Work
集里怎么说它
- 《OpenAI 产品负责人谈：AI时代怎么做产品、写文档、抬野心》(00:00起):本集说 Work 模式底层就是 Codex 的能力，但移除了编码 UI，让非开发者也能用智能体完成复杂金融模型等知识工作
① 提到它的金句
284 条
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]
你可以在你的家庭故事中成为主角，但在你工作的公司故事中，你可能不是主角。
指向原始笔记的链接
You can be the protagonist in the story of your family, but in the story of your work of a company, you are probably not the protagonist.
—— Hilary Gridley · [51:09]
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]
所以我注意到的是，你可以通过你可以给它多长的”绳子”去工作来判断 AI 变好了多少。
指向原始笔记的链接
And so what I have noticed is that you can tell how much better AI is getting by how long a leash you can give it to do work.
—— Dan Shipper · [14:57]
如果你整天在工作中，你在救火，你会想，“好吧，我是要用我知道的方式做这件事，还是用可能不奏效的新方式做？“
指向原始笔记的链接
If you’re working in a job all day, you’re fighting fires, and you’re like, “Okay, am I going to do this in the way that I know how or am I going to do it in the new way that might not work?”
—— Dan Shipper · [30:44]
所以基本上，对于每一个工作单元，你应该让下一个工作单元更容易做。
指向原始笔记的链接
So basically, for every unit of work, you should make the next unit of work easier to do.
—— Dan Shipper · [42:10]
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]
我们已经从软件是付费访问转变为现在你是在为交付的工作付费。
指向原始笔记的链接
We have moved from software being a pay for access to now you’re paying for work delivered.
—— Madhavan Ramanujam · [29:51]
跟 Brian 的产品会议会让人们前一天晚上睡不着，不仅因为他们真的通宵工作做准备，还因为他们知道他们要通宵工作，因为他们可能因为期待这个而睡不着觉。
指向原始笔记的链接
A product meeting with Brian would keep people up the night before, not just because they were actually working all night long to get prepared, but also they knew they would work all night long because they probably wouldn’t sleep in anticipation for this.
—— Chip Connolly · [16:21]
所有的失败都是那些试图同时玩多个游戏且拥有稀缺资源的，如果你是早期初创公司，这往往是行不通的。
指向原始笔记的链接
all the failures are the ones that tried to play multiple games at once with scarce resources, and that just tends to never work if you’re an early-stage startup.
—— Brian Balfour · [51:49]
在过去20年里，我们就工作如何完成有着非常固定的比例。PM花费这么多时间，开发人员花费这么多时间，上市花费这么多时间。我认为这种契约基本上正在爆炸，因为构建过程，封闭的求解过程，构建过程正在加速得如此之快。
指向原始笔记的链接
For the last 20 years, we’ve had very fixed ratios in terms of how work gets done. PM takes this amount of time, developers takes this amount of time, go-to-market takes this amount of time. And I think that contract is basically exploding because the build process, the closed solutionary process, the build process is accelerating so fast.
—— Oji · [08:37]
我想劝阻的一件事，就是在你开始做评估之前试图做大量的规划和准备工作。
指向原始笔记的链接
And one thing I would sort of discourage is trying to do a lot of planning and prep work before you start doing evals.
—— Ankur Goyal · [11:17]
我认为有很多它表现不太好的情况，但我觉得如果你把 LLM 当裁判的用途从一个基准重新定义为一个异常检测器，那它其实是可以的。
指向原始笔记的链接
Well, I think there’s a lot of cases where it doesn’t work very well, but I think if you reframe the utility of an LLM as a judge from being a benchmark to being an anomaly detector, then I think it’s actually okay.
—— Ankur Goyal · [25:10]
我认为如果你真的不清楚成功是什么样子的，提示词，你可能无法获得最棒的工作。我认为这对管理团队来说是真实的，对管理 AI 来说也非常真实。
指向原始笔记的链接
I think if you’re really unclear about what success looks like, the prompt, you’re probably not going to get the most amazing work. I think that’s true for managing teams and it’s very much true for managing AIs.
—— Julie Zhuo · [13:50]
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]
防守者面临的动态是，要做的工作远多于做这些工作的人。
指向原始笔记的链接
And the dynamics for defenders are there’s a lot more work to do than people to do it.
—— Vineet Edupuganti · [06:09]
我认为正向工作与倒推工作之间的重大区别是，是人类在启动 AI，还是 AI 在启动人类，对吧？
指向原始笔记的链接
I think the big difference between kind of the work forward versus work backwards is, you know, are humans initiating the AI or is AI initiating kind of the humans, right?
—— Evan Reiser · [26:27]
工作会膨胀以填满可用于完成它的所有时间。
指向原始笔记的链接
Work expands to fill the time available for its completion.
—— Stewart Butterfield · [54:43]
而我认为反驳是，你本可以用个人电脑做出同样的论点，而实际上计费反而上升了，因为工作变得更复杂了。
指向原始笔记的链接
And I think the counter is like you could have made the same argument with the PC and it’s like actually billables went up because the work got more complex.
—— Gabe · [16:59]
所以我认为我们在五年内希望看到的是，那样的工作是由一个由律所和那个客户训练的模型来完成的。
指向原始笔记的链接
And so I think the thing we see, we would want to see in five years is like that work is done by a model trained by a law firm and that client.
—— Gabe · [44:22]
我曾在一些大型科技公司工作,我总觉得我们可以解雇 90% 的人,我们会走得更快,因为最优秀的人不会有这些干扰。
指向原始笔记的链接
I used to work at a bunch of the big tech companies and I always felt that we could fire 90% of the people and we would move faster because the best people wouldn’t have all these distractions.
—— Edwin Chen · [00:14]
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]
我们实际上需要 AI 发挥作用，因为我们将需要机器来做所有我们没有人力去做的那些工作，因为在未来 100 年里，我们在字面上将使地球人口减少。
指向原始笔记的链接
And we actually need AI to work because we’re going to need machines to do all the jobs that we’re not going to have people to do because we’re literally going to depopulate the planet over the next 100 years.
—— Marc Andreessen · [10:59]
在软件工程中，独创性在于偏离，但世界上正在做的很多工作实际上只是运行这些程序和运营。
指向原始笔记的链接
In software engineering, the ingenuity is in deviating, but a lot of the work being done in the world is actually just running through these procedures and operations.
—— Sherwin Wu · [55:14]
想要被喜欢我不认为是 CEO 的一个好特质，想要成为最佳工作场所可能也不是正确的路。
指向原始笔记的链接
Wanting to be liked I don’t think is a good feature of a CEO, and wanting to be the best place to work probably isn’t the right way to go.
—— Brian Halligan · [55:52]
你会意识到杠杆最高的做法其实是把出问题的地方捕获进一条评测，这样你做的任何改动你都知道是真正有效的，不会破坏你之前努力取得的所有其他成果。
指向原始笔记的链接
You realize that the highest leverage thing you can do is actually try to capture what’s going wrong in an eval so that any change that you make you know is actually going to work and it’s not going to break all the other stuff that you’ve worked towards.
—— Ankur Goyal · [05:49]
当你真正接受它时，你会意识到你在评测上投入的工作，不会在你每次尝试新模型、改一些措辞或稍微改变用例时就蒸发掉。
指向原始笔记的链接
When you really embrace it, you realize that the work that you put into doing evals doesn’t evaporate every time you try a new model or you change some wording or you change your use case a little bit.
—— Ankur Goyal · [06:00]
因为当基础设施不工作时，人会死。
指向原始笔记的链接
Because when the infrastructure doesn’t work, people die.
—— Jeetu Patel · [54:47]
这就是我在这些工具上如此成功的原因是因为我有 20 年以上的管理经验。我知道如何让员工成功。这就是你需要让这些智能体工作的东西。
指向原始笔记的链接
that is why I have been so successful with these tools is because I have 20 years plus of management experience. I know how to make an employee successful. That is what you need to make these agents work.
—— Claire Vo · [86:29]
唯一会记得你工作到很晚的人是你的孩子。
指向原始笔记的链接
The only people that will remember that you stayed late at work are your kids.
—— Lenny · [102:32]
如果一个自动化不能 100% 的时间工作,它就不是一个真正的自动化。
指向原始笔记的链接
if an automation doesn’t work 100% of the time, it’s not really an automation.
—— Kat Wu · [70:06]
所以我认为在早期，我的观点不是我们不需要 PM，我的观点是设计师应该做那项工作。
指向原始笔记的链接
I think in the early days, my view was not that we don’t need PMs, my view is that designers should do that work.
—— Evan Spiegel · [32:29]
如果他们知道如何调整 UI 的小细节，但他们不理解智能体循环是如何工作的，我会更倾向于那种对理解智能体循环如何工作有深厚亲和力并且能够设计它们的设计师或产品经理，而不是那些能编写传统软件并调整样式的人。
指向原始笔记的链接
If they know how to tweak small details of the UI, but they don’t understand how an agent loop works, I would much rather take the designer or PM that deeply has an affinity for understanding how agent loops work and can design those than someone who can write traditional software and tweak styles.
—— Max Schoening · [09:23]
我们已经有了全民基本收入。它叫作知识工作。
指向原始笔记的链接
We already have universal basic income. It’s called knowledge work.
—— Max Schoening · [01:10]
我们没有关于人类智能是什么的理论，我们没有关于为什么这些模型工作得这么好的理论，我们也没有关于它们会变得好多少的理论。
指向原始笔记的链接
we have no theory of what human intelligence is, we have no theory of why these models work so well, we have no theory of how much better they will get.
—— Benedict Evans · [26:50]
没有那么多发表的东西。学术会议的主要工作单元是论文。就像，参加学术会议的全部意义就是你发表论文。它在 Europe 2025 上被引用，你在 Google Scholar 上有出版物。这不再发生了。
指向原始笔记的链接
There’s not that much not much published. The primary unit of work of the academic conference is the paper. Like, the whole point of getting to an academic conference is you publish the paper. It is cited in Europe 2025, and you have the publication in Google Scholar. That is no longer happening.
—— Swyx · [09:12]
就像，所有有趣的工作都闭源在产品中，有时你甚至得不到个人引用。就像，你只是所有东西都汇总到 Gemini three 里面。有一千个贡献者，你不知道里面有什么。
指向原始笔记的链接
it all the interesting work is closed source in in products, and sometimes you don’t even get an individual citation. Like, you just everything gets rolled up into Gemini three. And there’s a thousand contributors, and you don’t know what is in there.
—— Swyx · [09:28]
我们在法国、在欧洲更倾向于对自己说，这种新技术让我们害怕。我们要先尝试监管它，然后再去完全理解它将如何运作。
指向原始笔记的链接
we tend a little bit more in France, in Europe, to say to ourselves, There is this new technology that scares us. We will try to regulate it first, before completely understanding how it is going to work.
—— Charles Gorintin · [25:56]
你真的需要把所有这些切成不同的碎片,并与这么多不同的供应商和平台合作,为了完成一个非常简单的可视化吗?
指向原始笔记的链接
do you really need to chop all of this into different pieces and work with so many different vendors and platforms in order to get like a very simple visualization done?
—— Reynold Xin · [13:31]
而我们看到的是，组织在尝试采用一种颠覆性技术时正在重复同样的错误，就是直接把技术扔进去，不改变任何关于他们工作方式的事情，然后对他们没有变得更快感到惊讶。
指向原始笔记的链接
And what we see is that organizations are repeating the same kind of mistakes when trying to adopt a disruptive technology of just plopping the technology in, not changing anything about how they work and being surprised that they’re not going faster.
—— Daniel Jones · [03:55]
往墙上扔屎，直到粘住为止，希望这东西能起作用。
指向原始笔记的链接
Throwing shit at the wall until it sticks, hoping that the thing is going to work.
—— Kitsa · [02:49]
一个拥有强烈信任和融洽关系的领导团队几乎能解决任何问题,而一个没有信任或融洽关系的领导团队,任何事情都可能成为问题。
指向原始笔记的链接
a leadership team with strong trust and rapport can work through most anything a leadership team without trust or rapport like anything can become an issue
—— Adam Mosseri · [33:26]
当我们与现场技术员合作时，他们告诉我们，如果我必须脱下安全手套来在一个小玩意上打字来使用你的小玩意，我就不会去做。
指向原始笔记的链接
When we work with field technicians, they tell us, if I have to take my safety glove off to type into a thingy to use your thingy, I’m just not going to do it.
—— Kriti Sharma · [05:16]
现在的魔力在于，以前这需要很多人，可能甚至数百名工程师，数年的努力，才能使某种技术或软件在制造业供应链中工作。
指向原始笔记的链接
The magic now is previously it would have taken a lot of people, maybe possibly even hundreds of engineers, years worth of effort to make some sort of technology or software work for manufacturing supply chain.
—— Kriti Sharma · [12:43]
人们基本上告诉我们，「我的大脑正在腐烂。我的工作感觉更糟糕了。」
指向原始笔记的链接
People essentially told us, “My brain is rotting. My work feels worse.”
—— Noam Segal · [47:48]
生产力的提升是真实的，但工作的质量以及生产它的人的敏锐度正在受到打击。
指向原始笔记的链接
The productivity gains are real, but the quality of the work and the sharpness of the person producing it are taking a hit.
—— Lenny · [52:31]
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]
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]
这里的魔力不在于甚至很少是他们建议他们想要在产品里的东西。而是他们揭示出来的东西，那些对他们不起作用的、不显而易见的、只是个痛点的、或者与他们使用的语言不产生共鸣的东西。
指向原始笔记的链接
The magic here is not or even rarely what they’re suggesting that they want in the product. It’s what they’re revealing that doesn’t work for them or isn’t obvious or it’s just a pain or doesn’t resonate with the language that they’re using.
—— 嘉宾 · [03:59]
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]
有一线微光。我没法发布微光。它必须有最终的固体形状。否则，我只是在发布气体。那行不通。
指向原始笔记的链接
There’s a glimmer. I can’t ship a glimmer. It’s got to have a final solid shape. Otherwise, I’m just shipping gas. That doesn’t work.
—— 嘉宾 · [21:30]
我们有过竞争对手,他们把一切都献祭在速度的祭坛上,结果没有成功,完全没有。
指向原始笔记的链接
I think we’ve had competitors. That sacrificed everything at the altar of speed, and it didn’t work out, and not at all.
—— Eric Sager · [14:43]
你得想明白:当每四到六周排行榜榜首就出现另一个新模型时,可持续的竞争差异化路径到底是什么?
指向原始笔记的链接
You’ve got to work out, well, what is the path to sustainable competitive differentiation when every four to six weeks there’s another model at the top of the leaderboard?
—— Benedict Evans · [49:34]
所以杠杆不在于权重，而在于你如何连接工作。
指向原始笔记的链接
So the leverage is not in the weights, it’s in how you wire the work.
—— Garry Tan · [03:17]
我认为这些是将来我们回头看会说，天哪，真不敢相信我们当时让人做那种工作。
指向原始笔记的链接
I think these are jobs that will look back and say, God, I can’t believe we had people doing that kind of work.
—— Fred Turner · [65:32]
这说明了成为领域专家、拥有垂直专业知识的重要性。这种知识至少现在仍是真正被需要的，否则你无法重新想象这项工作。
指向原始笔记的链接
And so it shows you the importance of like being a subject matter expert and having vertical expertise. That knowledge, at least right now, is really needed. Otherwise you can’t reimagine the work.
—— Alex Lieberman · [13:14]
是的，百分之百。作为创始人，你的根本工作是两件事：第一，创造一个让人们感到被看见、被倾听、并被赋能去做出最佳工作的环境；第二，给业务降低风险。
指向原始笔记的链接
Yeah, a hundred percent, right? Your fundamental job as a founder is two things. Number one, create an environment where people feel seen, heard, and empowered to do their best work. And number two, to de-risk the business.
—— Damien Lewke · [20:24]
当领导层开始质疑时，我是否需要一名工程师加上每月 5,000 美元才能让他们完成工作？
指向原始笔记的链接
When leadership start questioning, do I need an engineer plus 5,000 a month just for them to do their work?
—— Heitor Lessa · [01:14]
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]
但是当模型只能做大约一小时的工作时，异步作为一种体验是很糟糕的。
指向原始笔记的链接
But when models can only do like an hour of work, async as an experience is kind of bad.
—— Lance Martin · [01:48]
有时候你确实需要 996,但对我来说那就像是，呃，也许你没有护城河。或者也许你需要做一些如此非同寻常的事情，以至于它需要你和你周围所有人都做出那种牺牲。但很多时候那只是表演性质的。
指向原始笔记的链接
And sometimes you do need to work 996, but to me that was like, well, maybe you don’t have a moat. Or maybe you need to do something so extraordinary that it requires that sacrifice from you and everybody around you. But a lot of the times it’s just performative.
—— Kareem Amin · [11:53]
如果你提供一个有价值的产品,资本主义会找到一种方法让供应链为你工作。
指向原始笔记的链接
if you’re providing a product of value, capitalism will find a way to make the supply chain work for you.
—— Sriram Krishnan · [19:24]
它是我最讨厌、讨厌的同事，然而它……它做得工作最好。
指向原始笔记的链接
It is my most loathed, loathed colleague, and yet it… It does the best work.
—— 嘉宾 · [22:54]
你不可能为家里机器人工作 24 小时支付 100 美元的算力。
指向原始笔记的链接
You cannot pay $100 of compute for 24 hours of your home robot to work.
—— Alex · [10:51]
我认为卓越的工程师想做高影响力的工作，而那往往恰好是难的。
指向原始笔记的链接
I think great engineers want to do high impact work, which happens to be hard.
—— 嘉宾 · [22:56]
但实际上,你只需要把任务交给模型,给它一种验证自己工作产出的方式,这样它就不会卡住,它就会一直做下去。
指向原始笔记的链接
But really, all you need is give the model the task, give it a way to verify the output of its work so it doesn’t get stuck, and it’ll just go.
—— Boris Cherny · [22:40]
智能体不介意烦人的刁钻规则，它们可以很开心地去撞墙让事情跑起来。
指向原始笔记的链接
Agents are fine with annoying Infiniki, they can very happy to go bang their heads against the wall in order to make something work.
—— Rob Willoughby · [27:31]
知识是可变的，热情不是，而且从来没有比现在更有理由去做你热爱的事情了。
指向原始笔记的链接
Knowledge is changeable, passion is not, and never has there been more reason to work on what you love.
—— Blake Scholl · [33:03]
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]
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]
所以突然间，你的产品路线图对每个单个客户来说完全是一堆雪花，这是行不通的。
指向原始笔记的链接
So suddenly your product roadmap is a complete snowflakes for every single customer, which is not going to work.
—— Lindsey Scrase · [10:52]
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]
这帮人都擅长的一点,比如 Ilya,这点真的被低估了,就是让这些大计算机运转得非常好。
指向原始笔记的链接
One thing these people are all good at, like Ilya, which is really underappreciated, is making these big computers work really well.
—— Reid Alberghati · [33:41]
而一个可运行的演示最多只是你必须做的工作的 1%。
指向原始笔记的链接
And a working demo is 1% at best of the work that you have to do.
—— Dmitri Dolgov · [08:21]
我们可以让 AI 为我们工作，或者我们可以让 AI 让我们去工作。
指向原始笔记的链接
we can put AI to work for us or we can have AI put us to work.
—— Claire · [20:37]
为了让这行得通，有两个因素需要汇聚。一个是发射成本下降，人们必须相信这一点。第二是，在地面建设基础设施正变得极其困难。
指向原始笔记的链接
To make this work, there’s a convergence of two factors. One is the launch costs coming down, and people have to believe that. The second is it becoming way harder to build stuff terrestrially.
—— Philip Johnston · [15:02]
真正杰出的人，是那些比任何人都更用心打磨自己手艺的人。他们做别人不愿做的事，因为他们热爱。
指向原始笔记的链接
Because the truly great people are the ones that work on their craft more than anyone else. They do things that no one else will do because they love it.
—— Jesse Cole · [28:16]
我认为实际上这是一种非常反直觉的动态,更智能的模型实际上在委派工作方面变得越来越擅长。
指向原始笔记的链接
I think actually there’s this really unintuitive dynamic where smarter models actually get better and better at delegating work.
—— Walden · [03:54]
这些深科技公司失败的原因很少是因为技术行不通。
指向原始笔记的链接
It’s uncommon that these deep tech companies fail because the technology doesn’t work.
—— Max Hodak · [19:54]
所以走得更快意味着一个小小的错位可能滚雪球成大量浪费的工作,而且那工作消耗代币,代币现在要花真金白银。
指向原始笔记的链接
So going faster means that a small misalignment can snowball into a ton of wasted work, and that work costs tokens, and tokens cost real money now.
—— Idan Gazit · [04:15]
你雇佣的每一个你曾经一起工作过的优秀的人，都会让这个地方对你曾经一起工作过的人更有吸引力。
指向原始笔记的链接
Every good person that you hire that you used to work with makes the place more attractive to the people that you used to work with.
—— Colin Zima · [47:01]
我认为这实际上是从过去产品开发的运作方式到现在的产品开发方式的完全颠倒。
指向原始笔记的链接
I think it’s actually a complete inversion of how product development used to work to how product development happens now.
—— Eric Vishria · [13:58]
更重要的不再是你出货多快，而是你决定构建什么、以及怎么构建，才能让你建的系统持续组合在一起、互相协作运作。
指向原始笔记的链接
How fast you ship, but what you decide to build and how you decide to build it in order to be able to continue composing the systems that you build together and having them interoperate and work together.
—— Ali Haghani · [14:31]
我们与最大的律师事务所和企业合作，他们的法律数据极其敏感，受特权保护。所以你不能把它放进通用模型里——我们甚至不能把它放进我们自己的模型里。
指向原始笔记的链接
We work with the largest law firms and enterprises, and their legal data is incredibly sensitive. It’s privileged. And so you can’t put it in generic models. We can’t even put it in our models.
—— Gabe Pereyra · [03:21]
通常，如果你和他们合作却没能得到更好的结果，那很可能是你的数据集有问题。
指向原始笔记的链接
And usually, if you work with them and you’re not able to get better results, there’s probably something you’re doing wrong with your data set.
—— Gabe Pereyra · [06:50]
但我认为找到某种形式的持续学习——这里的终局不是让我们构建最好的法律模型，而是帮助每家律所或企业客户把它定制成他们所做工作的类型。
指向原始笔记的链接
But I think figuring out some form of continual learning is I think the endgame here is not for us to build the best legal model. It’s for us to help every law firm or enterprise customer customize it to the type of work they’re doing.
—— Gabe Pereyra · [26:15]
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]
因此，他们想投钱进来，而我们交付了如此多的价值，以至于他们非常迫切地想和我们合作，而不是反过来。
指向原始笔记的链接
So therefore, they want to put money in and we’re delivering so much value that they’re really desperate to work with us, not the other way around.
—— Adam · [21:25]
我认为三年后，公司本质上就像董事会，20% 的工作是战略和思考，然后智能体只做 80% 的执行。
指向原始笔记的链接
I think in three years, companies will essentially be like board of directors, where 20% of the work is the strategy and the thinking, and then agents just do 80% of the execution.
—— Matt Swulinski · [60:41]
当我们实际测试让客户尝试直接用智能体通过我们的平台交付工作成果时，智能体的失败率高得惊人。
指向原始笔记的链接
When we’ve actually done testing of clients trying to use agents directly to deliver work products through our platform, the failure rate of agents is incredibly high.
—— Hayden Brown · [10:24]
我们最近做的一项调查显示，我们自己的客户中已有 23% 实际上已经把工作从 AI 移回给人类，或者即将这么做。
指向原始笔记的链接
A recent survey we did showed that 23% of our own clients have actually already moved work back to humans from AI or are about to do so.
—— Hayden Brown · [10:12]
而且坦率地说，它推进了这种精英选拔的理念：不该看谁拥有最好的智能体，或者谁在纸面上听起来最光鲜，而是真正看谁拥有数据和经验来证明他们能完成所需的工作。
指向原始笔记的链接
And frankly, advancing this meritocratic idea of it shouldn’t be about who has the best agent or who sounds the most shiny on paper, but really who has the data and the experience to demonstrate that they can do the work that’s required.
—— Hayden Brown · [14:50]
这些 AI 工具不仅仅是生成媒体的工具,它们实际上是一种新的讲故事媒介,而且还没人知道什么会成功。
指向原始笔记的链接
These AI tools are not just tools for generating media. They are actually a new medium for storytelling and that no one knows what’s going to work yet.
—— 嘉宾 · [00:29]
我觉得我们现在正处于那样的拐点之一:这些 AI 工具不仅是生成媒体的工具,实际上是一种新的讲故事媒介,而且还没人知道什么会成功。
指向原始笔记的链接
And I think we are at one of those inflection points right now where these AI tools are not just tools for generating media, they are actually a new medium for storytelling and that no one knows what’s gonna work yet.
—— 嘉宾 · [18:06]
它们实际上是讲故事的新媒介，而且还没人知道什么会奏效。
指向原始笔记的链接
They are actually a new medium for storytelling and that no one knows what’s going to work yet.
—— 嘉宾 · [00:33]
但我认为我们发现的是，这里有一个巨大的鸿沟，实际上需要大量的手工人力工作才能跨越到人们觉得真正持久地引人入胜的东西。
指向原始笔记的链接
But I think what we found is that there’s this enormous chasm that is in fact a lot of manual human work to cross over to something that people find truly compelling in a durable way.
—— Elliot · [68:41]
最有价值的时机是在尽可能晚的时刻，因为那时你将拥有你试图对齐的实际系统可以与之合作。
指向原始笔记的链接
The most valuable time for that to happen is at the latest possible moment because then you would have the actual system that you’re trying to align to work with.
—— Nick Bostrom · [30:11]
你越能把这些体验引导到高价值、又不让用户费很大劲去使用，就越能建立对系统的信任，然后才谈得上委派和采取行动。
指向原始笔记的链接
The more you can steer these experiences to very high value, without causing customers to do a lot of work to use it, the more that builds trust with the system, and then it allows you to delegate and take actions.
—— Willem Avé · [21:26]
每个人都在把智能体放进生产环境，如今甚至 15 岁的孩子都在这么做。该问的问题是它们是否真的有效。
指向原始笔记的链接
Everyone is putting agents into production, even like 15-year-old kids these days. The question to ask is whether they actually work.
—— Ayush Bhardwaj · [03:17]
每个 AI 都能到达生产环境，只是它没能跑起来，或者没能证明自身的成本合理。
指向原始笔记的链接
Every AI reaches production, but it just fails to work or justify its own cost.
—— Ayush Bhardwaj · [16:37]
一种在这里能运作得很好的模式是，如果你在平台中强制执行一个更宽泛的 agent 定义，把 LLM 和人类都囊括在内，那么你就可以做到让任何 LLM 能执行的动作，人类也能执行。
指向原始笔记的链接
One pattern that can work very well here is if in your platform you enforce a wider definition of agent which encompasses both LLMs and humans, then you can make it such that any action that can be taken by an LLM could also be taken by a human.
—— Christopher Lovejoy · [13:49]
它们能在一次过夜运行中处理过去需要几周的工作量，早上你回来时就会得到这个为你准备好的漂亮成品包。
指向原始笔记的链接
They can handle what used to be weeks of work in one overnight run, and you come back in the morning with this beautiful package ready for you.
—— Hursh Agrawal · [10:37]
我不会接任何关键路径的工作。
指向原始笔记的链接
I would not take any critical path work.
—— Hursh Agrawal · [05:37]
所以我真的敦促你们，在隔夜的时间里，试着思考我如何给它越来越多的范围，让我们能够把几周的工作、几个月的工作在隔夜之间完成，看看这些模型到底能做到什么。
指向原始笔记的链接
So I’d really push you overnight, try to think of how do I give it more and more scope so that we can do weeks of work, months of work overnight and see what these models are really capable of.
—— Hursh Agrawal · [15:23]
你期望某一套应用或用户交互能很好地工作，因为它们可以由你设计好的编排来完成，但当用户意图需要超出编排所规定范围的其他事情时，系统就会开始在接缝处崩裂。
指向原始笔记的链接
You expect a certain set of applications or user interactions to work well because they can be fulfilled by the orchestration that you have designed, but when the user intent requires other things to happen beyond what’s specified in the orchestration, the system can start breaking at the seams.
—— Ameya Bhatawdekar · [10:55]
因为归根结底，evals 才是那种描述你的系统应该如何运作的持久资产。
指向原始笔记的链接
Because ultimately, it’s the evals that are sort of your durable asset that describe how your system is supposed to work.
—— Ameya Bhatawdekar · [18:14]
我多希望自己只发一个 prompt 就能解决软件里的所有问题，但现实是，真正的工程工作很少能装进一个 prompt 里。
指向原始笔记的链接
I wish that I could just send off one prompt and solve all of the problems that exist in my software, but the reality is that real engineering work rarely fits inside one prompt.
—— Safia Abdalla · [06:45]
所以是的，我想我们都想要这种自动改进循环，但它们不适用于产品。
指向原始笔记的链接
So yeah, I think we all wanted these sort of like automatic improvement loops, and they just don’t work with product.
—— Ryan Carson · [20:25]
所以价值从做操作性工作转移到了判断。
指向原始笔记的链接
So the value sort of shifts from doing operational work to judgment.
—— Srini Raghavan · [26:20]
在某个时候，人们开始开这个玩笑：安全已经变成了工作量证明。
指向原始笔记的链接
At some point, people started to make this joke that security has become proof of work.
—— Neil Movva · [10:05]
每当我与 AI 团队交谈时，感觉他们想构建的是智能体。他们想构建的是工具。但如果你真的走进公司内部去看那些工作，他们没在想 AI。他们在想眼前这个具体的问题。
指向原始笔记的链接
Whenever I talk to AI teams, it feels like what they want to build is agents. What they want to build is tools. But if you actually go look inside of the work at companies, they don’t, they’re not thinking about AI. They’re thinking about this particular problem in front of them.
—— Mike Lewis · [37:39]
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]
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]
所以一切对我来说，当你对信息和各种工作的相关信息有无限的需求时，其核心是一个围绕质量、成本和延迟的优化问题。
指向原始笔记的链接
So everything, to me, when you have an infinite appetite for information and relevant information for all kinds of work, it’s at its core an optimization problem around quality, cost, and latency.
—— Parag · [14:22]
不是为了让你下午 2 点就下班，而是为了让你真正花时间思考下一级别的任务、更高层级的上下文，变得有点创造力。
指向原始笔记的链接
Not so that you can just leave work at 2 p.m., so you can actually spend time thinking about next level tasks, higher level contexts, be a bit creative.
—— Will · [18:03]
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]
比如，如果你看我们的冲刺，也许 60%-70% 的工作是在添加新功能、提升质量和各种各样的事情，但有 30%-40% 的工作是我们在不断地用新技术重新架构。
指向原始笔记的链接
So, like, if you look at our sprints, like, maybe 60-70% of the work we are doing is adding new features, improving quality, and all kinds of things, but 30-40% of the work is that we are constantly re-architecting with the new technology.
—— Sait Izmit · [13:27]
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]
摆脱组织结构图周围的那套废话。拥有这个东西并强调这个叫做工作图的东西。
指向原始笔记的链接
Just get rid of that nonsense around the org chart. Have this thing and emphasize this thing called the work chart.
—— Jay Parikh · [20:32]
你知道，它们中的大多数并不真正工作。你必须走向世界。你必须实际上与现实世界的公司互动，为了真正建立一些重要的东西。
指向原始笔记的链接
You know, most of them don’t really work. You have to go to the world. You have to actually interact with real-world companies in order to really build something significant.
—— Ofir Ehrlich · [08:19]
这就是为什么你的智能体需要一个上下文引擎才能完成真正的工作。
指向原始笔记的链接
And that’s why your agents need a context engine to get real work done.
—— Peter Werry · [06:59]
我认为我们投资组合中累积的 90% 的价值来自于我们在投资后与公司所做的工作。
指向原始笔记的链接
I think 90% of the value that accrues in our portfolio comes from the work that we do with the companies after we’ve invested.
—— Chris Olsen · [20:38]
耶稣基督，如果我公司里30%的人离职去为Harvey工作，那我就完蛋了。
指向原始笔记的链接
Jesus Christ, if 30% of my company leaves to go work for Harvey, I’m dead in the water.
—— 嘉宾 · [60:58]
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]
你必须先做有意的工程工作,才能看到生产率提升的那条曲棍球杆曲线,因为我们必须先在代码库里做真正的工作,智能体才能在那里取得成功,尤其是在既有代码库中。
指向原始笔记的链接
You have to do intentional engineering work before you’re going to see that hockey stick curve in productivity improvement because we have to do real work in our code bases first for agents to be successful there, especially in Brownfield existing code bases.
—— Clare Liguori · [09:54]
所以我在这里给所有人的一个大要点是:前沿工程的核心是有意识地改变你的工作方式。
指向原始笔记的链接
So my one big takeaway for everyone here is that frontier engineering is about intentionally changing the way that you work.
—— Clare Liguori · [19:57]
所以当 AI 不工作时，这不是他们的问题。我们不是供应商。这是我们的问题。
指向原始笔记的链接
So when the AI doesn’t work, it’s not their problem. We’re not the vendor. It’s our problem.
—— Varun Shenoy · [04:19]
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]
有生产力的、高质量的工程工作,可以说删除代码甚至更好。
指向原始笔记的链接
Productive and high-quality edge work, one can argue that removing code is even better.
—— Mingsheng Hong · [10:06]
我认为，如果你能把你的初创公司架构设计成能与这样的公司合作，你基本上能够满足其他所有人。
指向原始笔记的链接
I think if you can architect your startup to work with companies like this, with this kind of architecture, you’re probably going to be able to satisfy basically everybody else.
—— Brian Lewis · [17:43]
任何时候你创造了一种激励来向人们展示你在工作，而不是真正去工作，你基本上就是在激励错误的事情。
指向原始笔记的链接
Anytime you create an incentive to show people that you’re working rather than to actually do work, you’re basically incentivizing the wrong thing.
—— Eno Reyes · [69:52]
美国只有不到 2% 的电工实际接受过直流电方面的培训，因为它非常危险，你可能会把自己弄死，而且操作起来非常不稳定。
指向原始笔记的链接
There’s less than 2% of electricians in the U.S. that are actually trained on DC power because it’s very dangerous and like you could potentially kill yourself and it’s very volatile to work with.
—— Jen Kha · [17:26]
在我构建了那个真正强大的东西并且对我工作的方式很友好之后，我很兴奋去分发它，我和几个 PM 进行了现场演示，我看着他们从零开始操作，我看到了多么不同。我以为它已经是一个完整的、完全成熟的工具，但我看到他们在使用它时很挣扎，因为它完全是按照我的工作方式调整的。
指向原始笔记的链接
And after I built something that was really powerful and good for the way I was working and I was excited to distribute it, I had live demos with several PMs that I watched them operated from zero, and I saw how different. I thought it was already this complete, fully baked tool, and I saw them using it and struggling because it was completely adjusted to how I work.
—— Daniel Blum · [36:53]
智能体需要像互联网一样运作的支付。
指向原始笔记的链接
The agents need payments to work like the internet.
—— Harshal Bhangale · [05:44]
如果你是一个 GP 并且你把基金的 1% 到 2% 投入某事，如果它归零了谁会在乎呢？这无所谓。但是如果你把基金的 20% 投入某事，那最好能行。
指向原始笔记的链接
If you’re a GP and you put 1% to 2% of your fund into something, who cares if it goes to zero? It doesn’t matter. But you put 20% of your fund into something, that better work.
—— Brian Singerman · [20:23]
所以这件事能行得通的唯一途径就是个性化。
指向原始笔记的链接
So the only way this can work is through personalization.
—— Jorge Conde · [15:33]
因为我认为，你从不奏效的事情里学到的东西，和从奏效的事情里学到的一样多。
指向原始笔记的链接
Because I think you always learn more from things that don’t work and things that work.
—— Stéphane Bancel · [36:07]
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]
我们可以让 AI 去干活,或者我们可以让 AI 以更有效的方式让我们去干活。
指向原始笔记的链接
we can put AI to work or we can have AI put us to work in a more effective way.
—— 嘉宾 · [23:23]
需求是永不满足的，这是由这些模型的运作方式决定的。
指向原始笔记的链接
The demand is insatiable, just given the way these models work.
—— Rene Haas · [15:04]
我们越来越感觉到,你想与智能体协作,而这些智能体在聊天框里并不太自在。
指向原始笔记的链接
it sort of increasingly felt that you wanted to work with agents, and those agents weren’t quite at home in a chat box.
—— Roshan · [02:05]
如果我们可以更像同事一样与智能体协作,而不是更像软件一样使用它们,会怎么样？
指向原始笔记的链接
what if we could work with our agents more like colleagues versus working with them more like software?
—— Roshan · [04:12]
因为那位工程师本来只被允许做 PR 的工作，但现在我可以用同一个智能体部署到生产环境？那太可怕了。
指向原始笔记的链接
Because the engineer was allowed to do the PR work, but now I can use the same agent to deploy to prod? That’s too scary.
—— Tanmai Gopal · [23:19]
那不是在做产品工作。那更像是把你的思考外包出去了。
指向原始笔记的链接
That’s not doing product work. That’s more kind of outsourcing your thinking.
—— Tyler Folkman · [47:27]
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]
我把自己亲手做的工作比作：拿着凿子和钉子，去凿医疗基础设施这堵混凝土墙，做集成。
指向原始笔记的链接
I liken the work that I did literally with my own two hands as taking a chisel and a nail against this concrete wall of integrating into the infrastructure of healthcare.
—— Julie Yoo · [05:07]
这是目前唯一——算是我们在医疗科技领域见到的第一波真正有机的采用浪潮——医生们就在用 AI 记录员，因为它们真的好使。
指向原始笔记的链接
This is currently is the only, like the first real organic adoption wave that we’ve seen in health tech where doctors are just using AI scribes because they freaking work.
—— Julie Yoo · [15:15]
我实际上认为这是把他们的工作蒸馏成了他们最擅长于世、最感兴趣的那件事，而围绕核心工作的一切行政事务现在都由别人替他们处理了。
指向原始笔记的链接
So, I actually think that this is a distillation of their job into the thing that they’re the best in the world at, that they’re the most interested in, and all the administration that goes around doing that core work is now handled for them.
—— Anish Acharya · [16:37]
即使有人复制了你的大想法，他们也永远看不到让大想法成立的那十几个小想法。
指向原始笔记的链接
And even if somebody replicates your big idea, they never see the small ideas that make the big idea work.
—— Anish Acharya · [57:40]
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]
如果你给人们设置太多摩擦，他们就会去做不安全的事情，因为人类就是这样运作的。
指向原始笔记的链接
If you put too much friction in front of people, they’re just going to do unsafe things because that’s how humans work.
—— Sharadh Krishnamurthy · [35:40]
我想让智能体让我为它们工作。
指向原始笔记的链接
I want the agents to put me to work.
—— Claire Vaux · [46:55]
人们其实很希望它能成，但他们不相信它会成。
指向原始笔记的链接
I think people really wanted it to work, but they didn’t believe it was going to work.
—— Aaron Schumm · [27:12]
我见过太多次有人尝试用 AI 来解析字符串，其实一个简单的正则表达式就能完成这项工作。
指向原始笔记的链接
Where a simple uh script uh the the number of times I’ve seen AI try being used to parse strings where a simple regist can just do the work.
—— Ran Arusi · [22:06]
管理上下文的能力——不只是在两小时的会话里，而是跨越数周或数月的工作——我认为这将是关键的，而且需要比今天长得多的上下文推理能力。
指向原始笔记的链接
The ability to manage context, not just through a two-hour session, but across weeks or months of work is something that I think will be critical and will also require much longer context reasoning than we have today.
—— Alexander Whedon · [32:19]
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]
打造一群彼此喜欢的人，他们努力工作是因为这是为他们信任和尊重、希望看到他们成功的同事在做，而不是每个人只为了自己那笔钱。
指向原始笔记的链接
Building a group of people that like each other, that work hard because they’re doing it for their coworkers who they trust and respect and they wanna see them succeed versus everybody out just for a payday.
—— Jordan Tigani · [05:34]
某种希望是，如果这成功了，它可以成为其他开源项目采用的一种模式。如果不成功，那它就是一个警示故事。
指向原始笔记的链接
And kind of the hope is that if this works out, then this can be a model for other open source to use. If it doesn’t work out, then it’s a cautionary tale.
—— Jordan Tigani · [08:13]
如果你在做某件事,而且你在搭建一个一万行的代码拐杖来绕过模型的缺陷,那你很可能在做错误的事情。
指向原始笔记的链接
If you’re doing something and you’re building like this 10,000 lines of code crutch to work around the model flaws, you’re probably doing the wrong thing.
—— Tibo Sottiaux · [42:29]
如果你无法解释你想要达成什么,如果你无法解释你的意图,如果你没有与一个社区的联结,如果你没有品味,做出出色的工作会难得多。
指向原始笔记的链接
Like if you can’t explain what you’re trying to achieve, if you can’t explain your intent, if you don’t have a tie to a community, if you don’t have the taste, it’s going to be much harder to do great work.
—— Tibo Sottiaux · [70:55]
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]
在我们的行业里，「大概对」就是错。「大概对」最终会得到不能用的机器。
指向原始笔记的链接
Probably right is wrong in our business. Probably right ends up with machines that don’t work.
—— Andrew Anagnost · [23:17]
我们消费者正在尝到这种禁果的感觉：哇，事情真的可以这么容易吗？它真的可以直接就用？
指向原始笔记的链接
tasting this, like, forbidden fruit of, wow, things can really be that easy? Like, it can just work?
—— Demetrios Brinkmann · [25:23]
如果我们在工作中做的事仍然感觉笨重，或者必须非常技术化才能获得 AI 带来的进步，那我们就做错了。
指向原始笔记的链接
If what we do at work still feels cumbersome or we have to be really technical in order to get the advances that AI brings, then we’re doing it wrong.
—— Demetrios Brinkmann · [30:11]
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]
我认为,就目前而言,AI 还不足以取代人类完成最后 5%、10%,甚至 20% 的工作,即把作品从好提升到伟大的那部分。
指向原始笔记的链接
I don’t think, right now, AI is good enough to replace humans for the last five, 10, maybe even 20% of the work to get it from good to great.
—— Paul Bakaus · [11:50]
我认为你可以借助工具来磨砺你的品味,并展现出你更多的品味。
指向原始笔记的链接
And I think you can work with tools to sharpen your taste and bring out more of your taste.
—— Paul Bakaus · [14:59]
如果你要把一个智能体卖给某个人，你也必须帮他们解决评分标准的问题，即我们究竟如何验证这个智能体在做好的工作
指向原始笔记的链接
if you’re going to sell somebody an agent, you also have to help them with the rubric of how do we actually verify that this agent is doing good work
—— Maximillian Piras · [15:44]
如果你能找出一个相当可重复的模式来验证它们的工作，你其实也可以直接把智能体扔到那个问题上。
指向原始笔记的链接
if you can figure out a pretty repeatable pattern for verifying their work, you can actually just throw agents at that problem as well.
—— Maximillian Piras · [20:04]
所以你不只是构建一个智能体，你或许要构建一个用来验证那个智能体工作的智能体。
指向原始笔记的链接
you don’t just build the agent, you perhaps build the agent that verifies the work of the agent.
—— Maximillian Piras · [20:13]
就是你不用它来做你的工作，而是通过与它互动来得到更好的工作成果。
指向原始笔记的链接
That you don’t use it to do your work, you interact with it to get better work.
—— 嘉宾 · [08:58]
工程师的时间只有大约 40% 用于写代码,其中只有约 10% 用于写新代码。也就是说,工程师有 60% 的时间花在非编码任务、非编码工作上。
指向原始笔记的链接
An engineer’s time is only like 40% writing code and only like 10% writing new code. So there’s 60% of an engineer’s time that is non-coding tasks, non-coding work.
—— Robert Lucero · [27:20]
人们有一种假设，认为用 AI 智能体会非常不一样，但其实是一样的。
指向原始笔记的链接
People have this assumption that it would work very differently with an AI agent, but it’s the same.
—— 嘉宾 · [44:38]
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]
现在出生的孩子将永远不会工作，至少不会为了钱而工作。
指向原始笔记的链接
That kids that are being born now will never work, at least for money.
—— Matteo Franceschetti · [60:18]
所以任何大致属于牛顿力学但表现不对的东西，可能离一次快速的 SFT 就差不远了。
指向原始笔记的链接
So anything that’s like roughly Newtonian that doesn’t work out is probably just a quick SFT away.
—— Justin Johnson · [20:53]
不要制定什么五年计划或者大跃进。中央计划对斯大林和毛泽东都行不通，对创业者同样行不通。
指向原始笔记的链接
Don’t devise a five-year plan or a great leap forward. Central planning didn’t work for Stalin or Mao and it won’t work for an entrepreneur either.
—— 嘉宾 · [15:53]
结果表明，拿到优质参考客户最简单的方法，就是把软件免费送给圈子里的一小部分人，然后努力工作直到他们满意。
指向原始笔记的链接
It turns out the easiest way to get good reference customers is to give your software away for free to a couple of people in that network and work hard until they’re happy.
—— Keith Peiris · [32:30]
你基本上是在向公司里的每个人传达：每个职位都是可以争取的，没有任何一个职位是他们只要足够努力、足够有才华就无法填补的。
指向原始笔记的链接
You’re basically communicating to everyone in the company that every position is up for grabs, that there’s no position that if they work hard enough and if they’re talented enough, they won’t be able to fill.
—— 嘉宾 · [33:47]
很大比例——我想说可能高达约 50%——的 Shopify 的 pull request（每次改动生产系统时写的东西）现在不是由传统意义上做工程工作的工程师创建的，而是来自我们公司公共聊天里的对话。
指向原始笔记的链接
It’s that a very large percentage, I want to say it’s probably up to about 50% of pull requests in Shopify, which again, pull requests every time you change the production system, you write a pull request, are created now not by engineers doing engineering work in the traditional sense, but out of conversations in our common company chat.
—— Tobi Lütke · [07:31]
我的人生哲学是，在生命的尽头我会遇见那个我本可以成为的人，而我一生的工作就是把我和那个我将要遇见的人之间的差异减到尽可能小。
指向原始笔记的链接
My life philosophy is that I will meet the person I could have been at the end of my life and the work of my life is to reduce the difference between the person I will meet to as little as possible.
—— Tobi Lütke · [45:12]
我们如何帮助你在工作之外和工作之中，删除尽可能多的你出于习惯、出于惯性、只是因为「我们一直是这么做的」而在做的事情，这样你才能放大你真正想做的事情。
指向原始笔记的链接
How can we help you outside of work and at work delete as many things that you’re doing out of habit, out of inertia, just because that’s the way we’ve always done it, so that you can amplify the things you actually want to do.
—— Jim VandeHei · [09:57]
你基本上必须把问题分解、带着检查点去委派、给出快速反馈、认真审查工作成果,并且知道什么时候该亲自介入。
指向原始笔记的链接
You basically have to decompose the problem, delegate it with checkpoints, give fast feedback, review the work seriously, and know when exactly to jump in.
—— Paula Dozsa · [14:23]
其实，把这些数据重新组织所需的工作，感觉才是最重要的工作。
指向原始笔记的链接
Actually, the work required to reorganize it felt like the most important work.
—— Keith Parris · [07:12]
要成为真正的企业级 CRM，我们就不能对人们的工作方式抱有宗教式的执着。
指向原始笔记的链接
Like a real enterprise CRM, we can’t be religious about the way people work.
—— Keith Parris · [26:10]
我们的态度是，数据是你的，你想怎么用都行，我们会努力工作，每天都去赢得你生产力的那份权利。
指向原始笔记的链接
I think our take is, you know, it’s your day to do what you want, and we’re going to work hard to earn the right to your productivity every day.
—— Keith Parris · [30:40]
智能体可以帮你深挖这些问题，帮你找到可能驱动客户那些感知的指标，但通常不太擅长真正理解这项工作的端到端上下文。
指向原始笔记的链接
And so, you know, agents can help you dig into those and can help you find um uh find the uh the the metrics that might be driving those perceptions from customers or driving those anecdotes, um, but generally aren’t as good at really understanding the end-to-end context of the work.
—— Marc Brooker · [19:28]
当围绕资源分配、围绕优先级排序、围绕我们把时间花在更多产品工作还是更多技术工作上没有任何辩论时，这会在我的脑子里亮起一盏灯，说明有什么地方不对劲。
指向原始笔记的链接
When there’s no debate around resourcing, around prioritization, whether we spend time on more product work or more tech work, that lights a bulb in my head that something is wrong.
—— Ben Celebicic · [05:36]
有些人在实际工作中已经开始使用这些工具，并开始说：「看，我在 iOS 上构建这个功能。为什么我不直接让工具帮我在 Android 上也构建它呢？」突然之间，我一次完成了两个人的工作。
指向原始笔记的链接
Some people on the ground are picking up these tools and already starting to say, “Look, I’m building this feature on iOS. Why don’t I just tell the tool to build it for me on Android?” And all of a sudden, I’ve done the work of two people at once.
—— Ben Celebicic · [21:09]
几百毫秒的写延迟。如果你像 Shopify 那样为名人闪购做库存预留，那可不行。但对搜索来说非常非常好。
指向原始笔记的链接
Hundreds of milliseconds. If you’re like Shopify and doing inventory reservations for a Kylie Jenner flash sale, not gonna work. Very, very good for search.
—— Simon Eskildsen · [08:17]
它们都停留在固定块大小的模式里，而我们采取了不同的做法，我们说：既然可以选定多个，为什么要只选定一个呢？
指向原始笔记的链接
All of them just stayed in the model of let’s work with a fixed chunk size, whereas we took a different approach, and we said, why commit to one where we can commit to several?
—— Yuval Belfer · [10:59]
我们现在拥有的超智能模型正在实际加速科学研究,让初创公司能够更早取得更大的研究突破,因此这些深科技公司会运转得更好。
指向原始笔记的链接
The super smart models that we have now are actually accelerating scientific research and making it possible for startups to have bigger research breakthroughs earlier and that therefore these deep tech companies will actually work better.
—— Jared · [05:10]
你必须成为人们不只是读写,而是真正在其内部开展工作的地方,你知道,在你的记录系统里。
指向原始笔记的链接
You have to be the way people not just read and write, but actually do their work inside, you know, your system of record.
—— 嘉宾 · [17:19]
你必须搞清楚怎么让它工作，否则就别卖它。
指向原始笔记的链接
You’ve got to figure out how to make it work or not sell it.
—— Steven Sinofsky · [15:58]
我对自己将要开创的事业有多大的规模没有任何愿景，但我始终有信心，只要我们把工作做好、对顾客好，我们就没有上限。
指向原始笔记的链接
I had no vision of the scope of what I would start, but I always had confidence that as long as we did our work well and were good to our customers, there would be no limit to us.
—— 嘉宾 · [32:02]
如果一项工作、其质量或更好质量能由机器完成,我今天就会雇佣一台机器,即使它比人更贵。
指向原始笔记的链接
If the work, the quality or better of a human can be done by a machine, I will hire today a machine even if it’s more expensive than a human.
—— Daniel Dines · [32:27]
当没有手册时,AI 就不神奇了,它就行不通。
指向原始笔记的链接
When there is no manual, AI is not amazing and it doesn’t work.
—— Daniel Dines · [44:50]
模型是可以互换的,但工作流、工作地图以及围绕工作地图的工作流才是真正价值所在。
指向原始笔记的链接
Models are interchangeable, but the workflow, the map of work and the workflows around the map of work is where the real value is.
—— Daniel Dines · [45:49]
对于企业工作,我的预测是 90% 的流量会流向非常划算的模型。
指向原始笔记的链接
For enterprise work, my prediction is that 90% of the flow will go to very cost efficient models.
—— Daniel Dines · [46:38]
即使事情没做成，我们仍然会投资你的下一家公司。
指向原始笔记的链接
If things don’t work out, we’ll still fund your next company.
—— Gagan Biani · [25:11]
如果你在你的软件工厂里交付了坏代码，它会降低未来每一件经过那个工厂的工作的质量，因为模型会看到更多坏模式。
指向原始笔记的链接
if you ship bad code in your software factory, it degrades every future piece of work that goes through that factory because there’s more bad patterns for the model to see.
—— Dex Horthy · [38:30]
只要我们还在使用基于 transformer 的 LLM，这就是它们工作原理的数学本质，使用更少的上下文你就会得到更好的质量。
指向原始笔记的链接
As long as we’re using transformer-based LLMs, that’s just like the math of how they work, and you will get better quality if you use less context.
—— Dex Horthy · [52:08]
我觉得至少 Opus 5.5 不是一个糟糕的对抗性审查者，所以我让它审查更多 GPT 的 PR 和工作，反过来也是。
指向原始笔记的链接
I feel like at least Opus 5.5 is not a bad adversarial reviewer, and so I’m having it review more PRs and more work from GPT and vice versa.
—— 嘉宾 · [10:40]
创业是一场十年的苦役，而且大多数都不会成功。
指向原始笔记的链接
Startups are a 10-year slog and most of them don’t work out.
—— Goggin · [21:57]
现在，人类才是瓶颈，因为他们注意力有限，因为有这么多并行的工作在进行。
指向原始笔记的链接
Now, humans are the bottleneck with their limited attention because so much parallel work is going on.
—— Dario Hamidi · [34:51]
我们做的很多工作和获得的经验，都处在一个更大的主题背景下：AI 应用的范畴最近发生了很大变化，从单纯的信息综合产品变成了真正在干活的智能体。
指向原始笔记的链接
And I think a lot of the work that we’ve done and a lot of the learnings we’ve had are in this sort of broader thematic context of the scope of AI applications has changed a lot recently from just like information synthesis products to actual agents doing work.
—— Adit Abraham · [01:59]
大多数人让 VLM 或 LLM 给标注打一个分数，但这些分数完全不管用，尤其是当你的模型和裁判规模不同的时候。
指向原始笔记的链接
most of the people ask the VLM or LLM to assign some score to it, but those scores absolutely don’t work, especially if your models are of different size with judges.
—— Merve Noyan · [06:28]
当你拥有能够 24 小时 7 天工作的机器、能够随电力而扩展的机器、能够完成任何个人或人群都无法做到的事情的机器时,正确的答案是思考如何重组一个组织应该是什么样的,以迎接当今技术所带来的机会。
指向原始笔记的链接
When you have machines that can work 24-7, that can scale with electricity, that can accomplish what no individual human or group of humans can do, the right answer is to think about how do you start to reorganize what an organization should be to meet the opportunity of what’s available today with technology.
—— Michael Lee · [05:40]
点子很廉价,执行才是真正困难的。而真正杰出的人总能找到办法把事情做成。
指向原始笔记的链接
Ideas are cheap and execution is really hard. And truly exceptional people will always figure out how to make something work.
—— Michael Lee · [40:26]
你可以委派工作,但你不能委派责任。
指向原始笔记的链接
you can delegate the work, but you can’t delegate the accountability.
—— Wade Foster · [00:43]
用大模型来调度一支更小模型的舰队，我认为这是我们相当坚定信奉的范式，它已经被证明是编码系统的运作方式，而且我认为在未来一年它会进入几乎每一个智能体。
指向原始笔记的链接
Over a fleet of smaller models is I think a paradigm that we believe in quite firmly and it’s already proving out to be how coding systems will work and I think it’ll find its way into almost every agent over the coming year.
—— Hamad Bashir · [24:29]
类似地,Gen2 几乎无法创建连贯的视频,但如果你把它扩展,从某种意义上说没有理由它不会成功。
指向原始笔记的链接
Similar Gen2 can barely create coherent video, but if you scale it up, there is no reason why it shouldn’t work in a way.
—— Anastasis · [26:20]
我们的工作是搞清楚如何把 Anthropic 或 OpenAI 拿出来的所有这些技术投入到为他们服务的工作中去,这样他们就不必花时间去想该用哪个模型或如何搭建工作流。
指向原始笔记的链接
Our job is to figure out how we take all that technology coming out of Anthropic or OpenAI and put it to work for them, so they don’t have to spend time thinking about which model to use or how to set up a workflow.
—— Maggie Crowley · [00:27]
如果你作为领导者不这样做,那么要不了多久你就会不理解这些工具如何工作,而你会变得有些无关紧要。
指向原始笔记的链接
If you don’t do that as a leader, there’s only so long until you don’t understand how these tools work, and you’re going to be somewhat irrelevant.
—— Maggie Crowley · [21:09]
而这是对政府运作方式的百分之百的彻底误解，事实是，当你和政府交谈时，他们其实知道这些东西是怎么运作的，他们知道他们只是在听你说话，他们在听所有人说话，没有人会得到自己想要的，因为他们知道，要在我们的体制里做成任何事情，那就是一种妥协。
指向原始笔记的链接
And it’s a complete 100% misunderstanding of how government works, which is when you talk to the government, they actually know how these things work, and they know they’re just listening to you, and they’re listening to everyone, and no one is gonna get what they want, because they know in order to do anything in our system, it’s a compromise.
—— Martin Casado · [10:54]
那么，你如何持续地把这种品味注入到这些工作智能体中，以及它们如何持续地自我验证并协同工作，是你应该关注的最重要的事情；你做得越快，你就能越快地建立起一条通往垂直 AI 公司的可防御路径。
指向原始笔记的链接
So, how do you continuously inject that taste into these workers, and how they continuously self-verify and work together is the biggest thing that you should focus on, and the faster you do it, the faster you build a defensible approach to becoming a vertical AI company.
—— Roland Gavrilescu · [17:08]
广告以目前的形式无法与智能体兼容。
指向原始笔记的链接
Ads don’t work with agents in their current form.
—— Parag Agrawal · [23:46]
而令人非常惊讶、表现突出的一点是，编程智能体可以一次性完成这件事。它们不需要额外的机器人数据就能用这些代码工作，因为它们已经在如此多的代码数据上训练过了。
指向原始笔记的链接
And what was very surprising, what excelled was that coding agents can do this very one-shot. They did not need additional robot data in order to work with this code because they’re already trained on so much coding data.
—— Ham · [10:11]
我无法想象还有什么比努力工作却知道你出不出席都无所谓更糟糕的事了——那份工作其实不重要，不会有什么贡献，不会带来什么改变。
指向原始笔记的链接
I can’t imagine there’d be anything worse than working really hard and knowing it doesn’t matter if you show up or not, like if that work doesn’t actually matter, if it’s not going to contribute, if it’s not going to make a difference.
—— Maggie Landers · [10:18]
通过把其他源代码系统和项目跟踪系统中正在发生的活动以 webhook 的形式喂给它，我们开始达到那种自主性水平，我们的工厂在执行产品工程工作，而不仅仅是代码工作。
指向原始笔记的链接
By feeding our web hooks activities that are happening in other source code systems and project tracking systems, we’re starting to get to that level of autonomy where our factory is performing product engineering work, not just code work.
—— Ryan Cooke · [05:07]
你可能会陷入我所说的“中智梗图”，就是把所有时间都花在打磨工作流上，而不做实际的工作。
指向原始笔记的链接
You can do what I call midwit memeing, where you’re spending all of your time working on your workflow and not doing actual work.
—— Charlie Holtz · [03:30]
给它们一个不会被杀掉的沙箱，在那里它们可以探索你的代码库，可以处理非常困难的任务，并且知道当你合上笔记本盖子时它们不会被关掉。
指向原始笔记的链接
Give them a sandbox that won’t get killed, where they can explore your code base, where they can work on really hard tasks, where they know that they’re not going to get shut down when you close your laptop lid.
—— Charlie Holtz · [08:50]
制作那段录像迫使它把东西真正跑通。你没法伪造一次通过的运行。
指向原始笔记的链接
Producing that recording forces it to make the thing work. You can’t fake a passing run.
—— Andrew Orobator · [10:50]
Scaling loss 起作用不只是科学概念。视频只是比 LLM 和编程晚了两三年，但很显然它最终在视频领域也会起作用，我就决定赌一把。
指向原始笔记的链接
It’s not just a concept from science that scaling loss work. Video just takes maybe two, three years longer than LLMs and coding, but it was clear that finally scaling loss should work in video as well, and I just decided just to take a bet.
—— Alex Mashrabov · [10:50]
确实，可能超过 60% 的客户支持请求，尤其是第一道防线，可以用 AI 处理。但尤其是在 B2B 方面，AI 就是不行。
指向原始笔记的链接
It’s true that probably over 60% of customer support requests, especially the first line of defense, can be handled with AI. But when it especially comes to B2B, like AI just doesn’t work.
—— Alex Mashrabov · [33:52]
我确实相信，答案就这么简单：雇最好的人去做最好的工作，并想清楚如何留住他们。其他一切坦率地说都是次要的。
指向原始笔记的链接
I do believe that it’s as simple as hire the best people to do the best work and figure out how to retain them. Everything else is frankly secondary.
—— Alex Mashrabov · [50:16]
我几乎试遍了市面上所有的 AI 语音工具，除了 Eleven Labs 之外没有一个能用。
指向原始笔记的链接
So I’ve tried almost every AI voice tool out there and none of them work except for Eleven Labs.
—— 嘉宾 · [22:33]
在我 30 分钟骑车上班的路上，它把我的邮件分类了，归到不同标签下，回复了不同的邮件，发出了日历邀请，到工位时，收件箱清零。
指向原始笔记的链接
And throughout my 30-minute bike ride to work, categorized my emails, submitted them to different labels, replied to different emails, sent out calendar invites, got to the desk, inbox zero.
—— David Pawlan · [18:32]
那你为什么要信任别人的智能体呢？检查别人的工作，可能和自己动手做一样费工夫
指向原始笔记的链接
why should you trust someone else’s agent where it might be just as much work to check the work over as if you had done it yourself?
—— Liz Fong-Jones · [40:20]
你的预训练模型已经知道答案，而你针对一个已经存在于预训练语料库中的答案做强化学习时，系统基本上可以伪造一些工作、得到正确答案，而你会强化这些策略。
指向原始笔记的链接
Your pre-trained model already knows the answer, and you do reinforcement learning derived against an answer that was already in the pre-training corpus, that the system can basically fake some work, get to the right answer, and you reinforce those strategies.
—— Doj Chubuk · [39:47]
现在要分辨哪些是真正由人写的、哪些是纯粹的垃圾内容,真的变得很难了。在某些情况下,它们反而增加了更多工作,而不是真正减少工作。
指向原始笔记的链接
It’s literally becoming hard to understand what’s actually written by a human versus pure slop. In some cases, they’re adding more work, not actually deducting.
—— Faisal Masud · [00:18]
在某些情况下,它们在增加更多工作。它们实际上是在给你的日常工作流程里引入更多工作,而不是减少。
指向原始笔记的链接
In some cases they’re adding more work. They’re actually introducing more work into your daily workflow, not deducting it.
—— Faisal Masud · [08:12]
如果你必须依赖一个经理向一个经理汇报再向一个总监汇报,我不确定这些层级能否与 AI 长期共存,因为 AI 可以做其中很多工作。
指向原始笔记的链接
If you have to rely on a manager reporting to a manager reporting to a director, I’m not sure all those layers scale long-term with AI, because AI can do a lot of that work.
—— Faisal Masud · [22:07]
如果一方在输,这笔交易长期就不会太好;短期真的很好,你可以整天击掌,但一年后你就在流失那个客户了。
指向原始笔记的链接
If one party is losing, the deal doesn’t work out that well long-term; it’s really good short-term, you can high-five all day long, but a year later you’re churning that customer.
—— Faisal Masud · [26:39]
你基本上是在充当一个路由器，一台 Cisco 路由器或交换机，在两个代表你工作的有状态智能体之间，但它们没有能力相互通信，所以你需要给它们提示。
指向原始笔记的链接
You are basically acting as a router, a Cisco router or a switch between two stateful agents that do work on your behalf, but they have no ability to communicate, hence you need to prompt them.
—— Vlad Luzin · [03:23]
我觉得 PKI 的关键教训其实是：如果你告诉人们这是 PKI，那它就行不通。
指向原始笔记的链接
I think the key lesson of PKI was actually, if you tell people it’s PKI, it’s not going to work.
—— Joel De La Garza · [15:39]
反过来讲，存在各种内部激励问题：如果你开始做更复杂的智能体，就会与现有的、提供工作流的产品产生内部冲突——因为你的产品是在解决工作本身。
指向原始笔记的链接
The flip side is there’s all these internal incentive issues, right, which is if you start getting into more complicated agents, it’s an internal conflict with an existing product that’s offering workflow versus you’re resolving the work.
—— Seema Amble · [09:54]
我们作为一家初创公司、一家规模化中的公司，必须赢得信任；在位者已经拥有信任，但这也意味着，如果他们过早发布什么产品，而产品不好用、很糟糕，他们就会毁掉信任。
指向原始笔记的链接
And like we as a startup, scale-up, we have to earn the trust, incumbents already have the trust, but this also means they can destroy the trust if they ship something like too early and the product maybe doesn’t work or it’s bad
—— Vlad Kyle · [11:46]
所以这可能意味着你有 70% 的性能，但你仍然要做 100% 的工作，因为 70% 并不算多。
指向原始笔记的链接
So this can mean that you have 70% of the performance, but you still need to do 100% of the work because 70% is not that much.
—— Vlad Kyle · [41:28]
我相信,世界上每家公司的 CEO 都会梦想拥有这样的能力——基于某个决定改变他们单位和员工的优先级,而这个决定第二天早上或一小时内就开始生效。
指向原始笔记的链接
I believe that every CEO of every company in the world will be dreaming about having such possibility to change the priorities of their units and their employees based on this decision that start to work tomorrow morning or within one hour.
—— Andrii Hrytseniuk · [24:23]
有时我们会感到震惊,当不同国家购买那些没有在乌克兰测试过、只是宣称有效的无人机。
指向原始笔记的链接
Sometimes we’re shocked when the different countries buy the drones that have not been tested in Ukraine but just declared that they work.
—— Andrii Hrytseniuk · [33:12]
我们已经从帮助 AI 掌握测试,转变为帮助 AI 掌握真实工作。
指向原始笔记的链接
We’ve gone from helping AI master tests to helping AI master real work.
—— Jonathan · [01:51]
今天这些智能体大概能可靠地连续工作两天左右,比如在编码这类任务上。
指向原始笔记的链接
Today these agents maybe reliably work for like two days at a stretch, like for tasks like coding.
—— Jonathan · [07:33]
MCP 就已经用 100 个工具占满了你的上下文。可能你三分之一的上下文就这么没了，却什么工作都没做。
指向原始笔记的链接
MCP would already eat 100 tools in your context. Maybe one third of your context would be gone without actually doing any work.
—— Jan Čurn · [03:00]
我认为是我们人类保持掌控的能力，我的意思是创意上的掌控，对网络安全风险的掌控，以及最终对如何让智能体为我们工作的掌控。
指向原始笔记的链接
I think our ability to stay in control as humans and by that I mean creative control, control over the risks on cybersecurity, control over how we make the agents work for us ultimately.
—— 嘉宾 · [00:37]
我认为用你的个人智能体做跨领域联接最糟糕的地方在于，你交给它做的工作越多，你牺牲掉的对当前情况的理解就越多。
指向原始笔记的链接
I think that the worst part about doing cross-domain joins with your personal agent is that the more work you give it to do, the more understanding of what’s going on you’re sacrificing.
—— Alex Atallah · [19:08]
如果你的智能体无法直接使用你的应用、你的产品来获得「一切是否正常」的反馈循环，那么这是一项非常值得的投资，每次你做 UI 工作时都会有回报。
指向原始笔记的链接
If your agent cannot use your application, your product directly to get this feedback loop of is it all working, then that’s a really big investment that pays off every time you work on UI.
—— Harald Kirschner · [08:48]
而且不只是调整你如何与智能体协作，还要调整它们如何获得反馈。
指向原始笔记的链接
And don’t just tune how you work with agents, but tune how they get feedback.
—— Harald Kirschner · [18:48]
这有点像你是高级工程师,把工作分派给初级工程师,让他们干重活,然后你来验证工作。
指向原始笔记的链接
So it’s kind of like if you’re like a senior engineer, you’re kind of farming out the work to a junior engineer, so they’re kind of doing the heavy lifting, and then you are verifying the work.
—— 嘉宾 · [14:53]
必须去搭建和摆弄你的 loops,这也许是让人们感到兴奋的事，但我不认为这会是它最终运作的方式。
指向原始笔记的链接
Having to set up and fiddle with your loops is something that maybe people get excited about, but I don’t think this is the way that it’s going to work.
—— Tibo Sottiaux · [00:46]
就是在 Apple 这里，你不信任 Siri 能行，因为它十多年来一直不行。
指向原始笔记的链接
It’s just you don’t trust that it’s going to, with Apple, you don’t trust that Siri’s going to work because it hasn’t for over a decade.
—— MG Siegler · [50:24]
当你决定了你是谁、你想做什么、你要达成什么，并且朝它努力，你就不可能失败。
指向原始笔记的链接
When you decide who you are, what you want to do, what you’re going to accomplish, and you work toward it, it is impossible for you to fail.
—— 嘉宾 · [00:58]
我不是在找 650 个杀手。但那些在你身边工作的人，他们得是杀手。
指向原始笔记的链接
I’m not looking for 650 fucking killers. But the ones that work close to you, they need to be killers.
—— 嘉宾 · [80:17]
他说，我们新的编码智能体能力默认都会流向「做错误的事情」。产品从好变坏的速度比以往任何时候都快。
指向原始笔记的链接
He says that the default place for our new coding agent abilities to go to is work on the wrong things. Products go from good to bad faster than ever.
—— Kent C. Dodds · [18:41]
基本上，他们能够把四个月的数据准备工作缩短到几天。
指向原始笔记的链接
So basically, they were able to go from four months of data prep work to down to days.
—— Jeff Koss · [14:16]
你知道，他们在大公司每天真正工作三个小时，然后五点二十九分打卡走人，站在电梯口等着。
指向原始笔记的链接
You know, they put in three hours a day of work at these big companies and then they punch out at 5.29 and they’re at the elevator banks.
—— Jason · [27:36]
那是一台货真价实的计算机,这相当了不起,也是让智能体为大多数人可用的唯一途径。
指向原始笔记的链接
That’s a real-deal computer, which is pretty remarkable, and also the only way to make agents work for most people.
—— Ben Thompson · [07:34]
当我聊到那次对话的结尾时，我常常会说，嘿，把这个做成一个技能，或者更好的是，把它做成一个插件，这样下次我需要做这件事时，你只需调用插件就能把活干完，不用再重走一遍整套流程。
指向原始笔记的链接
And when I get to the end of that conversation, I will often say, hey, make this a skill, or even better, make this a plugin, so that next time I need to do this, you can just invoke the plugin and get the job done, and we don’t have to step through all of this work again.
—— Charlie Guo · [12:59]
你正在给智能体更多的自主权。它会在没有你监督或抽象监督的情况下，工作得越来越久。
指向原始笔记的链接
You’re giving the agent more autonomy. It’s going to work longer and longer without your supervision or abstracted supervision.
—— Gabriel Chua · [50:11]
而且结果某种程度上证明，当你真正专注于开发者体验的核心时，你最终得到的东西同时也是非常好的原语，能让 AI 安全、可靠地与数据库协作并帮助它扩展。
指向原始笔记的链接
And it does kind of turn out that when you really focus on that core of the developer experience, what you actually end up with is also very good primitives for AI being able to work with a database safely and reliably and helping it scale.
—— Ben · [18:12]
我们真诚地相信，你在生产系统上所做的所有这些工作——比如修复问题或者日常运行你的软件——在大部分情况下都应该由智能体来完成。
指向原始笔记的链接
So we genuinely believe that all of this work that you do on production systems, like fixing issues or running your software on a day-to-day basis, should be done by agents in most part.
—— Varun Krovvidi · [05:57]
我认为，这里是工程工作如此关键的落点、作为工程师在这里工作如此酷的根本原因是：归根结底，训练是成本中心，推理是收入中心。
指向原始笔记的链接
I think the fundamental reason why this is such a critical place for engineering work to get done and why it’s such a cool place to work as an engineer is that in the end, training is a cost center and inference is a revenue center.
—— Charles Frye · [04:52]
人们普遍无法把直接向别人出售模型权重的生意做成，所以你没法真正把一个训练部门变成一个能自我支撑的系统。
指向原始笔记的链接
People have not generally been able to make businesses work where they sell model weights to people directly, and so you can’t really turn a training setup into a proper self-supporting system.
—— Charles Frye · [05:11]
你试图在 GPU 资源上管理锁或协调所带来的复杂度增加，远大于收益，因为即使你能把它降到微秒级，工作也要等到 GPU 上前一个工作完成后才会开始。
指向原始笔记的链接
The complexity increase that you get from trying to manage locks or coordination on the GPU resources, that complexity increase is much greater than the gain, because even if you were able to get it down to a microsecond, the work wouldn’t start happening until the previous work on the GPU had finished.
—— Charles Frye · [24:08]
我们碰巧通过代码让那个程序运行起来，代码其实是我们如何协作的一个副产品。
指向原始笔记的链接
The fact that we happen to make that program work through code, the code is really a side effect of how we work together.
—— Sam Newman · [17:23]
AI 本来应该把我们从苦役中解放出来，对吧？但对大多数软件开发者来说并不是，因为我们在做更多的工作，我们有更多的上下文切换，我们正在失去那种大局观。
指向原始笔记的链接
AI was supposed to free us from drudgery. Right? It isn’t for most software developers because we’re doing more work, we’ve got more context switching, we’re losing that big picture.
—— Sam Newman · [39:16]
AI 将代表我们做工作，它们将不得不互相交谈，以便代表我们解决任务。
指向原始笔记的链接
AI will be doing work on our behalf and they will have to talk to each other in order to solve tasks on our behalf.
—— Vlad Luzin · [01:02]
但这是不需要 50 万行你需要搬过来的 Python 或 TypeScript 代码的 loop engineering,那样你的智能体才能真正互相提示工作。
指向原始笔记的链接
But it’s loop engineering without 500,000 lines of a Python or TypeScript code that you need to bring over so your agents can actually prompt each other for work.
—— Vlad Luzin · [15:45]
那最后的 20%，那才是现在真正的工作所在，因为 Claude 能带你走完 80% 的路，而它基本上在一夜之间就完成了其中大部分。
指向原始笔记的链接
That last 20%, that’s where the real work is coming now because Claude can take you 80% of the way there and it basically did most of that overnight.
—— Aakash · [49:49]
你在工作中是的那种领导者，很快就会演变成你在家中、在个人生活中的那种领导者。
指向原始笔记的链接
leader you are at work quickly evolves to the leader you are at home and in your personal life.
—— Reef Frerichs · [15:12]
我们想要努力实现的目标或系统，是让追踪记录不再只是一份日志，而成为信号的一个来源。
指向原始笔记的链接
The goal or the system that we want to work towards is where a trace stops just being a log and becomes a source of signal.
—— Jake Broekhuizen · [04:01]
只把工具交到人们手里，是得不到高质量的学习或工作成果的。你需要给他们真正的学习和实验时间。
指向原始笔记的链接
Good quality learning or work out of just handing people the tools. You need to give them real time to learn and experiment.
—— Em Shreve · [01:01]
我留在那里是因为我有这样一种自尊心：我绝不为别人打工。
指向原始笔记的链接
I stayed on there because I had this ego that I will never work for someone else.
—— Jeetu Patel · [32:22]
如果你为此而构建，那么你就能构建出人们真正愿意托付实际工作的东西。
指向原始笔记的链接
If you build for that, then you can build for something that people will actually trust with real work.
—— Abhi Arya · [16:03]
因为计算机不是那样工作的，你不需要给它们打气让它们好好表现。
指向原始笔记的链接
Like it’s not the way computers work, that you need to give them a pep talk to get them to perform.
—— Nick Bostrom · [05:31]
因为我发现，通常要经历三四五次转向，才能要么做出点东西，要么你就觉得：“就这样了，没成功。”
指向原始笔记的链接
Because what I found is it takes three, four, five pivots before you either got something or you’re just like, “That’s it. Didn’t work.”
—— Josh Woodward · [16:25]
很多声称做到 RSI 的论文，当我们在我们的内部基础设施里测试时，我们发现它们不管用。
指向原始笔记的链接
A lot of the papers that claim to do RSI, when we test it in our internal infra, we find that they don’t work.
—— Daval Ajada · [14:11]
产品市场契合度不是那种你没有然后就有了的东西。它是在给定的微细分市场中你始终需要为之努力的东西。
指向原始笔记的链接
Product market fit is not something that you don’t have and then you have. It’s something that you always have to work for within given micro segments.
—— Daniel Saks · [00:00]
那就是，说到底，扩展是行得通的。但在机器人学中，扩展以及如何扩展要难得多，因为没有互联网。
指向原始笔记的链接
which is like, at the end of the day, scaling will work. But in robotics, to scale and how to scale just much harder because there is no internet.
—— Jason Ma · [42:38]
花了几个小时就超越了我们在有人参与的情况下做了两年的工作。
指向原始笔记的链接
It took a few hours to outperform the two years of work which we’d done with human in the loop.
—— David Silver · [31:02]
如果有人想去 AI 的前沿工作，我大多会告诉他们去 Anthropic 和 OpenAI，这是最容易下的注。
指向原始笔记的链接
If somebody wants to work at the frontier of AI, I mostly tell them to work at Anthropic and OpenAI, which is the easiest bet.
—— Nathan Lambert · [03:50]
在我们看来，很明显 Muse 是被基准测试刷出来的，也就是说他们训练它让它在公开基准测试上看起来非常好看，但这些提升不一定能转化到你的日常工作中。
指向原始笔记的链接
To us, it seems pretty clear that Muse is benchmarked, which means they trained it to look really good on public benchmarks, but those gains don’t necessarily translate into your everyday work.
—— 嘉宾 · [00:36]
② 出现在这些集
1 集
③ 关联
点进去有真内容 —— 本页主要出口
Tara Seshan · Lenny · OpenAI · Codex · ChatGPT · 智能体 · 知识工作 · Sites · Stripe · 写作即思考