalition of Altman. I think at this point, the goal of Sam Altman is to become such a large and entrenched part of everyone's revenue that everyone's vested interest is seeing OpenAI succeed. This is what I was talking about. Right? It's like, if Altman is the only private company with a bunch of public companies that are trading based on his his his revenue growth Yep.
对吧？而且他与各种合作伙伴的投入，是的，他们都有动力确保OpenAI拥有资源和基础设施，能够持续投入、投入再投入。对吧？萨姆需要大规模提升收入，以支持所有已完成的交易。
Right? And and that his, like, spend with various players Yep. They all have an incentive to make sure that OpenAI has the resources to and infrastructure to be able to continue to spend and spend and spend. Right? Sam needs to massively scale revenue in order to support all of the deals that have been done.
是的。对吧？所以
Yep. Right? So
嗯，如果你想与大公司做生意，你需要合规。你需要使用Vanta，自动化合规管理风险，持续提升信任。Vanta的信任管理平台将安全合规流程中的手动工作自动化，代之以持续自动化。
Well, if you wanna deal do a deal with a big company, you need to be compliant. You need to get on Vanta, automate compliance management risk, improve trust continuously. Vanta's trust management platform takes the manual work out of your security and compliance process and replaces it with continuous automation.
所以道格说，从这个角度看，NVIDIA的交易是最好的例子，还有刚宣布的韩国内存交易。是的。其中一些交易只是为了报出最大的数字，但我相信这正达到一个规模，以至于所有人都不经意地将利益与OpenAI对齐。这是一种疯狂的战略，因为它把所有人都绑在同一枚火箭上，不参与就意味着你的收入增长会更差或变得无关紧要。真正能抵抗的公司寥寥无几，但它们必须投入更多才能参与游戏。以Meta为例。
So Doug says the NVIDIA deal is best viewed from this light, but also the Korean memory deal that just got announced. Yes. Some of these are simply the factor of saying the largest number, but I believe it is getting to a scale that everyone is inadvertently aligning their interest in OpenAI and so This is kind of a crazy strategy because it straps everyone to the same rocket and to not take part of it means that you will have worse revenue growth or become irrelevant. There really is only a few companies who can resist, but they will have to spend even more to be a part of the game. Take Meta for example.
他们试图不像对App Store依赖苹果那样被OpenAI束缚，但他们仍将与大多数内存制造商、NVIDIA、Oracle以及较小程度的微软组成的联盟对抗。谷歌是另一个玩家，他们拥有所有合适的工具，但投入规模不同。我认为这很快就会改变。是的。与此同时，亚马逊在规模上排名第三，他们与Anthropic的合作策略加上最差的加速器计划，感觉较弱。
They are trying not to be tied to OpenAI like they were to Apple for the App Store, they but are still going to be fighting against the coalition of most of the memory makers, NVIDIA, Oracle, and to a lesser extent Microsoft. Google is another player and they have all the right tools but are not playing at the same magnitude. I think that changes soon. Yep. Meanwhile, Amazon is dwindling third in terms of scale and their Anthropic strategy paired with the worst accelerator program of all of of all feels weak.
Anthropic已经开始转向谷歌的TPU，而不是Tranium。
Anthropic is already starting to turn toward the Google TPU instead of Tranium.
艰难啊。
Rough.
但我们要清楚，这是OpenAI，以及Sam能向全世界筹集到的巨额资本——数字看起来相当庞大。对OpenAI的倾向性是一种强大的工具。这就像，如果你欠银行一千美元，那是你的问题。如果你欠银行一百亿美元
But let's be clear, it's OpenAI and as much capital as Sam can raise against the world, and the numbers seem to be a lot. The alignment toward OpenAI is a powerful tool. It's akin to, if you owe the bank a thousand dollars, it's your problem. If you owe the bank $10,000,000,000
你在说
You're talking
那就是银行的问题了。现在，英伟达和芯片制造商将被套牢，并可能投资并满足OpenAI更高的资本需求。整个供应链正在赚取他们有史以来最多的钱。而现在他们必须将部分利润回馈给驱动者。这就是英伟达的交易，我预计很快会有更多企业主导的融资。
it's the bank's problem. Now, Nvidia and the chip makers are going to be on the hook and can probably invest and fuel the capital of needs of OpenAI higher. The entire supply chain is making the most money they have ever made. And now now they will have to pay some of that back to the driver. That's the Nvidia deal and I expect more corporate driven fundraising soon.
他最后说，星辰已经对齐。每个人——我是说除了微软的Amy Hood之外的每个人——都在期盼一个人工智能泡沫。我不相信现在有任何人 actively 反对它。政府、行业和金融界都渴望尽可能快地增长。这几乎肯定会以不如期望的方式结束，但那还是很遥远的事情。
And he finishes it off by saying the stars align. Everyone and I mean everyone except Amy Hood at Microsoft is rooting for an AI bubble. I do not believe that any anyone is actively rooting against it today. The government, industry, and finance are all excited to grow as fast as possible. This will almost assuredly end not as good as hoped, but that is a long, long time from now.
在动物精神蓬勃复苏之前，我们将迎来辉煌的GDP增长。下一步是观察谷歌和Meta将支出推高至自由现金流之上，我认为这是下一阶段AI交易的火箭燃料。如果他们不选择迈出这关键一步，这可能是一篇毫无价值的帖子。嗯。但星辰感觉比那更加对齐。
We will have glorious GDP growth before as animal spirits roar into life. The next step is watching Google and Meta up the spending past their free cash flow, which I think is rocket fuel for the next stage of the AI trade. If they don't choose to make this critical step, this may be a worthless post. Mhmm. But the stars feel more aligned than that.
嗯，你得去看看graphite.dev。AI时代的代码审查。Graphite帮助GitHub上的团队更快地交付更高质量的软件。我迫不及待想看到接下来几个季度的财报季。我迫不及待想看看谷歌和Meta是否会现金流转负，是否会开始发行债务。
Well, you gotta get on graphite.dev. Code review for the age of AI. Graphite helps teams on GitHub ship higher quality software faster. I cannot wait for earning season over the next couple quarters. I cannot wait to see if Google and Meta should, like, dip cash flow negative, if they start issuing debt.
就像，我们会发现这些
Like, we're gonna find these
所有人都在关注Meta的财报，10月29日。是的。另外还有英特尔财报的相关消息。
All eyes on Meta earnings, October 29. Yes. In other Intel earnings. Related news.
英特尔财报发布在万圣节前夜。非常诡异。
Intel earnings are the night before Halloween. Very spooky.
诡异。10月3日。是的。乔什·沃尔夫
Spooky. October 3. Yeah. Josh Wolfe
是的。
Yes.
在做空。嗯。QQQ。是的。他说，虽然Lux在人工智能领域大举做多，从边缘推理到开源，再到开发工具基础设施、应用程序以及现实世界的人工智能应用。我相信顶级私营公司，我们和其他风投机构的估值还远未达到公允水平。
Is shorting Mhmm. QQQ. Yes. He says, while Lux is hugely long AI from edge inference to open source to dev tools infrastructure to applications to AI in the real world. I believe the top private companies, we and other VCs are not even close to fairly valued yet.
随着企业从使用60家SaaS提供商减少到仅两三家，因为人工智能在内部完成了大部分工作，对SaaS业务的破坏尚未完全显现。但很有可能我的看跌期权会一文不值地到期，而人工智能的共识将继续，散户也会涌入。唉，为这种盲目共识付出的代价很高。我交谈的每个人都对数据中心、能源需求、GPU需求以及新闻周期对任何涉及人工智能的交易公告的反应持乐观共识，而资产负债表上的股权投资后来被记为收入，推动超出有机需求或支付能力，这让人想起两千年代初泡沫时期的电信往返交易，现在杠杆正在进入系统。所以他几乎是在反驳自己的投资论点。
The destruction to SaaS business is still to come as companies go from using 60 SaaS providers to like two or three as AI does much of this internally. But it is very plausible my puts expire worthless and AI consensus continue and retail pours in. Alas, you pay a high price for cherry consensus. Everyone I talk to is bullish consensus on data centers plus energy needs plus GPU demand plus the news cycle's reaction to any deal that gets announced with AI involved, and the balance sheet equity investments later used to book as revenue driving beyond organic demand or ability to pay are reminiscent of early two thousands bubble telecom round trip deals and now leverage entering the system. So he's kind of almost making the case against his own investment.
对吧？嗯。就像是杠杆正在进入系统，所以我们拭目以待。
Right? Mhmm. Being like leverage is entering the system so we'll see.
可能是个早间节目。虽然早，但无论长短，对于认真对待投资的人来说，上public.com进行多资产投资，行业领先，受数百万人信赖。我发现一个有趣的Polymarket想重点说一下。显然，Polymarket上有一个关于2025年哪部电影首周末票房最高的市场。你能猜到哪部领先吗？
Could be an early call. Well early, but long or short, get on public.com investing for those that take it seriously, multi asset investing, industry leading, trusted by millions. I found an interesting Polymarket I wanna highlight. Apparently, Polymarket has a market up for which movie has the biggest opening weekend of 2025. Can you guess what is in the lead?
演播室里有人吗？有电影迷吗？你们认为今年最大的电影会是哪一部？
Anyone in this studio? Any movie buffs? What do you think is the biggest the biggest film of the year is gonna be?
我只知道今年要上映的一部电影是《一战接一战》。
The only movie I know that is coming out this year is one battle after another.
那完全不对。好吧，没人
That's not even close here. Okay. No one
挥棒落空。
Swing and a miss.
那是一部艺术片。也许有一天会成为 cult 经典。我给你个提示第二名。我们采访过导演。詹姆斯
That one's an art house film. Maybe a cult classic one day. I will give you a hint for the number two. We interviewed the director. James
卡梅隆。
Cameron.
什么电影？《阿凡达：火与灰》。它在《我的世界》电影之后排名第二。《我的世界》电影遥遥领先，有77%的概率赢得最大周末票房。
What movie? Avatar Fire and Ash. It's in second place after a Minecraft movie. Minecraft movie is running away with it. It's 77% chance of winning the biggest weekend.
嗯，说到市场
Well, speaking of markets
是的。
Yes.
安东尼·庞普里亚诺
Anthony Pompliano
哦，是的。
Oh, yes.
在时间线上发帖说，这想法有点疯狂，但我相信会非常受欢迎。Open应该创建一个让人们能在预测市场下注的方式，预测房屋的售价。
Hit the timeline says, this is a somewhat crazy idea, but I believe it would be incredibly popular. Open should create a way for people to wager in a prediction market Yeah. The price a home will sell for.
我在想是否
I wonder if
每个人都曾在网上看过房源信息，然后说那房子定价过高，或者说那房子简直是捡了大便宜。Keith Raboy插话说，好主意。EB在X上说，这是个有点疯狂的想法。我们怎么把赌博加到几乎所有的生活互动中？而且，我再次能理解为什么人们会
Everyone has looked at a listing online and said that home is overpriced or that home that house is a steal. Keith Raboy chimed in, great idea. EB on x said, this is a somewhat crazy idea. How do we add gambling to literally every life interaction? And and again, I can see why people are
准备好来赌一把吧，
Get ready to gamble,
我能理解为什么人们讨厌这个想法。是的。但同时，我真心觉得这很可能是个爆款产品。至于Open是否应该成为开发它的人
I can see why people hate this idea. Yeah. At the same time, I I I genuinely think that this this is probably a hit product. Whether or not Open Open should be the one to build it
是啊。
Yeah.
是另一个问题——他可能会和别人合作。但是，是的。不过还是
Is another Well, he'd probably partner with someone. But Yeah. But still
在Polymarket上押注百万赌哪部电影会爆红。你可以想象在Zillow上滑动浏览，然后说，哦，是的，或者在Open上滑动浏览，心想，哦，是的，对我来说那条街上的房子是个烂摊子。归零了。一毛钱也不会超过50万卖出。
million on that Polymarket about which movie is going to pop. And you can imagine scrolling Zillow and saying, oh, yeah, or scrolling open and thinking, oh, yeah, that house down the street for me, it's a dog. Getting to zero. Not gonna sell for a dime over 500 k.
是的。所以不管怎样，我认为这不幸地会是个爆款，如果我们能让街区的每个人，是的，你知道，押注
Yeah. So anyways, I I I think this is unfortunately a hit if we could get every person in the neighborhood Yeah. You know, betting on
我想市场流动性会相当稀薄，因为根本没多少人参与。但我的意思是，我想把钱投到《华尔街日报》豪宅版块上的一些项目里。那会很有趣。因为我们评估一些价值2000万美元的豪宅时，我们会说，是的，我们认为这价格被严重低估了。
The markets would be pretty thin liquidity wise, I would imagine because there there just aren't that many people. But, I mean, some of the Wall Street Journal mansion section, that's where I wanna put some money down. That'd be fun. Because we review some $20,000,000 mansion, we're like, yeah. We think this is drastically underpriced.
2050万简直是捡便宜。
It's a steal at 20 '5.
是的，我认为流动性会是个大问题。但是
Yeah. I think I think liquidity would be the big problem. But
没错。
Yeah.
而且，显然有充分理由说明为什么Opendoor——其使命是提高住房拥有率——应该专注于这一使命，而不是涉足对使命的投机。但是
And and, yeah, obviously, there's strong argument for why Opendoor whose mission is to increase homeownership, should probably stay focused on the mission and not not get into gambling on the mission. But
谁知道呢？
Who knows?
谁知道呢？奥兰多·布拉沃倒是想稍微关注一下
Who knows? Orlando Bravo was Do wanna watch a little
这个？它是一个
bit of this? It's a
很长的采访。你只是有问题。你看了
long interview. You just have question. You watched
整个视频。告诉我我有几个，
the whole thing. Tell me I have a couple,
是的，这个帖子中还有一些更多笔记。所以他上了CNBC谈论AI的影响。显然，没有人，他是Sizelords的最终老板。是的。完成了许多最大的企业软件交易。
yeah, some some more notes in this thread. So he went on CNBC talking about the impact of AI. Obviously, nobody he he's the final boss of Sizelords. Yes. Done many of the biggest enterprise software deals.
他还因此不得不面对SaaS末日，以及市场在过去一段时间里对普通企业SaaS公司并不友好。是的。所以他提出论点，你知道，作为记录系统，你仍然处于非常强势的地位。你可以在其上构建许多AI工作流。但他也接着说，AI抱歉，IPO窗口是敞开的。
He also, through that, is having to reckon with the SaaSpocalypse and how the markets have not treated the average enterprise SaaS company very well Yes. Over the last. And so he makes the argument that, you know, it's still you're still in a very strong position as a system of record. You can build a lot of AI workflows on top of that. But he also goes on to say that the AI sorry, the IPO window is wide open.
是的。而且他们在过去十二个月里，我认为，部署了大约80亿美元，并回报了约40亿美元左右。
Yep. And they have, I think, deployed around 8,000,000,000 in the last in the last twelve months and returned about 4,000,000,000 or something.
所以他们实际上并没有将其上市。他们现在可能处于周期的错误阶段。
So they're not actually taking it public. They're maybe at the wrong wrong piece of the cycle right now.
他们还向他施压，说他收购了一个大型呼叫中心业务。所以他正在论证公司状况良好，能够从中获得很多优势。但问题是，在公司私有化期间，他们是否必须重新调整商业模式，使其更类似于Sierra的模式？
They also they pressed him on he he bought like a big call center business. And so he's making the argument that the company is fine. They're gonna be able to get a lot of advantages out of this. But the question is, while the company is private, will they have to reinvent their business model toward that looks something more like Sierra?
是的。
Yep.
布雷特·泰勒本人也再次表示——显然带有偏见——但他说这非常困难。是的。他还说，他认为私募市场中的人工智能估值存在泡沫。他说一个年收入5000万美元的公司不可能价值100亿美元。要让投资者资金翻倍，必须产生10亿美元的自由现金流。嗯。
And again, Brett Taylor himself was saying, obviously biased, but saying how hard that is. Yep. He also was saying, you know, he says AI valuations in the private markets are a bubble. He said a $50,000,000 ARR company cannot be worth $10,000,000,000 To double investors' money, it must produce 1,000,000,000 in free cash flow. Mhmm.
这就好比他自己所处的位置——他拥有的公司能达到那样的规模。他说，是的，这些公司大约值200亿美元。
And that's that's like him he's in the position where he's got companies that do those kind of numbers. He's like, yeah, they're worth about 20,000,000,000.
嗯。
Mhmm.
对吧？所以如果你投资一家年收入5000万美元却仍在亏损的公司，估值却高达100亿美元——当我听到这种数字时，像Perplexity这样的公司就会脱颖而出。我几乎可以肯定，在某个时间点以200亿美元估值时他们是在亏损的。这样的业务终将面对现实。
Right? So if you're investing in a in a company that's losing money at 50,000,000 of ARR, at $10,000,000,000 valuation, again, companies that that stand out to me when I hear these kind of numbers are like perplexity. Right? I'd be almost certain that they're losing money at a $20,000,000,000 valuation at some point or another. That business will have to meet reality.
我现在有三点反驳。
I have three rebuttals now.
发给我。
Hit me.
第一点，如果奥兰多·布拉沃与Julius合作，一切都会改变。他想进行什么样的分析？他想找到优秀的公司。他应该与他的数据对话，获得专家级的见解。第二点，理论上，他正在将这些公司私有化，因此他比现有的上市公司更容易调整商业模式，因为它们不需要公布财报。
Number one, everything changes for Orlando Bravo if he gets on Julius. What analysis does he wanna run? He wants to find great companies. He should chat with his data and get expert level insights. Number two, he in theory, he's taking these companies private, and so it's easier for him to pivot the business model than an existing public company, like, because they're not reporting earnings.
所以你会认为，如果他收购了一个呼叫中心，并且确实想彻底改变商业模式，在重建某种以人工智能为重点、基于消费的模式时大幅降低现金流？他应该为此做好准备。第三点，他为什么不直接将他整个私人投资组合，连同那个呼叫中心，推出PhoneCoin，成为PhoneCoin的数字资产库，通过SPAC上市，炒作为 meme 股票，在顶峰时退出，这对他来说是个好策略。
And so you would think that if he buys a call center and he does want to completely change the business model and take the cash flow way down while he rebuilds in some AI focused, you know, consumption based model? He should be set up to do that. And third, why doesn't he just take his entire private portfolio, with that with that call center, spin up PhoneCoin, become an ass digital asset treasury of PhoneCoin, SPAC it, meme stock it, get out at the top, that would be a good strategy for him
有可能。有可能。
potentially. Potentially.
很多
A lot
价值在那里。
of value there.
应该私信他并跟进一下。
Should DM him and touch that.
也许这不是他的专长，但说不定他能成为那个那个那个网红股票翻盘专家。谁知道呢？
Maybe not his wheelhouse, but maybe he could become the the the meme stock turnaround guy. Who knows?
网红股票翻盘。
Meme stock turnaround.
就就把它当作在做一个
Just just take it at taking a
功能性的 泰·洛佩兹不是试过吗？
functional Didn't didn't Ty Lopez try that?
他试过。他试过。而且，
He did. He did. And,
这让他陷入了
it got him in a
一点小麻烦。
little bit of hot water.
是的。
Yeah.
这不是最好的。总之，我们我们我们有一些现场嘉宾要来上节目。我们应该请他们进来吗？你还有别的帖子要发吗？
It's not the best. Anyway, we we we have some in person guests coming onto the show. Should we bring them in? Do you have another post you wanna run?
没有。我们可以我们可以请他们进来。
No. We can we can bring them in.
他们准备好了吗？先生们。来自The Acquired的本和大卫。
Are they ready go? Gentlemen. Ben and David from The Acquired.
看起来很精神。欢迎
Looking sharp. Welcome
来到节目。欢迎来到直播。我们在TVP和Ultradome现场直播。找个座位坐下吧。
to the show. Welcome to the stream. We are live in the TVP and Ultradome. Grab a seat.
欢迎来到Ultradome。
Welcome to the Ultradome.
你怎么样？
How you doing?
看看我们。杠铃的两端，完全相反的两端。是的。而且有这么多要聊的。
Look at us. Opposite ends opposite ends of the barbell. Yes. And so much to talk about.
我本来希望在我走进来的时候能得到扎克伯格式的待遇，然后我
I was hoping to get the Zuck treatment while I was walking on and get I
我，我刚才在那儿读了一段很悲伤的广告，但我们
I I just dripped a really sad read in there, but we
可以再插播一条，如果你想的话。
can throw another one in there you want.
我相信我们会在这次采访中间插播广告的。我已经做
I'm sure we'll do ad reads in the middle of this interview. I've done
所以，所以你们能来真是太好了。是的。在你们的，你们的场外活动上，你说的。是的。对吧？团队场外活动。
So so great to have you guys here Yeah. On your on your off-site, you said. Yeah. Right? Team off-site.
这是
This is the
把整个公司的人都聚在一起了。是的。
Got the whole company together. Yeah.
整个公司都到齐了。
The whole company in place.
不是每天都有的。
Doesn't happen every day.
是的。公司现在是什么状态？给我们大概介绍一下背景情况。你们做这个多久了？
Yeah. What what what is the state of the company? Just kind of contextualize things for us. How long how long have you been doing it?
开车时的评论，是的。
Reviews on the drive yeah.
我的意思是，昨晚我们互相给了对方很真实的
I mean, last night we were giving each other like real
是的。是的。
Yeah. Yeah.
很好的批评。漫长的散步。
Good criticism. Long walk.
现在是第八年了吗？第十年。第十年。
Is it year eight now? Year 10. Year 10.
我们刚刚庆祝了十周年纪念。
We just hit our ten year anniversary.
敲锣。我们走吧。嘿。是的。
Gong. Let's go. Hey. Yeah.
我们需要一个锣吗？你应该。
Do we need a Gong? You should.
不。不。不。这太直接抄袭了。
No. No. No. It's too direct to the copy.
你需要受到启发。
You need to be inspired.
哦，好吧，我们会我们会给你们寄一个，是的，就是备着用。我妻子应该不会喜欢那样。我们不我们不我们在家里录音，你知道，不在
Oh, well, we'll we'll send you we'll send you guys one just to Yeah. Just to have it handy. Don't think my wife would like that. We don't we don't we record in our homes, you know, not in
我想它会发展成一个图书馆，包含你收集的所有书籍、照片之类的东西。我其实有个问题想问这个。但首先，我想从第一集的研究过程故事开始讲起，然后我想听听最近的研究过程，因为我觉得它们应该很不一样。带我了解一下，当时是用Google文档随手记笔记吗？你是说
I would imagine it develops into, like, a library of all all the books that you've sourced and all the photos and stuff. I actually do have a question about that. But first, I mean, we should I wanna start with, like, the story of the research process for the very first episode, And then I wanna hear about the most recent research process because I imagine it's very different. But take me through, was it a Google Doc that you were just throwing notes in? Were you I mean
是的。
Yeah.
所以当时大语言模型还不存在？
So LLMs didn't exist?
嗯，你知道，我们的一个秘密是我们从未我们从来没有分享过
Well, you know, one our secrets is we we never we've never shared
好的。
Okay.
我们的研究哦，从来没有。好吧。人们问我们，听起来你们上节目时就像，是的，你们一定是很好的演员，因为你们假装不知道对方要说什么。
Our research Oh, never. Okay. People ask us, sounds like you guys get on the show and like you're like Yeah. You must be really good actors because you're pretending like you don't know what each other's gonna say.
我们从不互相分享。
We never share with each other.
我们真的不知道对方会说什么。
We genuinely don't know what each other is gonna say.
那在第一集的时候是这样的吗？不是。
And was it like that at the very first episode? No.
好吧。第一集，十年前，我们俩都在西雅图的Madrona办公室工作，因为那是我们当时的工作地点，算是个副业。是的。下班后我们聚在一起，然后我们说，好吧，我们要录皮克斯那一集。那Instagram呢？
Okay. The first episode, ten years ago, we were both working out of Madrona's office in Seattle because it's where we worked at the time, side project. Yeah. And we got together after work and we were like, alright, we're gonna record the Pixar episode. And Instagram?
皮克斯？
Pixar?
嗯，Instagram是试播集，然后皮克斯是，是的，是我们发布的第一集。
Well, Instagram was the pilot and then Pixar was Yeah. First one we released.
那些是
Those are
都像是宏大的故事。那不是你随便走进去就能说，嘿，不，我们就即兴发挥吧。
both like huge stories. That's not something you just walk into and say, hey, no, let's just freestyle.
嗯，我们确实这样做了三十七分钟。所以我们
Well, For thirty seven minutes we did. So we
当时就想，好吧，我们要录制了，你觉得怎么样？比如，一小时后开始？然后我们俩都去网上刷了刷。我们当时说，好了，你准备好了吗？我准备好了。我准备好了。
were like, okay, we're gonna record What do you think? Like, start in an hour? And so then we both like went and scrolled the Internet for a We were like, alright, are you ready? I'm ready. I'm ready.
而现在
And now
但我不认为我们有一个共享的谷歌文档。我想我们各自有各自的。是的。我觉得我们一直都是分开的，你知道，不是那种我们把东西都放进去的。
But I don't think we had a like one Google doc. I think we had our own Yeah. Like I think we've always had separate, you know, it's not like we putting stuff into.
酷。酷。
Cool. Cool.
我想在第一年结束时，我们大约有400名听众。
And I think by the end of year one, we had about 400 listeners.
哇哦。太棒了。敲响那个锣。我觉得人们没有意识到约翰是如何传达这个信息的。
Wow. Let's go. Hit that Gong. Which I think people don't realize how how John John dropped the message.
是的。就是这样
Yeah. That was how
我们也是这么觉得的。是的。我认为是的。我们私下里也聊过这个。有趣的是，你们在Acquired现在拥有的规模优势，作为一个企业要与你们竞争，实际上需要有两个主持人能够花费数周时间研究，就为了一个单一主题，这在节目早期阶段并不太经济。
how we felt too. Yeah. I think yeah. We we were kinda talking about this offline. It's interesting how the the kind of scale advantages that you have at Acquired now being in a position as a business where it's to compete with you guys, you would effectively somebody needs to have two hosts that can spend weeks researching weeks researching just, you know, a single topic which is which is not something that's super economical in the early stages of a show.
是的。所以十年前就开始的优势变得越来越极端。
Yeah. And so the advantages of just starting, you know, ten years ago are are just become more and more and more extreme.
或者至少从一个比现在窄得多的假设开始，这样你实际上可以做一点工作来制作节目，然后让它随着时间的推移慢慢扩展。每当有人问我，今天如何开始？我总是想，找出你能做的最独特的事情，并把范围限定在那一点上。然后随着时间的推移，看看你是否能成长，并真正证明投入大量资源做大事情的合理性。
Or at least starting with a much narrower and hypothesis than we have now where you actually can do a little bit of work to make the episode and then just sort of like letting it expand over time. Whenever people ask me like, how do I start today? I I always think like, figure out the most unique thing you can do and scope it to, like, just that. And then over time, see if you can grow and and actually warrant all the investment that it takes to make something big.
我们喜欢那样。这并不是战略性的，但我们幸运地参与了多个复合游戏。当然。这就是我们走到今天的原因。
We liked that. We it wasn't it wasn't strategic, but we lucked into playing multiple compounding games. Sure. And that's how we got here.
我是说，AirPods是在我们开始后一年推出的。哦。所以，就像，在世界上边走边听播客的概念，是的，当时还不是一回事。是的。
I mean, AirPods came out the year after we started. Oh. So, like, the the concept of wandering around in the world listening to a podcast Yeah. Wasn't really a thing. Yeah.
然后突然
And suddenly
你真的认为AirPods特别是一个
You see really, you think AirPods specifically were a
催化剂是AirPods，COVID。COVID一开始很糟糕，但对播客来说后来很棒。
catalyst for AirPods, COVID. COVID was horrible at first, but then great for podcasts.
一开始很糟糕，但你只是哦，是的。是的。是的。
It was horrible at first, but you just Oh, yeah. Yeah. Yeah.
是的。东西。第一个，就像，什么？
Yeah. Stuff. The first, like, what?
没人想听习惯改变了。
No one wants listen to habits changed.
所以大家习惯了在通勤时收听。哦，有意思。我们的数据有两三个月都在下滑。
So everyone was used to listening while they commuted. Oh, interesting. We had, two or three months of numbers falling off.
真的吗？
Really?
我们当时觉得，哦，我们要完蛋了。哇。但后来大家都养成了新习惯。
And we were like, oh, this is the end for us. Wow. But then everyone finds new habits.
是啊。是啊。是啊。然后它然后就增长了。
Yeah. Yeah. Yeah. And then it and then it grew up.
是的。在新冠疫情之前，我觉得人们不会在锻炼、散步或洗碗时听那么多东西。这些都是新冠疫情时期养成的习惯，后来就保留下来了。
Yeah. Before COVID, I think people didn't really listen as much while like working out or, you know, going for a walk or doing the dishes. That was all COVID behaviors that then stuck.
是啊。
Yeah.
是啊。
Yeah.
关于更现代的生产流程，我觉得变化很大。现在你可以接触到公司内部。你也非常擅长挖掘信息。我看到你刚刚分享了原始的Waymo设备，感觉你正在获取互联网上已经不复存在的照片，实际上是在发掘新的图像，我认为这非常酷。我觉得这可能是你在构建历史资料库时的专长。
On the more modern production workflow, I feel like it's changed a lot. Now you have access to the companies. You're also really good at digging up. I I saw you just shared the original Waymo rig and, like, it feels like you're getting photos that don't exist on the Internet anymore, and you're actually surfacing new images, which I think are super cool. I feel like that might be your gong when you build the the library to, you know, history.
对吧？但是，但是，再详细说说现在的情况，因为人们会打电话和你交流。你可以进行背景访谈，当然你也可以做正式的AQ2 ACQ2采访，但这并不总是其中的一部分。你是怎么考虑的，比如你愿意花多少时间与公司实际人员、可能已经离职的员工交流呢？嗯。还有书籍、第三方研究，你是如何将这些融合在一起的？
Right? But but but walk me through a little bit more about what happens now because people will pick up the phone and talk to you. You can do an interview on background, even though, of course, you can do a proper AQ2 ACQ2 interview, but that's not always part of the How do you think about, like, how much do you wanna spend with the actual company, with people that might have left the company Mhmm. With just the books, the third party research, every like, how do you blend that?
因为你
Because you
必须选择。我们现在基本上每个月都在写一本书。比如，我们刚刚完成的谷歌系列，我们可能和30多个人交谈过。是的。你知道，这些都是长达一小时的研究电话，有时候人们会问，哦，这会被录音吗？我们说，不，不，不。
have to pick We're basically writing a book every month now. Like, for the our Google series that we just wrapped up, we probably talked to 30 plus people. Yeah. You know, those are all hour long research calls where, you know, sometimes people are like, oh, is this gonna be recorded? We're like, no, no, no.
这不是为了节目。是的。这就像我们在写一本书。
This isn't for the show. Yeah. This is just like we're writing a book.
这有点……我总是担心显得自大。比如，这个人大多数人会很乐意邀请上播客。当然。我说，不。你是在帮我做研究。
Which is a little bit I'm always worried of coming across as cocky. Like, this is a person that most people would happily have on as a podcaster. Sure. I'm like, no. You're helping me do research.
是的。是的。是的。
Yeah. Yeah. Yeah.
但实际上，我认为他们喜欢
But actually, think they like
它是因为他们，嗯，是的。这是规模优势的一部分，比如，如果十年前你打电话给这些人，说跟我聊聊，就给我一小时的阿尔法。是的。历史，他们会说，抱歉。
it because they're Well, yeah. That's part of the scale advantages of, like, if ten years ago you called up some of these people and you said talk to me for just give me alpha for an hour. Yeah. The history, they'd be like, sorry.
就像你可能会得到更多，比如
Like You probably get way more, like
听播客。
Listen to the podcast.
如果需要，你可以匿名化。如果有戏剧性事件，你可以将其情境化。
That you can anonymize if you need to. If there's drama, you can contextualize it.
而且这不像我们是对冲基金打电话给他们说，哟，我在找交易。不，不是这样的。
And it's not like where we're hedge fund calling them up and being like, yo, I'm looking for a trade. It's like, no, no.
我们想讲述你的故事。
We wanna tell your story.
是的。是的。是的。
Yeah. Yeah. Yeah.
我们想要把它做对。我们开始调研通话时总是这样，首要问题是：关于你们公司，最常被误解的事情是什么？
And we wanna get it right. That's always how we start the research calls is the the number one question is what's the most commonly misperceived thing about your company?
没错。
Yep.
第二个问题是：那很好。传统媒体在过去几年里搞错了什么？当然。如果我们必须讲述你们公司的标准故事，告诉我们如何纠正这些错误。然后我们当然会在此基础上继续调研。
And two is That's good. What did the traditional press get wrong Sure. Over the last several years? And if we had to like tell the canonical story of your company, tell us how to right the wrong. And then of course we go do research after that.
是的。比如，好吧，那么有哪些，你知道的，是的。
Yeah. Like, okay, well what are the, you know Yeah.
关于一家公司的主流书籍，有多少次会把叙事完全搞错？比如，有多少次不是完全错误，但即使是，你知道，公司里的某个人因为某个特定产品获得了很多赞誉，但当你和几个人交谈后，你会发现，哦，其实是这个人在产品发布前就离开了。
How often how often do does like the the the leading book about a company end up getting getting kind of the the narrative just completely wrong? Like, how often are you not completely. Not completely wrong. But even but even like, you know, somebody at a company getting a lot of credit for a specific product when you talk to a handful of people you realize, oh, it was actually this guy who ended up leaving right before the launch.
有时候是写书的那个人获得了所有的赞誉。他们离开了，然后说'我是如何壮大这家公司的'。而所有人都在想，你在这里的时候几乎一直在写书，根本没做什么实际工作。
Sometimes it's the guy that wrote the book that gets all the credit. And they leave and they're like, how I grew this company. And everyone's like, you were writing the book literally the entire time you were here. You didn't do anything.
我认为通常情况是，并不是说某个巨头部门的领导独揽功劳，而是另一个巨头部门的领导。更像是你把团队的工作整合起来，我们都默认，比如杰夫·迪恩完成了这项了不起的事情。实际上，是杰夫·迪恩在谷歌领导了一个团队完成了这项了不起的
I think you usually, it's not like giant, division leader gets the credit and it was the other giant division leader. It's like you sort of roll up the work of the team and we all take it as convention that, like, Jeff Dean did this amazing thing. It's like, well, Jeff Dean led a team at Google that did this amazing
成就。我也在想，当时我们在Metaconnect大会上，他们展示了神经手环，那是收购Control Labs的成果。是的。作为初创公司的人，我觉得应该100%归功于Control的创始人们。我觉得，他们确实做了很多，应该获得部分功劳。
thing. I was also thinking it was the we were at Metaconnect, and and they have the the neural band, and that's that was an acquisition from Control Labs. Yep. And as the startup guy, I'm like, give a 100% of credit to the Control founders. And I'm like, well, really, like, they did a lot, and they should get of some the credit.
但后来可能投入了更多资金、更多研究，还有从未在Control工作过的新人加入，推进并发展了这项
But then there's probably a ton more money and ton more research and completely new people who never worked at Control that came on and stepped and advanced that In
就Control而言，那些创始人做了很多
the case of Control, those founders did a lot of
工作他们做了
the work They did
很大一部分。
a lot of it.
那也是收购的功劳。
Acquisition too then.
是的。收购之后也是。但团队里还是有更多人，更多资源。没错。就像是，我们是被归功了30%，还是70%，大概就在这个范围内吧。
Yeah. After the acquisition too. But then there's still more people on the team, more resources. Yes. It's like, are are we given 30% of the credit, 70% of the credit, somewhere in there probably right
书籍更常见的错误是，它们把核心故事搞对了。
The the more common thing books get wrong is they get the core story right
对。
Yep.
大概占书的一半篇幅。但后来关于公司有些争议，或者在当时写书的时候是有争议的。
In like half the book. But then there's something controversial about the company or that it was at the time they were writing the book controversial.
他们的个别事件。
One shots of them.
那个，是的。就是你读这本书的时候，你会想，
That yeah. That you're reading this book and you're like,
为什么有80页都在讲这个？哦，对了。就这一件事？
why is there 80 pages on this Oh, yeah. This one event?
是的。是的。
Yeah. Yeah.
哦，是的。完全正确。是的。剑桥分析公司。是的。
Oh, yeah. Totally. Yeah. Cambridge Analytica. Yeah.
你知道，或者有太多事情让你觉得，这其实并不是公司官方故事的一部分。是的。而且当时感觉就是那样。
You know what Or on there are so many things where you're like, this is actually not a part of the company's canonical story Yeah. And it felt like in the moment.
是的。我是说，现在Facebook就正在经历这种事。我认为下一个社交网络会完全围绕剑桥分析公司之类的事情展开，但大家好像都已经翻篇了，我们却在想，他们在元宇宙上是不是投入太多了？AI方面的赌注现在怎么样了？对吧。
Yeah. I mean, is happening with Facebook right now. I I think the next social network is gonna be entirely about Cambridge Analytica or something, which, like, everyone's kinda moved on from and we're like, well, did they overspend on the metaverse? What what what's going on with the AI bets? Like Right.
就像，我们当时在做一场《社交网络2》的搞笑剧本朗读会，完全是虚构的版本。我们完全聚焦在打造AI人才大战上，因为我们非常活在当下。是的。我敢肯定，像Yan Lakun根本不会出现在《社交网络2》里。是的。
Like, the the we we were we were doing a, like, a joke table read of a fake version of the social network too. We were focusing it entirely on building, like, the AI talent wars because we live very in the moment. Yeah. I'm sure, like, Yan Lakun is not gonna be in the social network too at all. Yeah.
对吧？但他应该出现。Alex Wang不会。是的。Nat Friedman也不会，但我很希望
Right? But it should be. Alex Wang is not. Yeah. Nat Friedman's not, but I would love
《社交网络3》。
The social network three.
是的。社交网络三将完全围绕这一点展开。
Yeah. The social network three will be all
关于这个。我想就是这样，对我们来说也是一样。我相信你们也经常遇到这种情况，比如，我们不是记者。通常写书的人是的。他们是以记者的视角来处理的
about that. I think that's it, the same thing for us. I'm sure you guys get this all the time too is, like, we're not journalists. Usually the people that are writing the books Yep. They're approaching it as
是的。
Yep.
记者。通常，他们一直在某个出版物上报道这家公司是的。然后他们写了这本书。是的。但他们不是实践者。
Journalists. Usually, they've been covering the company at, you know, a publication Yep. And then they write the book. Yep. And so but they're not practitioners.
你知道，本和我已经不再是实践者了是的。但我们来自那个世界。我们在公司工作过是的。你知道，我们做过风险投资。是的。
And you know, Ben and I are no longer practitioners Yep. But we come from that world. We've worked at companies Yeah. You know, we've been VCs. Yeah.
就像，我们有着完全不同的视角。
Like, we we just have a totally different perspective.
Senra？我我我一直都在说，我不介意创作者或影响者之类的说法。新闻播音员，我想，是广播员。广播员。大卫·森拉用了爱好者这个词。
Senra? I I I I was always just saying, like, I don't mind the phrase creator or influencer or whatever. Newscaster, I guess, is Broadcaster. Broadcaster. David Senra used the phrase enthusiast.
他说，我告诉CEO们，我不是记者，我是爱好者。
He says, I'm not he tells CEOs, I'm not I'm not a journalist. I'm an enthusiast.
嗯，任何认识Senra的人都知道。
Well, anybody who knows Senra knows.
那是最贴切的词。
That is that is the best word.
而爱好者是一个恰当的形容
And an enthusiast is a proper description
方式。
for it.
而我
And I
想知道它是否会流行起来。我想知道我是否应该采用这个说法。我想知道它是否适合我。它确实感觉没有人能
wonder if it will grow. I wonder if I should adopt that phrase. I wonder if it fits me. It it it certainly feels Nobody can
像戴夫那样充满热情
be as much of an enthusiast
都不想在这方面和他竞争。
as Dave wouldn't wanna compete with him on that.
我觉得你们应该采用一个非常戏谑的称号，那种明显很老派的，就像电视主持人那样。当然。
I think you guys should like adopt a really tongue in cheek moniker that like that's like so obviously old timey. Like television host. Sure.
当然。是的。
Sure. Yeah.
就这么做。现在，我们只是主持人。
Do that. Right now, we're just hosts.
我们只是主持人。是的。
We're just host Yeah.
是的。正在那里广播。
Yeah. Broadcasting there.
乔迪？在哪里
Jordy? Where
我们应该去哪里？我想我本来打算买一个
should we go? I think I wanted to get a
你是一直计划让这一切都保持活跃吗？比如，等等，
Were you always planning for this all to be alive? Like, wait,
怎么乔迪？不。我的意思是，我猜大卫是的。
how Jordy? No. I mean, I guess David Yeah.
你传给乔迪。我来做这是直播。
You tee it up to Jordy. I'll do This is live.
传回来。我们会传回给你们。不。所以这是一个非常自然的进展。我们一开始是每周一次的节目，就像许多两个技术兄弟聚在一起。
Throw it back. We'll throw it back to you guys. No. So it it was very natural progression. We started out with a weekly show as many two technology brothers get together.
他们我们应该开始一个播客。
They we should start a podcast.
是的。
Yeah.
约翰最初设想的节目形式是没有嘉宾，只专注于高速讨论各种话题。最终，节目确实反映了时间线的变化，对吧？算法在筛选有趣内容方面做得非常好。我们录制了前几期节目。
John's idea for the initial format of of no guests and just focused on high velocity of topics. It ultimately, the show ended up reflecting the timeline. Right? And algorithms were doing a really good job sorting what was interesting. And we went, we recorded the first couple of shows.
我们基本上只发给了森拉。我想他是唯一真正收听的人。他说，这很不错，我们非常喜欢。然后我们又做了一期，你知道，我们开始每周做几期，并意识到每次录制结束——这还只是预录阶段。
We pretty much only sent it to Senra. I think he was the only person that actually listened. He was like, this is good. We enjoyed it a lot. Then we did another, you know, we started doing like a couple a week and realized that every time we would turn and this was just prerecord.
是的。每次节目结束后，我们打开手机就会意识到，哦，
Yeah. Every time we would go off the air, we'd open our phones and realize like, oh,
这又是一笔交易
this is one more deal
我们想谈谈这个。是的。所以我们开始增加播出天数。我想到了年底我们达到了三天或四天。然后进入一月份时，我们就知道要增加到每周五天了。
we wanted to talk about this. Yeah. And so we just started adding days. I think we got to three days or four days by the end of the year. And then we knew going into January that we wanted to go to five days a week.
最终我们决定进行直播有很多原因。比如第一，它能让我们对正在发生的事情高度敏感。经常在我们直播的三个小时期间
Then we ultimately made sense to do live a bunch of reasons. Like one, it it just allows us to be highly reactive to what's happening. Oftentimes during the three hours that we're live
很棒的事情。
Something great.
事情正在发生。而且这样也更高效。我们可以。是的，我们不会花，你知道，节目播出后花好几个小时编辑。我们在播出时也很高效，你会注意到几乎没有冷场。
Stuff is happening. And so it it's also a lot more efficient. We can Yeah. We're not spending, you know, hours and hours editing the show afterwards. We're also efficient on the air where you'll notice like there's very little dead air.
就像可能每期节目有一次，会有那种，哦，我们该聊什么？然后我们很快
Like maybe once a show, there's like, oh, what should we talk about? And then we quickly
是的。
Yeah.
你知道，想出极端的
You know, figure extreme
极端的一端，就是把杠铃尽可能推离你们。
extreme end of just pushing as far down the barbell away from you guys.
对吧？所以，比如，让我们对比一个月，
Right? So, like, let's contrast a month,
然后我们就想，我们就想，我们每周做五次，然后每天做，然后然后节目在我们录制后两小时，直播后一小时就出来了。
and then we're like, we're like, we'll do it five times a week and then daily, and then and then the episode comes out two hours after we record, one hour after we record live.
然后做
And do
你而你没法直播或实时观看。
you And you can't get live or live.
你们会剪辑任何内容吗？比如，你们的
Do you ever edit anything? Like, your
片段我们甚至没有这个选项。我的意思是，我我们会做的是，在节目开始时有一个五分钟的倒计时，是的。为了让直播观众进来并知道节目即将开始，我们会为播客订阅源剪掉那部分。
clips We don't even have the option to. I mean, I I what we will do is we have a five minute countdown at the start of the show Yep. For the live feed to let people come in and know that the show's about to start, and we clip that out for the podcast feed.
是的。所以我们每集做一千次剪辑
Yeah. So we make a thousand cuts
我们只做一次。所以就是这样。那就是我们只做一次剪辑。
per episode. We make one. So there you go. That's We make one.
但你们制作的节目数量可能比我们多一千倍。所以
But you probably make a thousand times more episodes than we do. So
是的。我的意思是，我们这周已经做了一千次采访了。
Yeah. I mean, we we we've done a thousand interviews this week.
你们下周会做八个
You'll do eight next
我们会做50个。
We'll do 50.
对，对。而且只有大概
Yeah. Yeah. And and only like
嗯，你们所有的采访节目也是。
Well, you'll all of the interview show as well.
但是，是的。我不认为我们在这方面有区别。
But, yeah. I don't think we're differentiated there.
我们的访谈节目不是一个
Our interview show is is not a
我们只是在它
We just do it when it
出来的时候做。当它特别的时候。有巴尔默。
comes out. When it's special. There's Balmer.
但那是主要的节目。就像，我们八个中的两个。当然。这就像是有史以来最受关注的事情。今年我们做了12集。明年我们要做八集，而且会比以往任何时候都更好。
But that's the main show. Like, two of our eight Sure. This is like the most acquired thing ever. This year, we did 12 episodes. Next year, we're doing eight, and they're gonna be better than ever.
是的。
Yeah.
我正要提到这一点，但我不确定你是否
And I was about to mention that, but I wasn't sure if you
已经泄露了，但我很高兴你宣布了
were leaking it yet, but I'm glad that you announced
是的。我想我认为这是突发新闻。
Yeah. Think I think breaking news.
突发新闻。突发
Breaking news. Breaking
收购是我做得更少了
Acquired is I making fewer
我认为这是科技界的重大新闻。
think this is this is major news in the tech world.
是的。我认为我们很早就意识到的一点是，很多人听访谈播客是为了获取新闻。没错。但这其实体验并不好，因为你有没有想过，你知道吗？我真的想知道四天前CNBC在讨论什么。
Yeah. I think one thing we realized early on was that a lot of people were listening to interview podcasts for news. Yep. And that's actually not a great experience because have you ever in your life thought, you know what? I really wanna know what CNBC was talking about four days ago.
没错。对吧？你想确切知道正在发生什么。对吧？所以对我们来说，就像是保持那种状态，你们有没有
Yep. Right? You wanna know exactly what's happening. Right? And so for us, it's just like staying on that Well, did you guys
有什么灵感吗？比如，有没有什么，这相当创新。你知道，你们正在做，是的。
have any inspirations? Like, was there any like, this is quite innovative. Like, you know, you're doing it Yeah.
直播。所以我们当时在这个健身房锻炼，帕特·麦克菲的节目会在背景里播放。我们开始关注帕特在做什么，他最初是从播客开始的，录播形式，后来逐渐发展到每天基本上做三个小时的直播电视节目。所以当我们做到一半的时候，我们开始更多地以帕特·麦克菲为榜样，看看新媒体在直播领域能做什么。
Live. So we were we were working out at at this gym and Pat McAfee would on would be on in the background. And we started looking at what Pat was doing in the sense that he was started as a podcast, recorded, and then eventually wound up doing basically live TV for three hours every day. And so once we kind of were halfway there, we started looking more to Pat McAfee as as an example of kind of what new media could do in a live space.
他在ESPN交易之前就开始直播了吗？
Did he go live before the ESPN deal?
我想是的。我觉得他直播有一段时间了，但最初是播客。他把节目做大了，有了更多，是的。然后最终每期节目会请多位嘉宾。你甚至不会觉得这是个嘉宾节目，但他还是会做，比如和勒布朗对话。
I think so. I think he was live for a while, but it started as a podcast. He grew the show, had more yeah. And then eventually was doing, you know, multiple guests per show. It's that you don't even think about it as a guest show, but he'll still do the LeBron versation with LeBron Yeah.
就像那种情况一旦发生，就很特别。
Like once when that happens, it's special.
太有趣了。我商学院的一个同学和他在小马队一起打过球。
So fun. One of my business school classmates played with him on the cults.
哦，不会吧。
Oh, no way.
而且，你知道，我们毕业几年后，我们创办了Choir，他当时……是的。他就发短信给我说，你知道，我有个朋友……是的，回溯到我在小马队打球的时候。他现在在做……在做一档节目。
And, you know, there's a couple years after we graduated and, like, we had started choir and he was Yeah. And he just texted me. He's like, you know, I've got this this friend Yeah. Back from when, you know, I played on the Colts. He's he's doing a doing a show.
是啊。就像，你知道的，就像，哦，是的。真为他高兴。是的。和帕特在一起。
Yeah. Like, you know, like, oh, yeah. Good for him. Yeah. Was with Pat.
是的。
Yeah.
有个有趣的故事。我以前会赞助帕特·麦克菲，因为我曾经
Here's a funny story. I I would sponsor Pat McAfee back in the day because I used to
哦，是的。
Oh, yeah.
帮了很多
Help a bunch
公司做播客广告，后来更多地专注于YouTube。当时我在想，哇，他在NFL的职业生涯很短，然后成了播客主。那时我没意识到，把自己的工作变成谈论自己热爱的
of companies with podcast advertising and then ended up focusing more on YouTube. And at the time I was thinking what like, wow, he had this sort of like short career in the NFL, then he became a podcaster. I didn't realize at the time that, how fun it was to have your job be just talking about the thing that
事情是多么有趣。
you love.
对吧？所以我们最终进入了，我觉得帕特报道的重要之处在于他就在联盟内部。是的，对吧？这影响了他的报道方式。
Right? And so we ended up we were in the like, I I think what what's important about Pat's coverage is that he was in the league. Yeah. Right? And that informs his coverage.
这正是它有趣的部分原因。约翰和我也一样，比如在科技领域，播客通常地位不高。对吧？就像是人们
That's part of what makes it interesting. And John and I in the same way, like podcasting is generally low status in tech. Right? It's like people have
曾经是。
Used to be.
是的，曾经是。我觉得现在有点变化，但还是这样，很多人想成为创始人或什么的，而播客通常是他们主要业务的内容营销手段。所以我们很早就意识到，嘿，我们就在联盟里。
Yeah. Used to it used to be. I think it's changing a little bit but it's still like you A lot of people wanna be a founder or an And podcasts are are usually this like content marketing for the main thing that they're doing. And so we realized early on it's like, hey, we're in the league.
是的，而现在
Yeah. And now
我们意识到谈论联盟是
we realize like talking about the league is
很多关键洞察。我们十年前也有同样的情况，当时我们说，如果这是内容营销，它必须有效。它必须是主业。如果不是主业，你永远无法，你知道的
a lot of key insight. We had the we had the same thing ten years ago where we were like, if this is content marketing, it'll Yes. It's gotta work. It's gotta be the main thing. If it's not the main thing, you're never gonna be, you know
而且我们在接下来的八年里是各种类型的专业风险投资家，是的。开始做播客后，从未一次被诱惑过应该被XYZ公司播客收购。当然。那样的话，就会毁掉整个事情。
And we were professional venture capitalists in various flavors for eight years after Yeah. Starting the podcast and never once were tempted by should acquired Be The x y z firm podcast. Sure. Like, that that would kill the whole thing.
当然。
Course.
是的。这很难，因为如果你是一个活跃的投资者，并且你有40家投资组合公司，你能真正提供准确的市场报道吗？对吧？是的。如果你在谈论一个类别，并且你在这个赛道上有一匹马，你可以提供一点
Yeah. It's hard because you you can't if you're an active investor and you have, you know, 40 portfolio companies, can you actually give accurate coverage on a market? Right? Yeah. If you're talking about a category and you have a horse in the race, you can provide a little
公平地说，我们有一匹马
to be fair, we have a horse
在比赛中，我们
in the We
当然在比赛中有一匹马。我们，你得告诉那匹马。我
certainly have a horse in the race. We we you gotta tell the horse. I
所以我们有
so we we have
他们正在大量购买
They're ramping a
很多人会
lot of people would
不是他们。
It's not them.
人们在买GPU。你们在买马。是的。
People buying GPUs. You guys are buying Horses. Yeah.
不。人们会经常这样说——他们会批评记者是马匹爱好者，而我们为他们辩护了。对吧？我们说马应该被赞美。它们是不可思议的动物，所以在某种程度上，这是对技术的一座纪念碑
No. People people would say that a lot of They would they would critique journalists for being horse people and we came to their defense. Right? We said horses should be celebrated. They're incredible animals and so this is a mon in some ways a monument to technology
品牌就像是，科技品牌的反面是什么？嗯，就像是老钱、马术风格
brand was just like, what's the opposite of tech branding? Well, it's like old money, equestrian
你就像七十年代的迈阿密。
You're like seventies Miami.
'70. 没错。
'70. Exactly.
我们 TPN 是一种生活方式。
We were TPN is a lifestyle.
是的。完全正确。它是一个生活方式品牌。而那匹马就是完美的例子。所以我们一直很有趣。
Yeah. Exactly. It's a lifestyle brand. And the horse is like a perfect example of that. So we've been fun.
什么，你觉得
What what do you do you think
风险投资公司最终会大量在播客上做广告吗？
that venture capital firms will eventually advertise on podcasts significantly?
已经在做了。已经在做了。过去我们有过，哦，是的。他们和我们合作过广告。
Already do. Already do. We've had in the past, we've had them Oh, yeah. Advertise with us.
好的。还有
Okay. And
我认为这通常可以是一种比雇佣播客制作人和编辑、然后占用GP时间投资更好的资源利用方式。是的，所有这些事情。你可以直接购买，你可以创建一个
I think that's a I I think it's any any often can could oftentimes be a much better use of resources than saying like, okay, we're gonna hire the podcast producer and a podcast editor and then we're gonna take the GP's time away from investing Yeah. And all these things. And you can just buy, you can create a
你可以想想红牛在一级方程式中的做法，其他所有人都是被赞助的。
You can think of Red Bull of the f one where everyone else is sponsored.
风险投资界的红牛。
Red Bull of venture.
是的，也许这才是正确的
Yeah. Maybe that's the right
正确的策略就是变得非常聪明和有趣。是的，就去其他播客上做一些免费媒体宣传。
The right strategy is just be really smart and interesting Yeah. Just go do a bunch of free media on other podcasts.
但我有一些聪明有趣的投资人朋友就是不喜欢谈论播客。他们不喜欢做大量公开采访。对吧？他们应该直接购买赞助。是的。
But they're but I but I have investor friends that are smart and interesting and just don't like talking about podcasts. Talk they don't like doing a bunch of public interviews. Right? They should just buy sponsorships. Yeah.
你如何看待不同剧集之间的互动关系？显然，如果你被关在房间里一年然后只发布一集，效果肯定不如在不同主题间切换，比如发现好市多和谷歌之间的联系。是的。你在谈论微软时会回调采访过谷歌剧集的嘉宾吗？总是这样。
How do you think about the interplay between the different episodes? Obviously, if you were locked in a if you were locked in a room for like a year and then you published one episode, it wouldn't be as good as bouncing between seeing a connection between Costco and Google, for example. Yeah. Have you do do you do you ever call back to someone you interviewed for the Google episode if you were talking about Microsoft? All the time.
那么，你有没有考虑过对他进行一次独家采访，你可以从你和鲍尔默的对话中截取一段，下次你讲述微软的故事或另一个故事时放进去。
And and have you ever thought about exclusive interview with him and you could take a clip from your conversation with Balmer and put that in the next time you visit the story of Microsoft or another story.
这是我们做事方式的一个方面。
This is one of the areas where we have a way of doing things.
是的。
Yeah.
我不确定我们坚持自己的做事方式，是否是让Acquired节目与众不同的部分原因
And it's not clear to me that us sticking to the way that we do things is, like, part of what makes Acquired special
或者
or
还是说我们只是固守成规。是的。每次我们考虑那样做时，是的。我们就会想，嗯，我们到目前为止还没这么做过。
if it's like we are just stuck in our ways. Yeah. And every time we've thought about doing that Yeah. We're like, well, we haven't done it so far.
是的。
Yeah.
这样做让它更类似于其他类型内容的运作方式。完全正确。那么我们不这样做，实际上我们并不具备插入片段的那种制作水准，这是否导致了我们的差异化？
And doing that makes it more similar to the way that other types of content work. Totally. So is the fact that we don't do that and we're actually not the production value where we would insert the clip, does that lead to our differentiation?
我们起步于2015年。
We came up so we started in 2015.
是的。是的。
Yeah. Yeah.
我们出现的时候，独立播客已经存在很久了，但播客主要还是NPR的天下。你知道，在2014、2015年想到播客，你就会想到NPR。是的。
And we came up right as like Indie podcasts have been a thing forever, but like podcasts were mostly the NPR. You know, when you thought podcasts in 2014, 2015, you thought NPR. Yeah.
15人的团队，精心制作的《Serial》。
15 person team, Well crafted Serial.
是的。从制作角度来说非常精良，他们会剪辑片段，里面还会有过渡音乐。
Yeah. Highly produced in the sense of, like, they would splicing clips and there would be transition music in there.
你甚至可能不记得主持人是谁。
You might not even remember who the host was.
没错。
Right.
你并没有在建立关系。
You weren't developing a relationship.
你会切换到现场记者那里，他正在外面拍摄素材，与人交谈。对。
You'd cut to the reporter in the field who's out getting tape talking to the person. Right.
而且为了更好拥有
And for better have
持续蓬勃发展。是的。绝对是。我我忘了是哪个顶级收入网络，但它就像一场恐怖秀。Wondery被收购了。
continued to thrive. Yeah. Absolutely. I I forget one of the top grossing networks, but it's like a horror show. Wondery was acquired.
是的。有一个是
Yeah. One was
做了很多那样的事。有一个投入了大约4.45亿美元在
doing a lot of that. There was one that was putting up like $445,000,000 in
息税折旧摊销前利润。是的，我记得这个。
EBITDA. Yeah. I remember that.
正在制作，像是恐怖电影。恐怖镜头。是的。所以
On on just making, like, horror films. Horror shots. Yeah. So
恐怖题材很有趣，你把两个很棒的东西结合起来了。播客，如果你想的话，可以成为运营利润率极高的行业。当然。比传统娱乐好得多，
horror it's funny that you're smashing two amazing things together. Podcasting, which can, if you want it to be, a extremely high operating margin business. Sure. Like, much better than traditional entertainment,
好得多
much better
比好莱坞好。
than Hollywood.
我知道你最近在往这个方向走。恐怖题材，是的。就像是电影赚钱的方式。
I know where you're going with this these days. And horror Yeah. Is like the way to make money Movies.
电影。是的。
Movies. Yeah.
是的。你，你知道我们娱乐界有个朋友告诉我们，恐怖片的一个酷炫之处在于，如果你是个资本家，是的，你不需要明星阵容。谋杀题材的好处是，因为人们愿意去看恐怖片，不在乎里面是谁演的。
Yeah. You you the crazy thing we got a buddy at entertainment who was telling us that the cool thing about horror, if if you're a capitalist Yeah. Is you don't need a list. Good thing about murder. Because people are willing to go see horror movies without carrying who's in this.
你不需要去多个星球拍摄。你可以说，整个剧情就是，我们被锁在这个房间的地下室里，就像在一个录音棚里拍摄一样。你可以全程制作，非常吓人。
You don't need to go to multiple planets. You can be oh, the whole plot, we're locked in the basement of this room, and it's like you're filming in one sound You can go make entire time, it's terrifying.
比如，最高票房6000万美元，跟大片比起来不算什么。是的，是的。但成本对我来说大约800万美元。
Like, $60,000,000 top line, which is nothing compared to the big movies. Yeah. Yeah. But it costs, like, $8,000,000 to me.
没错。没错。没错。是的。是的。
Yep. Yep. Yep. Yeah. Yeah.
看看它会如何发展非常有趣。
Very interesting to see where it goes.
给我们最新一集的预告片。是的。我不想剧透太多。就像你在图书巡回宣传时，
Give us a give us a trailer for the most recent episode. Yeah. I wanted to give too much away. It was like you're on a book tour and
你把整本书的情节都讲了，没人需要买了。这是一集四个小时的节目。所以别担心，我们不会
you tell the whole plot of the book and nobody needs to buy It's the it's a four hour episode. So don't worry. We won't
透露太多信息了。请用两分钟时间讲讲谷歌。
give too much away. Google in two minutes, please.
嗯，我认为最大的亮点在于，谷歌的历史实际上就是整个人工智能领域的发展史。是的。几乎所有在基础模型公司从事创新工作的领导者——没错，几乎都能追溯到与谷歌的渊源，他们中绝大多数在2015、2016年期间都曾供职于谷歌。
Well, the I guess the biggest hook is the history of Google is actually the history of the entire AI landscape. Yeah. Everyone almost everyone doing interesting things in foundational model companies at at the leadership level. Yeah. You can trace a lineage back to Google, and almost all of them were there 2015, 2016.
没错。你展示了伊利亚在AlexNet团队的那张照片。
Yeah. You showed that picture of Ilya on the AlexNet team.
是的。而且不仅仅是伊利亚，比如Anthropic的达里奥。我的意思是，所有人。基本上开创了这个领域的杰夫·辛顿。他们都在那里待过。
Yeah. And it's not just Ilya, like Dario from Anthropic. Like, I mean, everybody. Jeff Jeff Hinton who basically invented the field. Like, they're all there.
塞巴斯蒂安·特伦、吴恩达，比如卡帕西。所有人。卡帕西、杰夫·迪恩。
Sebastian Thrun, Andrew Ng, like Carpathi. Everybody. Carpathi, Jeff Dean.
太多了。
So many.
你知道的，每一个你熟知的主要AI领导者，无论他们现在在哪家公司——是的。唯一的例外是Facebook的扬·勒昆。他是唯一一个并非来自谷歌的人。
You know, every single major leader in AI that you know of, no matter what company they are at Yep. With the one exception of Jan Lakun and Facebook. He's the only one who, like, didn't come from Google.
是的。而且我会给你提供关键信息。
Yeah. And I'll give you the hook.
他们创造了他们自己最可怕的噩梦。
They created their own worst nightmare.
是的。然后他们发表了它。Transformer论文。Transformer，没错。他们他们
Yes. And then they published it. The transformer paper. The transformer, yeah. They they
但这可能是让他们免于被拆分的关键。是的。比如，反垄断案中的法官指出，有这么多前谷歌员工在AI领域带来了激烈竞争。
But it might be the thing that saved them from getting broken Yes. Like, the judge in the antitrust case cited there's so much competition in AI from all these former Googlers.
但代价是什么？
But at what cost?
对。没错。
Right. Yeah.
这太神奇了。但是，是的。
That's amazing. But the Yeah.
我的意思是，我们一直回归到公司的创立使命——组织世界的信息。感觉他们完成了任务，创造出了比我们见过的任何东西都更好地实现这一目标的工具。
I mean, we've always come back to the founding mission of the company is to organize the world's information. And it feels like I'll like, they did their job to create the thing that does that better than anything we've seen
是的。
Yeah.
作为人类，对吧？每个人都喜欢启动，比如，从Gemini、Grok、ChatGPT或任何这些不同的大型语言模型得到的结果，比传统搜索更让人享受阅读和理解世界的过程。
As humans. Right? Everybody enjoys firing up, you know, jet like, the the results you get back from Gemini or Grok or ChatGPT or any of these different LLMs is much more enjoyable to to to just read through and understand the world than traditional search.
关于谷歌故事的另一点，我觉得没人理解，我直到我们做了整个三集系列节目才明白。它一直都与微软有关。嗯。微软从一开始就是生存威胁。嗯。
The other thing about the Google story that, like, I don't think anybody understood. I certainly didn't understand till we did our whole, you know, three episode series on it. It's always been all about Microsoft. Mhmm. Microsoft has always been, at first, the existential threat Mhmm.
然后目标是，我们要成为下一个微软。我们要主导他们。我们要创建Gmail和Docs，是的。应用程序，就像微软做的一切，我们都要做。
And then goal of, like, we're gonna become the next Microsoft. We're gonna dominate them. We're gonna create Gmail and Docs Yep. Apps, like, everything Microsoft does, we're gonna do.
因为记住，搜索，他们建立了这个极其荒谬的，你知道的，有史以来最赚钱的业务。
Because remember, search, they built this ridiculously, you know, the the most profitable business of all time.
在，你知道，他们曾经是
On, you know, they were
除了石油。除了沙特阿美。
Except for oil. Except for Saudi Aramco.
他们是微软资产的租户。是的。在Internet Explorer上。一切都在Internet Explorer上，而Internet Explorer又都在Windows上。
Which is They were tenants of Microsoft's property Yes. On Internet Explorer. It was all on Internet Explorer, which all was on Windows.
Internet Explorer曾占据70%的市场份额，而Windows则有大约90%的市场份额。所以在任何时候，如果微软想动摇谷歌那台荒谬的印钞机，有那么几年他们确实可以做到。是的。
Internet Explorer had 70% market share, and Windows had, like, 90% market share. And so at any given point, if Microsoft wanted to destabilize Google's ridiculous cash printing machine, there was a few years where they really could have Yeah.
那就是Chrome。那就是Chromebooks。那就是Android。那都是关于现在，OpenAI，微软，好像总是围绕着微软。
That's Chrome. That's Chromebooks. That's Android. That's all And about then now, OpenAI, Microsoft like, it's always all about Microsoft.
所以当
So when
这很有趣。
That's interesting.
OpenAI在埃隆投入微软怀抱之后加入。是的。谷歌的反应就像是，你在开玩笑吧。是的。我们花了公司头二十年才摆脱他们的控制，现在微软又卷土重来了。
OpenAI went in after Elon went into the arms of Microsoft Yeah. And Google is like, you gotta be kidding me. Yeah. We just spent the first twenty years of our company getting out from under their thumb, and now here's Microsoft coming back in.
是的。等等。你刚才说的是指OpenAI和Sam Sam吗？进入了，是的。
Yeah. Wait. When you said did you mean OpenAI and Sam Sam? Went into Yeah.
不。在埃隆离开之后，是的。并且撤回了他的资金，是的。然后OpenAI需要一个资本合作伙伴
No. After after Elon left Yeah. And pulled his funding Yeah. And OpenAI needed a capital partner
投入微软温暖的怀抱。温暖的拥抱，再次，谷歌的
In warm arms Microsoft. The warm embrace who Again, Google's
你知道，在ChatGPT发布时，那是谷歌最糟糕的时刻，微软拥有OpenAI 49%的股份。所以他们就像，天哪，就像
you know, at the at the worst moment for Google when ChatGPT came out, Microsoft owned 49% of OpenAI. So they're like, holy gita, like
它回来了。
It's the back.
他们又回来了。是的。房子里的怪物。这是他们自己的恐怖片。
They're back again. Yeah. The monster in the house. It's their own horror film.
是的。这是一部恐怖片。
Yeah. It's a horror film.
你最近对“七力模型”有什么看法？我记得很多集节目最后都有七力模型的分析。你觉得这个框架是否需要更新？你认为有没有一本新书，在战略思维方面有望达到那种影响力，就像NBA级别看待科技公司或更广泛业务的方式？还是在你看来，这已经是历史的终点了？
I how are you thinking about the the Seven Powers these days? I remember a lot of the episodes end with analysis from Seven Powers. Do do you think there's do you think any any of that framework needs an update? Do you think there's a a new book that is on a trajectory to have that level of influence in terms of strategic thinking that would be kind of like the NBA level way to think about tech companies or businesses broadly? Or is that kind of the end of history in your mind?
我想这取决于——我还没仔细想过这个问题。但我确实认为七力模型仍然是分析企业、判断其能否持续盈利的有效方法。
I guess it depends if I haven't thought about this. I do think seven powers is still like the applicable way to analyze a business and figure out if it will be durably
是的。
Yeah.
相较于竞争对手。一个新变化是，由于规模效应，AI模型现在拥有比以往任何时候都更强的数据网络效应和数据护城河。飞轮效应。
Profitable versus its competitors. The one new thing is AI models have because of scaling laws have much stronger data network effects and data moats than we've ever seen in the past. Flywheel.
没错。MidJourney就是个例子。
Yeah. You see this with mid journey.
数据越多越好，而且这种趋势有增无减。谷歌之所以下令所有团队停止使用定制模型，转而使用Gemini，是因为我们需要尽可能多地给Gemini喂数据——不仅来自谷歌所有平台，还包括谷歌云客户的每一个界面。这就像是……现在只有两种
Just more more data is better, and that continues unabated. Mean, the reason Google had an edict that all teams need to stop using these bespoke models and start using Gemini is we gotta feed Gemini as much data as we can from not only every Google surface, but then every Google Cloud customer surface. It's like There's only two
模型中的中等规模经济，因为公司在模型工作中的任何碎片化，都需要集中起来，全部输入到一个模型中。
mid scale economies in models because you you like any fragmentation you have in your work with your models across your company, like, you need to centralize that and feed it all into one.
所以，如果有什么不同的话，这感觉更像是加倍投入七种力量的理由。
So If anything, that feels like a a reason to double down on the seven powers.
是的。我我认为它仍然适用，但它是
Yes. I I think it's still applicable, but it's
它正在失效。
It's breaking.
这些模范公司真的，
These model companies have just really,
真正的力量。对业务有深刻的理解，他们就像是，这些词没有一个出现在七种力量中。是的。但是
really power. Doing a understanding of business, they're like, none of these words appear in seven powers. Yeah. But
我要说，我们已经非常了解汉密尔顿了。是的。还有他的公司 Strategy Capital，非常非常熟悉。我是他基金 Strategy Capital 的顾问委员会成员。而且，你知道，他们总是在寻找和工作，就像他，你知道，他们并不认为七种力量是万能的。
I will say, we we've gotten to know Hamilton Yeah. And his firm Strategy Capital really really well. I'm on his advisory board for Cool. At his fund Strategy Capital. And, you know, they're he's always looking and working, and like he's, you know, they're not he doesn't believe that seven powers is the be all end all.
是的。
Yeah.
他们正在寻找并为Sure工作。你知道，接下来就是Sure。我相信肯定会有
They're looking and working for Sure. You know, the next thing. Sure. I'm sure there will
更多。是的。
be more. Yeah.
科技史上哪些关键时刻让人们过度使用了杠杆？嗯。因为感觉我们可能正处于
What points throughout tech history stand out where people got over their skis on with leverage? Mhmm. Because it feels like we're potentially
你要带我们去Oracle吗？
Are you taking us to Oracle?
是的，Oracle。但感觉就像，你知道，我们刚读了Semi Analysis的Doug O'Laughlin的一些东西，他确实认为下一步将是，你知道，超大规模企业的负自由现金流，并且真的，你知道，为了获胜而加大杠杆。对吧？每个人都只想赢。
Yeah. Oracle. But it feels like, you know, we were just reading something from Doug O'Laughlin at at Semi Analysis and he really feels like the next step is going, you know, negative free cash flow for the hyperscalers and really, you know, levering up in order to just win. Right? Everybody just wants to win.
所以，是的，只是好奇有没有这样的时刻。显然，电信业是债务驱动的，我们也看到了那里的结果。但在现代，感觉超大规模企业从未说过，让我们真的加大我们的
And so, yeah, was just curious at any kind of point. Obviously, the telecom was very debt fueled and we saw what happened there. But it doesn't feel like in the modern era, we've you know, the hyperscalers have ever said, let's really lever up our
嗯，而且这周我们一起参加的那个活动，你们昨天错过了。有更多的讨论。你可以更广泛地定义杠杆。比如，杠杆不一定只是债务资本。如果你看看合同和公司，系统里其实有很多杠杆，比如
Well, and there's there's the event that we were at this week together, you guys missed it yesterday. There was more discussion. There's you could define leverage more broadly. Like, leverage isn't necessarily just debt capital. Like, there's a lot of leverage in the system if you look at the contracts and company, like
当然。
Sure.
看看微软的开盘价对吧？就像，你知道，或者这些交易，比如，有多少资本，无论是，
Look at opening at Microsoft. Right? Like, you know, or or any of these deal, like, how much of the capital, whether it's,
你知道我的意思是，很多这些合同实际上在资产负债表上作为负债存在。
you know I mean, a lot of these contracts literally live on the balance sheet as liabilities.
完全正确。你有高超的技能。资金只是在圈子里转来转去，这就在系统中建立了杠杆。是的。
Totally. You've got the hyper skills. You've got money's all just going around and around the circle that builds leverage in the system. Yeah.
因为XAI交易是一个SPV，我认为是由XAI主导的，但他们好像有85亿美元现金，然后还有125亿美元的GPU？是基础债务。但这基本上是在SPV层面发生的。好吧。我认为这有点值得注意，因为在但
Because the XAI deal is an SPV that I think is led by XAI, but they're doing like 8 and a half billion of cash and then like 12 and a half of of GPUs? Of Basic debt. But it's happening at the basically happening at the SPV level Okay. Which I thought was I mean, I think is somewhat notable because in but
是的。怎么，比如，怎么
Yeah. How how, like, how
系统里的杠杆比资产负债表上显示的要多得多。是的。会显示出来。
far are there's more leverage in the system than the balance sheets would Yeah. Would show.
你希望历史回溯的范围有多大？荷兰东印度公司这个话题有意思吗？
How how much range are you looking for in terms of how far back you go in history? Is Dutch East India Company interesting?
经常出现
Comes up all
这种情况。
the time.
是的。我我没错。因为有很多争论，比如，按今天的美元市值计算实际价值是多少？还是大家都喜欢追溯
Yeah. I I Yeah. Because there's debates on, like, what was the real Everyone loves to market cap in in in dollars today? Or is it Everyone loves to go
到19.99美元这个价位，但其实还有太多其他例子。而且在我看来你们就像是科技史领域的关键领袖。所以我觉得肯定还有更多内容。从数据中你们也看到LVMH表现惊人——我一直把你们视为科技史播客，而LVMH的成绩好得不可思议。这是你们当初预料到的吗？
to $19.99 right now, but there are so many other examples. And and you guys are like the the, like, the key leaders of tech history in my mind. And so I would imagine that there's there's more to it. And you've seen in the data LVMH performed like, I think of you as a tech history podcast, and LVMH just does incredibly well. Like, was that something you predicted?
这其中有什么玄机吗？
Is there something there?
最精彩的剧集往往包含三个关键要素。我们刚开始做科技播客时，我以为只会覆盖科技公司。后来在进入主流前的过渡期，我们开始向科技受众普及非科技现象。嗯。
So the best episodes Yeah. Are the ones that have these three key ingredients. And I always thought when we started over a tech podcast, we cover tech companies. Then we were in this middle phase before we sort of became more mainstream, which was, educating a tech audience about non tech phenomena. Mhmm.
就像，没有科技公司擅长品牌建设。所以当我们开始研究奢侈品公司时，简直让所有科技界人士大开眼界，哇，这就是为什么这很有价值。
Like, no tech companies are good at brand. And so when we started studying the luxury companies, it's like blowing the minds of all these tech people, like, woah. That's why this is valuable.
是的。是的。
Yeah. Yeah.
这包括我自己。就像，我在研究过程中学到了，然后我想，嘿，观众们，我必须和你们分享这个。猜猜我刚发现了什么。所以三个关键要素是
Which included myself. Like, I learned during the research and I'm like, hey, audience. I I gotta share this with you. Guess what I just figured out. And so the three key ingredients
这个东西不仅仅是戴在你手腕上的一块金属。对吧。你知道，就像
that This thing is more than a hunk of metal your wrist. Right. You know, like
这三件事是：第一，你需要一个英雄主角嗯。他有一个伟大的故事，我们可以把所有教训都挂在这个惊人的英雄旅程上。
The three things are, one, you need a hero protagonist Mhmm. That has a great story where we can really hang all the lessons on this amazing hero's journey.
人们关心的是人。
People care about people.
是的。
Yes.
他们不只是想读一份新闻稿的事实清单。
They don't just wanna read a a fact sheet of press releases.
否则，卖方分析师就会成为很棒的播客主持人了。是的，但他们大多不是。
Otherwise, sell side analysts would be great podcasters Yep. And they're mostly not.
是的。
Yep.
第二点是你需要一个隐藏在显而易见之处的秘密。我们需要能够发现一些非常巧妙的东西。好市多（Costco）的低SKU数量以及它如何导致，基本上是库存
Two is you need a secret hiding in plain sight. We need to be able to find something very clever. Costco's low SKU count and how it leads to, basically inventory
周转率。供应商。所以，就像，好市多所有惊人的好处。
Turnover. Suppliers. So, like, all the all the amazing benefits of Costco.
是的。我差点就开始讲整个好市多了。好市多的低SKU数量。这些东西之一就像是隐藏在显而易见之处的秘密
Yeah. I almost launched into the whole Costco. Costco's low SKU count. How how one of these things is this, like, secret lurking in plain sight
我猜他们5%的东西是手工制作的。
I imagine makes 5% of their stuff by hand.
是的。没错。然后第三点是，我之前发表过这个竞选演说，但我忘了。
Yes. Yeah. And then three is I've given this stump speech before, but I forgot
你通常都会发表这个演说。
You it usually give the speech.
关于 Bayt 播客
Bayt 提供中文+原文双语音频和字幕，帮助你打破语言障碍，轻松听懂全球优质播客。