🟠 YC演示日，Paul Graham加盟，AWS会从谷歌购买TPU吗？ | Harj Taggar、Paul Graham与Jessica Livingston、Richard Wang、Philip Ho、Ali Attar、Kurush Dubash等

这也像是一年中最忙碌的时段。

It's also just like it's a punishing time of the year.

我猜可能是因为临近假期什么的，我们简直忙得不可开交。

I guess it's, like, right before the holidays or something because we've just been complete torn.

我们显然很想去YC演示日。

We we obviously wanted to go to YC demo day.

我还想去正在举办的NeurIPS——顶级AI会议。

I also wanted to go to NeurIPS, which is going on right now, the premier AI conference.

还有DealBook峰会。

There's also DealBook Summit.

Andrew Sorkin正在做各种最棒的访谈。

Andrew Sorkin's doing, like, all the greatest interviews.

同时还有Reinvent大会。

At the same time, there's Reinvent.

我本来也想去参加的。

I wanted to go to that.

来自DealBook的访谈

Interviews coming out of DealBook.

我今天早上刚看到一些片段

I just saw some clips this morning.

我们会播放一些

You got We'll play some.

Scott Besson表现得很强势

Scott Besson just going hard.

是啊

Yeah.

Alex Karp也很强势

You got Alex Karp going hard.

这些方面都没有真正的意外，不过是啊

No real surprises on either of those fronts, but Yeah.

很期待

Excited to

嗯

Well

获取最新进展

get the update there.

让我来介绍一下Cognition

Let me tell you about Cognition.

Devin背后的团队——这款AI软件工程师，用你的个人AI工程团队来攻克积压任务

The team behind the AI software engineer, Devin, crush your backlog with your personal AI engineering team.

让我们用这篇Zephyr的帖子结束训练覆盖率的讨论，文中提到谷歌的TPU取得了这类成功

Let's let's let's close out the training coverage with this Zephyr post who says Google is having this kind of success with TPUs.

那亚马逊的Tranium呢？

What about Amazon's Tranium?

Tranium是新推出的但性能不足，仅667万亿次浮点运算（BF16）

Tranium is new and underpowered, just 667 t flops b f 16.

它配备了大量HBM，但带宽低于H100 TPU v6e

It has lots of HBM, but the bandwidth is lower than the h 100 TPU v six e.

这款具有竞争力的h100，既不需要HBM也不依赖带宽，而Ironwood在FLOPS带宽和HBM容量方面与Blackwell旗鼓相当。

This competitive h 100, non HBM or bandwidth, and Ironwood is competitive with Blackwell on FLOPS bandwidth and HBM capacity.

随着产能爬坡，预计Ironwood将快速抢占市场份额——从吞吐量/总体拥有成本对比可见，NVIDIA对Tranium的优势明显。

I expect Ironwood to quickly gain market share as it ramps up as you can see from throughput slash TCO, NVIDIA versus Tranium.

Ruben Moggs数据显示，Tranium3在TCO训练FLOPS上比Tranium2更难匹敌Blackwell，但在TCO内存带宽方面将差距缩小了5%。

Ruben Moggs, Tranium three harder than Blackwell versus Tranium two on TCO training FLOPS and reduces the gap by 5% on TCO MEM bandwidth.

因此NVIDIA与Tranium之间的差距实际上正在扩大而非缩小。

So the gap between NVIDIA and Tranium is actually increasing rather than decreasing.

顺便说明，这些计算是在CPX发布前完成的。

By the way, this math was done before CPX was introduced.

即使CPX加上Rubin方案，其推理成本低于Tranium我也不会感到意外。

I won't be surprised if CPX plus Rubin is cheaper than Tranium for inference.

我认为确实存在这样的可能性：某些专用场景——就像Descartes的案例那样，会出现专门针对Tranium优势领域优化的特殊模型，从而进一步细分市场。

So I I do think that there's a world where there's, where there's something specialized, like what's going on with Descartes, some sort of special model that's that that that excel that that that thrives in what Tranium is good at, and they can further niche down.

不过我们拭目以待。

But but we'll see.

我是说，也许他们会后来居上，彻底击败TPU，明年我们都会讨论Trainium。

I mean, maybe they come from behind and they just destroy t p TPU, and we're all talking about Trainium next year.

总之，让我来介绍一下Linear。

Anyway, let me tell you about Linear.

认识一下现代软件开发的系统。

Meet the system for modern software development.

Linear能简化从路线图到发布的整个开发周期的工作流程。

Linear streamlines work across entire entire development cycles from road map to release.

我们要稍微默哀一下。

We're gonna say a little rest in peace.

为小队默哀。

Rest in peace to squad.

旧金山心爱的白化短吻鳄去世了，享年30岁。

San Francisco's beloved albino alligator has passed away at age 30.

这个年纪很不错了。

That's a good age.

我不知道鳄鱼通常能活多久，但我很高兴正在查资料。

I don't know how long alligators typically live, but I'm glad Looking it up.

感觉像是三十到五十年

Feels like Thirty to fifty years

对于美洲短吻鳄来说。

for the American Alagina.

稍微短了点，不过克劳德当然

Little bit short, but, Claude was, of course

通常能活到70岁或更久。

Often supporting reaching 70 or more.

是的。

Yes.

总之，安息吧，你知道，显然人们立刻开始猜测了。

Anyways, RIP, there was, you know, obviously, people started speculating immediately.

Anthropic当然是赞助商

Anthropic, of course, was the sponsor

是的。

Yes.

克劳德。

Claude.

对。

Yes.

而且，你知道，人们都在怀疑这其中是否存在什么阴谋？

And, you know, people were wondering was was there foul play involved?

嗯哼。

Mhmm.

有没有可能这只可怜的恐龙...不是恐龙。

Was it possible this this poor dinosaur not dinosaur.

恐龙。

Dinosaur.

鳄鱼去世的那天正好是公告发布的那天。

Alligator passed the day that they that that it that it got announced.

是啊。

Yeah.

他们聘请了IPO律师。

That they've hired IPO lawyers.

有人猜测，克劳德是否被作为祭品献给了资本市场的众神，就像某种仪式一样？

Some people were speculating could is it possible Claude was sacrificed to the capital markets gods in some type of ritual?

不过话说回来，看看他脸上的这个表情。

But anyways, he look at the look at this expression he has on his face.

我们能稍微放大一点吗？

Can we zoom in a little bit?

真是个酷炫的家伙。

What a what a what a cool guy.

他会被人们铭记。

And he will be remembered.

没错。

Yeah.

丹·普里马克正在谈论X Lite。

Dan Primak here is talking about, x lite.

我想我们很快就能邀请到CEO上节目。

I think we might have the CEO on the show soon.

特朗普政府将向一家名为X Lite的光刻初创公司投资1.5亿美元。

The Trump administration will invest a $150,000,000 into a lithography startup called X Lite.

今早与X Lite CEO交流了其首个芯片法案奖项的情况。

Its first Chips Act award chatted this morning with X Lite CEO.

现在有几家光刻公司了。

There's a few lithography companies now.

我们节目曾邀请过其中一些。

We've had some on the show.

这感觉像是全新的投资层级，非常有趣，比如政府投入的1.5亿美元，看起来像是B轮融资规模。

This feels like an entirely new it's a it's a very interesting tier of investment, like a $150,000,000 from the government that feels like a series b.

他们确实在今年夏天完成了B轮融资，由Playground Global领投，Playground合伙人兼英特尔前CEO帕特·基辛格出任X Lite执行董事长。

They did raise a they did raise a series b this past summer led by Playground Global with Playground partner and former Intel CEO, Pat Gelsinger, becoming X Lite's executive chairman.

哇哦。

Woah.

所以政府投资英特尔是合理的。

And so makes sense that the government's investing in Intel.

帕特·基辛格，当然是英特尔的前任CEO。

Pat Gelsinger, of course, former Intel CEO.

现在他涉足X Lite，筹集了4000万美元资金，又从政府那里获得了1.5亿。

Now he's getting involved in X Lite, marshaled 40,000,000 of capital, went and got a 150 from the government.

故事还在继续。

The the story continues.

还有另一家AI初创公司，想要重塑8000亿美元的芯片产业。

There's also another AI startup, that wants to remake the $800,000,000,000 chip industry.

这家公司登上了《华尔街日报》，由前谷歌研究人员创立，名为递归智能，获得了红杉资本支持的3500万美元融资，用于自动化芯片设计。

This one's in The Wall Street Journal founded by ex Google researchers, Recursive Intelligence, raised 35,000,000 with backing from Sequoia to automate chip design.

显然，这不是光刻技术。

Obviously, this is not lithography.

这是设计流程。

This is the design process.

但公司们仍在Dylan Patel

But still, companies are Dylan Patel

稍微谈到了一些

was talking a little bit about

这个。

this.

堆栈。

The stack.

哦，他确实提到了。

Oh, he did.

我没听说这事。

I didn't hear about that.

非常酷。

Very cool.

这是用于AI芯片设计的人工智能。

This is AI for for AI chip design.

哦，没错。

Oh, that's right.

是的。

Yes.

用于AI芯片设计的人工智能。

AI for AI chip design.

我们所需的一切。

Everything we need.

在距离斯坦福大学几个街区的一条安静住宅街上，两位前谷歌研究员正在创办一家初创公司，他们希望这将重塑价值8000亿美元的芯片行业。

On a quiet residential street a few blocks from Stanford University, two former Google researchers are launching a startup they hope will remake the $800,000,000,000 chip industry.

安娜·戈尔迪和阿扎利亚·梅尔霍斯尼正试图开发能够自动化设计尖端芯片的软件，这一愿景将使每家公司都能从零开始打造自己的芯片。

Ana Goldie and Azalea Merhosni are trying to build software that can automate the design of cutting edge chips, a prospect that would allow every company to build their own chips from scratch.

两人在郊区住宅的顶层工作，最近筹集了3500万美元资金，在红杉资本等投资方的支持下启动了递归智能项目

Working from the top floor of a suburban home, the duo recently raised 35,000,000 to kick start recursive intelligence with funding from Sequoia Capital and

抱歉。

strike Sorry.

我们

The We

明白了。

got it.

递归的我们得加上这个。

The recursive We gotta add that.

把它加进名字里。

Putting it in the name.

我们得把这个加到列表里，因为已经有标准资本、现代资本、标准智能、现代智能了。

We gotta add that to the list of because there's standard capital, modern capital, standard intelligence, modern intelligence.

原始智能。

Raw intelligence.

原始智能是那个自由谎言。

Raw intelligence was the the lying free.

应用型是另一种。

Applied was another one.

Cap intelligence。

Cap intelligence.

那另一个是什么来着？

And then what was the other one?

洛克希德·格鲁姆的公司叫什么来着？

There's what what what's Lockheed Groom's, company?

物理智能。

Physical intelligence.

有物理智能，物理资本。

There's physical intelligence, physical capital.

所以这是个矩阵，比如资本，它是什么来着？

So it's the matrix of, like, capital, what was it?

资本智能和直觉之类的。

Capital intelligence and intuition or something like that.

你把它们全部相乘，就能得到完整的体系。

And you and you multiply them all out and you get the whole thing.

最终，我们会用完这些词。

Eventually, we're gonna run out.

对吧？

Right?

某种程度上是有限的。

There's it's some somewhat finite.

不。

No.

永远会有更多名称出现。

There will always be more names.

那就创造些新词吧。

Start up a new new words.

哇，真厉害。

So wow.

这家公司3500万，估值7.5亿。

The company, 35,000,000, for a valuation of 750,000,000.

稀释比例非常低。

That's a very low dilution.

什么？

What?

5%左右？

5% or something like that?

相当了不起。

Pretty remarkable.

确实

Definitely

风投确实很厉害。

VC is remarkable.

高。

High.

是的。

Yeah.

我...我原本以为这会是一个非常资本密集型的行业，但如果他们只是开发软件的话，也许...也许他们在这方面有更多控制权。

I I I would I would have assumed this would be a very a very capital intensive business, but I I I suppose if it's just a software that they're developing, maybe maybe they have more control here.

诸如亚马逊和谷歌等公司已为AI和数据中心用途开发了定制芯片，而苹果通过内部研发设备芯片（包括帮助重振其MacBook笔记本电脑的M系列芯片）节省了数十亿美元。

Companies such as Amazon and Google have developed custom chips for AI and data center use, and Apple saved billions of dollars by insourcing chips for its devices, including the M series chips have helped revitalize its MacBook laptops.

这类硅基方案可能更便宜，更...

Such silicon options can be cheaper, more I had

昨天有个有趣的瞬间。

a funny moment yesterday.

我们收到了一个亚马逊包裹，没错。

We got an Amazon package Yep.

包装上全是外星人（Alienware）品牌的标志。

And it was covered with, like, Alienware, like, Alienware branding.

而我...

And I

我问莎拉

I asked I asked Sarah.

我说，你是不是买了外星人的东西？

I was like, did you did you get something from Alienware?

这是怎么回事？

Like, what is going on?

结果发现是个广告

And it turned out to be an ad.

不是

No.

他们在宣传说这是由英特尔驱动的

They were advertising that it's powered by like Intel.

哦，有意思

Oh, interesting.

这并没有让我立刻想买外星人设备

Which which didn't make me necessarily want to immediately buy an Alienware device.

你确实，你你你把钱直接放回自己口袋了，因为你是纳税人。

You do, you you you put the money straight back in your pocket because you're a taxpayer.

你拥有英特尔。

You own Intel.

确实如此。

That's true.

确实如此。

That's true.

你应该支持英特尔。

You should support Intel.

不。

No.

英特尔在游戏领域无疑表现出色。

Intel Intel is undisputably great for gaming.

这一点毋庸置疑。

There's no question there.

问题是，他们能否建造一个能与台积电竞争的晶圆厂？

The question is, are they gonna, you know, be able to build a fab that competes with TSMC?

我想这完全是另一个问题了。

I guess it's a completely different question.

我可能会去组装一台外星人Intel电脑。

I I might go build an an Alienware Intel PC.

哦，我们要为办公室的赛车装备准备。

Oh, we're going to for the for the the office racing rigs.

模拟赛车肯定需要Intel处理器。

The sim racing needs the Intel Inside for sure.

肯定的。

For sure.

这节目要变成模拟赛车秀了，我们会一边模拟赛车一边看其他播客节目并实时吐槽。

This is just gonna turn into a sim racing show where we watch other podcasts while sim racing and reacting to it.

没错。

Yeah.

所以我喜欢这样。

So I like it.

Vanta，自动化合规与安全，人工智能驱动从证据收集、持续监控到安全审查和供应商风险的一切。

Vanta, automate compliance and security, AI that powers everything from evidence collection and continuous monitoring to security reviews and vendor risk.

德瓦卡什·帕特尔写了一篇重磅文章，彻底颠覆了时间线预测。

Dwarkash Patel has a, a massive essay shaking up the timeline.

对AI发展的看法。

Thoughts on AI progress.

他说自己短期适度看空，但长期极度看涨。

He says he's moderately bearish in the short term, but explosively bullish in the long term.

嗯，非常有趣。

Well Very interesting.

他表示不理解为什么有些人持短期观点。

So he says he's confused why some people have short timelines.

他们说通用人工智能即将到来。

They say AGI is coming soon.

但与此同时，他们对RLVR（可验证奖励的强化学习）持乐观态度。

But at the same time, they're bullish on RLVR, which is reinforcement learning with verifiable rewards.

因此他表示，如果我们真的接近拥有类人类学习能力的系统，那么整个方法就注定失败。

And so he says, if we're actually close to a human like learner, this whole approach is doomed.

目前，实验室正试图通过中期训练将这些技能批量植入模型。

Currently, the labs are trying to bake in a bunch of skills into these models through mid training.

有一整套企业供应链正在构建RL环境，用于教导模型如何使用Excel编写财务模型。

There's an entire supply chain of companies building RL environments, which teach the model how to use Excel to write financial models.

例如，我们正在讨论的Crunch——一家YC孵化公司推出的AI Excel分析师工具，专为Excel高级用户设计。

For example, I think we're actually talking to an AI Excel analyst for Excel power users called Crunch at twelve fifty YC company.

我认为这些都是不错的构想。

I think that these are good ideas.

实际上，我十分看好这个模型。

I I'm actually very bullish on on, this this, this model.

但在通用人工智能何时到来、超级智能何时出现的问题上，我理解德瓦基什的观点。

But in the context of when does AGI arrive, when does superintelligence arrive, I understand Dwarkish's point.

他说，要么这些模型很快就能以自主的方式在工作中学习，使所有这些预先训练变得毫无意义；要么它们做不到，这意味着通用人工智能（AGI）不会很快到来。

He says either these models will soon learn on the job in a self directed way, making all of this prebaking pointless, or they won't, which means AGI is not imminent.

人类不需要经历一个特殊的训练阶段，在那里我们必须预先练习可能使用的每一款软件。

Humans don't have to go through a special training phase where they need to rehearse every single piece of software we might ever use.

巴伦在最近的一篇博客文章中对此提出了有趣的观点。

Barron made interesting points about this in a recent blog post.

当我们看到前沿模型在各种基准测试中进步时，我们不应该只想到规模扩大和巧妙的机器学习研究思路，还要想到花费数十亿美元聘请博士、医学博士和其他专家来编写问题和提供示例。

When we see frontier models improving at various benchmarks, we should not, we should think not just of increased scale and clever ML research ideas, but billions of dollars spent paying PhDs, MDs, and other experts to write questions and provide examples.

让我们为博士们鼓掌。

Let's give it up the PhDs.

以及推理目标。

And reasoning targets.

让我们为专家们鼓掌。

Let's give it up for the experts.

这些精确的能力。

These precise capabilities.

从某种程度上说，这像是专家系统时代的大规模重现——只不过现在不是直接付钱让专家将他们的思维编码成程序，而是让他们提供大量经过形式化追踪的推理过程案例，然后我们通过行为克隆将这些案例提炼到模型中。

In a way, this is like a large scale reprise of the expert systems era where instead of paying experts to directly program their thinking as code, they provide numerous examples of their reasoning and process formalized and tracked, and then we distill them into models through behavioral cloning.

这让我略微倾向于延长AI发展时间线，因为既然我们需要投入如此巨大努力来为前沿系统设计极高品质的人类行为轨迹和环境，这说明它们仍然缺乏真正AGI必须具备的核心学习能力。

This has updated me slightly towards longer AI timeline since we since given we need such effort to design extremely high quality human trajectories and environments for frontier systems implies that they still lack the critical core of learning that an actual AGI must possess.

这种矛盾在机器人领域显得尤为突出。

This tension seems especially vivid in robotics.

从根本意义上说，机器人技术是算法问题，而非硬件或数据问题。

In some fundamental sense, robotics is an algorithm's problem, not a hardware or data problem.

人类只需极少训练就能学会远程操作现有硬件完成有用工作。

With very little training, a human can learn how to teleoperate current hardware to do useful work.

因此如果我们拥有类人类学习能力的系统，机器人技术大部分难题就能迎刃而解。

So if we had a human like learner, robotics would, in large part, be solved.

但正因为缺乏这种学习能力，我们才不得不走进成千上万不同家庭和工厂，学习如何收拾餐具或叠衣服。

But the fact that we don't have such a learner makes it necessary to go out into thousands of different homes and factories and learn how to pick up dishes or fold laundry.

我听过五年内实现突破派的反对观点：我们做这些GRL是为了构建超级AI研究者，然后让百万个自动化Ilia的副本可以解决如何从经验中学习如何解决。

One counterargument I've heard from the takeoff within five years crew is that we have to do this clue GRL in service of building a superhuman AI researcher, and then the million the million copies of automated Ilia can go figure out how to solve robust and efficient learning from experience.

这给人一种'每笔交易都在亏钱，但我们会通过销量赚回来'的感觉。

This gives the vibes of we're losing money on every sale, but we'll make it up in volume.

这个自动化研究者竟然要解决人类近一个世纪都未能突破的通用人工智能算法问题，而它连儿童的基本学习能力都不具备？

This automated researcher is somehow going to figure out the algorithm for AGI, something humans have been banging their heads against for the better part of a century while not having the basic learning capabilities that children have?

这在我看来极其不可信。

That seems super implausible to me.

况且，即便你相信这种说法，它显然也不符合实验室研究强化学习的现实路径。

Besides, even if you even even if that's what you believe, it clearly doesn't describe how the labs are approaching RLVR.

要自动化伊利亚的工作，根本不需要预先掌握制作PPT演示文稿这类咨询顾问技能。

You don't need to prebake the consultant's skills at crafting PowerPoint slides in order to automate Ilia.

显然，实验室的做法暗示着这样一个世界观：这些模型在泛化和在职学习方面将持续表现不佳，因此必须预先植入那些他们认为具有经济价值的技能。

So clearly, the lab's actions hint at a world where, at a world view where these models will continue to fare poorly at generalizing and on the job learning, thus making it necessary to build in the skills that they hope will be economically valuable beforehand.

我想谈谈经济扩散的部分。

I wanna go to the section on economic diffusion.

不过在此之前，我要先讲讲Privy的故事。

But first, I'm gonna tell you about Privy.

Privy让在加密轨道上构建应用变得简单，能安全地启动白标钱包、签署交易，并通过一个简单的API集成链上基础设施。

Privy makes it easy to build on crypto rails, securely spin up white label wallets, sign transactions, and integrate on chain infrastructure all through one simple API.

所以你一直在询问经济扩散的问题。

So you've been asking about economic diffusion.

我们目前的扩散速率是多少？

What is the rate that we're diffusing?

让我们看看德瓦卡什对经济扩散有何见解。

Let's see what Dwarkash has to say about economic diffusion.

他认为经济扩散滞后是对能力缺失的自我安慰。

He says that economic diffusion lag is cope for missing capabilities.

这似乎也受到泰勒·考恩观点的影响，即通用人工智能已经到来。

And so this is also seems informed by the Tyler Cowen take that, AGI is here.

模型表现良好，只是需要时间来适应它们。

The models are good, but it just takes time to adopt them.

对此我深有共鸣，因为当我去诊所时他们递给我一张纸，我就知道网页表单已经足够好了。

And I I'm very sympathetic to this because when I go to the doctor's office and they hand me a piece of paper, I know that a web form is good enough.

比如，数字表单的功能已经完备。

Like, the capabilities of the digital form are complete.

并不是说表单缺少什么功能或不够可靠。

It's not that the form is lacking in something or it's not reliable enough.

并不是他们说的那种情况，比如‘哦，是的’。

It's not like they're like, oh, yes.

比如网站有20%的时间会宕机，所以在这种情况下纸质表单仍然更有意义。

Like, the website goes down 20% of the time, and so paper makes more sense still in this case.

其实不是这样的。

It's like, no.

这只是一个普及度的问题。

It's just a diffusion problem.

只是那个诊所的负责人还在用老方法做事。

There's just someone who runs that doctor's office is like, like doing it the old way.

对吧？

Right?

这就是经济扩散滞后问题，我认为在许多场景中确实存在这种情况。

And that's the and that's the economic diffusion lag problem that I think is real in a lot of scenarios.

但说到缺失的功能，我举个非常具体的例子。

But The missing capability thing, I mean, just just to give a pretty concrete example.

目前，AI在生成文本方面表现非常出色。

Right now, AI is great at generating text.

对吧？

Right?

它非常擅长分析一段内容并据此生成文本。

It's great at kind of analyzing a piece of content and then generating text based on that.

然而，我们TPPN团队仍有多个成员的工作是寻找节目中精彩瞬间，然后围绕这些内容创建字幕，并分享到X、Instagram、YouTube等平台。

And yet, we still have multiple people on the team at TPPN whose job is to find interesting moments of the show and then create captions around that and share it to X and Instagram and YouTube and other platforms.

还有乔拉·凯什

And and Jorah Kesh

经历过这种情况

experienced it

同样，他试图通过一个大型Gemini提示来找出完整播客中最有趣的部分。

too, where he was trying to find the most interesting pieces of a full podcast with one big Gemini prompt.

他尝试了所有不同的模型，却无法真正找到最突出和最具传播力的要点。

And he was trying all the different models and couldn't get it to actually find like the most salient and viral points.

是的。

Yeah.

另一个突出的问题是，看似缺失的能力之一是无法识别幽默，甚至近乎情感的东西。

So one of the other thing that stands out is one of the seeming missing capabilities is ability to identify humor or even something like it's almost emotional.

Ilya和Dwarkash讨论过这个，我记得Ilya举了个例子，科学家研究过那些因脑损伤而丧失情感体验能力的人。

So Ilya and Dwarkash talked about this where I think Ilya was giving the example of scientists studied people who had had various brain injuries that limited their ability to experience emotion.

当他们失去情感时，有人可能需要花两小时才能决定选哪双袜子。

And when they took out emotion, it took them, it can take somebody two hours to figure out which pair of socks to choose.

他们震惊了。

And they were stunned.

不过就是选双袜子而已。

It's just a pair of socks.

你知道自己一天中发生了什么。

You know what's going on in your day.

为什么你需要情感来做这种决定？

Why do you need emotion in order to make that kind of decision?

所以至少在AI领域，似乎缺失的能力是，比如在Parkesh的案例中，如何找出播客中有趣的瞬间。

And so it seems like, at least in AI, a missing capability is like, Okay, finding out what's an interesting moment of a podcast in Parkesh's case.

是能让听众产生共鸣的内容吗？

Is it something that makes the audience member feel feel something?

对吧？

Right?

是

Is it

我是说，需要梳理的内容实在太多了。

I mean, there's just so much to pull through.

比如我记得在Carpathi的采访中，我当时和Tyler一起在看。

Like, I remember during the Carpathi interview, I was watching it and Tyler was watching it.

就在那一刻，卡帕西说，那些编程模型确实神奇又惊艳，但它们产出的却是垃圾。

And there's this moment where Carpathi says, like, the coding models are are are amazing and they're magical, but what they produce is slop.

而‘垃圾’这个词简直成了年度热词，或者说可能是去年的年度词汇。

And it's like that word slop is so it's the it's it's like the word of the year or maybe the word of last year.

这个词影响巨大。

Like, it's a huge word.

它承载了太多含义。

It has a huge amount of weight.

从他口中说出来，简直不可思议。

Coming from him, it's crazy.

愤怒诱饵击败了‘垃圾’成为年度词汇

That rage bait beat out slop for the word of

的称号。

the year.

‘垃圾’很可能会成为2024年的年度词汇之类。

Slop is probably the twenty twenty four word of the year or something like that.

不过话说回来，重点是，当我听到泰勒用'垃圾'这个词形容时，所有人都震惊了。

But, anyway, the point was, like, when when when I heard that, when Tyler heard that that that word, Carpathi calling it slop, everyone was like, woah.

我当时就想，我们该把这段剪出来。

And I was like, we should clip that.

结果一看，已经被别人抢先剪辑了。

And we looked, and it had already been clipped by a no.

被一个真人剪辑的。

By a human.

是啊。

Like Yeah.

时间线上也有人发现了这个值得关注和回应的疯狂瞬间。

Someone on the timeline had also identified that it was was the crazy moment that we should be reacting to and taking in.

这

It

确实，太疯狂了。

was Yeah, it's crazy.

另一个值得注意的是，在WAP平台上，人们最常做的工作之一，或者说他们在线赚取第一桶金的方式，就是为各种内容创作者和媒体公司剪辑视频。

The other thing that's notable is on WAP, one of the top jobs that people do on WAP, or way they make their first dollar online, is just clipping for various content creators and media companies.

而且他们制作的一些剪辑片段非常粗糙。

And some of the clips that they make are so sloppy.

那真的只是节目中的随机片段，他们却用20个不同的账号疯狂发布。

It's literally just a random segment of the show, and they're blasting it out from 20 different accounts.

我们至今还在花钱雇人做这种事，这本身就值得注意。

And the fact that we're still paying humans to do that, still, I mean, it just feels notable.

是啊。

Yeah.

好吧，我们来读读德瓦克·赫斯关于经济扩散实验室滞后的观点——这是对缺失能力的自我安慰。

Well, let's read Dwark Hess's take on economic diffusion lab lag being cope for missing capabilities.

所以有时候'Copium'会是个很棒的命名，

So sometimes Copium would be a beautiful name for

顺便说，可以用于AI芯片。

an AI chip, by the way.

确实如此。

It would.

确实如此。

It would.

你们有Tranium。

You got Tranium.

也许他们需要

Maybe they need

你们有Copium。

You got Copium.

Copium。

Copium.

有时人们会说，AI没有更广泛地部署在各公司中，并且除了编程之外尚未提供大量价值的原因是技术扩散需要很长时间。

Sometimes people will say that the reason that AIs aren't more wide widely deployed across firms and already pro providing lots of value outside of coding is that technology takes a long time to diffuse.

Dorkash认为这只是自我安慰。

Dorkash thinks this is cope.

他说人们用这种自我安慰的说辞来掩盖这些模型缺乏创造广泛经济价值所需能力的事实。

He says people are using this cope to gloss over the fact that these models just lack the capabilities necessary for broad economic value.

史蒂文·巴恩斯对此问题及其他多个观点发表了一篇精彩的文章。

Steven bar Burns has an excellent post on this and many other points.

他表示新技术需要很长时间才能融入经济体系。

He says new technologies take a long time to integrate into the economy.

那么试问，那些高技能、经验丰富且具有企业家精神的移民人类，又是如何能立即融入经济体系的呢？

Well, ask yourself, how long how do highly skilled, experienced, and entrepreneurial immigrant humans manage to integrate into the economy immediately?

当你回答完这个问题后，请注意多克什说，通用人工智能也将具备这些能力。

Once you've answered that question, note that a AGI will be able to do those things too, Dorkesh says.

如果这些模型真像是服务器上的人类，它们将以惊人的速度扩散。

If these models were actually like humans on a server, they'd diffuse incredibly quickly.

事实上，它们会比普通人类员工更容易融入和上岗。

In fact, they'd be so much easier to integrate and onboard than a normal human employee.

它们能在几分钟内读完你所有的Slack记录，并立即掌握其他AI员工拥有的全部技能。

They could read your entire Slack and drive in minutes and immediately distill all the skills that your other AI employees have.

此外，招聘市场非常像一个柠檬市场，很难事先判断谁是优秀人才，而雇佣一个不合适的人成本相当高。

Plus, the hiring market is very much like a lemons market where it's hard to tell who the good people are beforehand, and hiring someone bad is quite costly.

这种动态因素在你只需启动另一个经过验证的AGI模型实例时完全无需担心。

There's there this is a dynamic that you wouldn't have to worry about when you just wanna spin up another instance of a vetted AGI model.

基于这些原因，我预期将AI劳动力融入企业会比雇佣人类容易得多，而企业每天都在大量招人。

For these reasons, I expect it's going to be much easier to diffuse AI labor into firms than it is to hire a person, and companies hire lots of people all the time.

如果能力真达到AGI水平，人们会愿意每年花费数万亿美元购买代币。

If the capabilities were actually at AGI level, people would be willing to spend trillions of dollars a year buying tokens.

知识工作者 是啊。

Knowledge workers Yeah.

想想看。

Think about that.

我们雇佣一个人 是啊。

We we hire someone Yeah.

就像，我们雇佣一个AI 是啊。

That like, we hire an AI Yeah.

或者说我们在利用一个AI，它已经听过了TBPN的每一分钟内容。

Or or we're leveraging an AI, and they've listened to every single minute of TBPN Yeah.

每一分钟。

Ever Yeah.

并且观看了每一个片段。

And watched every clip.

是的。

Yeah.

而现在，你需要对模型进行微调之类的操作。

And right now, you'd have to fine tune that into the model or whatever.

这不是一开始就能直接获得的。

You you don't just get that out of the gate.

没错。

Yeah.

我只是想说，我们最终确实雇佣了很多这样的人。

And I'm just saying, like, the the we we do end up hiring a lot of people that Yeah.

那些是

That are

是的。

Yeah.

比如之前只是听众的人。

Like previously just listeners.

对。

Yeah.

但要找到一个了解节目中每一刻的人，确实。

But getting somebody that knows every single moment that has ever happened on the show Yeah.

没错。

Yeah.

那将会非常强大。

Would be super powerful.

但话说回来，目前还缺乏一些能力，使得智能体无法在内部创造太多价值。

But again, there's just like missing a missing capability set that doesn't allow agents to deliver a lot of value internally.

目前实验室收入与预期相差四个数量级的原因在于，模型的能力远不及人类知识工作者。

The reason that lab revenues are four orders of magnitude off right now is that models are just nowhere near as capable as human knowledge workers.

是的。

Yeah.

我同意这一点。

I I agree with that.

我有一点不太认同的是，他引用了斯蒂芬·伯恩斯的话说：'问问你自己'。

I the the one thing that I don't necessarily agree with here, he says, well, ask yourself, this quote from Stephen Burns.

那些技能高超、经验丰富且具有创业精神的移民人类是如何做到立即融入经济体系的？

How do highly skilled, experienced, entrepreneurial immigrant humans manage to integrate into the economy immediately?

他们确实某种程度上能立即融入经济体系，但移民流动其实是个缓慢的过程。

I mean, they do sort of integrate into the economy immediately, but, like, the the immigration flow is, like, a slow process.

这并非一蹴而就的事情。

Like, it doesn't just happen immediately.

移民数量不会突然从零激增到——比方说——百万人之类的规模。

It's not just, like, you know, the amount of immigration went from, like, zero to, like, I don't know, a million people or something.

就像是，四处流动。

Like, it's, like, move around.

这里存在一定的滞后效应。

There is, like, a there is a bit of a drag.

但我理解他在这里表达的观点，确实有道理。

But I understand what he's saying here, and it does make sense.

总之，让我来介绍一下public.com，为认真对待的人提供的投资平台。

Anyway, let me tell you about public.com, investing for those who take it seriously.

他们提供多元资产投资服务，深受数百万用户信赖。

They got multi asset investing, and they're trusted by millions.

Verge杂志也想掺一脚，试图用标题攻击David Sachs。

The verge is trying to get it on the action, trying to attack David Sachs with a headline.

这简直太可笑了，先是《纽约时报》盯上David Sachs，现在Verge也想来插一脚。

It's like it's like so funny that the New York Times went after David Sachs, then the verge was like, we wanna go after him too.

我们也想

We wanna get

一些仇恨。

some of hate.

等等。

Wait.

等等。

Wait.

让我们...让我们好好发挥。

Let us let us cook.

我们听到所有人

We heard everybody

在科技界都讨厌这篇文章。

in tech hates this this article.

确实认为

Do think

也讨厌。

hate too.

嗯，虽然我不赞同这种新闻报道方式，但这个标题确实挺搞笑的。

Well, I while I don't agree with this journalistic approach, it is a pretty funny headline.

是啊。

Yeah.

哦，没错。

Oh, yeah.

太滑稽了。

It's hilarious.

标题是'硅谷正在团结支持一个糟糕的家伙'。

The the the headline is Silicon Valley is rallying behind a guy who sucks.

这就像是在说，这是什么意思？

It's like, what does that mean?

比如

Like

纯粹就是。

Just pure.

纯粹就是，定性式的，纯粹在骂人。

Pure, like, qualitative, like, just name calling.

他们就是，我们不喜欢这家伙。

They're just like, we don't like this guy.

纯粹的人身攻击。

Pure ad hominin.

你

You

知道吗，尽管去吧，如果你的粉丝喜欢，如果你的观众想要这样。

know, go off if people if people if your fans like it, if that's what your your audience wants.

这就是在煽动愤怒情绪。

It's it's it's rage bait.

肯定会火起来的。

It's gonna go hard.

已经有一千个赞了。

It already got a thousand likes.

在一篇关联文章中，《The Verge》每篇链接通常得不到一千个点赞，所以这表现已经超常了。

On a linked article, the verge is not putting up a thousand likes per link, so this is outperformance.

而且，它的付费墙设置得很严。

And, it's heavily paywalled.

不订阅那玩意儿，你就没法知道David Sacks有多烂。

You cannot learn how David Sacks sucks without subscribing to that thing.

他们干得漂亮。

They did a good job.

你得付钱。

You gotta pay.

你得付钱。

You gotta pay.

你想知道他为什么这么烂。

You wanna know why he sucks.

我做到了。

I did it.

你付钱了吗？

Did you pay?

所以我不知道他为什么糟糕。

So I don't know why he sucks.

但那真的会

But That'd be really

很搞笑，如果付费墙后面写着：我们只是开玩笑的。

funny if behind the paywall is like, we're just kidding.

实际上我们认为《纽约时报》这次搞错了。

He's actually We think the New York Times missed on this one.

谁

Who

知道呢？

knows?

保罗·格雷厄姆。

Paul Graham.

是的。

Yeah.

在时间线上。

On the timeline.

他说，一家初创公司告诉我，他们的一位投资人不喜欢他们向新成立的初创公司销售产品，而希望他们向资金更雄厚的大公司销售。

He says, a startup told me that one of their investors didn't like that they were selling to newly founded startups and wanted them to sell to bigger companies who have more money.

如果投资人这么告诉你，就把他们当白痴忽略掉。

If investors tell you this, write them off as idiots.

向初创公司销售是你能做的最好的事情。

Selling to startups is the best thing you can do.

我相信今天我们讨论的许多公司未来都会向同批的其他公司销售产品。

I'm sure many of the companies we're talking with today will be selling to other companies in the batch.

很多人，很多人会说这样不好。

A lot of people a lot of people, like, say that's bad.

对。

Yeah.

他们试图说，YC就像一个循环经济。

They try to say, like, a YC is a circular economy.

但你必须忽略YC已经创建的数百家非常真实的企业，它们与世界上各类公司都有合作。

But you have to ignore the the hundreds of, very real businesses that have been created through YC and gone on to work with every kind of company in the world.

是的。

Yeah.

目前看来，初创企业似乎确实更聪明、官僚主义更少、更能代表未来趋势。

It certainly seems, at this point, startups tend to be smarter, less bureaucratic, more representative to future trends.

即使创业生态系统中存在某种封闭的循环经济，实际上也有巨大压力要求你交付真正有价值的东西，因为每一分钱都很宝贵。

Like, even if there's a, you know, some sort of insular circular economy in the startup ecosystem, like, there's a pretty immense amount of pressure to actually deliver something that's valuable because every dollar is precious.

没错。

Yeah.

这些是

These are

这些每个创始人，是的。

these every found yeah.

他们很理性。

They're they're being rational.

并不是说...是的。

It's not like Yeah.

我确信存在一些小案例，公司确实有过一些不良行为。

It's not like, I'm sure there's been small instances where companies were actually, you know, had had somewhat bad behavior.

但总的来说，就像如果我要为你的SaaS产品付费...是的。

But in general, it's like if I'm gonna pay for your the SaaS Yeah.

或者你正在运行的测试版工具，它必须足够好。

Tool or the beta that you're running, it has to be good.

是的。

Yeah.

必须能用。

Has to work.

你看到斯图尔特·布兰德了吗？

Did you see, did you see Stewart Brand?

他说，Anthropic因在其AI训练中使用了48万本Brand的书籍，被判赔偿15亿美元。

He says, so there's a $1,500,000,000 judgment against Anthropic for including 480,000 Brand.

书籍用于训练他们的人工智能。

Books in training their AIs.

其中有五本是我的书。

Five of my books are among them.

据我的经纪人马克斯·布罗克曼透露，每本书可能获得1500美元的赔偿。

Word is there might be a $1,500 payout per book according to my agent Max Brockman.

这名字不错。

That's a good name.

他说，我给我的经纪人马克斯写了下面这段话。

He said, I wrote them I wrote to my agent Max the following.

如果有任何款项支付给我，请将其退回Anthropic公司，并感谢他们将我的书籍纳入他们的AI训练。

If any payment comes to me, please send it back to Anthropic with my thanks for including my books and their AIs.

判决网站提供了退出赔偿的选项，但我发现操作起来很繁琐。

The judgment website offers a way to opt out of the payment, but I found it cumbersome.

所以我没有这么做。

So I didn't.

我有原则，但懒得做高度原则性的人。

I'm principled, but too lazy to be highly principled.

我真的很喜欢这个。

I really like this.

他是Long Now基金会的联合创始人，该基金会保持中立立场。

This is a he's the cofounder of the Long Now Foundation, which takes no sides.

在这个论坛上，作为个人，我偶尔会表明立场。

In this forum, as a private person, I do take sides occasionally.

所以我觉得这件事挺有趣的。

So I thought that was a funny thing.

二级市场上欺诈行为层出不穷。

There are secondary market fraud going on left and right.

但首先，让我介绍一下graphite.dev，这是AI时代的代码审查工具。

But first, let me tell you about graphite dot dev, code review for the age of AI.

Graphite帮助GitHub上的团队更快地交付更高质量的软件。

Graphite helps teams on GitHub ship higher quality software faster.

带我们了解一下这个。

Take us through this.

是的。

Yeah.

阅读这段内容时，Matt Graham提到二级市场充斥着欺诈和不良行为者。

Reading through this, Matt Graham says secondary markets are rife with fraud and bad actors.

看到这些底层投机者利用Andoril的增长牟利，同时通过不合理或不透明的费用结构剥削散户投资者，这让我感到痛心。

And it pains me to see these bottom feeders profiting off of Andoril's growth while fleecing retail investors through unreasonable or opaque fee structures.

本周的荒谬事件中，Ignite VC这家我们从未会面或有过任何接触的基金。

In this week's episode of nonsense, Ignite VC, a fund we've never taken a meeting with or had any contact with whatsoever.

由我们素未谋面的Brian创立，正在通过公开的Google文档向投资者募集资金，投资于一个SPV，该SPV又将投资于另一个SPV，后者可能最终与一位据称但未具名的早期Anderol员工签订远期合约。

Founded by Brian, who we've never met, is soliciting investors via public Google Docs to invest in SPV that will in turn invest in another SPV that will in turn potentially enter into a forward contract with a supposedly though unnamed early Anderol employee.

嗯。

Mhmm.

这里有几个问题。

Few problems here.

首先，所谓的远期合约在私营企业中 notoriously 难以结算，且对手方风险极其真实。

First off, so called forward contracts are notoriously hard to settle in private companies, and counterparty risk is extremely real.

那些复杂边缘情况怎么办？比如收购时股票无法交易，或是婚姻、离婚、死亡导致标的股票所有权复杂化的情况？

What about the many complicating corner cases like acquisitions where shares don't trade, or marriages, divorces, or deaths where ownership of the underlying shares is complicated?

总体来说这是个风险很高的交易结构，我认为大多数人其实并不真正理解。

Just generally a risky structure to close that I don't think most folks actually understand.

所以没错，如果你签订远期合约，本质上购买的是某些股票未来价值的权利，然后当有人结婚、离婚、去世，或是破产——这些情况下你可能根本无法收回投资，即便本该产生回报。

So yeah, if you enter into a forward contract and you basically buy the right to the future value of some shares, and then somebody gets, again, married or divorced or passes away, or bankruptcy is another situation where you might not be actually able to collect, even your investment should have generated some return.

Matt指出第二点：这份交易备忘录基本没有包含Andoril的业绩细节，没有任何营收数据，也没有产品具体信息。

Matt says, second, this deal memo includes basically no details about Andoril's performance, no revenue figures whatsoever, no product specifics.

我想这算是好事，对吧？

I guess that's good, right?

如果他们真掌握了这些信息却随意散播的话。

Like, if they were if they were just floating around information that they had acquired.

但不管怎样，继续讲，这几乎像是在怂恿投资者靠炒作和势头而非基本面来投资。

But anyways, continuing, almost as if it's soliciting investors to invest on hype and momentum and not fundamentals.

总的来说，我会建议人们对任何缺乏基本细节的交易备忘录保持怀疑态度。

Generally, I'd advise folks to be skeptical of any deal memo lacking basic details.

第三，远期合约被安德洛尔公司的股票计划和章程明确禁止，这意味着安德洛尔绝不会同意Team Ignite的SPV在我们私有期间实际持有这些股份，可能性为零。

Third, forward contracts are explicitly disallowed by Anderol's stock plan and bylaws, which means that Anderol will never consent to Team Ignite's SPV, actually taking possession of these shares while we are privately held, zero chance.

最后，备忘录大部分时间都在讨论结构和费用问题，这些内容简直疯狂。

And finally, the memo spends most of its time talking about the structure and fees, which are insane.

一个双层SPV结构，除了8%的前期费用、两年内3%的年费、20%的附带权益外，还要承担所有法律和行政成本通过，而最疯狂的部分是其隐含的每股价格简直荒谬至极。

A double layered SPV with all legal and admin costs pass through in addition to an 8% upfront fee, 3% annual fee for two years, 20% carried interest, and the craziest part, an implied price per share that is completely insane.

在这种情况下，隐含每股价格比九个月前最近一次优先股融资高出115%。

In this case, the implied PPS is a 115% higher than the most recent preferred raise from nine months ago.

虽然受宠若惊，但这也让这些投资者处于近乎荒谬的境地——他们支付的每股价格是我们最近交易价格的两倍多。

Flattered, I suppose, but also puts these investors in an almost absurd position by paying more than double the price per share of our most recent transaction.

正如开头所说，我完全不认识布莱恩或Ignite团队。

As stated at the top, I don't know Brian or Team Ignite at all.

也许他们是些善良的人，这一切都是个大误会。

Maybe they're kind of wholesome people, and this is all a big misunderstanding.

但如果我是投资者看到这个所谓的'机会'，我会逃之夭夭。

But if I were an investor looking at this, quote, opportunity, quote, I'd run for the hills.

我相信创始人回复说：感谢提醒。

And I believe the founder the founder replied and said, appreciate the heads up.

该文件是用于与现有有限合伙人讨论的内部草案，并非为公开传播而准备。

The document reference was an internal draft prepared for discussion with an existing LP and was not intended for public circulation.

似乎是有人未经授权分享了它，我们正在调查此事是如何发生的。

It appears someone shared it without authorization and we're looking into how that happened.

但你看到了吗？然后有七个人分享了他们收到的这份备忘录的邮件截图。

But do you see what And then There's like seven people that share a screenshot of like a direct email So they got with this exact memo.

好的。

Okay.

另外他们声明：不为任何与Andoril相关的工具征集投资。

And the other thing is they say not soliciting investment for any Andoril related vehicle.

马特说，真的吗？

Matt says, really?

这份草案是由你们的创始人和管理合伙人撰写的。

The draft was written by your founder and managing partner.

我亲眼看着他实时编辑这份文件。

I literally watched him edit the doc in real time.

而且他还有截图，显示谷歌文档里创始人的名字，你知道的，基本上

And he has a screenshot screenshot of like the the the founder's name in Google Docs like, you know, basically

真是一团糟。

What a mess.

什么

What

总之。

a Anyways.

好吧

Well

所以别这么干。

So don't do this.

别这么做。

Don't do it.

不如你创办一家公司并申请加入Y Combinator？

Instead, why don't you start a company and apply to Y Combinator?

去建立一个真正的企业，而不是到处兜售那些不愿出售股份的SPV和公司。

Build an actual business instead of going around hustling, SPVs and companies that don't wanna sell shares.

但我们继续前进

But we are moving on

将双层SPV融入我们的Y

Double to layer SPVs into our Y

覆盖范围。

coverage.

Harsh Tigar正在我们的重播等待室里。

We have Harsh Tigar here in the restream waiting room.

让我们把他接入电视和超级直播。

Let's bring him into the TV and Ultra.

哈什，非常感谢你在Y Combinator演示日这么忙的时候抽空来和我们交流。

Harsh, thank you so much for taking the time on a busy Y Combinator demo day to come talk to us.

你最近怎么样？

How are doing?

我挺好的。

I'm doing good.

谢谢邀请我。

Thanks for having me.

太棒了。

Fantastic.

给我们说说，今天进展如何？

Take us through, how's the day going?

日程安排是怎样的？

What is the schedule like?

我很想深入了解你观察到的一些趋势，以及那些表现突出的公司。

And then I'd love to dig into some of the trends that you're seeing, some of the standout, companies.

我相信我们会和其中很多公司进行交流。

I'm sure we're gonna be talking to a lot of them.

不过，今天活动的流程是怎样的？我们现在处于整个毕业过程的哪个阶段？

But, what's the what's the run of the show today, and where are we in the course of the process of, you know Yeah.

这些公司的毕业流程？

Graduating these companies?

我们大约两小时前就开始了，早上十点左右。

So we got we got started, like, almost a couple of hours ago, ten in the morning.

所以创始人和投资者们都聚集在一起。

And so the founders kind of investors all gathered together.

他们进入YC办公室的主会场后，创始人开始进行演示，快速介绍他们的进展、产品、背景等，每人限时一分钟。

They get into the main room here at the YC office, and then the founders start giving presentations, talking about, like, the progress, what they build, themselves, their background, the pretty quick fire presentations, one minute each.

然后在每组演示之间会有休息时间，投资者可以趁机和一些创始人交流见面，当然，希望最终能投资其中一批项目。

And then there's sort like a break in between sort of blocks of presentations where the investors can hang out and talk to some of the founders and get to meet them and, you know, obviously, hopefully invest in a bunch of them.

所以我们差不多快到午餐时间了。

So that's kind of we're, like, we're just about approaching lunch.

现在这个时段，大家已经听了一堆公司的介绍，可能对其中一些感兴趣的项目有了初步印象。

So it's kinda like that part of the day where people have, like, listened to a bunch of companies, probably got, like, a sense of some of the stuff that they're interested in.

我看到现在人们正聚在一起谈生意。

I see people right now, like, just hanging out doing deals.

所以氛围还挺欢快的。

So it's it's kinda like a fun vibe.

现场气氛很活跃。

It's, like, live.

这就是我们最棒的状态。

It's the best of us.

集体投资环节。

Party rounds.

我们超爱这种氛围。

We love it.

真希望我们能在现场。

I wish we could be there.

今年YC团队有没有分享什么关键指标或统计数据来开启演示日？

Are are there any, like, hero metrics or or stats that, the YC team shared this year to kick off demo day?

你们最近是如何展示YC的整体情况的？

How are you how are you sharing, like, the shape of YC these days?

是的。

Yeah.

我是说，我们这次没有

I mean, we didn't go

过多强调统计数据。

too stats heavy this time.

大概来说，延续了我们全年看到的趋势，就是这批公司的收入增长更快，能签下与大公司的合同。

Around, I think, I mean, at a high level, it's just the the continuation of the theme we've seen this whole year, which is just like the company during the batch are just getting faster revenue growth, assigning, like, contracts with, like, big companies.

没错。

Yeah.