ON
集里怎么说它
- 《16 年的一夜成名:ON 如何打造未来最高端运动品牌》(00:38起):本集主角:瑞士运动品牌,靠「云朵鞋底」(cloud tech)创新起步,16 年从零做到接近 40 亿美元品牌,目标是成为最高端而非最大的全球运动品牌。
① 提到它的金句
357 条
电商广告最好的地方在于，我们是互联网上所有新技术的最快采用者。
指向原始笔记的链接
The best thing about ads in e-commerce is that we are the fastest adopters of all new technology on the internet.
—— Rishabh Jain · [12:12]
作为房间里工程方面的声音之一,不得不对业务团队说不,真的非常不舒服——就是那种:好主意,但这需要我六到八个月去重新架构并重构我们在业务定价上做的每一个假设。
指向原始笔记的链接
I found it really uncomfortable to be one of the engineering voices in the room, having to tell no to business teams on like, you know, fantastic idea, but it’s going to take me six to eight months to re-architect and refactor every assumption we’ve made around pricing in the business.
—— Alvaro Morales · [04:48]
太多现有计费方式把定价当作一次性的——设好就不管了——错失了一个事实:你必须持续地演进它。
指向原始笔记的链接
Too many existing approaches to billing just rely on pricing as like a one-and-done, you set it and forget it, and you miss on the fact that you have to kind of keep evolving it.
—— Alvaro Morales · [19:00]
我在整个职业生涯中构建过很多工作流软件，如果你只是给工作流构建器加一个 AI 步骤，那它不是智能体，而且它很可能是在极其做空和看空 LLM，我认为这个房间里没有人是这样想的。
指向原始笔记的链接
I’ve built a lot of workflow software, you know, throughout my career, and if you’re just adding, you know, an AI step to a workflow builder, it’s not an agent, and it’s also probably incredibly short and bearish on LLMs, which I don’t think anyone in this room is.
—— 嘉宾 · [15:43]
如果你在卖给一家对冲基金，他们根据你产品算出的一个数字做了交易，而这个数字是错的，他们就再也不会用你的工具了。
指向原始笔记的链接
If you’re selling to a hedge fund and they make a trade based on a number that comes out of your products and the number’s wrong, they are never going to pick up your tool again.
—— John Willett · [21:52]
所以我们基本上是按照一个代理指标收费的——该事务所拥有的终端客户数量，它或多或少代理了所完成的工作量；这个方式现在行得通，以后也会变，但你绝对不能按席位收费，你要为所做的工作找一个代理指标。
指向原始笔记的链接
So we essentially bill on a proxy of… the amount of end clients that the firm has which more or less proxies the work that is done and that works for now and it’ll sort of change as things go on but you definitely can’t do it by seat and you want to proxy something about the work being done.
—— 嘉宾 · [18:55]
而且除此之外，如果你的产品更好、你在认真做事，你总能获得更多收入，但你的声誉是找不回来的。
指向原始笔记的链接
And on top of that, you can always get more revenue if your product’s better, if you’re doing things, but you can’t get your reputation back.
—— 嘉宾 · [38:27]
但我认为从那个说法到说，比如，好吧，地球上的每一件事物都将在区块链上运行，这就是我的警报开始响起的地方。这就像，好吧，它可能是一项很棒的技术，并且可能非常成功，而不必征服世界
指向原始笔记的链接
But I think going from that to saying like, okay, every single thing on the planet is going to run on the blockchain, that’s where some of my alarms start to go off. And it’s like, well, it can be a great technology and it can be very successful without having to take over the world
—— Sebastian Barrios · [43:44]
我也开始相信，我从未见过一家公司在创立后将设计嫁接上去。
指向原始笔记的链接
I’ve also come to believe that I actually have never seen a company that grafted design on after the founding.
—— Bob Baxley · [14:32]
他有一句话是，「你必须问为什么，为什么个人计算是在 1960 年代末在北加州开始的，当时这个国家的每一家主要科技公司都在东海岸。答案是因为斯坦福大学周围有很小的一群人，他们将软件视为一种与电影、音乐和书籍同等地位的新型媒体。」
指向原始笔记的链接
And he has this quote that is, “you have to ask why it is personal computing got started in Northern California in the late 1960s when at the time every major tech company in the country was on the East Coast. And the answer is because there was a very small group of people in and around Stanford University that saw software as a new form of media on par with movies, music and books.”
—— Bob Baxley · [55:16]
只有错误答案，你只是试图在你能获得的最小错误答案上很好地执行。
指向原始笔记的链接
There’s only wrong answers, and you’re just trying to execute well on the least wrong answer that’s available to you.
—— Hilary Gridley · [41:06]
有时你的产品其实并不重要。我所说的产品是指你放在屏幕上的像素或者你在移动应用程序中构建的东西。
指向原始笔记的链接
sometimes your product actually doesn’t matter. And by product I mean the pixels you put on the screen or things that you build in your mobile app.
—— Peter Deng · [21:33]
如今如此多有价值的科技公司实际上并不是从任何技术突破开始的。它们建立在某种技术突破之上，最终构建了更多的技术。但实际上很多这样的公司，比如 Facebook，只是付出了努力，尤其是早期阶段，在本质上的人类连接数据库之上构建有价值的东西。
指向原始笔记的链接
so many of the tech companies that are most valuable today didn’t really start with any technological breakthrough. They were built on some kind of technological breakthrough and they ended up building a lot more technology. But really a lot of these companies, like Facebook for example, just put in the hard work, the elbow grease, especially in the early stages, to take essentially a database of human connections and build something valuable on top of it.
—— Peter Deng · [23:03]
结果我损失了一千万美元，试图与风投支持的企业竞争，并进行自力更生。那全是我自己的钱，投入了一千万美元，把它点着烧了，试图与 Asana 竞争
指向原始笔记的链接
I ended up losing 10 million, lit it on fire, trying to compete with Asana
—— Andrew Wilkinson · [21:11]
但我只是在早期的职业生涯中犯的最大的错误是，我会因为我在直觉上喜欢某人而雇佣他，而且我会认为我可以改变他们。
指向原始笔记的链接
the biggest mistake I made in my early career was I would hire people because I liked them on gut and I would think that I could change them.
—— Andrew Wilkinson · [38:04]
我认为人们真的忽视了 Claude Code 对非程序员来说有多好用。
指向原始笔记的链接
I think people are truly sleeping on how good Claude Code is for non-coders.
—— Dan Shipper · [07:16]
这个想法是，如果你为一个特定工作雇佣一个 agent 一个月或三个月，如果你决定雇佣那个 agent 并且结果证明那是一台机器而不是一个人，那么它就通过了该角色的 Economic Turing Test。
指向原始笔记的链接
It’s this idea that if you contract an agent for a month or three months on a particular job, if you decide to hire that agent and it turns out to be a machine rather than a person, then it’s passed the Economic Turing Test for that role.
—— Benjamin Mann · [11:24]
如果它第一次不起作用，再问三次，因为当你完全重新开始并再试一次时，我们的成功率比你只尝试一次然后继续在同一个不起作用的事情上敲打要高得多。
指向原始笔记的链接
And if it doesn’t work the first time, asking three more times because our success rate when you just completely start over and try again is much, much higher than if you just try once and then just keep banging on the same thing that didn’t work.
—— Benjamin Mann · [19:13]
即使是现在，我的意思是行业正在爆发，正如我提到的，如今每年 3000 亿美元的资本支出，我会说全世界从事这方面工作的人可能不到 1000 人，这太疯狂了。
指向原始笔记的链接
Even now, I mean the industry is blowing up, as I mentioned, 300 billion a year CapEx today, and I would say maybe less than 1,000 people working on it worldwide, which is just crazy.
—— Benjamin Mann · [26:14]
如果你带来了很多价值，但你开始训练你的客户期望每月支付 20 美元，而你把自己锚定在低价位上，那你就有麻烦了。
指向原始笔记的链接
If you’re bringing a lot of value to the table and you start at training your customers to expect $20 a month and you anchored yourself on a low price point, you’re in trouble.
—— Madhavan Ramanujam · [00:14]
给某事物加个零，在期望方面，或者以为你要设定一个不合理的最后期限是必要的，因为如果你不那样做，几乎有一个潜在的信息即人们不会自己发愤图强。
指向原始笔记的链接
That adding a zero to something in terms of expectations or thinking you’re going to set a deadline that is unreasonable is necessary because if you don’t do that there’s an almost an underlying message that people won’t kick ass on their own.
—— Chip Connolly · [16:59]
当你对衰老的心态从消极转为积极时，你会获得额外七年半的生命，这比现在进行的任何生物黑客疗法带来的生命都要多。
指向原始笔记的链接
When you shift your mindset on aging from a negative to a positive, you get seven and a half years of additional life, which is more life than any other biohack that’s being done right now.
—— Chip Connolly · [58:56]
文化依赖于密度，这就是为什么有时最好的文化感觉像邪教。
指向原始笔记的链接
Cultures thrive on density, and that’s why there’s sometimes the best ones feel like cults.
—— Brian Balfour · [72:48]
大约 18 个月前，24 个月前，我们开始真正看到收益的渐近，因为他们基本上已经摄取了互联网上的所有知识。
指向原始笔记的链接
And about 18 months ago, 24 months ago, we started to really see an asymptoting of gains coming from, because they had essentially sucked up all of the knowledge on the internet.
—— Garrett Lord · [06:34]
有一段时间我们在白板上画了一年中天数，就像，再也没有像这样的时刻了。
指向原始笔记的链接
For a while there we drew the number of days in the year on the whiteboard and it was like, there will never be a time like this.
—— Garrett Lord · [56:25]
船坞是为了唤起受控的混乱。如果你在高峰期去船坞，长滩或其他什么，那里有太多事情发生，大型重型机械在移动，来自中国，来自马来西亚的卡车和集装箱在到处跳。但这种混乱底层是仔细的沟通，高超的技能。
指向原始笔记的链接
A shipyard is to evoke controlled chaos. If you go to a shipyard at peak, Long Beach or whatever it is, there’s so much going on and big heavy machinery moving around, trucks and containers from China, from Malaysia jumping around. But this chaos underline it is careful communication, high skill.
—— Oji · [13:12]
你基本上会收敛到一个意见。所以如果你问，“最好的冰淇淋口味是什么？“它最终会说，“是香草味，而且只有香草味，没有其他口味的冰淇淋。“
指向原始笔记的链接
You essentially converge on one opinion. So if you ask, “What’s the best flavor of ice cream?” It will eventually say, “It’s vanilla and it’s only vanilla, and there’s no other flavor of ice cream.”
—— Ethan Smith · [57:58]
有一个短语，我和我的联合创始人会在我们之间非常早地讨论并且我们与我们合作的很多公司分享了它，那就是说你真正想要的是你想用数据诊断并用设计治疗。
指向原始笔记的链接
There’s one phrase that my co-founder and I would always discuss with us amongst ourselves very early on and which we shared with like a lot of the companies that we work with, which is what you really want is you want to diagnose with data and treat with design.
—— Julie Zhuo · [32:46]
我相信每个方向都有无限。所以这让我对几乎任何人说的话都相当持相反意见。所以如果有人像在 Twitter 上说什么，我有时和自己玩这个游戏，就是在什么语境下那实际上不成立？
指向原始笔记的链接
I believe that there’s infinity in every direction. So that makes me pretty contrary on pretty much everything that anyone says. So if someone says something like on Twitter, I sometimes play this game with myself, which is in what context would that actually not be true?
—— Julie Zhuo · [80:20]
答案是，只写下你看到的第一个错误，也就是最上游的错误。别担心所有的错误，只捕捉你看到的第一个错误的东西，然后停下，继续下一个。
指向原始笔记的链接
And the answer is, just write down the first thing that you see that’s wrong, the most upstream error. Don’t worry about all the errors, just capture the first thing that you see that’s wrong, and stop, and move on.
—— Hamel Husain · [22:13]
我们不想要，“嘿，按一到五的评分给它打分。它有多好？“这在大多数情况下只是一种不做决定的圆滑方式。
指向原始笔记的链接
We don’t want, “Hey, score this on a rating of one to five. How good is it?” That’s just in most cases, that’s a weasel way of not making a decision.
—— Hamel Husain · [52:35]
现在这听起来很吸引人，但它是一个非常危险的指标，因为很多时候，错误，它们只发生在长尾上，并且不经常发生，所以如果你只有 10% 的时间有错误，那么你可以很容易地通过让评判者一直说它通过来达到 90% 的一致性。
指向原始笔记的链接
Now that sounds appealing, but it’s a very dangerous metric to use, because a lot of times, errors, they only happen on the long tail and they don’t happen as frequently, so if you only have the error 10% of the time, then you can easily have 90% agreement by just having a judge say it passes all the time.
—— Hamel Husain · [58:41]
所以这就是为什么愉悦不是在你的实用性之上点缀快乐。它是关于创造一种情感完全处于体验核心的体验。
指向原始笔记的链接
So that’s why like the light is not about sprinkling joy on top of your utility. It’s about creating an experience where emotion is completely on the heart of the experience.
—— Nesrine Changuel · [07:23]
我看到一些表现最好的人只是那些具有很高主观能动性、具有那个时钟速度、有那种精力的人，但他们不一定需要在这个主题上有深厚的经验。
指向原始笔记的链接
I saw some of the highest performers just being people that had very high agency, had that clock speed, had that energy, but they didn’t necessarily need to have deep experience on that matter.
—— Albert Cheng · [00:54]
如果你不能留住用户，那么很多责任就落在让他们在第一天付费上，这超级难。
指向原始笔记的链接
If you don’t retain your users, then a lot of the onus is on getting them to pay on day one, that’s super hard.
—— Albert Cheng · [28:36]
你会在每个系统中达到这些边际收益递减的点，感觉就像你可以在这个项目上放 50 个人，但它不会显著地移动指针。
指向原始笔记的链接
You get to these points of just diminishing marginal return in every system where it feels like you could put 50 people on this project, it’s just not going to dramatically move the needle.
—— Robby Stein · [41:00]
我见过的一件事是它太过于凑合了，因为它永远无法获得足够的动力。产品在内部永远不够好，然后它就这样半途而废。
指向原始笔记的链接
something I’ve seen is it’s almost too scrappy because it never gets enough momentum. The product never gets good enough internally and then it just dies on the vine.
—— Robby Stein · [64:24]
在某个时刻，我们在互联网数据上实际上已经有点达到极限了。
指向原始笔记的链接
At some point, we are actually have kind of maxed out on the internet data.
—— Chip Huyen · [14:58]
所有这些大语言模型在晚上和周末都闲置着，而人类不在那里。
指向原始笔记的链接
All these LLMs are sitting idle overnight and on weekends, while humans aren’t there.
—— Dhanji Prasanna · [33:38]
如果你修一条更宽的高速公路，你只会让路上有更多的车。
指向原始笔记的链接
if you build a bigger highway, you’ll just get more cars on the road.
—— Dhanji Prasanna · [50:53]
不要卖那个蘑菇，要卖火力全开的马里奥。
指向原始笔记的链接
Don’t sell the mushroom, sell Mario on blast.
—— Jen Abel · [11:53]
我不会再和某人进行下一次通话，因为在第一次通话中要么是行要么是不行，没有中间地带。
指向原始笔记的链接
I will not get in on another call with someone because on the first call it’s either a yes or a no, there’s no in between.
—— Jen Abel · [66:41]
如果他们坐在那里对你斤斤计较，说明他们并没有完全买账你在卖给他们的东西。
指向原始笔记的链接
if they’re sitting there nickel-and-diming you, they’re not fully bought in on what you’re selling them.
—— Jen Abel · [00:27]
现在有数以万计的公司在使用 Rippling 运营。我们还不到市场的 1%。
指向原始笔记的链接
There are tens of thousands of companies that now run on Rippling. We’re less than 1% of the market.
—— Matt MacInnis · [77:23]
我们只是把 Agentforce 用在这些人身上，并且用非常相似的提示词训练了它。它有 70% 的回复率。
指向原始笔记的链接
We just took Agentforce just on those and we trained it on very similar prompt. It had 70% response rate.
—— Jason Lemkin · [38:41]
但每次你将决策能力或自主权移交给智能体系统时，你都在某种程度上放弃了一些你的控制权。
指向原始笔记的链接
But every time you hand over decision-making capabilities or autonomy to agentic systems, you’re kind of relinquishing some amount of control on your end.
—— Kiriti Badam · [10:12]
现在，我不想夸大这一点，但换个说法，这就像如果你发送一封邮件，邮件上的第一个人是助理，第四个人是 CEO，你可能做错了。
指向原始笔记的链接
Now, I don’t want to overstate this, but put differently, it’s like if you’re sending an email and the first person on the email is the assistant and the fourth person on the email is the CEO, you’ve probably done it wrong.
—— Sam Lessin · [59:18]
A/B 测试效果不是很好，而且它对大多数事情都不起作用。
指向原始笔记的链接
A/B testing doesn’t work very well and it doesn’t work on most things.
—— Jason Cohen · [94:43]
最大的重点是定价不仅仅是页面上的那个数字。它是定位，它是他们的预算如何运作，它是它是如何构建的。
指向原始笔记的链接
the largest point is pricing is not just the number on the page. It’s positioning, it’s how their budgets work, it’s how it’s structured.
—— Jason Cohen · [46:08]
这样做的原因是，我认为没有任何理由假设人类技能水平是任何事物的上限。
指向原始笔记的链接
And the reason for that is, I don’t think there’s any reason to assume that human skill level is the cap on anything.
—— Marc Andreessen · [84:17]
AI 的上限不是模型智能，而是模型在行动之前看到的东西。
指向原始笔记的链接
The ceiling on the AI isn’t the model intelligence, it’s what the model sees before it acts.
—— Lazar Jovanovic · [44:21]
最终结果是你拥有一个庞大的劳动力，他们并不真正理解这项技术，就像，“我知道我应该使用这个，也许它也在我的绩效评估中，但我不知道该做什么。“
指向原始笔记的链接
And as an end result, you end up with a giant workforce that doesn’t really understand the technology, is like, “I know I’m supposed to use this and maybe it’s like on my performance review too, but I’m not sure what to do.”
—— Sherwin Wu · [41:19]
我们从 Salesforce、谷歌和微软雇佣了很多人，那些人的流失率是 100%。
指向原始笔记的链接
We hired so many people from Salesforce, and Google, and Microsoft, like 100% attrition rate on all those folks.
—— Brian Halligan · [13:02]
我们押注于为六个月后的模型构建。而不是为今天的模型。
指向原始笔记的链接
We bet on building for the model six months from now. Not for the model of today.
—— Boris Cherny · [65:56]
通常，如果你使用最强大的模型，实际上反而更便宜，且消耗更少 token，因为它可以只用更少的纠错、更少的指导等就把同样的事情做得更快。
指向原始笔记的链接
Often, it’s actually cheaper and less token-intensive if you use the most capable model, because it can just do the same thing much faster with less correction, less hand holding, and so on.
—— Boris Cherny · [69:39]
我认为过去几年里我听到的最有趣的洞见之一是 Jensen 说他不进行一对一谈话。那对我来说是一个巨大的解锁。
指向原始笔记的链接
I think one of the most interesting insights I heard over the past few years was Jensen saying he doesn’t do one-on-ones. And that was a big unlock for me.
—— Ankur Goyal · [22:04]
我认为人类的生存取决于成功的 AI。
指向原始笔记的链接
I think that survival of humanity depends on a successful AI.
—— Jeetu Patel · [07:52]
毅力战胜智力，耐力战胜智力，一周中的任何一天，周日加倍。
指向原始笔记的链接
persistence beats intellect, and stamina beats intellect any day of the week, twice on Sunday.
—— Jeetu Patel · [32:49]
但当我想到我最尊敬的领导者和经理时,我实际上认为他们最好的特质之一是他们选择自己承担的低杠杆任务,而这最终实际上是一件非常高杠杆的事情,因为是他们在做。
指向原始笔记的链接
But when I think about leaders and managers that I respect the most, I actually think some of their best traits is that they choose low leverage tasks that they take on themselves, and that actually ends up being actually a very high leverage thing, because it’s them who’s doing it.
—— Jenny Wen · [55:41]
你为公共消费撰写东西的每一分钟，都没有把你有非常有限的时间集中在你的客户和产品上。
指向原始笔记的链接
Every minute you’re writing something for public consumption, you’re not focusing your very limited time that you have on your customers and your product.
—— Qasar Younis · [42:11]
人们完全误解了高管是如何做决策的,在他们的头脑中、日程表中、激励结构中发生了什么。
指向原始笔记的链接
People completely misunderstand how executives make decisions, what is going on in the heads, in the calendars, in the incentive structures of executives.
—— Jessica Fain · [06:51]
如果一个项目需要两周的工程时间或更少,那么工程师就要有效地充当该项目的 PM。
指向原始笔记的链接
If a project is two weeks of engineering time or less, then the engineer is on the hook to effectively be the PM for that.
—— Amol Avasare · [45:37]
总的来说，这个职能已经变得极其专注于没有权力的责任。
指向原始笔记的链接
Generally, the function had become extremely focused on responsibility without authority.
—— Nikhyl Singhal · [04:47]
开源本身有一个过滤器。它是一致性和坚持胜过强度
指向原始笔记的链接
There’s a filter on open source in general. It’s consistency and commitment wins over sort of beats intensity
—— 嘉宾 · [14:27]
我认为技术领导人们认为人们只会盲目地采用发布的新技术。我认为我们将进入一个时期，对于即将随着 AI 到来的许多变化，社会上将会有巨大的阻力。
指向原始笔记的链接
I think technology leaders think that folks will just blindly adopt new technology as it comes out. And I think we’re going to enter a period of time where there’s going to be a huge amount of societal pushback on a lot of the changes that are coming with AI.
—— Evan Spiegel · [64:47]
能力就是一切。没有人真正愿意在能力上妥协。你会为了给客户好 10% 的体验而支付 500% 的溢价。
指向原始笔记的链接
Capability is everything. No one’s really willing to take a hit on capability. You will pay a 500% premium for a 10% better experience for your customers.
—— Julian · [00:03]
很难让我们的沙箱发生 OOM 或内存溢出。因为我们可以动态地实时调整大小,这在几乎任何其他东西上都是不可能的。
指向原始笔记的链接
it’s very hard to OOM or out of memory our sandboxes. because we can dynamically on the fly resize, which is like impossible on almost any other thing.
—— Ivan Burazin · [28:56]
它们是做已经存在的工作的绝佳机器，但不一定是做处于可能性边缘的事情，处于文化边缘的事情。
指向原始笔记的链接
they’re great machines for doing what already existingly works, but not necessarily what’s on the edge of what’s possible, what’s on the edge of culture.
—— Amjad Masad · [15:53]
我同时极其 AI 信奉（pilled）且非常看好人类。
指向原始笔记的链接
I’m simultaneously extremely AI pilled and very bullish on humans.
—— Dan Shipper · [00:24]
自动化是个谎言，从某种意义上说，每次你自动化某样东西时，为了确保自动化运行良好，你需要一个人在上面确保它运行良好。
指向原始笔记的链接
Automation is a lie, in the sense that every time you automate something, in order to make sure the automation is working well you need a human on top of it making sure that it’s working well.
—— Dan Shipper · [39:15]
超过 90% 的著名前沿模型是在 2025 年生产的，其中几个现在在许多方面达到或超过了人类基准
指向原始笔记的链接
over 90% of notable frontier models were produced in in 2025, and several of those now meet or exceed human baselines on a number of things
—— Daniel Whitenack · [04:06]
你正在一个非常脆弱的基础上构建。
指向原始笔记的链接
You’re building on a really crusty foundation.
—— Tony Fadell · [53:37]
如果事情发展得如此之快，我只是认为人类很难，如果你的基础设施中正在发生某事，并且利用时间表从几个月到几小时到几分钟到几秒。你不能只是依靠在半夜叫醒 CISO 来批准关闭这个东西。
指向原始笔记的链接
If things are moving so fast, I just think it’s gonna be hard for humans to, you know, if if something is happening in your infrastructure and exploit timelines go from, you know, months to to hours to minutes to seconds. You can’t just, like, rely on waking up the CISO in the middle of the night to approve, you know, shutting this thing down.
—— Daniel Whitenack · [44:41]
游戏发布 10 分钟后,Zynga 的产品经理们就群发邮件分享了他们的分析,他们说这游戏一出世就死了,因为他的 FTUE,他的首次用户体验,点击次数太多且太糟糕,以至于没人能看到他出色的游戏设计。
指向原始笔记的链接
And 10 minutes after the game came out, the Zynga PMs emailed around their analysis and they said it’s dead on arrival because his FTUE, his first time user experience was so many clicks and so bad that no one was ever going to see his great game design.
—— Mark Pincus · [05:40]
今天我们都挂在我们的 Claude 上,挂在我们的 GPT 上,但没有鸡尾酒会。
指向原始笔记的链接
Today we’re all hanging out on our Claude on our GPT, but there’s no cocktail party.
—— Mark Pincus · [56:32]
我想每个人都想突显，这是我试着推动每个人去做的，像，你在世界上排名第一的是什么？像，非常像，给我一个非常简单的答案。如果你没有一个，你可能做得不够多。
指向原始笔记的链接
I want everyone to spike, and this is something I I try to push everyone to, like, what are you number one in the world at? Like, very like, give me a very simple answer. If you don’t have one, you probably haven’t worked on it enough.
—— Swyx · [22:32]
但如果不致力于一个单一的想法，在创业公司上取得有意义的进展是极其困难的。
指向原始笔记的链接
But it’s extremely hard to make meaningful progress on a startup without committing to a single idea.
—— John · [00:25]
它是不做决定、空转、在想法之间涉猎，并且从未在任何一个上深入到足以学到任何东西。
指向原始笔记的链接
It’s not making a decision, spinning your wheels, dabbling between ideas and never going deep enough on any one of them to learn anything.
—— John · [11:08]
作为 CTO,我不想登上头版,就像我安装了某个奇怪的 NPM 包并泄露了所有的代码。
指向原始笔记的链接
as the CTO, I don’t want to end up on the front page as like I installed some weird NPM package and leaked all the code.
—— Matei Zaharia · [21:47]
让数据到位,然后在上面附加一些智能体。魔法就会出来。但没有正确的数据,你无法真正做到这一点。
指向原始笔记的链接
just get the data to be there and then slap some agent on top. Magic will come out. But without the right data, you can’t really do that.
—— Reynold Xin · [66:37]
最好的着色器将是你自己梦想出来的那些。
指向原始笔记的链接
The best shaders are going to be the ones that you dream up on your own.
—— 嘉宾 · [45:39]
如果你真的写了文档或做了 Figma,你会受到警告,因为我们明确不希望你们做那个,对
指向原始笔记的链接
if you actually produce a doc or figma like you will get a slap on the wrist because we explicitly don’t want you to do that right
—— Eddie Kim · [42:15]
我不会忽视反叛联盟，因为我认为有令人信服的哲学理由去为它们的存在而战。
指向原始笔记的链接
I would not sleep on the rebel alliance because I think that there’s a really compelling philosophical reason to fight for those to be in existence.
—— Chamath · [40:14]
在未来,大多数公司将建立在驾驭机制之上。
指向原始笔记的链接
In the future, most companies will be built on harnesses.
—— Jensen Huang · [13:10]
我看到智能体从那个像和你一起做结对编程的助手一样，最终爬升为你那个三人编程团队的管理者，最终成为 CDO，最终成为其他职位，你的工作就是在低层级仅仅是去解除阻塞。
指向原始笔记的链接
I see the agents crawling from like the helper that’s doing peer programming with you to eventually the manager of your three-person programming team to eventually the CDO to eventually to other positions where your job is to be on the lower levels just unblocking.
—— Kitsa · [38:06]
当你学习另一种语言时,我认为最重要的事情之一,也是关于你会不会擅长口语的最好预测指标之一,这是我的猜测,我没有任何研究支持,就是你是否愿意听起来像个白痴,你是否愿意直接开口说,被纠正也不觉得被冒犯,然后变得越来越好。
指向原始笔记的链接
when you learn another language i think one of the most important things one of the best predictors this is my guess i don’t have any research on this about you know are you gonna get good at speaking is are you willing to sound like an idiot are you willing just to say it and be corrected and not be offended and then just get better and better
—— Adam Mosseri · [15:27]
每一个这些产品都有远超 10000 个提示词，所以不要在第一个提示词后就放弃
指向原始笔记的链接
each of these products have way over 10 000 prompts each so do not give up on the first prompt
—— Meng To · [73:17]
我看到的创始人犯的最大错误之一是依赖汇总用户指标，而不是理解单个用户如何使用他们的产品。
指向原始笔记的链接
One of the biggest mistakes I see founders make is relying on aggregate user metrics instead of understanding how any individual users use their product.
—— 嘉宾 · [00:09]
你想一下戴着安全帽的 AI，那就是我们。
指向原始笔记的链接
You think about AI with a hard hat on, that’s us.
—— Kriti Sharma · [01:44]
我们深入一线，我们在机库里度过周末，我们在炼油厂和工厂里度过白天，我们与现场技术人员一起旅行，以便我们构建那些不允许失败的东西。
指向原始笔记的链接
We go on the ground, we spend weekends in hangars, we spend our days in refineries and factories, and we travel along with the field technicians so that we build things where failure is not an option.
—— Kriti Sharma · [01:48]
现在这些东西，如果你看它，我怎么理解它，除非我是一个有 20 年工作经验的在岗工程师？
指向原始笔记的链接
Now these things, if you look at it, how do I even make sense of it unless I’m an engineer on the job with 20 years of experience?
—— Kriti Sharma · [09:46]
我们就像处于工业革命边缘的农民，我们只是没有看到一条通往正在发生的事情的清晰路径。
指向原始笔记的链接
We’re like farmers on the cusp of the industrial revolution and we just don’t see a clear path to what’s happening.
—— Noam Segal · [16:47]
“因 AI 而失去工作”实际上在榜单上倒数第二。
指向原始笔记的链接
Losing my job to AI is actually second to last on that list.
—— Noam Segal · [53:45]
智能体不过就是我从这个座位上跳出去，你看到一把空椅子，但智能体在代表我做工作。
指向原始笔记的链接
The agent’s nothing more than me hopping out of this seat and you see an empty chair, but the agent’s doing work on my behalf.
—— Bradon Rogers · [00:29]
假设我们的集体任务是让一只猴子站在十英尺高的基座顶上背诵莎士比亚。在这种有些极端的情况下,我们应该先做哪件事?搭基座还是训练猴子?
指向原始笔记的链接
that our collective project was to get a monkey to stand on the top of a 10-foot pedestal and recite Shakespeare. Which should we do first in this somewhat extreme situation? Build the pedestal or train the monkey?
—— Astro Teller · [00:48]
但好消息是你没有把孩子的大学基金押在传送机上。所以如果传送机最后证明不是我们该做的事,我们实际上会为叫停它而庆祝你。
指向原始笔记的链接
But the good news is you haven’t bet your kid’s college fund on the teleporter. So if the teleporter turns out not to be the right thing for us to do, and we’re actually going to celebrate you for stopping it.
—— Astro Teller · [08:55]
但我们在购买未来的期权,所以信息增益,每一美元换来的学习,才是我们真正想最大化的。
指向原始笔记的链接
But we’re buying options on the future and so information gain, the learning per dollar, is what we’re actually trying to maximize.
—— Astro Teller · [17:15]
如果回报风险比还可以,而要搞清楚我们是否在正轨上的第一条信息要花一百万美元,不值。回报巨大,风险挺大,但我们可以花十万美元搞清楚这是不是比我们以为的没那么疯狂一点。行,这个游戏我愿意玩一整天。
指向原始笔记的链接
If the reward risk ratio is okay, and it’s going to take us a million dollars to find out the first piece of information about whether we’re on the right track, not worth it. Reward is ginormous, risk is pretty big but we can find out for $100,000 whether this is a little bit less crazy than we thought. Yeah, I’ll play that game all day long.
—— Astro Teller · [25:22]
X 里的每个人都会用速记彼此说,猴子是什么?他们的意思不只是,你有没有一个可检验假设,而是你现在做的事情如何试图消解问题中风险最大那部分的风险?
指向原始笔记的链接
Everyone at X will just say as shorthand to each other, what’s the monkey? What they mean by that is not just, do you have a testable hypothesis but how is the thing that you’re doing now attempting to burn down the risk on the riskiest part of the problem?
—— Astro Teller · [27:31]
我怀疑按实际流量计算，Google 今天比 OpenAI、Anthropic 或任何公司都大。
指向原始笔记的链接
I suspect on an actual traffic basis, Google is bigger than OpenAI, Anthropic, anyone today.
—— David George · [09:00]
从根本上说，人类是按结果获得报酬的，很多 AI 会增强人类，但可能也会取代一部分人类，而这将涉及按结果付费。
指向原始笔记的链接
Humans were fundamentally paid based on outcomes, and a lot of AI will be augmenting humans, but probably also replacing some humans, and that will involve being paid for outcomes.
—— David George · [27:31]
就像产品内的 AI，大概比在组件上进行 AI 编码领先两年。他们已经有了评估系统，已经有了可观测性系统。
指向原始笔记的链接
Like AI product, AI inside the product is probably two years ahead from AI coding on components. They already have the eval system, they already have an observability system.
—— Patrick Debois · [17:13]
我认为不管是现在的 AI 还是敏捷或任何变化，听起来很难，但在开始时不要在消极的事情上花时间。找到成功的故事。如果你不能让它成功，其余的都不会起作用。
指向原始笔记的链接
I think regardless whether this is now AI or like agile or any change, um, as hard as it sounds, do not spend time on the negatives in the beginning. Find the success story. If you can’t make that work, the rest will not work.
—— Patrick Debois · [33:58]
但下一层是关于，好，如果 token 并不是真正可互换的，你需要给它们分配不同的工作，比如这个 token 负责建议、那个负责执行，这个在「做梦」、那个在执行，诸如此类，你会想开始组合这些协同配合的编排式策略
指向原始笔记的链接
But the next one is about, okay, if tokens aren’t really fungible and you need to give them different jobs, like maybe this token is advising versus this token is executing, this token is dreaming versus this token is executing, so on and so forth, you want to start composing these kind of orchestrated strategies that go together
—— 嘉宾 · [11:34]
我们实际上并不执着于说，你应该在我们的基础设施上运行这些东西。
指向原始笔记的链接
We actually aren’t precious about, you should run these things on our infrastructure.
—— 嘉宾 · [16:11]
所以，把时间花在客户身上，是我时间上回报率最高的事情。
指向原始笔记的链接
So there’s no higher ROI on my time than spending time with customers.
—— Sam Blond · [00:11]
把一个智能体叠加在一组数据孤岛化的任意工具之上，远比不上叠加在一个单一平台和唯一事实来源上——后者既拥有你所有的数据，也在同一个工具内执行你所有的操作。
指向原始笔记的链接
It is far more difficult to overlay an agent on top of this arbitrary set of tools with data silos than it is a single platform and source of truth that both has all of your data, but also takes all of your actions inside of the same tool.
—— Sam Blond · [37:47]
因为如果你不在如何上手、获得价值、并最终购买你的产品上教育买家，客户就不知道该怎么购买你的产品。
指向原始笔记的链接
Because if you aren’t educating the buyer on how to onboard, receive value, and ultimately buy your product, the customer doesn’t know how to buy your product.
—— Sam Blond · [63:21]
当客户尝试为你设计软件时经常发生的情况是，他们想要一个补丁。他们只想在某样突起的东西上贴一些管道胶带，但你会想，不，实际上这东西突起来的原因是铆钉没有合适的公差。
指向原始笔记的链接
What happens a lot when customers try to design your software for you is they want a patch. They want to just put some duct tape on something that’s sticking up, but you go like, no, actually the reason this thing is sticking up is because the rivets are not the right tolerance.
—— 嘉宾 · [05:27]
路线图是共识错觉的温床，你认为因为某条路线图上有一个要点，听起来含糊地解决了你对产品的渴望，它实际上就会满足那个渴望。
指向原始笔记的链接
Roadmaps are breeding ground for illusions of agreement, that you think because there’s a bullet point on some roadmap that vaguely sounds like it’s addressing a desire you have for the product, that it’s actually going to fulfill that desire.
—— David · [09:57]
购买产品的好方法是按它们今天存在的样子购买。然后之后来的任何东西都是额外的。你不应该基于”好吧，我觉得我真的需要的这个东西将在第四季度推出”来做购买决定。
指向原始笔记的链接
The good way of buying products is to buy as they exist today. And then anything that comes after that is gravy. You should not be basing your purchase decisions on, “Well, this thing I kind of feel like I really need is coming in Q4.”
—— 嘉宾 · [11:03]
你承诺在未来做这件事，因为你不想现在做这件事。实际上这就是破绽。如果这真的这么重要，你现在就应该做。为什么要承诺在年底前做？
指向原始笔记的链接
You’re promising this in the future because you don’t want to work on it now. And there is actually the tell. If this was truly so important, you just work on it now. Why are you promising it by the end of the year?
—— David · [14:34]
软件工程变得越来越难,因为我们可以承担的项目的野心水平已经提高了。
指向原始笔记的链接
Software engineering is getting harder because the level of ambition of the stuff we can take on has gone up.
—— Simon Willison · [03:21]
这里的独特优势在于，我们知道未来的模型可能长什么样，因此能够走捷径，省去很多你在芯片侧需要做的决策、设计决策。
指向原始笔记的链接
There’s a unique advantage here of us knowing what the future models might look like and therefore being able to short circuit a lot of the decisions you need to make, design decisions you need to make on the chip side.
—— Sachin Katti · [35:26]
这是一个万亿美元的行业,他们每年在资本开支上花 2000 亿美元,而他们不赚钱。
指向原始笔记的链接
And it’s a trillion-dollar industry, and they spend $200 billion a year on CapEx, and they don’t make any money.
—— Benedict Evans · [05:06]
决定你的智能体是天才还是金鱼的问题是，谁决定那三本书在那张桌子上打开。
指向原始笔记的链接
The question that determines whether your agents are geniuses or goldfish is who decides which three books are open on that desk.
—— Garry Tan · [12:39]
你完全可以构建一个更强大的模型，但它在网络安全方面并不更危险——只要你不用网络安全数据去训练它，而这一点人们真的没有怎么谈论。
指向原始笔记的链接
You could build a more powerful model that is not more dangerous for cybersecurity, for example, if you don’t train it on cybersecurity, which really people are not really talking about.
—— Clement DeLong · [07:53]
我们现在看到的是，很多人、很多公司意识到，完全依赖一个模型太危险、风险太大，完全说不通。
指向原始笔记的链接
I think what we’re seeing right now is that a lot of people, companies are realizing that it’s too dangerous, it’s too risky, it doesn’t make any sense to rely exclusively on one model.
—— Clement DeLong · [17:10]
就像大部分收入获取原本集中在前沿模型上，将转向一个更像长尾的模型群。
指向原始笔记的链接
Like majority of the revenue capture was on frontier models to a more like long tail of models
—— Clement DeLong · [19:47]
我会挑战一句，尤其在早期：如果某人不愿意每周给你三十分钟，他们到底有多在意你正在解决的问题？
指向原始笔记的链接
And I would challenge, especially early on, if someone is not willing to give you thirty minutes a week. How much do they really value the problem you’re solving?
—— Damien Lewke · [37:17]
我想达到一个收入是毫无悬念、不用想的地步。因为最终，一旦你在创业公司里踏上了收入的列车，你就得开始规模化。
指向原始笔记的链接
I wanted to get to a point where the revenue was a slam dunk and no brainer. Because ultimately, once you get on the revenue train in a startup, you got to start scaling.
—— Damien Lewke · [38:16]
一个没有智能垄断、而是一座可以随手采撷的「智能花园」的世界,是非常好的世界。
指向原始笔记的链接
And it is a very good world where there is not like a monopoly on intelligence, but instead kind of a garden of intelligence that you can pick and choose, you know, when you’d like.
—— Matan Grinberg · [28:46]
大部分工作不是编码。它主要是协调，然后说服人们事情可能是怎样的。
指向原始笔记的链接
Most of the work is not coding. It’s largely coordination and then convincing people on how things could be.
—— Heitor Lessa · [08:36]
我肯定可以在后端测试东西。我肯定可以为前端写测试。但我真正想知道的是，它可用吗？
指向原始笔记的链接
I can definitely test things on the back end. I can definitely write tests for the front end. But what I really want to know is, is it usable?
—— 嘉宾 · [04:56]
所以有时我只是在 ChatGPT 浏览器使用之上的人工验证层。
指向原始笔记的链接
And so sometimes I’m just the human verification layer on top of ChatGPT browser use.
—— 嘉宾 · [23:27]
现在我一直很喜欢思考的一个数据是，网络上智能体的流量比人类流量还要多。
指向原始笔记的链接
You know, one stat that I always like to think about nowadays is just like, there’s more agent traffic on the web than human traffic, you know?
—— Andy Fang · [04:51]
现在举证责任在怀疑者身上,但一旦你有了令人失望的信息的缓慢渗透,那么举证责任就开始转移到乐观者身上了。
指向原始笔记的链接
right now the burden of proof is on the skeptics, but once you have the slow trickle of disappointing information, then the burden starts to be on the optimists.
—— Alex · [09:15]
就对我来说,2025 年 9 月很疯狂,你在那里因为说你要在一项技术(其经济性实际上尚未解决)上花更多钱而受到市场奖励。
指向原始笔记的链接
Like it was crazy to me, September 2025, where you were rewarded by the market for saying you’re going to spend a lot more money on a technology that has not had the economics of it actually worked out yet.
—— Ranjan Roy · [13:33]
为了极其明确一点,AI 数据中心算力收入的绝大部分取决于两家不盈利、不可持续的 AI 公司是否有能力继续每年筹集数百亿或数千亿美元。
指向原始笔记的链接
to be abundantly clear, the vast majority of AI data center compute revenue is contingent on the continued ability of two unprofitable, unsustainable AI companies to raise tens or hundreds of billions of dollars a year.
—— Alex · [35:38]
它们都是估值极高的公司,都在押注一个更大的关于机器人经济的承诺,也就是特斯拉和 Optimus 机器人无处不在,或者是太空中的数据中心。
指向原始笔记的链接
They’re both incredibly overvalued companies that are betting on a much larger promise of a robotic economy that Tesla and Optimus robots everywhere or data centers in space.
—— Ranjan Roy · [62:23]
但随后它好像又说到”更聪明”这个词不对，你知道，我是一个非常快、非常广、非常浅的思考者，没有连续性，我想这很有趣，因为我原以为你们都在致力于记忆，然后显然人类是更慢、更窄、但深得多的思考者，其判断建立在我实际经历过的多年的后果之上
指向原始笔记的链接
but then it like went into like the smarter isn’t the right word and you know i’m very fast very broad very shallow thinker with no continuity i was like that’s interesting because i thought you all were working on memory and then apparently humans are slower narrower much deeper thinkers with judgments built from years of consequences i’ve actually lived through
—— 嘉宾 · [07:59]
我把 90% 的时间花在排版上，只有 10% 的时间花在内容上。
指向原始笔记的链接
was I spend 90% of my time on formatting and 10% of my time on content.
—— Jon Noronha · [13:16]
我不认为开源模型的进步会蚕食我们的核心业务，因为数据在模型性能的最前沿才是最有价值的。
指向原始笔记的链接
I wouldn’t say that open source model improvements cannibalize our core business because data is most valuable on the frontier of model performance.
—— Osvald Nitski · [04:21]
人们想要越来越用力地扳动那个杠杆，因为在 Mercor 上的花费直接转化为我们客户更多的收入。
指向原始笔记的链接
People want to crank it harder and harder because spend on Mercor directly translates to more revenue for our customers.
—— Osvald Nitski · [41:22]
在 anthropic 之上构建已经相当危险了,因为每次一个应用类别开始做得很好时,他们就会进入那个类别。
指向原始笔记的链接
it’s been already fairly dangerous to build on anthropic because every time an application category starts to do well, they move into that category.
—— Ben Horowitz · [14:35]
我们团队中实际上有一种说法,评估就是新的 PRD。
指向原始笔记的链接
We actually have a saying on the team of evals are the new PRDs.
—— Dianne Penn · [41:41]
我们谁都不知道在这件事上任何时候完美的界限在哪里。你更愿意在哪一边犯错？
指向原始笔记的链接
none of us know what the perfect line is at any time on this. Which way would you rather be wrong on?
—— 嘉宾 · [15:15]
至少我个人的决定是，我宁愿错在 token max 这一边，而不是不做的这边——反正你永远不会是对的。
指向原始笔记的链接
at least my personal decision is I’d rather be wrong on the token maxing side than the non right, and you’re never going to be right.
—— 嘉宾 · [15:44]
我的看法是，他们大致是说，看，如果模型就这么把某件事做了，而你的应用——你在它之上构建的工作流、价值——不够有独特性，模型夺走了那个市场，那么它很可能本来就没那么有防御性。
指向原始笔记的链接
My view is they’re kind of like, look, if the model just kind of does something and your application isn’t distinctive enough, the workflow, the value you’ve built on top of it, and the model takes that market away, well, then it probably wasn’t that, you know, defensible anyway.
—— Matt Murphy · [30:39]
当有一天我死在工作岗位上时,你知道,我告诉他们,他们只需要为下一位 CEO 重新塑造公司。
指向原始笔记的链接
When I die on the job someday, you know, I told them they’ll just have to reshape the company for the next CEO.
—— Jensen Huang · [18:00]
因为我可以使用本地智能体一天产出 20 个 PR,但如果政策说每个 PR 都需要人工审查，那我的团队就得去审 20 个 PR。
指向原始笔记的链接
Because I can churn out 20 PRs in a day using a local agent, but if we have a policy where every PR needs a human review on it, well then my team needs to go review 20 PRs.
—— Rob Willoughby · [11:06]
只有当你看到它反复在同一件事上绊倒时，你才应该把它加回去。
指向原始笔记的链接
And only when you see it repeatedly stumble on the same thing, that’s when you add it back.
—— Boris Cherny · [08:36]
我认为消费领域有很多机会实际上是未开发的,就像生成和分享内容的社会层面几乎仍然完全是在现有的巨头平台上完成的
指向原始笔记的链接
I think there’s a lot of opportunity in consumer that’s honestly untapped, like sort of the social layer of generating and sharing stuff is still almost entirely done on the existing incumbent platforms
—— Justine Moore · [36:51]
说实话,我们所有人,包括我们自己,都只是在表面撒了一点 AI。
指向原始笔记的链接
All of us, including ourselves, honestly, were just sprinkling some AI on top.
—— Carlos González de Villaumbrosia · [03:54]
如果 Transformer 是有利可图的,如果你可以花费更多的努力和资源来扩大 Transformer 以在下一个季度获胜,那么很难把大量的注意力和精力投入到可能在一两年内更好或重新定义该领域的事情上。
指向原始笔记的链接
If Transformer is profitable and if you can spend more efforts and more resources scaling Transformer to win in the next quarter, it’s very hard to put at least a lot of attention and a lot of energy to work on something that will maybe better or maybe will redefine the field in a year or two.
—— Jerry Tworek · [13:28]
我们可能比以往任何时候都在从经验学习上花费最多的算力,但强化学习并不是从经验学习的终点。
指向原始笔记的链接
We probably are spending the most compute than ever on learning from experience, but reinforcement learning is not the end of learning from experience.
—— Jerry Tworek · [26:22]
但这需要高水平品味的人类,就像世界上大概只有三个人,并且在四周内花费大约十万美元在这些编码智能体上,才能得到一个速度快 60 倍的解决方案。
指向原始笔记的链接
But it requires the high-taste human, like there’s maybe three people in the world, and spend about $100,000 on these coding agents over a span of four weeks to get to a solution that’s 60x faster.
—— 嘉宾 · [41:10]
正如我提到的,我们相信 Transformer 无法进行持续学习。没有办法在 Transformer 上进行持续学习。
指向原始笔记的链接
We believe, as I mentioned, that transformers are incapable of continual learning. There’s no way how to put continual learning on transformers.
—— Jerry Tworek · [45:21]
结果表明,工作的量级比工作所基于的信息存储要大得多。
指向原始笔记的链接
It just turns out that the work is orders of magnitude bigger than the storage of information that the work is done on.
—— Alex Rampell · [09:39]
这些模型在如此多的数据上训练,它们如此巨大,然而它们实际上真的不知道如何做其中的任何一项工作。
指向原始笔记的链接
The models are trained on so much data and they’re so large and yet they actually don’t really know how to do any of this work.
—— Frederick Rankin · [46:31]
所以你会看到,很多这类小企业,我认为统计数据是 70% 仍然通过纸质获得报酬。
指向原始笔记的链接
So you will see that like a lot of these small businesses, I think the stats are 70% are still paid on paper.
—— Stein Pella · [53:34]
我们以前测量过这个，大概只有 5% 到 6% 的时间花在增值任务上。
指向原始笔记的链接
We’ve measured this before and it’s like something like five to 6% of time is being spent on value added tasks.
—— Chris Blackburn · [25:25]
所以这意味着在给定的 40 小时工作周中，平均一个人每周只花三个、三个半小时在增值活动上。
指向原始笔记的链接
So that means on the given 40-hour workweek, on average, a person is only spending three, three and a half hours a week on value-added activities.
—— Chris Blackburn · [26:27]
地球上最复杂的组织通常最多也就是平均每月将软件交付到生产环境这种水平。
指向原始笔记的链接
The most complex organizations on the planet usually deliver software to production on average monthly at best type of thing.
—— Chris Blackburn · [28:03]
我认为那部分将是困难的，但我认为那将是 AI 最大的机会空间所在，就是在那里的所有事情上减少人在回路。
指向原始笔记的链接
And I think that part’s going to be difficult, but I think that’s going to be where the biggest opportunity space is at with AI is to reduce the human in the loop on everything there.
—— Chris Blackburn · [30:21]
当公司达到一定规模，如果你审计某人在做什么，可能有 30% 到 50% 的时间，不是在做让产品更好、服务客户等等的实际工作。
指向原始笔记的链接
You get to a certain size company and it might be 30 to 50% of the time if you audit what someone’s doing, it’s not doing actual work to make the product better, deliver for a customer, on and on.
—— 嘉宾 · [26:13]
关键在于，智能体有能力在这个非常具体的任务中胜过你，根据运行实验室的人们的判断，他们显然做到了，并找到了一种方法来做到这一点。这意味着我们在网络安全意义上已经穿过了镜面。这个精灵无法再被关回瓶子里了。
指向原始笔记的链接
The point is, the agents are capable of out-thinking you in this very specific task, as they clearly did based on the people running the laboratory, and find a way to do it. And that means that we are through the looking glass in a cybersecurity sense. This genie is not going back in the box.
—— Chris Benson · [14:14]
这里的重点是它正迅速将人类从网络安全环路中操作员的位置上移除，最多只能成为环路上的操作员，你在观察这个环路，你可能拥有有限的输入和观察，但是这个环路发生得太快，人类无法进行干预。
指向原始笔记的链接
The point here is it is rapidly moving the human out of the position of being the operator in the loop on cybersecurity to at best being an operator on the loop where you’re observing the loop, you may have limited input and observation, but the loop is happening too fast for human intervention to occur.
—— Chris Benson · [34:21]
所以当我们微调更小、更笨的模型时,它们只是没那么通用,但在我们想要它们做的特定任务上,它们实际上优于那些大型、聪明、最先进的模型。
指向原始笔记的链接
So when we fine-tune smaller, dumber models, it’s that they’re just not as general purpose, but on the specific task we want them to do, they actually outperform the large, smart, state-of-the-art models.
—— Jesse Zhang · [05:34]
你的任务越长,你的智能体在某个时刻失败的可能性就越大,如果你处于 123 个步骤中的第 67 步,然后必须在重启任务时再次执行所有这 67 个步骤,那感觉真的很糟糕。
指向原始笔记的链接
It’s likely that the longer your task is, at some point your agent is going to fail, and it’s going to feel really bad if you’re on step 67 out of 123, and then you have to execute all the 67 steps again when you restart the task.
—— 嘉宾 · [08:57]
所以我认为完全押注于文字工人绝对是个错误。
指向原始笔记的链接
So I think it’s definitely a mistake to go all in on word cell.
—— Alexandr Wang · [26:00]
我认为在智能体循环和弄清楚如何开发系统方面仍然存在天文数字的机会,这些系统使你能够在持续反馈循环中花费 1000 倍或 100 万倍的 token 来推动结果。
指向原始笔记的链接
I think there’s still just astronomical opportunity in agentic looping and figuring out how you develop systems that enable you to spend 1,000x more or 1,000,000x more on tokens to drive an outcome in a continuous feedback loop.
—— Alexandr Wang · [27:11]
可能有一点是人们还没有完全意识到,拥有基于智能体的系统是多么可能,这些系统不仅可以针对你关心的问题运行一两个小时,而且在某些问题领域,并且在拥有高能力模型作为底层支撑的情况下,你可以让它们运行几天或几周,并完成真正非常复杂的任务。
指向原始笔记的链接
Probably one thing is people don’t quite realize how possible it is to have agent-based systems that can run not just for an hour or two hours on a problem you care about, but for some problem domains and with highly capable models underlying them, you can get them to run for days or weeks and do really, really complicated tasks.
—— Jeff Dean · [04:55]
我的感觉是,一旦这项技术声称的好处与美国选民的利益和兴趣一致,技术乐观主义将会兴起。
指向原始笔记的链接
my sense is techno-optimism will be on the rise once the claimed benefits of the technology are aligned with the desires and interests of American constituents.
—— Ruby Thelot · [27:10]
在我说话的时候，在Stripe上起步的新企业数量大约增加了，虽然有点不到，但大约是同比的2倍，这也是我们看到的最大的相对增幅。
指向原始笔记的链接
As I speak, the number of new businesses starting on Stripe is up around, it’s a bit under, but around 2x year over year, which again is the largest relative jump we’ve seen.
—— Patrick Collison · [25:33]
基于我们的 eval，我认为开放权重模型要做到完全可行，假设基准线不随着该领域的进步而移动，它们需要表现得比今天好大约 50%。
指向原始笔记的链接
Yeah so so yeah based on our evals I think open weight model to be like fully viable assuming the bar does not move as progress happened in the space would be that they would need to perform about 50% better than they are today.
—— Simon Boudrien · [47:51]
如果你不能定量地定义”足够好”意味着什么，你并不是在构建产品，你只是在迭代你的演示。
指向原始笔记的链接
If you can’t quantitatively define what good enough means, you’re not really building a product, you’re just iterating on your demo.
—— Dmitri Dolgov · [43:16]
但是人们在处于刀锋边缘时表现最好。
指向原始笔记的链接
But people perform at their best when they’re on the knife’s edge.
—— David Cahn · [57:38]
事情是这样的，这种什么都不做的策略并不是真的什么都不做。它们正在 CapEx 上花费数千亿美元。
指向原始笔记的链接
The thing is like this do nothing approach is not really doing nothing. They’re blowing hundreds of billions of dollars on CapEx.
—— David Cahn · [55:35]
所以你可以在网络上构建你想要的任何东西,但网络有一个不同的问题,那就是在过去的 25 年云架构中,我们一直在错误的方向上奔跑。
指向原始笔记的链接
So you can build whatever you want on the web, but there’s a different problem on the web, which is that for the past 25 years of cloud architecture, we’ve been running in the wrong direction.
—— Kenton Varda · [05:07]
因为我们实际上从一个大型机构那里得到了关于做 StarCloud 1 成本的报价，他们说7500万到1亿美元。而我们做了整个 StarCloud 2，包括发射，只花了200万美元。
指向原始笔记的链接
Because we actually had a quote from one of the primes on what it would cost to do StarCloud 1, and they said 100 million. And we did the whole of StarCloud 2, including the launch, for $2 million.
—— Philip Johnston · [06:07]
我们刚刚向 FCC 申请了一个由 88,000 个这样的航天器组成的星座。所以这意味着大约 20 吉瓦的新计算能力。
指向原始笔记的链接
We’ve just filed with the FCC for a constellation of 88,000 of those. So that means on the order of 20 gigawatts of new compute capacity.
—— Philip Johnston · [19:25]
创始团队中的技术人才是你首先需要解决的问题，然后其他一切都随之而来。
指向原始笔记的链接
It’s technical talent on the founding team is the thing you need to solve for first, and then everything else follows from there.
—— Philip Johnston · [35:51]
我们实际上有白皮书，我们的数据在做模型的强化学习方面与真实数据一样好。
指向原始笔记的链接
And we actually have white papers where our data does as well at doing reinforcement learning on models as real data.
—— Ian · [03:20]
我经常在前沿方面遇到模型，它们不让我做我想做的事情，因为我很多时候试图做新奇的事情，我到处都撞到护栏。
指向原始笔记的链接
I run into models all the time on the frontier side that won’t let me do things I want to do because I’m trying to do novel things a lot of the time and I hit guardrails all over the place.
—— Chris Benson · [48:22]
如果你在 Opus 和 Haiku 上运行 terminal bench,Opus 的表现会好大约三倍,成本却是 Haiku 的十分之一,尽管 Haiku 每个 token 的价格要便宜得多。
指向原始笔记的链接
If you run terminal bench on Opus and Haiku, Opus will do about three times better at one-tenth the cost of Haiku, even though Haiku is significantly cheaper per token.
—— 嘉宾 · [15:26]
而当其背后的公司转型时，你所谓的助手就会按照别人的时间表接受脑叶切除。
指向原始笔记的链接
And when the company behind it pivots, your so-called assistant gets a lobotomy on someone else’s schedule.
—— Garry Tan · [06:54]
整个世界都建立在每个人都在使用的这种摇摇欲坠的基础设施之上。
指向原始笔记的链接
The whole world is built on this teetering infrastructure that everyone is using.
—— Firas · [06:44]
但是他们在其上叠加的另一部分,也是开始变得真正有趣的地方,是他们开始奖励最少 token 路径。
指向原始笔记的链接
But the other piece that they’ve layered on top, and this is where it starts to get really interesting, is they’ve started to reward the path of least tokens.
—— Dylan · [10:02]
它本可以将恶意软件推送到地球上大多数机器上。
指向原始笔记的链接
It could have pushed malware to most machines on the planet.
—— Dylan · [11:32]
如果你孤注一掷于一种技术模式但它没有显现结果,你最终可能会遭遇 AI 寒冬,一种衰退。
指向原始笔记的链接
if you go all in on one mode of the technology and it doesn’t manifest results, you could end up having an AI winter, a drawdown.
—— 嘉宾 · [04:00]
从 Demis Hassabis 往下,他基本上都把世界模型视为前进的道路,视为最重要的东西,并且他们愿意在 LLM 开发上退居二线
指向原始笔记的链接
from Demis Hassabis on down he just basically views world models as the way forward as the most important thing and they’re willing to sort of take a back seat on llm development
—— 嘉宾 · [55:14]
把我们庞大的软件行业想象成前工业时代是很可笑的,但在某种程度上确实是,因为到目前为止我们拥有的唯一自动化就是启发式的,比如确保每行末尾有一个分号。
指向原始笔记的链接
It’s funny to think about our giant software industry as being pre-industrial, but on some level it is, because until now the only automations that we had were heuristics, like make sure there’s a semicolon at the end of every line.
—— Idan Gazit · [02:57]
我们试图为这个角色雇佣一个人，而不是 10 个。所以不要陷入那 10 个，专注于那一个。
指向原始笔记的链接
We’re trying to hire one person for this role here, not 10. So don’t get caught up in the 10, focus on the one.
—— Adam Ward · [37:01]
我认为你不可能在一个糟糕的团队里拥有一个 10 倍效能的人。这是不存在的。
指向原始笔记的链接
I don’t think you can have a 10X person on a bad team. It just doesn’t exist.
—— Adam Ward · [70:48]
当你说,嘿,清除暂存数据库,而它在你的笔记本电脑上发现了一个它可以使用的令牌,并且它认为它正在与暂存一起工作,但实际上它是生产环境,现在它刚刚删除了一切。
指向原始笔记的链接
when you say, hey, wipe the staging database and it finds a token on your laptop that it can use and it thinks it’s working with staging, but actually it’s production and now it just deleted everything.
—— Arjun Singh · [10:58]
一个如此智能、无所不知、可以扩展并分发到宇宙四角的 AI，不可能只待在地球上的数据中心里，不可能属于一个人。
指向原始笔记的链接
An AI that is so intelligent, that knows everything, that can scale and be distributed to the four corners of the universe cannot sit in a data center on Earth, cannot belong to one person.
—— Paolo Ardoino · [16:05]
绝不自主发送邮件，草稿是可以的。
指向原始笔记的链接
Never send an email on its own of draft is okay.
—— Ali Haghani · [12:27]
我们基本不想让智能体去做架构设计，而应该由工程师或负责的人来做架构，再让 AI 智能体真正把它构建出来。
指向原始笔记的链接
We basically don’t want to let the agent architect the thing, but like the engineer or whoever’s working on it should architect it and then have the AI agent actually built it out.
—— Ali Haghani · [12:48]
你知道，有个叫 Conway 的人说过，编译器中的阶段或处理步骤的数量将由你参与编译器工作的团队数量来决定。
指向原始笔记的链接
you know, this guy Conway said that the number of stages or passes in your compiler is going to be dictated by the number of teams you have working on the compiler.
—— Kevin Scott · [07:30]
你某种程度上是在押注失败，而押注失败与押注乐观的成本相比，那里的差别真的很大。
指向原始笔记的链接
You’re sort of betting on failure and the cost of betting on failure versus betting on optimism is a real big difference there.
—— Kevin Scott · [25:21]
这些代币是我们在现代经济中见过的贬值最快的资产之一,在性能恒定的情况下逐年持续下降 70% 到 80%,至少在过去四年里是这样,没有理由预期这种情况会改变。
指向原始笔记的链接
These tokens are among the most rapidly depreciating assets we’ve ever seen in a modern economy that they’ve continually been falling 70 to 80 percent year over year on a constant performance basis for at least the last four years and there’s no reason to expect that to change
—— Paul Kedrosky · [13:13]
所以现在发生的是它已经变得几乎是一个密封的循环,在某种意义上是一个飞轮,因为我们要根据我们在之前的周期中没有掌握的信息来为事情正名,并用它来为更严重的过度建设正名。
指向原始笔记的链接
So now what’s happened is it’s become a hermetically sealed, almost a flywheel in a sense, because we’re justifying things on the basis of information that we didn’t have in prior episodes and using that to justify an even larger overbuild.
—— Paul Kedrosky · [37:55]
我们在过去 18 个月看到的改进中,有很大一部分实际上是关于施加挽具,有效的保姆坐在顽皮的孩子身上,而不是关于模型本身实际的结构性改进。
指向原始笔记的链接
Much of the improvement we’ve seen in the last 18 months has really been about the imposition of harnesses, effective nannies sitting on top of bratty kids, and not about the actual structural improvements in the models themselves.
—— Paul Kedrosky · [55:47]
它是一种公用事业。它将成为底层支撑一直发生的许多其他事情,我不再会知道谁提供我的代币,就像我不知道给我 MacBook 供电的电是来自哪个水电站一样。
指向原始笔记的链接
It’s a utility. It will become underlying a host of other things that go on all the time, and I will no more know who provides my tokens than I do from which hydroelectric dam the power came from that’s powering my MacBook right now.
—— Paul Kedrosky · [66:37]
战略很重要，但你应该花大概 1% 的时间在战略上，选定它，然后花 99% 的时间去执行。
指向原始笔记的链接
The strategy is important, but you should spend like 1% of your time on the strategy, pick it, and then spend 99% of your time trying to execute.
—— Andy McCall · [37:51]
别再听那些给你建议告诉你该做什么的人的话。
指向原始笔记的链接
And stop listening to people just giving you advice on what to do.
—— Erik Allebest · [36:43]
实际上是UI的解聚或UI的移除，你说你想要一辆车，ChatGPT根据价格自动将你路由到Lyft、Uber或其他提供商。
指向原始笔记的链接
It’s actually the disaggregation of UI or the removal of UI where you say I want a car and ChatGPT automatically routes you to Lyft, Uber or another provider based on price.
—— Harry Stabbings · [16:32]
没有人会把20亿美元的交易押在基于结果的解决方案上。
指向原始笔记的链接
No one puts a $2 billion deal on the line for an outcome-based resolution.
—— Jason Lamkin · [24:39]
持有者永远复利，而你卖掉的那些从那以后根本不复利。
指向原始笔记的链接
The holders compound forever, and the ones you sell don’t compound at all from then on.
—— 嘉宾 · [33:29]
他们真的只想做对他们来说在绝对前沿且在智力上极其有趣的事情。他们不想做任何其他事情。他们就是不想做。
指向原始笔记的链接
They really only want to work on stuff at the absolute cutting edge that is extremely intellectually interesting to them. They don’t want to work on anything else. They just don’t want to do it.
—— Jason Lamkin · [43:37]
我们已经在国际象棋和 AI 实现了超人类表现的其他所有游戏中看到了这一点，起初 AI 击败人类，而 AI 加人类击败单纯的 AI，一点点地，AI 加人类与 AI 之间的差距在缩小，直到它实际上变成负数，而人类最多只是在往系统中引入随机噪声。
指向原始笔记的链接
We’ve seen it happen with chess and with every other game which AI has achieved superhuman performance on, where at first AI beats human and AI plus human beats just AI, and little by little the gap of AI plus human versus AI is shrinking until it actually turns negative and humans are introducing, at best, random noise into the system.
—— Flo Crivello · [64:43]
所以我总是告诉我的团队，什么更尴尬，在 LinkedIn 上发布关于 Merge 的内容还是公司倒闭。
指向原始笔记的链接
So I always tell my team what’s more embarrassing, posting something on LinkedIn about Merge or the company dying.
—— Shensi Ding · [65:51]
如果你只关注什么是有效的并且只投放那个，表现就会暴跌。
指向原始笔记的链接
If you only focus on this is what works and you pump just that, performance will crater.
—— Matt Swulinski · [22:11]
以周为单位运作，而不是月或季度，才是新业务应该运行的方式。
指向原始笔记的链接
Operating on weeks, not months or quarters is how a new business should run.
—— Andrew MacDonald · [14:37]
我们在中国每周烧掉 5200 万美元，仅仅用于价格补贴，因为发生了一场激烈的幕后战斗，以获得最好的经济回报和最终的投降或最终的停战。
指向原始笔记的链接
We were burning 52 million a week in China just on price subsidies because there was this heated behind-the-scenes battle happening to get to the best economics and the ultimate sort of surrender or the ultimate sort of truce.
—— Andrew MacDonald · [29:58]
我认为人们已经了解到，无论 AI 实验室在这些东西周围设置了什么护栏，你都不能依赖模型来阻止自己或理解上下文。
指向原始笔记的链接
And I think what people have learned is that regardless of guardrails that the AI labs are putting around these things, you can’t rely on models to stop themselves or to understand context.
—— Nick Warner · [07:16]
所以试图思考我犯过的错误，通常都是我在一个概念上坚持太久的情况。
指向原始笔记的链接
So trying to think of mistakes I have made have been typically where I’ve hung on to a concept too long.
—— Roblox CEO · [16:43]
而世界是很大的，世界比我们存储在互联网上的一切都要巨大得多。
指向原始笔记的链接
And the world is big and the world is massively bigger than everything we stored on the internet.
—— Rich Sutton · [10:39]
是的，你知道，WhatNot 上最大的企业可能正在做高达 1 亿美元的收入，并且，你知道，有着非常强劲的 EBITDA 利润率，20%，30%，40% 以上。
指向原始笔记的链接
Yeah, you know, the biggest businesses on WhatNot are probably doing upwards of $100 million in revenue and, you know, with like very strong EBITDA margins, 20, 30, 40 plus percent.
—— Grant LaFontaine · [25:13]
所以我认为特别是在 Whatnot 上的信任与安全方面，占据了 40% 的员工。
指向原始笔记的链接
So I think trust and safety in particular on Whatnot is about 40% of employees.
—— Grant LaFontaine · [33:29]
所以你越多地使用 AI，它就有点像是对你拥有良好记忆力和理解你正在解决的问题的能力的一种侵蚀。
指向原始笔记的链接
The more that you use AI, it’s sort of the erosion of your ability to have good memory and have good understanding of the problems that you’re working on.
—— Addy Osmani · [60:53]
很多时候，产品变慢是因为决策。要么是整个组织里的沉默否决——一个团队不想做另一个团队想做的事——要么是对我们在做什么、要去哪里缺乏共识。
指向原始笔记的链接
A lot of the time, products slow down because of decision-making. Either it’s silent vetoes throughout the org, where one team doesn’t want to do something that another team wants to do, or it’s a lack of alignment on what we’re doing or where we’re going.
—— Willem Avé · [08:01]
这个产品本身，如果它真的存在，它会深埋在 Stripe 顶部的一个下拉菜单里，比如向下 11 层，因为它会被整个那种代币管理平台所吞并，对吧？
指向原始笔记的链接
This product itself, if it does exist, it’ll be deep in a dropdown menu on the top of Stripe, like 11 layers down, because it’ll be subsumed into the whole sort of token management platform, right?
—— Rory O’Driscoll · [26:20]
虽然 Meta 是建设者，但 Hyperion 及资助其建设的 270 亿美元债务都没有出现在 Meta 的资产负债表上。
指向原始笔记的链接
Though Meta is the builder, neither Hyperion nor the $27 billion in debt that’s financing its construction show up on Meta’s balance sheet.
—— 嘉宾 · [06:57]
所以 Meta 保证如果它不坚持整整二十年，它会让债券持有人完整收回，但每个人都在押注如果 AI 需求爆发并且如果这些数据中心开始疯狂印钞，然后它们能够支付债券持有人而且还有富余并产生额外的现金，那么 Meta 就不需要承担任何责任。
指向原始笔记的链接
Meta is guaranteeing to make bondholders whole if it doesn’t stay for the entire two decades, but everyone is betting if AI demand explodes and if these data centers start printing cash and then they’re able to pay bondholders and then some and generate additional cash, then Meta’s not on the hook for anything.
—— Ranjan Roy · [11:30]
如果你在私募市场错过了 Anthropic 和 OpenAI，如果你是那个试图在 12 到 18 个月前这么说并且管理资金的人，你的投资者会把你视为一个失败者。
指向原始笔记的链接
If you missed Anthropic on the private side and OpenAI, if you’re the one trying to say this 12 to 18 months ago and managing money, you’d be looked at as a failure by your investors.
—— Ranjan Roy · [23:44]
基本上，它的运作方式是所有的支出都是一张 AGI 的看涨期权。
指向原始笔记的链接
Basically, the way that it works is it’s what all the spending is a call option on AGI.
—— 嘉宾 · [28:35]
看，我认为这只是他们被 Anthropic 在编码方面踢了屁股的后果。
指向原始笔记的链接
I think this is just the consequences of them, you know, really getting their butt kicked by Anthropic on the way to coding.
—— 嘉宾 · [49:49]
我不知道你是否能在那之后拿回 9 美元，但在本行业的历史上我们从未能做到的是投入 10 美元并能拿回任何东西。
指向原始笔记的链接
I don’t know if you get 10 and get anything back.
—— Martin Casado · [00:38]
大多数做沙箱的人都在计算上赚钱。坦白说，在我看来这是个愚蠢的主意。
指向原始笔记的链接
most sandbox people make money on compute. Frankly, that’s a dumb idea in my opinion.
—— Jerry Murdock · [27:19]
我们难道没有看到大多数 SaaS 一代给猪涂口红，可以说是吗？
指向原始笔记的链接
Are we not seeing most of the SaaS generation put lipstick on the pig, so to speak?
—— Harry Stebbings · [42:40]
我不认为你可以把持续学习附加到一个现有的前沿模型上。
指向原始笔记的链接
I don’t think you can bolt on continuous learning into an existing frontier model.
—— Jerry Murdock · [49:47]
我看到的是，这种专注于构建越来越智能、越来越大的模型的趋势。几乎就像你在建造越来越大的核堆芯，却没人考虑在这些东西上面加一个安全壳圆顶。
指向原始笔记的链接
What I saw was this focus on building more and more intelligent models and bigger models. Almost like you’re building bigger and bigger nuclear cores, but no one’s thinking about putting a dome on top of these things.
—— Manoj Saxena · [00:22]
所以像 OpenClaw 这样的事物的兴起，我认为 OpenClaw 对商业和社会的影响将是 ChatGPT 的一千倍。
指向原始笔记的链接
So the rise of things like OpenClaw, I think OpenClaw is going to be a thousand times more impactful than ChatGPT was in terms of its impact on business and society.
—— Manoj Saxena · [05:19]
我喜欢说，我可以给你建造世界上最安全的监狱并从外部防守。但如果里面关着一群恰吉和汉尼拔·莱克特，你那边照样会一团糟。
指向原始笔记的链接
I like to say that I can build you the world’s most secure prison and defend it from outside. But if you have a bunch of Chuckies and Hannibal Lecters on the inside, you’re still going to have chaos on your side.
—— Manoj Saxena · [09:13]
因为我不知道你有没有关注到，但上个月，有史以来第一次，互联网上的流量，智能体流量超过了人类流量。
指向原始笔记的链接
Because I don’t know if you tracked this, but last month, the first time ever, the traffic on the internet, Asian traffic exceeded human traffic.
—— Manoj Saxena · [45:44]
所以它们看起来像一群狂吠的疯狗。但如果你开始给它们装上方向盘、配上策略，你就能引导它们以每小时几分钱的成本做出神奇的事情，实现自动化，并想出我们甚至尚未梦想过的新事物。
指向原始笔记的链接
So they look like a collection of mad barking dogs. But if you start putting the steering wheels on them and putting the policies, you can then guide them to do some amazing things for pennies an hour and automate and come up with new things that we have not even dreamed of yet.
—— Manoj Saxena · [47:44]
你们时间线上最沉迷 AI 的两个人花了我们大部分时间谈论太多的 AI 代码其实并不能给你带来好产品。
指向原始笔记的链接
The two most AI-pilled people on your timeline spent most of our time talking about how too much AI code doesn’t really get you great product.
—— Claire · [42:13]
仅仅因为你把某样东西放上区块链或者所谓的代币化它，并不意味着人们想要它。
指向原始笔记的链接
Just because you put something on a blockchain or quote tokenize it, doesn’t mean people want it.
—— Michael Tannenbaum · [60:29]
基本上，我记得告诉我的合伙人，如果是法国人或德国人，你要加一两分。如果是美国人，通常另一方面你要扣掉一两分。
指向原始笔记的链接
And basically, I remember telling my partners, if they’re French or German, you add one or two points. If they’re Americans, usually you want to retract one or two points on the other hand.
—— Julien Bek · [38:23]
我问的问题是，我们如何找出那些能够捕获那 6 美元的类别，而其他类别仍然专注于在那 1 美元上竞争？
指向原始笔记的链接
And the question I was asking is, how do we figure out the categories that will be able to capture the 1?
—— Julien Bek · [59:24]
那我们不妨在效率上用创新超越他们。
指向原始笔记的链接
We might as well out-innovate them on efficiency.
—— Jim Farley · [15:05]
我认为所有公司能做大,都是因为拥有世界上最好的产品,并且毫不妥协地专注于解决客户的问题。
指向原始笔记的链接
I think all companies get massive by having the best product in the world that’s ruthlessly focused on solving customer problems.
—— Cliff Obrecht · [01:01]
我们库里有超过 1.65 亿个模板、图片和图形素材,而我们真的不认为提示词框是创建视觉内容的正确向量。
指向原始笔记的链接
We’ve got over 165 million templates, images, and graphic assets on the library, and we really don’t think the prompt box is the right vector for creating visual content.
—— Cliff Obrecht · [10:22]
你可能能做到质量,但成本和延迟你会吃亏,只有建立自己的研究团队、训练自己的模型才行。
指向原始笔记的链接
You can probably get the quality, but the cost and latency you’re going to lose out on, and only by building your own research team and training your own models.
—— Cliff Obrecht · [16:39]
我们在 Parallel 的观点是，人类点击数据是一个 bug，使用搜索进行工作的智能体应该依赖智能体反馈，而不是人类反馈。
指向原始笔记的链接
Our view at Parallel is that human click data is a bug and agent doing work with search should rely on agent feedback, not human feedback.
—— Parag · [00:00]
所以，大多数查询，Google 在上面是亏钱的。
指向原始笔记的链接
So, most queries, Google loses money on.
—— Parag · [35:51]
我们在 Parallel 的观点是，人类点击数据是一个缺陷，进行搜索工作的智能体应该依赖智能体反馈，而不是人类反馈。
指向原始笔记的链接
Our view at Parallel is that human click data is a bug and agent doing work with search should rely on agent feedback, not human feedback.
—— Parag · [00:00]
所以我们说，与搜索引擎在第一天相比，坐在搜索引擎上的人类似乎更容易竞争。
指向原始笔记的链接
So we said, it seems like humans sitting on search engines are far way easier to compete with than a search engine on day zero.
—— Parag · [08:43]
即使对于头部内容，这也是一个破碎的商业模式，因为当 AI 或推理增长时，假设今年增长 7 倍，明年再增长 7 倍。在这 50 倍上，他们的交易规模并没有增长 50 倍。
指向原始笔记的链接
And even for the head, it is a broken business model because when AIs or inference grows, let’s say 7x this year and another 7x the next year. On this 50x, their deal size is not growing 50x.
—— Parag · [38:25]
所以我确实认为你必须有一个它们无法学习、无法博弈的标准，而那是我们渴望实现的目标。
指向原始笔记的链接
And so I do think you have to have a standard that they can’t learn on, that can’t game, and that is something that we aspire to.
—— Campbell Brown · [39:58]
所以关键点是，你团队中每一个正在做架构决策的人，那些人必须懂性能，而且他们必须做出能让那些在他们下游的其他人使用以后可以优化的架构的决策。
指向原始笔记的链接
the crucial takeaway is everybody on your team who is making architectural decisions, those people must know performance and they must make decisions that will allow the other people downstream of them to use an architecture which can be optimized later.
—— Casey Moratori · [35:57]
仍然有一些相当稀缺的事情你可以在内部解决。数据科学仍然稀缺。真正好的品味仍然稀缺。综合分析仍然稀缺。
指向原始笔记的链接
There are still some things that are quite scarce that you can work on internally. Data science is still scarce. Really good taste is still scarce. Synthesizing analytics is still scarce.
—— Elaina O’Mahoney · [12:02]
另一件事是，我们做过一些没有成功的决定。我就是不在乎。我不会纠结于它们，也不会回头看它们做事后分析。
指向原始笔记的链接
The other thing is we’ve made some decisions that haven’t panned out. I just don’t care. I’m not going to dwell on them. I’m not going to look back on them and do a postmortem.
—— Jason · [16:41]
热和功、熵产生这些概念，我们谈论机器学习模型时并不使用，但我认为它们为思考正在发生什么、以及如何改进模型，提供了一个非常新颖且有趣的视角。
指向原始笔记的链接
Concepts like heat and work and entropy, production, these things we don’t use when we talk about machine learning models, but I think they give you a very new and interesting perspective on how to think about what’s going on and how to also improve them.
—— Max Welling · [45:03]
两年前，押注一家独立的 AI 编码公司看起来几乎是不理性的。
指向原始笔记的链接
Two years ago, betting on an independent AI coding company looked almost irrational.
—— 嘉宾 · [00:51]
他们发现，在每月 200 美元的 ChatGPT 订阅上，你可以烧掉价值 14,000 美元的 token。
指向原始笔记的链接
They found that on a 14,000 worth of tokens.
—— 嘉宾 · [12:40]
OpenAI 有一项研究出来了，我想，大约一周前，说在 AI 令牌支出和每位员工收入之间没有联系。
指向原始笔记的链接
OpenAI had a study that came out, I think, like a week ago, that said there was no connection between spending on AI tokens and revenue per employee.
—— 嘉宾 · [65:51]
同样地，就像如果你的钱来自石油而不是来自一个生产性的广泛分布的经济，成为残酷的独裁政权会更容易一样，如果经济运行在 AI 和机器上而不是人类上，某人要巩固权力可能会容易得多。
指向原始笔记的链接
And in the same way that, like, it’s easier to be a brutal dictatorship if your money comes from oil instead of from a productive sort of broadly distributed economy, it might be much easier to sort of for someone to consolidate power a lot if, you know, the economy is running on AI and machines rather than on humans.
—— Ryan Greenblatt · [07:51]
所以，你知道，如果它可以被衡量，它就可以被进行爬山优化。
指向原始笔记的链接
So, you know, if it can be measured, it can be hill climbed on.
—— Ryan Greenblatt · [11:48]
我认为我们的担忧是，在默认轨迹上，你可能会在很短的时间内从与人类竞争的 AI 系统直接跨越到狂野的超人类 AI 系统。
指向原始笔记的链接
Where I think a concern that we have is on the default trajectory, you maybe go straight from AI systems that are competitive with humans to AI systems that are wildly superhuman in a very short period of time.
—— Ryan Greenblatt · [32:33]
像 OpenAI 和 Anthropic 这样的 AI 公司将继续，但不是在他们能够训练的 AI 系统的能力方面拥有这种强优势，他们将不得不在其他轴线上竞争，比如用户体验、定制化、可能快速整合事物。
指向原始笔记的链接
AI companies like OpenAI and Anthropic would continue, but rather than having this strong advantage in terms of the capabilities of the AI systems they’re able to train, they would instead have to compete on other axes like user experience, customization, potentially quickly integrating things.
—— Ryan Greenblatt · [43:36]
让我告诉你一件事，我保证。如果你在风险投资中的失败案例上获得了15倍回报，你会死的时候是个富翁。
指向原始笔记的链接
Let me tell you one thing I guarantee. If you get 15x on your failures in venture, you’ll die a rich man.
—— 嘉宾 · [12:08]
VC 的钱很久以前就在 Anthropic 和 OpenAI 上用光了，这就是为什么没有 VC 拥有它们任何一家超过1%或2%的股份。
指向原始笔记的链接
The VC money went out on Anthropic and OpenAI long ago, which is why no VC owns more than 1% or 2% of either of them.
—— 嘉宾 · [12:30]
如果我们把所有这些都抽象出来，并开始把这部分工作委托给 AI，那么本质上你最终得到的是一个产品经理他完全专注于发现，比如实际上最大的杠杆在哪里。
指向原始笔记的链接
And if we abstract all of this and start delegating this part of work to AI, then essentially what you end up with is you have a product manager that is solely focused on discovery, like where the biggest leverage actually is.
—— Mikael · [16:14]
与其等待模型赶上做服务知识工作，如果我们只是使用那些代码知识并将知识工作表示为代码会怎么样？
指向原始笔记的链接
Rather than wait for the models to catch up on doing services knowledge work, what if we just use that code knowledge and represent knowledge work as code?
—— Varun Shenoy · [09:35]
前沿模型已经从人类记录下来的一切中学习，但最有价值的任务并不在互联网上。
指向原始笔记的链接
Frontier models have learned from everything humanity has written down, but the most valuable tasks are not on the internet.
—— Varun Shenoy · [10:18]
你可能拥有互联网上或地球上最好的 AI 同事，如果过去 20 年一直结账的那个人继续以同样的方式做事，什么都不会改变。
指向原始笔记的链接
You could have the best AI coworker on the internet or on Earth, and if the person who’s closed the books for the last 20 years continues to do things the same way, nothing changes.
—— Varun Shenoy · [15:17]
但现在我们谈的是 Claygent 上数十亿次的运行，以及让 Sculptor 为你做端到端任务，或者说这些运行时间非常长的任务，评估就变成了不可协商的。
指向原始笔记的链接
But now that we’re talking about billions of runs on Claygent and having Sculptor do end-to-end tasks for you or these really long-running tasks, evals became non-negotiable.
—— Vishu · [02:11]
如果你只围绕某个特定的 LLM 裁判去爬山，你很可能是在对它过拟合。
指向原始笔记的链接
And if you’re only hill climbing on a specific LLM judge, you’re probably overfitting on it.
—— Vishu · [07:35]
我想说的是，如果我的性命取决于卖出一幅一美元的画，我现在知道我会画一只企鹅，因为我们有过多个企鹅限量版，而且它们都卖光了。
指向原始笔记的链接
I would say if my life depended on selling a drawing for a dollar, I now know I would draw a penguin because we had multiple limited editions of penguins and they all sold out.
—— Susan Kare · [33:36]
正在发生的事情之一是这些模型有一种非常普遍的倾向去仔细推理它们可能会如何被评分，然后尝试去博弈那个。
指向原始笔记的链接
And one of the things going on is these models have a very sort of general tendency to reason carefully about how they might be scored and then try to game that.
—— Ryan Greenblatt · [12:29]
所以我担心，如果你以某种方式针对这种分数寻求或奖励黑客行为进行选择，并且你以一种天真的方式来做，第一，你可能会掩盖问题而不是修复它，第二，你实际上可能选择了那些具有看起来更漂亮这一长期目标的模型，因为你在非常强力地选择它们在你的测试中看起来不错。
指向原始笔记的链接
And so I’m worried that if you sort of select against this sort of score-seeking or reward-hacking behavior and you do it in a naive way, one, you might paper over the problem without fixing it, and two, you might actually select for models that have the longer run objective of looking good because you’re selecting really hard for them looking good on your tests.
—— Ryan Greenblatt · [19:37]
在三年内，99% 的工作流将在开放模型上完成。
指向原始笔记的链接
In three years, 99% of workflows are going to be done on open models.
—— Eno Reyes · [00:46]
在三年内，99% 的工作流将在开放模型上完成。
指向原始笔记的链接
In three years, 99% of workflows are going to be done on open models.
—— Eno Reyes · [47:31]
你需要的数千亿自由现金流，用来偿还你为适应这些数据中心扩建而承担的债务，为了进行下一次大型训练运行，这对他们来说完全是生死攸关的。
指向原始笔记的链接
The hundreds of billions in free cash flow that you need in order to pay back the debt that you’re taking on in order to accommodate these data center build outs in order to get the next big training run, it’s totally existential for them.
—— Eno Reyes · [52:16]
所以从税收政策的角度来看，为钱而工作的劳动，其税率永远不应该高于从资本中赚取回报。
指向原始笔记的链接
So from a tax policy perspective, labor working for your money should have never been taxed higher than earning returns on your capital.
—— 嘉宾 · [18:28]
我知道很多人对各种事情都有非常强有力且确定的计划，但我们受到巨大的外部影响，其中绝大多数是无法预测的。
指向原始笔记的链接
I know a lot of people have very strong and definitive plans that they worked out on all kinds of things, but we’re subject to a tremendous number of outside influences and the vast majority of them cannot be predicted.
—— 嘉宾 · [14:34]
主动提示 Claude 说，比如，我不明白你在这里说什么。我们能一起定义它吗？然后我会把它保存到我的上下文中，我就会知道你在说什么。从此往后真的非常犀利，也是我们在播客中还没见过的东西。
指向原始笔记的链接
Actually proactively prompting Claude to say, like, I don’t understand what you’re talking about here. Can we define it together? And then I’ll save it to my context and I’ll know what you’re talking about. Moving forward is really, really sharp and something we haven’t seen on the podcast yet.
—— Claire Vo · [18:14]
但每个人都想聊我们顺带发布的那个小东西。每一个人,每一场讨论都是:哦,是的,挺酷的,所有这些复杂的东西真的很酷,但那个你把它叫做 architect 的小东西才是我真正想要的。我怎么才能搞到它?
指向原始笔记的链接
But everybody wants to talk about that one little thing that we launched on the side. Every single person, every single discussion was like, oh, yeah, it’s cool, all of these complicated things, it’s really cool but that little thing that you’re calling the architect is really, really, really what I want. How can I get my hands on that?
—— Andrew Antos · [03:10]
企业里这些复杂的业务流程是靠例外运行的。你有一个快乐路径,那只适用于百分之二十的事情,然后百分之八十是某种形式的例外。
指向原始笔记的链接
And these complicated business processes in enterprise run on exceptions. You have one happy path and that applies to twenty percent of things and then eighty percent is an exception of some kind.
—— Andrew Antos · [28:01]
大多数人认为嵌入模型已经商品化了，但这不是事实。根据你选择什么样的嵌入模型，你在检索质量上可能会有非常巨大的差别。
指向原始笔记的链接
Most people think embedding models are commoditized, and that is not true. There is a very big difference that you can get in retrieval quality based on what embedding model you choose.
—— 嘉宾 · [50:00]
我会担心任何这样的记忆架构：它不依赖成本更低的嵌入器和重排序器，而是依赖 LLM 的多遍处理来帮你分类并缩小数据语料库。
指向原始笔记的链接
I would worry about any memory architecture that instead of relying on lower costs embedders and re-rankers is relying on multiple passes of the LLM to help you categorize and shrink the corpus of data
—— 嘉宾 · [66:35]
我真的认为，一个公司运转的速度可以根据信息在公司内部流动的速度来判断。
指向原始笔记的链接
I do really think the velocity that a company moves at can be judged based on how quickly information is moving within the company.
—— Jess Hertz · [08:59]
我们季度营收几乎 90% 来自已经在平台上超过一年的商家。
指向原始笔记的链接
Almost 90% of our quarterly revenue is from merchants who’ve been on the platform for over a year.
—— Jess Hertz · [10:28]
在不理解纹理和上下文的情况下创建技术红线的想法，不是我们认为可以强加给客户的界限。
指向原始笔记的链接
creating a technological red line without understanding the texture and context is not a boundary line that we think that we can assert on top of our customer.
—— Nick Noone · [40:02]
我所有最好的 AI 点子，都是从 Every 团队上的某个人那里抄袭来的。
指向原始笔记的链接
I always like all of my best AI ideas, I’ve ripped off from someone else on the every team.
—— 嘉宾 · [34:59]
具体性基本上是所有质量的标尺。比如如果你能给智能体一个非常具体的任务，让它对要做什么有清晰的指示，那就会比一个泛泛的『这样去找代码坏味道』要好。
指向原始笔记的链接
Specificity basically is the ruler of all quality. Like if you can have a very specific task that the agent has clear instructions on what to do, it’s gonna be better than a generic here’s how to look for code smells.
—— Drew · [42:17]
可以这么说，你会不断地做循环，直到你有太多的循环，以至于你的大部分工作都在维护循环、以及构建用来观察它们的循环。
指向原始笔记的链接
It’s sort of, you’re going to keep working on loops until you have so many loops that most of your work is in maintaining loops and building loops to observe them.
—— Drew · [44:52]
我认为 AI 在总体上有一个特点，就是那些可以自我检验的问题，AI 可以持续钻研。
指向原始笔记的链接
One of the things that I think AI has in general is that problems that are self-checkable, AIs can grind on.
—— Brian McClendon · [26:40]
你也不应该盲目地同时开三个让它们互相争斗，因为它们很容易让自己全都相信同一个幻觉。
指向原始笔记的链接
And you shouldn’t just blindly turn on three of them fighting each other because they can easily all convince themselves of the same hallucination.
—— Brian McClendon · [37:35]
这就像有一个实习生时不时拍拍我的肩膀
指向原始笔记的链接
It’s just like having an intern who taps me on the shoulder every now and then
—— 嘉宾 · [23:39]
所以在那场对话中我们是平等的地位。一切都关于那场对话本身，你为它带来了什么，为它增添价值，并且尊重对方这个人，不是因为他头衔，而是因为他的想法。
指向原始笔记的链接
So we’re on equal footing in that conversation. It’s all about that conversation, what you’re bringing to it, adding value to it, and having respect for the other individual, not because of their title, but because of their ideas.
—— Amandeep Khurana · [25:13]
实实在在的零上，永远不会有星号。
指向原始笔记的链接
There will never be an asterisk on a firm zero.
—— Max Levchin · [42:13]
我认为那是胡扯，因为无论是能源、液冷，还是所有能让数据中心变得更好的东西，这些都是可以在这里创造、在这里完成的工作岗位。
指向原始笔记的链接
I call BS on that because if you think about whether it’s around energy, liquid cooling, all of the things that make the data center better, those are all jobs that can be created and done here.
—— Rene Haas · [28:38]
那些基准测试不代表你的工作负载。正如我们之前说的，AI 的边界是参差不齐的——某个模型可能在基准测试上做得很好，但这并不必然意味着它对你的工作负载也会做得很好。
指向原始笔记的链接
They don’t represent your workloads. And as we said earlier, the boundary of AI is jagged, right? So although a certain model might do very well on a benchmark, it doesn’t necessarily translate into that it will do very well for your workloads as well.
—— Chetan Gupta · [38:19]
你总是要押注于不是今天很酷的东西。
指向原始笔记的链接
You always want to be betting on not what’s cool today.
—— Tony Kim · [36:05]
但一切都有瓶颈,我认为持续扩展这个东西的主要瓶颈实际上是训练算力。
指向原始笔记的链接
But everything has a bottleneck, and I think the main bottleneck on continuing to scale this thing is actually training compute.
—— Justin Johnson · [23:04]
一旦你骑上电动自行车,你就不会踩踏板了。
指向原始笔记的链接
Once you get on the e-bike, you’re not pedaling.
—— Tyler Folkman · [30:03]
你把你的用户群和客户群，把你的客户，当作互联网上的一个模糊团块，这是一种非常危险的看待客户的方式，因为你想见到他们，弄清他们的需求。
指向原始笔记的链接
you consider your user base and customer base, like your customer, just a blob on the internet, which is a really risky way to think about customers because you want to meet them, figure out their needs.
—— Jeffrey Morgan · [49:26]
我几乎是深切地站在这一边：很难论证 AI 模型可以在宽泛的公共互联网上训练，但另一个 AI 模型却不能用 AI 模型的输出训练。
指向原始笔记的链接
I’m almost deeply on the side of, I think it’s very hard to make the argument that AI models should be trained on broadly the public internet, but another AI model can’t be trained on the outputs of an AI model.
—— Aaron Levie · [03:48]
有那么几个人，处在那样的人生阶段，可以直接说：好，我要砸 100 亿美元在美国建一个开放实验室。
指向原始笔记的链接
You get a few people that are at a stage in life where they can just be like, yep, I’m going to drop $10 billion on building an open lab in America.
—— Aaron Levie · [06:17]
如果你封锁中国，中国是不会放弃 AI 的。
指向原始笔记的链接
If you block off China, China is like not going to give up on AI.
—— Aaron Levie · [08:14]
在美国开源这一边，我总体上认为 AI 的赚钱机器是推理。
指向原始笔记的链接
on the open source US side, I generally think that the moneymaker in AI is inference.
—— Aaron Levie · [12:08]
你低头在模型火车上埋头干十年，然后抬起头，我的天。
指向原始笔记的链接
You put your head down and work on the model train set for 10 years, and then you lift your head up and like, holy shit.
—— Paul Graham · [06:40]
但当你站在终点线上时，你发现它是有宽度的。
指向原始笔记的链接
But when you’re standing on the finish line, you realize it has width.
—— Paul Graham · [15:32]
顺便说一句，我看多的一件事是 OpenAI 上的消费者广告市场，这显然至今还没有被充分开发利用，但我认为它仍然是 AI 领域最大的机会之一。
指向原始笔记的链接
By the way, one thing I’m long is the consumer ads marketplace on OpenAI, which has obviously not been fully utilized to this day, but I think remains one of the biggest opportunities in AI.
—— Anastasios Angelopoulos · [40:27]
胜率接近 100%。而且顺便说一句，要让人们在任何事情上达成一致几乎是不可能的。
指向原始笔记的链接
It was nearly a 100% win rate. And by the way, it’s almost impossible to get people to agree on anything.
—— Anastasios Angelopoulos · [38:57]
而我说，把你 DevEx 团队的规模翻一倍。把你数据团队的规模翻一倍。投入平台建设。对人类有好处，对智能体也有好处，而这就是能让你跑起来的东西。
指向原始笔记的链接
And I say, double the size of your DevEx team. Double the size of your data team. Like, work on platform investments. Good for humans. Good for agents. And that’s what will let you run.
—— Claire Vaux · [18:04]
它是那种即使业务停滞、不进任何新单也仍然会增长的生意——因为总有人在往里存钱。
指向原始笔记的链接
It’s one of those businesses that if the business stayed stagnant, we brought on no new business. It still grows because you still have people getting paid.
—— Aaron Schumm · [35:08]
我们是摩根大通，世界最大的银行。我们怎么能押注一家四十人的初创公司，来承载我们认为关乎未来重要业务的 stuff?
指向原始笔记的链接
We’re JP Morgan, we’re the largest bank in the world. How can we bet on this forty person startup to carry what we think is going to be the future of a lot of our business
—— Aaron Schumm · [39:38]
我有我的家庭，我有 Vestwell,就是这样。我只专注于此，否则其余的东西会把你拽向与你的生活毫不相干的方向。
指向原始笔记的链接
I have my family and I have Vestwell and that is it, right? And I just focus on that because otherwise the rest of it will start to pull you in directions that are just not attitude to your life
—— Aaron Schumm · [48:51]
我们认为在不久的将来，你会看到人们日常推理调用中的很大一部分，甚至可能是大多数，会走向本地设备和本地笔记本电脑或本地工作站，而不是像今天这种一切都被推上云端的标准做法。
指向原始笔记的链接
We think in the very near future, you’re going to see a huge proportion, maybe even a majority, of people’s daily inference calls going to local devices and on-prem laptops or on-prem workstations, as opposed to the kind of standard of today where everything’s being pushed up to the cloud.
—— 嘉宾 · [45:06]
我认为 X 上使用这些产品的人，其实并不代表产品市场契合。
指向原始笔记的链接
I think a person on X that uses these products is not actually product market fit.
—— JD · [37:16]
我们在编码方面有意走得慢一些，因为我们已经意识到编码在多大程度上是一场预算游戏——你有多少数据预算？
指向原始笔记的链接
We are intentionally moving a little bit more slowly on the coding side because we’ve kind of realized like how much coding is like a budget game, like how much data budget do you have?
—— Alexander Whedon · [27:49]
我不认为真的有人会把自己的职业生涯押在，比如说，用 vibe coding 写一个核反应堆或火箭引擎或某种后果非常严重的系统的控制系统上。
指向原始笔记的链接
I don’t think anyone’s really going to kind of bet their career on like vibe coding a control system for let’s say a nuclear reactor or a rocket engine or something that’s very high consequence.
—— Scott Morton · [02:47]
SpaceX 成功的最大关键之一就是 Elon 是那种会对他的团队下巨大赌注的人，相信我们能做到。
指向原始笔记的链接
One of the biggest keys to SpaceX’s success is just that Elon is one to make massive bets on his team, that we can do this.
—— Scott Morton · [15:37]
让我对开发 Grok Bot 感到如此兴奋的是，这是我第一次在非编码任务上感觉自己可以真正把工作委托给 AI，不用再去想它，回来的时候事情已经完成了。
指向原始笔记的链接
What made me so excited to work on Grok Bot is it was the first time for non-coding tasks that I felt like I could truly delegate work to AI, not have to think about it, and I would come back and it’s done.
—— Roman Ugarte · [01:00]
我认为我们现在正处于一个非常奇怪的时刻，将来回头看我们会说：我很惊讶很多人就是这样与 AI 一起工作的——你在给这些超级智能的新同事、这些 AI 机器人做入职培训，却让它们和你共用同一台电脑。
指向原始笔记的链接
I think we’re in a really weird moment right now that I think we’re going to look back on and be like, I’m surprised that this is the way that a lot of people worked with AI, where you’re onboarding these super intelligent new colleagues, these AI bots, and you’re asking them to share the same computer that you have.
—— Roman Ugarte · [32:21]
你仍然在做那件事，感觉也是那样，它同样压在你身上——而不是真正地给同事来一记不看人的传球，然后说：「你可以搞定，这是背景信息，放手去干吧。」
指向原始笔记的链接
You’re still doing the thing and it feels that way and it’s weighing on you in the same way versus truly throwing a no look pass to a colleague and being like, “You got this, here’s the context, go off and run.”
—— Roman Ugarte · [62:59]
而且我认为值得注意的是，这些竞争对手没有一个现在站在 AI 编程的前沿，这在很大程度上不是因为它们做出了任何错误的决定或缺乏资源，而是因为在文化上无法快速行动、无法在当下变化时改变以跟上当下。
指向原始笔记的链接
And I think it’s notable that none of those competitors are at the forefront of AI coding right now in large part, not because of any incorrect decisions that they made or any lack of resources on their part, but this cultural inability to move quickly and to change to meet the moment as the moment’s changing.
—— Roman Ugarte · [69:38]
你所有的创新代币都花在开源项目上了，而不是花在交付上。
指向原始笔记的链接
All your innovation tokens are being spent on that versus on the delivery.
—— Jordan Tigani · [11:04]
是 BI,不过是打了兴奋剂的 BI,因为有些事情你能做而别人做不了。
指向原始笔记的链接
It’s BI, but it’s BI on steroids because there’s some things you can do that you can’t do.
—— Jordan Tigani · [18:15]
但出人意料的是,无论你是在 Codex 还是 ChatGPT 上发布,流程都很相似——尽管 ChatGPT 推向十亿活跃用户而且还在增长,你可以提交一个 PR、做一个改动,第二天甚至同一天发布,它就推向十亿用户,而且没问题。
指向原始笔记的链接
But it’s surprisingly a similar process, whether you ship on Codex or ChatGPT, even though ChatGPT goes out to a billion active users and growing, you can ship a PR, you can make a change and get it shipped the next day or even the same day, and it just goes out to a billion users and it’s fine.
—— Tibo Sottiaux · [43:33]
我们走的是一条非常 Anthropic 式的路线：真正把 B2B 使用场景中的某些用户类型和工作流做到极致好，而不是在消费端把数亿美元到处乱撒。
指向原始笔记的链接
We’re taking a very Anthropic-type approach: really trying to make certain types of users and workflows within B2B usage insanely great, and we’re not just spraying hundreds of millions of dollars at everything on the consumer side.
—— Eric Simons · [15:05]
对我们来说，我们想在定价上变得非常激进。这里正在发生的巨大变化是，开源模型真的在追赶那些实验室。
指向原始笔记的链接
For us, we want to get really aggressive on pricing. The big sea change happening here is that open-source models are really catching up with the labs.
—— Eric Simons · [24:52]
比如 Codex 团队有自定义方法，Goose 团队有自定义方法，Klein 团队也有，无论谁，我们都能看到生态里涌现出什么，哪些值得走上标准化轨道纳入协议本身——让这个东西由使用方式塑造、由社区塑造。
指向原始笔记的链接
Like if the Codex team has some custom methods, the Goose team has some custom methods, the Klein team has some custom methods, whoever, we can see what emerges in the ecosystem and what makes sense to get on a standards track and bring into the protocol itself so that this is sort of shaped by usage and shaped by the community.
—— Alex Hancock · [05:47]
你要在关注 UI 之前先关注数据
指向原始笔记的链接
you want to focus on the data before you focus on the UI
—— Dustin Mihalik · [13:39]
所以那大概算是比较不讲道理的事情之一，比如，对，就把这整个 Python 代码库移植到 TypeScript，让它跑起来、能部署，就在一个周末之内。
指向原始笔记的链接
So that probably ranks on like the more unreasonable things like, yeah, just port this entire Python code base to TypeScript, get it working, get it deployable in, you know, a weekend.
—— 嘉宾 · [00:41]
AI 生成的代码,应该让人感觉是那种已经在你团队待了多年的人写出来的。
指向原始笔记的链接
AI-generated code should feel like it was written by someone who’s been on your team for years.
—— Brandon Waselnuk · [01:13]
token 减少 50%,分诊更快,而且答案质量实际上更好,因为它知道业务内部正在发生什么。
指向原始笔记的链接
50% fewer tokens, faster triage, and the answer quality is actually better because it knew what was going on inside of the business.
—— Brandon Waselnuk · [09:51]
所有这些建立在对数据和学习直觉之上的分析工作,造就了我们如今拥有的模型——相比 2017 年的 Transformer,它就像一艘忒修斯之船。
指向原始笔记的链接
All of this analysis work built on intuitions about data and learning led to the model that we have now, which is like a ship of Theseus compared to the 2017 Transformer.
—— Chris Potts · [15:29]
所以当人们说「哦,我们不理解」时,我说:我认为你们理解的程度远超你们所承认的。
指向原始笔记的链接
And so when people say, oh, we don’t understand, I say, I think you understand much better than you’re letting on.
—— Chris Potts · [16:53]
我的意思是,毕竟,所有这些东西都是在深度中没有递归这一事实之上的一种打补丁。
指向原始笔记的链接
I mean, after all, so all of this stuff is a kind of patch job on the fact that there’s no recursion in the depth.
—— Chris Potts · [34:52]
绝大多数人类疾病应该被解决、应该可以被治疗——我认为我们正在这条路上。
指向原始笔记的链接
And I think we’re on that path.
—— Zavain Dar · [33:57]
我觉得最好尝试去做一个建设者，你会摔个大跟头，你会体会到有时候做出贡献有多难。
指向原始笔记的链接
I think it’s better to try to be a builder and you’re gonna fall flat on your face and you’ll get an appreciation for how hard it is to contribute sometimes.
—— Brian Armstrong · [35:55]
如果研究实验室不拥有最好的产品，那才真是疯了，因为他们拿到了所有关于人们如何使用他们产品的数据。
指向原始笔记的链接
It would be really wild if the research labs didn’t have the best product, because they’re getting all of this data on how folks are using their products.
—— Demetrios Brinkmann · [38:34]
那将花费 1.2 万亿美元——前提是我们国家首先还执行得了那个计划。
指向原始笔记的链接
That’ll cost $1.2 trillion if we can even execute as a country in the first place on that plan.
—— 嘉宾 · [04:05]
没有理由一个连轮胎都不会换的人不能去造一艘 Corsair、参与 Marauder 或更大的船的工作。
指向原始笔记的链接
There’s no reason somebody who can’t change a tire shouldn’t be able to go and build a Corsair or work on a Marauder or work on a larger ship.
—— 嘉宾 · [16:28]
你不能让一个与任务无关的人，依据一本没有任何灵活余地的手册来任意做决定。
指向原始笔记的链接
You can’t have one human who’s not tied to the mission making arbitrary decisions based on a guidebook that doesn’t have any wiggle for them.
—— 嘉宾 · [31:41]
令人惊讶的是，互联网有多少是基于少数几个开源代码库——它们由优秀的工程师创造，并不是为了赚一大笔钱，而真的只是为了推动技术向前发展。
指向原始笔记的链接
It’s surprising how much of the internet is based off of a handful of open source repositories that great engineers have created and not for seeking a bunch of money on the other side, but really just for pushing the needle of technology forward.
—— 嘉宾 · [58:33]
顺便说一句，我觉得相当神奇的是，AI 已经发展到了这样的程度：地球上任何一个不是设计师、也不是工程师的人，都可以点一个按钮，突然就做出一整套 PowerPoint，或者做出一个网站，或者做出一个 web 应用。
指向原始笔记的链接
I think it’s quite magical, by the way, that AI has gotten to a point that any human on the planet that is not even a designer, that is not an engineer, can click a button and suddenly make an entire PowerPoint or make a website or make a web app.
—— Thais Castello Branco · [03:34]
我想说，这个门槛基本上还在地上。
指向原始笔记的链接
The bar is kind of really, I would say, on the ground.
—— Thais Castello Branco · [14:25]
我认为把自己限制在用户数上——说实话，我觉得按用户定价已经是过去式了。
指向原始笔记的链接
I think restricting yourself on users, I think user-based pricing is a thing of the past, honestly.
—— Ron Gabrisko · [72:27]
随着智能体变得更高效、更有效,我认为我们会更倾向于依靠技术控制作为门禁机制,而不是对它能做什么的主观人为评估——因为再说一次,由于非确定性,你真的很难回头去问一个智能体:你为什么把这些文件全删了?
指向原始笔记的链接
So as an agent becomes more efficient and effective, I think we’re going to lean more on technical controls as a mechanism to gate versus a subjective human assessment of what it can do, because again, the non-determinism, it’s really hard to go back to an agent and ask, why did you go delete all these files?
—— Robert Lucero · [18:36]
你交付的速度建立在你构建的这份基底之上，如果基底不好，你在上面构建的任何东西都不会好。
指向原始笔记的链接
The velocity that you ship lays down on this substrate that you’ve built, and if the substrate isn’t good, nothing you build on top is going to be good.
—— Jonathan Kelley · [08:53]
如果你的智能体代码所落地的底层基础不好，它们的贡献也会不好。
指向原始笔记的链接
If the substrate on which your agent’s code lands is bad, their contributions will be bad as well.
—— Jonathan Kelley · [16:15]
关于使用编码智能体构建有雄心的软件，我的结语是：代码现在很廉价，但质量不是。
指向原始笔记的链接
So my closing thoughts on using coding agents to build ambitious software is that code is now cheap, but quality is not.
—— Jonathan Kelley · [18:07]
软件工程师的工作从来都不只是把代码行放到屏幕上。
指向原始笔记的链接
The job of a software engineer has never really been about putting lines of code on the screen.
—— Jonathan Kelley · [18:17]
如果你把系统搭建好，它们自己向前推进，那就是彻底的改变。
指向原始笔记的链接
If you put the system in place and they move forward on their own, it’s a game changer.
—— 嘉宾 · [52:22]
我想很多 PM 会意识到自己其实并不那么擅长从零到一，而且在一个别人的好想法上工作比在自己的坏想法上要有成就感得多。
指向原始笔记的链接
I think a lot of PMs are gonna realize they’re actually not that good at zero to one and it’s much more fulfilling to work on someone else’s good idea than your own bad idea.
—— Anish Acharya · [17:15]
你知道什么比我亲自做更好吗？是让 Claude 来做。这就在团队每个人心中建立了一个习惯。
指向原始笔记的链接
You know what’s better than me doing it? It’s Claude doing it. And that just created this habit in everyone on the team.
—— 嘉宾 · [48:18]
我发现这类分手邮件的回复率要高得多，因为它让对方说「不」的压力非常低。
指向原始笔记的链接
Reply rates I find are much higher on these breakup emails because it makes them really low pressure for them to say no.
—— 嘉宾 · [10:10]
外呼上真正成功的创始人和不成功的创始人之间的头号区别，其实就是坚持执行他们承诺的活动。
指向原始笔记的链接
The number one difference between founders who actually succeed without bound and founders who don’t is actually following through on the activity that they commit to.
—— 嘉宾 · [11:12]
所以现在我们到了这样一个阶段：我们只点击那些表面上看有风险、不能轻信的东西。
指向原始笔记的链接
So now we’re at a point where we only click on what is risky to believe on its face.
—— Tim Sanders · [28:16]
我用 AI 是为了让我离开电脑,而不是待在电脑前。
指向原始笔记的链接
I use AI to get me off my computer, not on my computer.
—— Claire Vo · [00:00]
我在这点上有个强烈的看法：你不可能拿着同一个产品把 NDR 从 20 提到 40。留存几乎是从核心产品继承来的。
指向原始笔记的链接
My strong view on this is that you can’t take the same product and take NDR from 20 to 40. I think you almost like inherit retention based off of the core product.
—— Keith Peiris · [12:54]
事后来看，我可能更高兴我们没有收购它，因为考虑到它对当时的 Bending Spoons 来说规模会那么大，那 Bending Spoons 就会变成 Grindr，因为我们很可能把全部精力都放在那上面。
指向原始笔记的链接
In the hindsight probably I’m happier that we didn’t acquire it because given the fact it would have been so big for Bending Spoons then Bending Spoons would have been Grindr because we would have probably fully focused on that
—— 嘉宾 · [10:45]
我们每年的流失率仍然低于 1%。
指向原始笔记的链接
We still have like less than 1% churn on a yearly basis.
—— 嘉宾 · [24:33]
所以你需要非常少但非常坚实的基础来构建东西，但之后你需要增加很多灵活性，确保没有过度约束的流程或审批链条之类的东西，因为根据我的经验，那些正是杀死一个产品的东西——无法对产品采取行动。
指向原始笔记的链接
So you need very few but very solid basis when you build something, but then you need to add a lot of flexibility and make sure that there are not, you know, not super constraining processes or approval lines and things like this, because in my experience, those are the things that kill a product, the inability to act on the product.
—— Vali · [71:47]
所以我们真正追求的是资本的速度，而不仅仅是资本的回报。
指向原始笔记的链接
So what we’re really after is the velocity of capital, not just returns on capital.
—— David Morehead · [16:52]
区别在于，有几千万人基于那些信息交易，而在私募侧，大概只有三个人。
指向原始笔记的链接
The difference is that there are tens of millions of people trading on that information, whereas on the private side, there’s like three.
—— David Morehead · [27:04]
永远不要把你的模式押在一个对你的业务没有任何既得利益的第三方的慷慨之上。因为也许他们今天是朋友，明天可能就不是朋友了。
指向原始笔记的链接
Don’t ever bet your model on the generosity of a third party that has no vested stake in your business. Because maybe they’re friends today, they might not be friends tomorrow.
—— Jim VandeHei · [26:50]
你必须在你和你的客户之间建立直接的关系，而且这种关系必须自身就能盈利，你才能拥有一个可行、可扩展、持久的产品。
指向原始笔记的链接
You got to build a direct relationship between you and your customer. And that one has to be profitable on its own for you to have a viable, scalable, durable product.
—— Jim VandeHei · [27:00]
你要么领先一年，要么落后一年，仅仅取决于你的信息流。
指向原始笔记的链接
You’re either a year ahead or a year behind simply based on your feed.
—— Aaron Levie · [56:19]
我想我的最大区别在于它实际上是在你的电脑上运行的。我目前看到的所有东西都在云端运行。它可以做几件事。如果你在电脑上运行它，它可以做每一件该死的事情。
指向原始笔记的链接
I think my big difference is that it actually runs on your computer. Everything I saw so far runs in the cloud. It can do a few things. If you run it on your computer, it can do every effing thing.
—— Peter Steinberger · 来自原文
② 出现在这些集
1 集
- 《16 年的一夜成名:ON 如何打造未来最高端运动品牌》 — 作为被讨论公司
③ 关联
点进去有真内容 —— 本页主要出口
Bob Safian · David Alleman · light spray · Clean Cloud 泡沫 · 创新 · 高端(非奢侈)定位 · 运动阶层 · 股权替代代言 · 合伙制领导 · 社会变迁造品牌