Brendan Foody on Teaching AI and the Future of Knowledge Work (Ep. 267)
Hello everyone, and welcome back to Conversations with Tyler.
大家好，欢迎回到与泰勒对话。
Today, I’m sitting here, chatting with Brendan Foody at the offices of Mercor.
今天，我坐在这里，在 Mercor 的办公室与 Brendan Foody 聊天。
Mercor is an AI company — we’ll get into more detail soon enough — which dates from early 2023.
Mercor 是一家人工智能公司 — 我们很快就会了解更多细节 — 其历史可以追溯到 2023 年初。
Brendan is the CEO and co-founder.
布伦丹是首席执行官兼联合创始人。
I believe he’s the youngest unicorn founder ever.
我相信他是有史以来最年轻的独角兽创始人。
Mercor, by some estimates, is the fastest-growing company ever, for instance, the quickest speed to $400 million.
据估计，Mercor 是有史以来增长最快的公司，例如，最快达到 4 亿美元。
Brendan also, at age 22, is the youngest Conversations with Tyler guest ever.
22 岁的布兰登也是《与泰勒对话》有史以来最年轻的嘉宾。
My proudest achievement.
我最自豪的成就。
There’s more we’ll get to soon enough, but Brendan, welcome.
我们很快就会讲到更多内容，但布兰登，欢迎。
Thank you so much for having me, Tyler.
非常感谢你邀请我，泰勒。
Excited to be here.
很高兴来到这里。
Now, I saw an ad online not too long ago from Mercor, and it said $150 an hour for a poet.
不久前，我在网上看到了 Mercor 的一则广告，上面写着一位诗人每小时 150 美元。
Why would you pay a poet $150 an hour?
你为什么要付给诗人 150 美元每小时？
That’s a phenomenal place to start.
这是一个非凡的起点。
For background on what the company does — we hire all of the experts that teach the leading AI models.
有关公司业务的背景 ——我们聘请了所有教授领先人工智能模型的专家。
When one of the AI labs wants to teach their models how to be better at poetry, we’ll find some of the best poets in the world that can help to measure success via creating evals and examples of how the model should behave.
当其中一个人工智能实验室想要教他们的模型如何更好地诗歌时，我们会找到一些世界上最好的诗人，他们可以通过创建模型应如何表现的评估和示例来帮助衡量成功。
One of the reasons that we’re able to pay so well to attract the best talent is that when we have these phenomenal poets that teach the models how to do things once, they’re then able to apply those skills and that knowledge across billions of users, hence allowing us to pay $150 an hour for some of the best poets in the world.
我们能够支付如此高的薪水来吸引最优秀的人才的原因之一是，当我们有这些杰出的诗人教模型如何做一次事情时，他们就能够将这些技能和知识应用于数十亿用户，因此我们可以每小时为世界上一些最优秀的诗人支付 150 美元。
The poets grade the poetry of the models or they grade the writing?
诗人给模特的诗歌打分还是给写作打分？
What is it they’re grading?
他们的评分是什么？
It could be some combination depending on the project.
根据项目的不同，它可能是某种组合。
An example might be similar to how a professor in English class would create a rubric to grade an essay or a poem that they might have for the students.
一个例子可能类似于英语课上的教授如何创建一个评分标准来对他们可能为学生准备的论文或诗歌进行评分。
We could have a poet that creates a rubric to grade how well is the model creating whatever poetry you would like, and a response that would be desirable to a given user.
我们可以让一位诗人创建一个标题来对模型创建您想要的任何诗歌的效果进行评分，以及给定用户所需的响应。
How do you know when you have a good poet, or a great poet?
你怎么知道你有一个好诗人，还是一个伟大的诗人？
That’s so much of the challenge of it, especially with these very subjective domains in the liberal arts.
这就是它的巨大挑战，尤其是在文科这些非常主观的领域。
So much of it is this question of taste, where you want some degree of consensus of different exceptional people believing that they’re each doing a good job, but you probably don’t want too much consensus because you also want to get all of these edge case scenarios of what are the models doing that might deviate a little bit from what the norm is.
其中很大一部分是品味问题，你希望不同的杰出人士达成某种程度的共识，相信他们每个人都做得很好，但你可能不想要太多的共识，因为你还想获得所有这些模型正在做什么的边缘情况，这些场景可能会稍微偏离正常情况。
So, you want your poet graders to disagree with each other some amount.
所以，你希望你的诗人评分者彼此之间存在一定程度的分歧。
Some amount, exactly, but still a response that is conducive with what most users would want to see in their model responses.
确切地说，有一定数量，但仍然是有利于大多数用户希望在模型响应中看到的响应。
Are you ever tempted to ask the AI models, “How good are the poet graders?”
您是否曾想问人工智能模型：“诗人评分员有多好？”
We often are.
我们经常是这样。
We do a lot of this.
我们做了很多这样的事情。
It’s where we’ll have the humans create a rubric or some eval to measure success, and then have the models say their perspective.
在这里，我们将让人们创建一个标题或一些评估来衡量成功，然后让模型说出他们的观点。
You actually can get a little bit of signal from that, especially if you have an expert — we have tens of thousands of people that are working on our platform at any given time.
实际上，你可以从中得到一些信号，特别是如果你有专家的话——“在任何给定时间，我们都有数以万计的人在我们的平台上工作。”
Oftentimes, there’ll be someone that is tired or not putting a lot of effort into their work, and the models are able to help us with catching that.
通常，会有人感到疲倦或没有在工作中付出很多努力，而模型能够帮助我们捕捉到这一点。
You’ve had a recent project lately.
你最近有一个最近的项目。
You hired Larry Summers, I believe, for finance and economics.
我相信，您聘请了拉里·萨默斯（Larry Summers），负责金融和经济工作。
That was a little bit of a unique deal.
这是一个有点独特的交易。
He’s been a guest on this podcast.
他是这个播客的嘉宾。
Cass Sunstein for law.
卡斯·桑斯坦 (Cass Sunstein) 担任法律职务。
He’s been a guest twice on this podcast.
他曾两次担任该播客的嘉宾。
Eric Topol for medicine.
埃里克·托波尔（Eric Topol）从事医学工作。
I’ve been a guest on his podcast.
我是他播客的嘉宾。
How do you pick those people?
你如何挑选这些人？
Obviously, they’re highly accomplished, but what makes them good at doing this other than just being smart, productive people?
显然，他们非常有成就，但是除了成为聪明、高效的人之外，是什么让他们擅长这样做呢？
I’ll step back and provide a little bit of context on APEX, or the AI Productivity Index, and why we chose them to help with it.
我将退后一步，提供一些有关 APEX（即人工智能生产力指数）的背景信息，以及我们选择他们来帮助解决这一问题的原因。
The largest disconnect that we were seeing in AI research is that everyone was focused on academic evals, like GPQA for PhD-level reasoning or IMO for Olympiad math, which were wholly disconnected from the outcomes that customers actually care about, of how do we get the model to automate a medical diagnosis or a legal draft?
我们在人工智能研究中看到的最大脱节是，每个人都专注于学术评估，例如博士级推理的 GPQA 或奥林匹克数学的 IMO，这与客户真正关心的结果完全脱节，即我们如何获得模型来自动化医疗诊断或法律草案？
Or preparing a certain financial analysis of a company.
或者准备公司的某些财务分析。
So, we chose legal experts, medical experts, finance experts — people that have a broad economic perspective to see what is the right methodology to think about measuring success across each of these domains.
因此，我们选择了法律专家、医学专家、金融专家——具有广泛经济视角的人来了解衡量每个领域成功的正确方法。
Working with them on segmenting: What are all the different industries within law?
与他们合作进行细分：法律范围内有哪些不同的行业？
What are all the different types of law?
法律有哪些不同类型？
How do we leverage our marketplace of all of these experts to best capture and measure how well models have automated all of those domains?
我们如何利用所有这些专家的市场来最好地捕获和衡量模型对所有这些领域的自动化程度？
So, it’s because they’ve had real-world experience, and they’re not only academics?
那么，是因为他们有现实世界的经验，而且他们不仅仅是学者？
Is that the way to think about it?
是这样思考的吗？
I think that’s part of it.
我认为这是其中的一部分。
I think a lot of them obviously have meaningful real-world experience, but also this broad vantage point of the entire industry, of not just someone that specializes in a particular type of law or a particular industry in big law, but rather having this very large perspective in how we should structure the project, how we should think about the rigorous processes associated with curating the datasets, setting up the reviews, et cetera.
我认为他们中的很多人显然拥有有意义的现实世界经验，而且对整个行业具有广泛的优势，不仅仅是专门研究特定类型的法律或大型法律中的特定行业的人，而且在我们应该如何构建项目、我们应该如何考虑与管理数据集、设置评论等相关的严格流程方面拥有非常大的视角。
On measuring real-world AI progress
关于衡量现实世界的人工智能进展
The paper you did with that group of people as your researchers — and many others, I should add — what’s the main thing you all learned from that exercise and from the paper?
你和那群人作为你的研究人员一起写的论文 ——以及许多其他人，我应该补充一点—— 你们从这次练习和论文中学到的主要内容是什么？
I think the largest takeaway is that the rate of model improvement at economically valuable tasks is incredible.
我认为最大的收获是模型在具有经济价值的任务上的改进速度令人难以置信。
If you look at the level that GPT-4o scored on this model, a frontier model a year ago, and that against GPT-5 today, the delta is profound.
如果你看看 GPT-4o 在这个模型（一年前的前沿模型）上的得分水平，以及今天与 GPT-5 的得分水平，就会发现差距是巨大的。
It often gets —
常常会得到 —
Can you put a number on that or somehow break it down for us?
你能给出一个数字或者以某种方式为我们分解一下吗？
Yes, call it a 25 percent–, 30 percent–improvement.
是的，可以称之为 25%、30% 的改进。
Per year?
每年？
Per year, exactly.
确切地说，每年。
Now, GPT-5 is at 64 percent, so maintaining that would definitely be challenging.
现在，GPT-5 的利用率为 64%，因此保持这一水平肯定具有挑战性。
It gets my mind wondering, what will this technology be able to do in another year or two?
这让我想知道，这项技术再过一两年能做什么？
And how will that have this profound impact on the economy that so many of us have been wondering about for a while?
这将如何对我们许多人长期以来一直想知道的经济产生如此深远的影响？
When you give these numbers, to what extent are you measuring how well they do on the test versus how much economic value are they creating?
当你给出这些数字时，你在多大程度上衡量他们在测试中的表现以及他们创造了多少经济价值？
I’ll walk through the methodology and how we derive that.
我将介绍该方法以及我们如何得出该方法。
Essentially, within each industry, we start out with surveys of hundreds of experts.
本质上，在每个行业内，我们都是从对数百名专家进行调查开始的。
Within consulting, we get experts that were previously at McKinsey, Bain, BCG, and other top consulting firms.
在咨询领域，我们拥有曾在麦肯锡、贝恩、BCG 和其他顶级咨询公司工作过的专家。
Then we survey, how do they spend their time?
然后我们调查一下，他们如何度过时间？
What percentage of their time is in customer meetings, is in online research, is in analysis, preparing deliverables for customers?
他们花在客户会议、在线研究、分析、为客户准备可交付成果上的时间比例是多少？
Then, within each of those buckets, we ask them to write the corresponding prompts and rubrics associated with how they spend their time.
然后，在每个桶中，我们要求他们写下与他们如何度过时间相关的相应提示和规则。
Using their time is the best proxy we have for the economic value associated with their salary, or what customers are willing to pay for, and it’s incredible to see.
利用他们的时间是我们对与他们的工资相关的经济价值或客户愿意支付的费用的最佳代理，这是令人难以置信的。
The model scoring 64 percent on that is pretty profound.
该模型在这方面的得分为 64%，相当深刻。
Obviously, there is some complexity in mapping that to economic impact because in certain industries like medicine, you can’t have a 30 percent failure rate.
显然，将其映射到经济影响存在一定的复杂性，因为在医药等某些行业，失败率不可能达到 30%。
You need to have near-perfects or similar to driverless cars in some ways.
你需要在某些方面拥有近乎完美或类似于无人驾驶汽车的产品。
In other industries, like an initial legal draft or a consulting analysis, this technology is already starting to have a profound impact, and it’s only accelerating.
在其他行业，例如最初的法律草案或咨询分析，这项技术已经开始产生深远的影响，而且只会加速。
But isn’t there something about switching from task to task which the models can’t do at all?
但是，模型根本无法完成从一个任务到另一个任务的切换吗？
So, the model would beat me on a test.
因此，该模型在测试中击败了我。
The model might even run better podcast questions than I do.
该模型甚至可能比我更好地回答播客问题。
But somehow, combining those all in a single entity, I can do.
但不知何故，将所有这些结合在一个实体中，我可以做到。
Even the best model — it’s still basically at zero as far as I can tell.
即使是最好的模型——据我所知，它仍然基本上为零。
The economic value is, in a way, still at zero?
从某种程度上来说，经济价值仍然为零？
It’s interesting.
这很有趣。
I think what you’re getting at is there’re two key things the models struggle at that humans tend to be very good at.
我认为你的意思是，模型在人类往往非常擅长的两件事上遇到了困难。
The first is these longer horizon tasks of not just something that we could do in a few hours, but something that might take us 50 or 100 hours to do.
第一个是长期任务，不仅是我们可以在几个小时内完成的任务，而且可能需要我们 50 或 100 小时才能完成的任务。
Then the second thing is integrating multiple tools with our response and going about doing these things, maybe interacting with people as one of those elements.
然后第二件事是将多种工具与我们的响应相结合并着手做这些事情，也许与人互动作为这些元素之一。
I think that that’s coming very soon.
我认为这很快就会到来。
The next version of APEX —
APEX 的下一版本——
What does “very soon” mean?
“很快”是什么意思？
Your best guess.
你最好的猜测。
I’ll talk about it in terms of APEX, and then I’ll talk about it in terms of model advancement because there’s a large correlation between the two.
我先从APEX方面讲，然后再从模型推进方面讲，因为两者之间有很大的相关性。
We’re doing a lot to measure all of those capabilities and how models interact with the entire workspace, and how models do these very long-horizon tasks in an eval that we’re launching in the next couple of months.
我们正在做很多事情来衡量所有这些功能以及模型如何与整个工作区交互，以及模型如何在我们将在接下来的几个月中启动的评估中执行这些长期任务。
Very quickly, once researchers are able to measure those capabilities, they’ll be able to help flag them.
很快，一旦研究人员能够测量这些能力，他们将能够帮助标记它们。
I would be shocked if we don’t have enormously capable models across those dimensions of lots of tool use with very long horizon tasks in the next six to twelve months.
如果我们在未来六到十二个月内没有在大量工具使用和长期任务的维度上建立强大的模型，我会感到震惊。
Let’s just take the body of knowledge alone.
让我们只考虑知识体系。
Forget about the long horizon.
忘记遥远的地平线。
Just an on-the-spot test.
只是现场测试。
Let’s say I’m Cass Sunstein.
假设我是卡斯·桑斯坦。
I know Cass.
我认识卡斯。
He has an incredibly impressive body of knowledge in many areas.
他在许多领域都拥有令人难以置信的令人印象深刻的知识体系。
When are we at the point where, basically, Cass cannot ask a question that the best models cannot answer?
我们什么时候才能基本上达到 Cass 无法提出最佳模型无法回答的问题的程度？
Wow, that’s an interesting question.
哇，这是一个有趣的问题。
I think it depends domain by domain, but in law, I think it’s going to be a long time.
我认为这取决于各个领域，但在法律上，我认为这将需要很长时间。
The reason is that there’s so much taste involved in legal responses that effectively getting all of the taste that Cass has into the model is going to be difficult.
原因是法律回应涉及太多品味，因此很难将 Cass 的所有品味有效地融入到模型中。
I do think we’ll very quickly get to the point where Cass has a really hard time finding a mistake the model makes, where he has to spend maybe a week just trying to probe it.
我确实认为我们很快就会达到这样的地步：卡斯很难找到模型所犯的错误，他可能需要花一周的时间来尝试探索它。
How far away is that?
那有多远？
That might be about two or three years, it wouldn’t surprise me, for a question and response.
对于一个问题和回应，这可能需要大约两三年的时间，这并不令我感到惊讶。
I would think it’s six months away, would be my guess.
我猜，我认为还有六个月的时间。
Maybe.
或许。
If he asked it 1,000 questions, I think he could induce an error.
如果他问它 1000 个问题，我认为他可能会引发错误。
Fifty questions — I think, in less than a year, we might be there.
五十个问题 — 我想，不到一年的时间，我们可能就能实现这一目标。
It depends also a little bit on how tightly you define an error.
它还在一定程度上取决于您定义错误的严格程度。
He might have all sorts of knowledge of niche areas of the law that the model isn’t strong at.
他可能拥有该模型不擅长的法律领域的各种知识。
There’s some question of how you measure this, but I hold Cass in very high regard with respect to his niche knowledge of the law and ability to stump the models.
有一些关于如何衡量这一点的问题，但我非常尊重卡斯，因为他对法律的专业知识和挑战模型的能力。
What would be an area where the human expert is relatively strong and an area where the human expert, compared to the model, is relatively weak?
与模型相比，人类专家相对较强的领域和人类专家相对较弱的领域是什么？
There’re interestingly a lot of areas in law where the right way of approaching something is not written down or codified.
有趣的是，法律中有很多领域没有记录或编纂处理某事的正确方法。
It exists more in the heads of experts — at least not explicitly.
它更多地存在于专家的头脑中——至少不明确地存在。
I think it’s those domains where there’s a lot of taste that isn’t well-documented that the models will struggle immensely with, because they either need those tokens in the pre-training data of doing these web-scale training runs.
我认为，正是那些有很多味道没有得到充分记录的领域，模型将面临巨大的困难，因为它们要么需要在进行这些网络规模训练运行的预训练数据中使用这些标记。
Or they need it in the post-training data, of having a legal expert from us to create those datasets.
或者他们需要训练后数据，需要我们的法律专家来创建这些数据集。
If they don’t have those, then the model will inevitably struggle with that particular problem.
如果他们没有这些，那么该模型将不可避免地遇到该特定问题。
Now, I’ve argued in economics that the leading economics journals should take their referee reports and the submissions and send them somewhere, arguably here.
现在，我在经济学中主张，领先的经济学期刊应该将他们的审稿报告和提交的材料发送到某个地方，可以说是这里。
Would that be useful to you?
这对你有用吗？
It certainly would.
肯定会的。
It’s something we’ve talked about a bunch in the past.
这是我们过去多次讨论过的事情。
I think that the largest way that these deep domain experts can help to contribute to the advancement of AI is defining the evals.
我认为这些深度领域专家可以帮助人工智能进步的最大方式是定义评估。
When we have these phenomenal tests for model capabilities — whether in economics, law, or other domains — it’s amazing how fast the researchers can hill climb them and optimize around them.
当我们对模型能力进行这些惊人的测试时——“无论是在经济学、法律还是其他领域”——“研究人员爬坡并围绕它们进行优化的速度是惊人的。
So, more help in building these tests and sending them to us and other labs is extremely impactful.
因此，在构建这些测试并将其发送给我们和其他实验室方面提供更多帮助是非常有影响力的。
Those are nonprofits, those institutions.
这些是非营利组织，那些机构。
Why don’t they just send it to you now for free?
他们为什么不现在免费发送给您呢？
Do you have a theory of this?
你有这方面的理论吗？
I’m not sure exactly.
我不太确定。
It would improve science, right?
这会改善科学，对吗？
It would improve science.
它将改善科学。
I think maybe two things.
我想也许有两件事。
One is awareness of this.
一是对此的认识。
I think that while evals are the thing that everyone’s talking about in Silicon Valley and the AI labs, it feels like most people in the rest of the country couldn’t quite describe exactly why you need an eval.
我认为，虽然评估是硅谷和人工智能实验室中每个人都在谈论的事情，但感觉美国其他地方的大多数人都无法准确描述为什么需要评估。
I think the second is a little bit of fear, where everyone worries about how is AI going to impact their jobs, their work, their ability to contribute to the economy, and be meaningful.
我认为第二个是有点恐惧，每个人都担心人工智能将如何影响他们的工作、他们的工作、他们为经济做出贡献的能力，以及他们的意义。
I think that that’s always top of mind, even for nonprofit organizations that want to contribute and preach this world of abundance.
我认为这始终是首要考虑因素，即使对于想要贡献和宣扬这个丰富世界的非营利组织也是如此。
On why rubrics are the new oil
为什么标题是新的石油
Let’s say we took the live economics or legal, whatever seminars — let’s say the top 10, top 20 schools — recorded them all, somehow anonymized the data, but you had the comments in transcript, and sent that to you.
假设我们参加了现场经济学或法律课程，无论研讨会 ——假设前 10 名、前 20 名学校 ——都将它们记录下来，以某种方式匿名化了数据，但你有记录中的评论，并将其发送给你。
Would that be useful?
那会有用吗？
It would be very useful.
这将非常有用。
One thing I will say, though, is that there are two kinds of data, is a good way of thinking about it.
不过，我要说的一件事是，有两种数据，这是一种很好的思考方式。
The first kind of data is just the output, where you have some curriculum that the model is reading and learning from.
第一种数据只是输出，其中有模型正在阅读和学习的一些课程。
The second kind of data is some way of measuring success, where you have the rubric for the response, you have the test question and answer, you have the unit test, and code.
第二种数据是衡量成功的某种方式，其中有响应的标题，有测试问题和答案，有单元测试和代码。
That second kind of data is the most valuable, where we’re able to have the models attempt the problem many, many times, score those responses, and learn from them, but both are incredibly impactful and things we would love to get support with.
第二种数据是最有价值的，我们可以让模型多次尝试解决问题，对这些响应进行评分，并从中学习，但这两种数据都具有令人难以置信的影响力，也是我们希望获得支持的东西。
On your wish list — just to make this more concrete — you can have some kind of data.
在你的愿望清单上 — 只是为了让这个更具体 — 你可以拥有某种数据。
Forget about realism, you just get it for free.
忘记现实主义吧，你只需免费获得它。
What is it you most want?
你最想要的是什么？
Let’s say social sciences, forget about realism.
就说社会科学吧，忘掉现实主义吧。
Interesting.
有趣的。
Forget about realism, social sciences I think that we tend to focus a lot on what’s economically valuable.
忘掉现实主义和社会科学吧，我认为我们倾向于更多地关注具有经济价值的事物。
If people have tests that the models are bad at, that map to a meaningful amount of economic value — and it could be an academic domain that can be applied to create a lot of value in other areas — that’s super exciting for us.
如果人们对模型不擅长的测试进行映射，得出有意义的经济价值 ——并且它可能是一个可以应用于其他领域创造大量价值的学术领域 ——这对我们来说非常令人兴奋。
Maybe a good heuristic is, if we could build a model that, without seeing this test and reading through it, could max out the test, how much economic impact would that add?
也许一个很好的启发是，如果我们可以建立一个模型，在没有看到这个测试并通读它的情况下，可以最大化测试，这会增加多少经济影响？
Whatever test is able to measure that the best is most helpful.
无论什么测试都能衡量出最好的才是最有帮助的。
Maybe in medicine, it’s a test around how well the model is doing a certain diagnosis in a particularly difficult domain where we think the models can add a ton of impact.
也许在医学上，这是对模型在一个特别困难的领域中进行某种诊断的效果的测试，我们认为模型可以在这个领域中增加大量的影响。
Maybe in economics, it’s areas of analysis and modeling of businesses that aren’t well codified but could meaningfully impact the way that we underwrite businesses.
也许在经济学中，这是对企业进行分析和建模的领域，虽然没有得到很好的编纂，但可能会对我们承保企业的方式产生有意义的影响。
Those types of things are what’s going through my head.
这些都是我脑子里想的事情。
Let’s say it’s poetry.
就说它是诗吧。
Let’s say you can get it for free, grab what you want from the known universe.
假设你可以免费获得它，从已知的宇宙中获取你想要的东西。
What’s the data that’s going to make the models, working through your company, better at poetry?
有哪些数据可以让你的公司的模型更擅长诗歌创作？
I think that it’s people that have phenomenal taste of what would users of the end products, users of these frontier models want to see.
我认为人们对最终产品的用户、这些前沿模型的用户想要看到的东西有非凡的品味。
Someone that understands that when a prompt is given to the model, what is the type of response that people are going to be amazed with?
有人明白，当向模型提供提示时，人们会惊讶于什么类型的响应？
How we define the characteristics of those responses is imperative.
我们如何定义这些反应的特征至关重要。
Probably more than just poets that have spent a lot of time in school, we would want people that know how to write work that gets a lot of traction from readers, that gains broad popularity and interest, drives the impact, so to speak, in whatever dimension that we define it within poetry.
可能不仅仅是在学校度过了很多时间的诗人，我们希望人们知道如何写出能够吸引读者的作品，获得广泛的受欢迎程度和兴趣，推动影响，可以这么说，无论我们在诗歌中定义什么维度。
But what’s the data you want concretely?
但你具体想要的数据是什么？
Is it a tape of them sitting around a table, students come, bring their poems, the person says, “I like this one, here’s why, here’s why not.” Is it that tape or is it written reports?
是一盘他们围坐在桌子旁的磁带，学生们过来，带来他们的诗，那个人说，“我喜欢这个，这就是为什么，这就是为什么不。”是那盘磁带还是书面报告？
What’s the thing that would come in the mail when you get your wish?
当你实现愿望时，邮件中会收到什么东西？
The best analog is a rubric.
最好的类比是标题。
If you have some —
如果你有一些 —
A rubric for how to grade?
如何评分的标准？
A rubric for how to grade.
关于如何评分的标题。
If the poem evokes this idea that is inevitably going to come up in this prompt or is a characteristic of a really good response, we’ll reward the model a certain amount.
如果这首诗唤起了这个提示中不可避免地出现的想法，或者是一个非常好的回应的特征，我们将奖励模型一定的金额。
If it says this thing, we’ll penalize the model.
如果它说这样的话，我们将惩罚该模型。
If it styles the response in this way, we’ll reward it.
如果它以这种方式做出回应，我们将奖励它。
Those are the types of things, in many ways, very similar to the way that a professor might create a rubric to grade an essay or a poem.
这些类型的事情在很多方面都与教授创建评分标准来对论文或诗歌进行评分的方式非常相似。
Poetry is definitely a more difficult one because I feel like it’s very unbounded.
诗歌绝对是更难的一门，因为我觉得它非常无界。
With a lot of essays that you might grade from your students, it’s a relatively well-scoped prompt where you can probably create a rubric that’s easy to apply to all of them, versus I can only imagine in poetry classes how difficult it is to both create an accurate rubric as well as apply it.
你可能会对学生的很多论文进行评分，这是一个相对范围较广的提示，你可以创建一个易于应用于所有学生的评分标准，而我只能想象在诗歌课上创建一个准确的评分标准并应用它是多么困难。
The people that are able to do that the best are certainly extremely valuable and exciting.
能够做到最好的人无疑是非常有价值和令人兴奋的。
To get all nerdy here, Immanuel Kant in his third critique, Critique of Judgment, said, in essence, taste is that which cannot be captured in a rubric.
为了让所有书呆子都在这里，伊曼努尔·康德在他的第三篇评论《判断力批判》中说，本质上，品味是无法用标题来捕捉的。
If the data you want is a rubric and taste is really important, maybe Kant was wrong, but how do I square that whole picture?
如果你想要的数据是一个标题并且品味真的很重要，也许康德是错的，但我如何调整整个图片呢？
Is it, by invoking taste, you’re being circular and wishing for a free lunch that comes from outside the model, in a sense?
从某种意义上说，通过援引品味，你是否在循环并希望获得来自模型之外的免费午餐？
There are other kinds of data they could do if it can’t be captured in a rubric.
如果无法在标题中捕获，他们还可以使用其他类型的数据。
Another kind is RLHF, where you could have the model generate two responses similar to what you might see in ChatGPT, and then have these people with a lot of taste choose which response they prefer, and do that many times until the model is able to understand their preferences.
另一种是 RLHF，您可以让模型生成两个类似于您在 ChatGPT 中看到的响应，然后让这些有很多品味的人选择他们喜欢的响应，并多次这样做，直到模型能够理解他们的偏好。
That could be one way of going about it as well.
这也可能是解决这个问题的一种方法。
On enshrining taste in LLMs
关于法学硕士的品味
I’m sure you know these studies where there’re some AI-generated poems and some human-generated poems.
我相信你知道这些研究，其中有一些人工智能生成的诗歌和一些人类生成的诗歌。
Often, the humans prefer the AI-generated poems, even though, to people with “taste,” they’re worse.
通常，人类更喜欢人工智能生成的诗歌，尽管与有“品味”的人相比，它们更糟糕。
Whose side do you take there?
你站在哪一边？
It depends what you’re optimizing for.
这取决于您要优化的目标。
I think that generally, we’re in the mindset of, for the power users of these AI products, what are the types of responses that they would want to see and be happy with?
我认为，一般来说，我们的心态是，对于这些人工智能产品的高级用户来说，他们希望看到并满意的响应类型是什么？
But it’s challenging because that sometimes deviates from the types of responses that the top 1 percent of experts in poetry might say is a broadly good poem.
但这是具有挑战性的，因为这有时会偏离前 1% 的诗歌专家可能认为是一首总体上好诗的回应类型。
So, striking that balance is really up to a lot of the researchers and product leaders at the labs.
因此，实现这种平衡实际上取决于实验室的许多研究人员和产品负责人。
What do they think good looks like?
他们认为好看是什么样的？
And how do we act as their partner in defining that?
我们如何作为他们的合作伙伴来定义这一点？
If you could model a much older poet — William Wordsworth, Blake, John Milton, Rilke — some of my friends say there are no truly great poets left anymore.
如果你可以模仿一位更老的诗人——威廉·华兹华斯、布莱克、约翰·弥尔顿、里尔克——我的一些朋友说，真正伟大的诗人已经不复存在了。
The best poets were way back when.
最好的诗人很久以前就出现了。
Is it a goal to model the older poets and figure out what they would think, and rather than having Larry Summers and Cass Sunstein come in, that you have some AI-generated model of John Milton?
你的目标是模拟老诗人并弄清楚他们会怎么想，而不是让拉里·萨默斯和卡斯·桑斯坦加入，而是让人工智能生成约翰·弥尔顿的模型？
Maybe.
或许。
I will say it ties back to the goal of APEX, which is that we saw people were too focused on a lot of these purely academic domains and not focusing enough on how will people actually use the models and the economy.
我想说这与 APEX 的目标有关，即我们看到人们过于关注许多纯学术领域，而没有足够关注人们实际如何使用模型和经济。
But I certainly do think that, especially as we start to automate more industries and there’re more liberal arts and these kinds of domains where people want to spend time on poetry, certainly building the tools to help them create phenomenal poems and make them happy and their readers happy is definitely the way we would go about it.
但我确实认为，特别是当我们开始实现更多行业的自动化，并且有更多的文科和人们想要花时间在诗歌上的领域时，当然，构建工具来帮助他们创作非凡的诗歌，让他们快乐，让他们的读者快乐，这绝对是我们要做的方式。
I’m not sure if it would be using the archetypes of these former poets.
我不确定它是否会使用这些前诗人的原型。
How would you go about it, Tyler?
泰勒，你会怎么做？
I don’t know.
我不知道。
I don’t trust contemporary poets, frankly.
坦白说，我不信任当代诗人。
There aren’t many of them I like to read.
我喜欢读的书并不多。
Maybe Geoffrey Hill would be one.
也许杰弗里·希尔就是其中之一。
Some are too postmodern.
有些太后现代了。
Some maybe are too woke.
有些人可能太醒了。
Some are too identity-driven.
有些过于以身份为导向。
I love older poetry, so it’s not that I don’t like poetry.
我喜欢古诗，并不是不喜欢古诗。
I worry about putting them — they’re not quite in charge, I get that, but giving them so much leeway.
我担心让他们负责——我明白，他们并没有完全负责，但给他们很大的余地。
It does evoke this really interesting idea of how we want to teach models and measure the success of these models.
它确实唤起了我们如何教授模型并衡量这些模型的成功的一个非常有趣的想法。
Is it via consensus?
是通过共识吗？
Is it via a handful of the top experts in that given domain?
是通过该领域的少数顶尖专家进行的吗？
There’s really no correct answer.
确实没有正确答案。
I think that different AI labs, different researchers will go down different routes, and that will frame the ways that these products feel and the things that they ultimately achieve.
我认为不同的人工智能实验室、不同的研究人员会走不同的路线，这将塑造这些产品的感觉以及它们最终实现的目标。
Maybe we should only enshrine the current age when the current age is at a peak.
也许我们应该只在当前年龄处于巅峰时才供奉当前年龄。
Scott Sumner says the best movies were maybe made in the 1960s and ’70s.
斯科特·萨姆纳 (Scott Sumner) 表示，最好的电影可能是 20 世纪 60 年代和 70 年代制作的。
Whether or not you agree, you could have movie evaluators be only from that time.
无论你是否同意，你都可以让电影评价者只从那个时候开始。
There are some still alive.
还有一些还活着。
If you think the best heavy metal, say, comes from the 1980s, well, you wouldn’t have the current evaluators.
比如说，如果你认为最好的重金属音乐来自 20 世纪 80 年代，那么你就不会有现在的评价者了。
You would pick evaluators from the ’80s.
你会选择 80 年代的评估者。
The best poetry does seem, by most people’s standards, to be really quite old, and we can’t resurrect those individuals.
按照大多数人的标准，最好的诗歌确实看起来非常古老，我们无法复活这些人。
But the notion that you enshrine current taste when taste changes so much — it’s a very interesting decision.
但是，当品味发生如此大的变化时，你要奉行当前的品味，这是一个非常有趣的决定。
It certainly is.
确实如此。
My guess is that in a long enough time horizon, we’ll enshrine taste from every different decade and every different era.
我的猜测是，在足够长的时间范围内，我们将铭记每个不同十年和每个不同时代的品味。
Then the model will be able to learn what taste do you have and how does it pull on each of those knowledge bases to best personalize it to your preferences.
然后，该模型将能够了解您的品味，以及它如何利用每个知识库来最好地根据您的喜好进行个性化。
On turning society into one giant RL machine
将社会变成一台巨型强化学习机器
How much of society, ideally, should become a big reinforcement learning machine?
理想情况下，社会中有多少人应该成为一台大型强化学习机器？
We tape everyone, everything, every debate people have over the coffee table.
我们记录了人们在咖啡桌上进行的每一个人、每件事、每一次辩论。
I think it will become an immense amount very quickly.
我认为它很快就会变得巨大。
There’s obviously still going to be the personal conversations over the coffee table that people don’t want recorded.
显然，咖啡桌上仍然会有人们不希望被记录的私人谈话。
But my firm belief is, especially for economically valuable tasks, we’ll move towards a world where people do things once.
但我坚信，特别是对于经济上有价值的任务，我们将走向一个人们只做一次事情的世界。
Instead of the investment banker redundantly analyzing a data room to prepare an analysis of a company every couple of weeks for a new project and a new customer, they’ll teach the model how to do that once in the particular domains that they operate in.
投资银行家不再冗余地分析数据室，每隔几周为新项目和新客户准备一次公司分析，而是教模型如何在他们运营的特定领域中做到这一点。
Similar to building software once, they’ll be able to use that many times as they use their agent.
与构建软件一次类似，他们可以在使用代理时多次使用该软件。
Instead of the customer support rep monotonously responding to tickets every day, they’ll find the mistake that the agent makes.
他们不会让客户支持代表每天单调地回复票证，而是会发现客服人员所犯的错误。
They’ll turn that into an RL environment, and then, all of a sudden, the agent will be able to solve that problem many times.
他们会将其转变为强化学习环境，然后，突然间，智能体将能够多次解决该问题。
I think, in many ways, the economic incentives and how knowledge work will change has a lot of similarities to software, and that we’ll move towards these fixed-cost investments of teaching an agent how to do something, building an RL environment for something, and then being able to use agents as many times as we want to perform that activity.
我认为，在很多方面，经济激励和知识工作的变化与软件有很多相似之处，我们将转向这些固定成本投资，教代理如何做某事，为某事构建 RL 环境，然后能够根据需要多次使用代理来执行该活动。
That’s why I believe that a huge portion of the economy will become an RL environment machine.
这就是为什么我相信经济的很大一部分将成为强化学习环境机器。
Do you think pendants or Meta-like glasses will be more important than that?
您认为吊坠或类似 Meta 的眼镜会比这更重要吗？
Are we going to do both?
我们要两者都做吗？
I think a lot of both.
我认为两者都有很多。
If I take myself, I don’t do that much small talk.
如果我自己的话，我不会说那么多闲话。
Say you attach the little pendant to me, and you got the tape of all my conversations, you could feed it in.
假设你把小吊坠戴在我身上，你得到了我所有谈话的录音带，你可以把它放进去。
What’s the social value of that?
这有什么社会价值？
Is it $5, $50, a bit more?
是 5 美元、50 美元还是多一点？
How valuable is that?
那有多大价值？
It certainly depends a lot by person.
这当然很大程度上取决于个人。
I would imagine yours are quite valuable.
我想你的东西一定很值钱。
Quite?
相当？
How much would you —
你愿意花多少钱 —
Could be a good side hustle.
可能是一个很好的副业。
I’m not asking for an offer, but how much, actually, would you pay?
我不是在要求报价，但实际上，您愿意支付多少钱？
I would pay a lot just out of pure curiosity.
出于纯粹的好奇心，我愿意花很多钱。
If I were trying to think about how valuable it would be to our customers and our business, I imagine it would be something in the order of — it’s hard because it changes over time — certainly tens of thousands, if not many hundreds of thousands of dollars a year, and how that evolves over time.
如果我试图思考它对我们的客户和我们的业务有多大价值，我想它会是这样的：“这很困难，因为它会随着时间的推移而变化”——“每年肯定有数万甚至数十万美元，以及随着时间的推移如何变化。”
But my guess is that for the vast majority of people, they’ll still care a lot about privacy, so maybe that data will be collected to personalize their individual agent, but they’re not going to be comfortable with that getting added to the broader model weights to customize the base model that billions of users are —
但我的猜测是，对于绝大多数人来说，他们仍然非常关心隐私，因此也许会收集这些数据来个性化他们的个人代理，但他们不会接受将其添加到更广泛的模型权重中以定制数十亿用户的基本模型 -
But that’s easy.
但这很容易。
You can tape me with my pendant.
你可以用我的吊坠把我粘起来。
I run it through my AI, and I say, “Take out anything I don’t want Mercor to hear.” And it will do that quite well, maybe not perfectly.
我通过人工智能运行它，然后说：“去掉我不想让 Mercor 听到的任何内容。”它会做得很好，但也许并不完美。
Then you get what’s left over: all the debates about elasticities and tax incidence.
然后你就得到了剩下的：所有关于弹性和税收归属的争论。
Maybe.
或许。
I suspect you’re probably more comfortable with it than most people.
我怀疑你可能比大多数人更适应它。
Most people would probably say, “Well, you’re asking the AI to be the layer of trust to remove the sensitive information, but it’s going to have a bias in doing so.”
大多数人可能会说，“好吧，你要求人工智能成为删除敏感信息的信任层，但这样做会有偏见。”
I think there’s always going to be some level of sensitivity around these topics.
我认为围绕这些话题总会有一定程度的敏感性。
I actually believe that some of the companies that have done a very good job around their brand of privacy are going to have an advantage in it.
事实上，我相信一些在隐私品牌方面做得很好的公司将会在这方面占据优势。
Apple — while maybe not totally at the frontier of AI yet — has done such a good job in their brand around privacy.
苹果——虽然可能还没有完全走在人工智能的前沿——在隐私方面的品牌做得非常好。
That’s going to allow them to have a lot of trust from users in a way that they’re able to collect all of this personalized information.
这将使他们能够收集所有这些个性化信息，从而获得用户的高度信任。
On when AI will stump experts
人工智能何时会难倒专家
Let’s say three to five years out, when the top models will be both clearly better than virtually all human experts or maybe all human experts and recognized as such — the latter we certainly don’t have — what do you think, in that world, the reputation of expertise is like?
假设三到五年后，顶级模型将明显优于几乎所有人类专家或可能所有人类专家，并得到认可 ——“后者我们当然没有”—— 你认为在那个世界里，专业知识的声誉是什么样的？
One view is no one respects the experts because the machines are better, but I think an alternative possibility is the machine, by not being tied to a personality, is less disliked.
一种观点是没有人尊重专家，因为机器更好，但我认为另一种可能性是机器，由于不与个性联系在一起，因此不那么不受欢迎。
People actually respect the experts more because they get this impersonal distillation of the experts.
人们实际上更尊重专家，因为他们得到了专家的客观升华。
It’s like, “Oh, the experts did that.
就像，“哦，专家做到了。
They’re so amazing.
他们太棒了。
They’re not annoying me like on the late-night TV show.” What will happen to the status of human experts?
他们不像深夜电视节目中那样烦我。”人类专家的地位将会发生什么变化？
I think so.
我想是的。
I think that I definitely am already at the point where there are certain domains where I trust ChatGPT or whatever model I’m using more than I trust a particular expert in that industry for a very quick medical perspective even in some cases, or whatever it is.
我认为我肯定已经到了这样的地步：在某些领域，我对 ChatGPT 或我正在使用的任何模型的信任程度超过了我对该行业某个特定专家的信任程度，即使在某些情况下，也能获得非常快速的医学观点。
I think that there’s some element of it being highly competent.
我认为其中有一些要素是非常有能力的。
There’s some element of it not having a face to it that causes us to place this high trust.
其中有一些不露面的因素导致我们给予如此高度的信任。
But I do think that the point you made at the beginning is around evoking the question of what is the point at which these models will be able to do everything that experts aren’t able to do.
但我确实认为你一开始提出的观点是围绕着引发这样一个问题：这些模型何时能够做到专家无法做到的一切。
My read on the market is that models are advancing very, very quickly in being able to automate — call it 50 percent or 75 percent of what humans and experts are able to do — but will really struggle with that last 25 percent.
我对市场的了解是，模型在自动化方面进展得非常非常快——可以说是人类和专家能够做到的 50% 或 75%——但在最后 25% 方面确实会很困难。
I think that for a very long time, human expertise will be imperative to help accomplish that last 25 percent as the ultimate bottleneck to more economic prosperity and productivity.
我认为，在很长一段时间内，人类的专业知识对于帮助实现最后 25% 至关重要，这是实现经济繁荣和提高生产力的最终瓶颈。
How long until the best models can write a poem as good as the median Pablo Neruda poem?
还要多久，最好的模特才能写出一首与中等的巴勃罗·聂鲁达诗一样好的诗？
Oh, I think that’s probably not too far off.
哦，我想那可能不会太遥远。
I think the models —
我认为模型 —
I would say less than a year.
我想说不到一年。
When you say —
当你说 —
Yes, less than a year.
是的，不到一年。
Less than a year.
不到一年。
How about the very best Pablo Neruda poems?
巴勃罗·聂鲁达最好的诗歌怎么样？
I’m not too calibrated on poetry, so I’d have a hard time saying.
我不太懂诗歌，所以很难说。
I think it’s much further out.
我认为这还很遥远。
Is that your intuition?
这是你的直觉吗？
I agree.
我同意。
I think that’s consistent with my intuition as well.
我认为这也符合我的直觉。
I think that this longer tail of advancement is generally the most difficult.
我认为这种较长的进步尾巴通常是最困难的。
The other heuristic I have for it is that, going back to this dimension of the time horizon of the task, models are in some ways superhuman with what you can do in a chat window with your chatbot, but they still can’t draft an email for us.
我的另一个启发是，回到任务时间范围的这个维度，模型在某种程度上是超人的，你可以在聊天窗口中与聊天机器人一起做些什么，但他们仍然无法为我们起草电子邮件。
They still can’t schedule a meeting.
他们仍然无法安排会议。
Those things will come, but I think that there’s a long way before we’re able to tell a model, “Go off and build a startup for 90 days.” There’s going to be an immense amount of human expertise associated with how do we get to that across every knowledge work vertical that we want the models to operate in?
这些事情终将到来，但我认为我们还有很长的路要走，才能告诉模型“用 90 天的时间建立一家初创公司。”将会有大量的人类专业知识与我们如何在我们希望模型运行的每个知识工作垂直领域实现这一点相关？
Insofar as we turn society into this big engine for reinforcement learning, what new jobs get created by doing that?
就我们将社会变成强化学习的巨大引擎而言，这样做会创造哪些新的就业机会？
I think the most interesting part of our business is that everyone else in Silicon Valley is talking about how we automate away jobs, versus we’re very focused on how do we build this new job category of people training agents, building RL environments to help teach models.
我认为我们业务中最有趣的部分是，硅谷的其他人都在谈论我们如何自动化工作，而我们非常关注如何构建人员培训代理的新工作类别，构建 RL 环境来帮助教授模型。
That’s what I believe it’ll converge to.
这就是我相信它会收敛到的。
Instead of the investment bankers doing the analysis, they’ll build RL environments and train agents.
他们将构建强化学习环境并培训代理，而不是由投资银行家进行分析。
It’ll be the same across consulting and software engineers, and customer support and pretty much every knowledge work vertical.
咨询和软件工程师、客户支持以及几乎所有垂直领域的知识工作都是如此。
It’s hard to say the exact pace at which that’ll happen, but I would not be surprised if, within five years, a majority of high-end knowledge workers are training models, whether in their full-time jobs or through our marketplace, to help improve agents at whatever workflows they want to automate.
很难说这种情况发生的确切速度，但如果在五年内，大多数高端知识工作者都在培训模型，无论是在他们的全职工作中还是通过我们的市场，以帮助改进座席他们想要自动化的任何工作流程，我不会感到惊讶。
On AI and employment
论人工智能与就业
To hold those jobs, how much technical AI will a person need to have?
为了保住这些工作，一个人需要拥有多少技术人工智能？
Or do they just have to know about the thing?
或者他们只需要知道这件事？
They just need to know about the thing.
他们只需要了解这件事。
The only element of technical AI that they’ll need is to find where the model makes a mistake.
他们需要的技术人工智能的唯一要素是找到模型出错的地方。
As long as they can find where the model makes a mistake and understand, in some ways, the frontier of the model and its capabilities — how you can push it to its limit — then it’s relatively easy to create some criteria or way of measuring that mistake so that the model can learn from it.
只要他们能够找到模型在哪里犯了错误，并在某种程度上理解模型的前沿及其能力——“如何将其推向极限”——那么就相对容易创建一些标准或方法来衡量该错误，以便模型可以从中学习。
I think we’ll have that across every different vertical, with every different tool, with these very long horizons, whether it’s 100 hours or 100 days that we want the model to work on something.
我认为我们将跨越每个不同的垂直领域，使用每种不同的工具，以及非常长的视野，无论是 100 小时还是 100 天，我们希望模型能够处理某些事情。
That’s going to very quickly become the primary bottleneck to model improvement.
这将很快成为模型改进的主要瓶颈。
Is the demand for software price elastic?
软件价格的需求是否有弹性？
I think it’s extremely price elastic.
我认为它的价格弹性非常大。
In fact, I think that the elasticity is the exact right thing to home in on with respect to how job displacement will evolve in these domains.
事实上，我认为弹性对于这些领域的工作替代将如何演变来说是完全正确的。
I think if we make software engineers 10 times more efficient, we’ll have even more software engineers.
我认为，如果我们让软件工程师的效率提高 10 倍，我们就会拥有更多的软件工程师。
Maybe we’ll have 10 times as many software engineers and build 100 times as much software, right?
也许我们将拥有 10 倍的软件工程师并构建 100 倍的软件，对吗？
Versus other domains, maybe that’s not the case.
与其他领域相比，情况可能并非如此。
Maybe we only need so much accounting in the world, or we only need so much customer support, but I think software engineers certainly will be able to do so much more.
也许我们在世界上只需要这么多的会计，或者我们只需要这么多的客户支持，但我认为软件工程师肯定能够做更多的事情。
Where else do you think of as price elastic?
您认为还有哪些地方具有价格弹性？
I think that building businesses is also.
我认为建立企业也是如此。
A lot of the product and distribution associated with software is certainly going to be something we see a lot more of.
我们肯定会看到更多与软件相关的产品和发行版。
I think there’re a lot of domains.
我认为有很多领域。
Even if you think about investing — obviously, it’s not as price elastic as software, but I do think that there’s still enormous inefficiency with respect to how we allocate capital in the economy.
即使你考虑投资——显然，它的价格弹性不如软件，但我确实认为我们在经济中分配资本的方式仍然存在巨大的低效率。
If I think back to the early days of Mercor, we were having a hard time getting our $10,000 of working capital for our initial seed investments.
回想 Mercor 成立初期，我们很难获得 10,000 美元的营运资金来进行初始种子投资。
Then very quickly, once you get to a reasonable scale, the markets are very, very capitalized.
然后很快，一旦达到合理的规模，市场就会变得非常非常资本化。
I think a lot of this early capital allocation, as well as even just better understanding how companies will develop over time, is going to be really interesting.
我认为很多早期的资本配置，甚至只是更好地了解公司将如何随着时间的推移发展，都将非常有趣。
Also, how that information and analysis manifests itself within companies.
此外，这些信息和分析如何在公司内部体现。
For an operator — they have this investing problem of what are all the different bets that they have within their companies?
对于运营商来说，他们面临着这样的投资问题：他们在公司内有哪些不同的赌注？
How do they allocate capital and resources associated with that?
他们如何分配与之相关的资本和资源？
I think that there’s so much elasticity with respect to how we build more products, how we distribute those products, and how we allocate resources within companies more effectively.
我认为，我们如何生产更多产品、如何分销这些产品以及如何更有效地在公司内部分配资源，都具有很大的弹性。
What will education look like five to ten years out?
五到十年后的教育会是什么样子？
I think education is one of the things I’m most excited about, where a good heuristic is if everyone has Sal Khan as their personal tutor, available 24/7 to teach them whatever topic they want to learn.
我认为教育是我最兴奋的事情之一，一个很好的启发是，如果每个人都有 Sal Khan 作为他们的私人导师，24/7 全天候教授他们想要学习的任何主题。
It’ll be that it’s much easier to motivate themselves.
这会更容易激励自己。
It’s much better access to information, much better ways of explaining that information, and that’ll be profoundly impactful.
这是更好的信息获取方式，更好的解释信息的方式，这将产生深远的影响。
But that seems less price elastic, right?
但这似乎价格弹性较小，对吗？
Yes.
是的。
Only so many hours of Sal Khan a day, no slight intended to him.
萨尔汗每天只有这么多时间，没有任何怠慢他的意思。
But it’s not going to be 27 hours a day, right?
但一天不会是27小时吧？
That’s true.
这是真的。
That’s true.
这是真的。
Employment for teachers, researchers might shrink?
教师、研究人员的就业可能会萎缩？
I think in some ways, areas of that may shrink.
我认为在某些方面，这一领域可能会缩小。
But I also think that there’s a large element to teaching that exists in personal relationships, of which the model will be able to do part, but not all of it, of how does the teacher act as guiding the student through their journey and helping them to improve both in their curriculum as well as their emotional development.
但我也认为，教学中有一个很大的因素存在于人际关系中，该模型将能够完成其中的一部分，但不是全部，即教师如何引导学生完成他们的旅程，并帮助他们提高课程和情感发展。
So, I think teachers will still play an important role in the economy, ideally, able to just provide higher touch points of contact with all the students in smaller class settings.
因此，我认为教师仍将在经济中发挥重要作用，理想情况下，能够在小班环境中为所有学生提供更高的接触点。
On why vibes-based hiring fails
为什么基于氛围的招聘会失败
This is October 2025.
这是 2025 年 10 月。
How many people work at Mercor?
Mercor 有多少人工作？
Right now, we have just over 300 people across the world as our full-time employees.
目前，我们在全球仅有 300 多名全职员工。
How did you hire so many good people so quickly?
你们是如何如此迅速地聘用这么多优秀人才的？
We used our technology and our platform to help with it a bunch.
我们利用我们的技术和平台来帮助解决这个问题。
The origin story of the company was automating all of the ways that we would review resumés, conduct interviews, and decide who to hire.
该公司的起源故事是自动化我们审查简历、进行面试和决定雇用谁的所有方式。
The ways that we assess talent, the ways that we optimize funnels to build out teams, is really ingrained in the DNA of the company and a top priority of me and my co-founders.
我们评估人才的方式，我们优化渠道以建立团队的方式，确实根植于公司的 DNA 中，也是我和我的联合创始人的首要任务。
I’m extremely grateful for everyone that we have on the team, and they make it look easy.
我非常感谢我们团队中的每个人，他们让这一切看起来很容易。
How do other people do interviews wrong?
其他人在面试中是如何做错的？
This is something we’ve talked about a bunch because you obviously wrote a phenomenal book on talent.
这是我们已经讨论过很多次的话题，因为你显然写了一本关于人才的非凡书籍。
I think one of the largest problems that people make is that they don’t measure the actual skills and capabilities that they want someone to exhibit on the job.
我认为人们犯的最大问题之一是他们没有衡量他们希望某人在工作中展示的实际技能和能力。
Instead of focusing on how do we measure how well this person does their investment analysis of the data room, they have this vibes-based conversation of where did the person grow up, how similar are they, do they think they would enjoy hanging out together.
他们没有关注我们如何衡量这个人对数据室的投资分析有多好，而是进行了基于共鸣的对话，讨论这个人在哪里长大，他们有多相似，他们是否认为他们会喜欢一起出去玩。
Obviously, that’s still important if you’re having a working relationship, but I think that they often over-index on that relative to the skills that people actually exhibit.
显然，如果你有工作关系，这一点仍然很重要，但我认为他们经常过度评价这一点，而不是人们实际表现出的技能。
Just give them a project and grade them, in essence.
实质上，只需给他们一个项目并给他们评分即可。
I think that’s the cleanest way to do it.
我认为这是最干净的方法。
Let’s say it’s not programming.
假设这不是编程。
As the company gets bigger — the major AI companies — a lot of them now are quite large, and most of the people who work there don’t do AI at all.
随着公司变得越来越大——“主要的人工智能公司”——现在很多公司规模都很大，而且大多数在那里工作的人根本不做人工智能。
They do jobs that are not so dissimilar from what they might do at Coca-Cola, which is fine.
他们所做的工作与在可口可乐的工作没有太大不同，这很好。
It’s just part of growth.
这只是成长的一部分。
They’re legal, they’re communications, they do events, whatever.
它们是合法的，它们是交流的，它们举办活动，等等。
When you’re trying to hire people like that, say, what’s the test?
当你试图雇用这样的人时，比如说，测试是什么？
What’s the project, or what is it you look for?
该项目是什么，或者您正在寻找什么？
I think that’s definitely more difficult.
我认为这肯定更困难。
You probably want to look for cases in their life where they’ve worked in similar roles because you can’t curate a project that’s as similar to exactly what they’ve done.
您可能想在他们的生活中寻找他们曾担任过类似角色的案例，因为您无法策划一个与他们所做的事情完全相同的项目。
You would see the best proxy for that, and then really drill in to understand the details of that working environment, how similar it is, how well they performed in that, talking to people that previously worked with them in that environment to get a gauge for it.
你会看到最好的代理，然后真正深入了解该工作环境的细节，它有多相似，他们在该环境中的表现如何，与以前在该环境中与他们一起工作的人交谈以对其进行衡量。
It definitely is more difficult to measure someone’s slope and how they’ll develop on the job over a six-month time horizon than it is to measure their Y-intercept.
衡量某人的斜率以及他们在六个月的时间范围内如何在工作中发展肯定比衡量他们的 Y 截距更困难。
I think that’s one trend that we’ve found in talent assessment.
我认为这是我们在人才评估中发现的一种趋势。
Do you think body language in an interview is predictive?
您认为面试中的肢体语言具有预测性吗？
I think it can be, but I also think it can be a false signal because I’ve definitely had cases where I over-index on, “Oh, this person feels a little bit awkward,” or whatever it is, but they do a phenomenal job at the actual work.
我认为可能是这样，但我也认为这可能是一个错误信号，因为我确实遇到过过度索引“哦，这个人感觉有点尴尬”之类的情况，但他们在实际工作中做得非常出色。
So, I think it’s important to be very cautious around which of these signals are actually correlated with performance and which ones aren’t.
因此，我认为重要的是要非常谨慎地了解哪些信号实际上与性能相关，哪些信号不相关。
Articulateness — overrated or underrated?
表达能力 — 被高估还是低估？
It depends a lot on the job.
这在很大程度上取决于工作。
It depends a lot on the job.
这在很大程度上取决于工作。
Let’s say 10 years from now, when we can really measure pretty well the performance of people we’re interviewing today — less than 10 years, but let’s say 10 years.
假设 10 年后，我们可以真正很好地衡量我们今天采访的人的表现 ——不到 10 年，但假设是 10 年。
Let’s say you have a company, such as Amazon, that does a very large number of interviews.
假设您有一家公司（例如亚马逊）进行大量采访。
Let’s say they’re all taped, and you run them through the best AI models.
假设它们都被录制下来，并且你通过最好的人工智能模型来运行它们。
How good a predictor do you think that will be, in your opinion?
您认为这将是一个多好的预测器？
I think that it will be certainly superhuman because humans aren’t very good at it, but it’s still such a difficult problem that there’s going to be variance.
我认为这肯定是超人的，因为人类不太擅长，但这仍然是一个困难的问题，会存在差异。
For roles like the one you described, what’s going through my head is there’re a lot of confounding variables.
对于像你所描述的角色，我的脑海中浮现出很多令人困惑的变量。
Did the person have an issue in their family that caused them to be off their game or not show up to work?
此人的家庭是否存在问题导致他们无法参加比赛或缺勤？
Did they get sick during the interview process and maybe weren’t full of energy?
他们是不是在面试过程中生病了，精神不振？
There’re all these things that just add noise to that problem, but I do believe that, as we’re able to get all of that data in context, to have all the notes from the manager around what was happening in this person’s life, both during the interview process as well as on the job, that will allow it to, over time, become phenomenal.
所有这些事情只会给这个问题增添噪音，但我确实相信，由于我们能够在上下文中获取所有这些数据，获得经理关于这个人生活中发生的事情的所有笔记，无论是在面试过程中还是在工作中，随着时间的推移，这将使它变得惊人。
Maybe we have that on a 10-year time horizon.
也许我们在 10 年的时间范围内就能实现这一目标。
On solving labor market matching problems
关于解决劳动力市场匹配问题
How can we make labor markets more efficient?
我们如何才能提高劳动力市场的效率？
I think that one of the largest inefficiencies in labor markets is that everything is disaggregated, and that when one of our friends is applying to a job, they would apply to a couple dozen jobs.
我认为劳动力市场最大的低效率之一是一切都是分类的，当我们的一个朋友申请一份工作时，他们会申请几十个工作。
When a company is considering who to hire, they’ll consider a fraction of a percent of people in the economy.
当公司考虑雇用谁时，他们会考虑经济中一小部分人。
It feels like there needs to be a structural change there, where there’s an aggregator that everyone applies to, and every company hires from facilitating this perfect flow of information.
感觉那里需要进行结构性变革，每个人都可以申请一个聚合器，每个公司都通过促进这种完美的信息流来雇用员工。
We need a very good AI for that to work?
我们需要一个非常好的人工智能才能发挥作用吗？
I think a very good AI will help with that working.
我认为一个非常好的人工智能将有助于实现这一目标。
The reason I think it doesn’t happen today is that there’s a very difficult matching problem.
我认为今天不会发生的原因是存在一个非常困难的匹配问题。
Let me give LinkedIn as an analog.
让我以 LinkedIn 作为类比。
LinkedIn has all the distribution to pretty much every company and every candidate, but at the very same time, it’s incredibly difficult to understand, based on someone’s LinkedIn profile, whether they’ll actually perform well at a given job.
LinkedIn 拥有几乎所有公司和每位候选人的所有分布，但与此同时，根据某人的 LinkedIn 个人资料，很难了解他们是否真的能在特定工作中表现出色。
I believe that, in that case, it’s very much a matching problem — less so a distribution and aggregation problem — to facilitate this effective flow of information and aggregation within knowledge markets.
我相信，在这种情况下，这很大程度上是一个匹配问题，而不是一个分配和聚合问题，以促进知识市场内信息的有效流动和聚合。
But I think it’s also in line with the fact that the nature of jobs is changing dramatically.
但我认为这也符合工作性质正在发生巨大变化的事实。
Previously, everyone would think about this problem in the context of full-time roles, but as we trend towards this world of everyone building our RL environments, and being able to do work remotely, and train models in this fractional way, that also will shift the dynamics of enabling more aggregation, enabling more globalized matching, and how that will impact the economy.
以前，每个人都会在全职角色的背景下思考这个问题，但随着我们走向这个世界，每个人都构建我们的 RL 环境，并能够远程工作，并以这种分数方式训练模型，这也将改变实现更多聚合、实现更全球化匹配的动力，以及这将如何影响经济。
Some of my friends think that mentors and nepotism will make a good comeback.
我的一些朋友认为导师和裙带关系将会卷土重来。
They say everyone will submit a perfect cover letter, have an optimized LinkedIn profile.
他们说每个人都会提交一份完美的求职信，拥有优化的 LinkedIn 个人资料。
They’ll even have practiced with an AI doing the interview.
他们甚至会练习用人工智能进行采访。
They won’t all get up to speed, but a lot of them will, and there’ll be this large mass of apparently pretty qualified candidates.
他们不会全部跟上进度，但他们中的很多人都会跟上进度，并且将会有大量显然相当合格的候选人。
What you’ll actually do is resort to the old tried and true.
你实际上要做的就是诉诸经过考验的旧方法。
Do you know this guy’s uncle, or someone else who can recommend them?
你认识这个人的叔叔，或者其他可以推荐他们的人吗？
Agree?
同意？
Disagree?
不同意？
I think in some companies and industries, that will happen.
我认为在一些公司和行业中，这种情况将会发生。
I agree with it.
我同意。
My hope is that we have models that are helping to run companies in a very thoughtful, efficient way that are data-driven about it, where the models have an eval set of all the performance reviews of people in that given company, and they’re able to make an accurate prediction over whether this reference, or that piece of nepotism should actually be considered, or maybe as a countersignal.
我希望我们拥有的模型能够帮助以一种非常深思熟虑、高效的方式来运营公司，这些方式是数据驱动的，其中模型有一组对特定公司人员的所有绩效评估的评估，并且它们能够准确预测是否应该真正考虑这个参考或那个裙带关系，或者也许作为一个反信号。
That’s my hope, but it’ll probably play out with some combination of both over time.
这是我的希望，但随着时间的推移，这可能会通过两者的某种结合来实现。
In the AI-run labor market, let’s say, it’s more efficient, but do you think there are fewer second chances, and late bloomers in that world?
可以说，在人工智能运行的劳动力市场中，它的效率更高，但你认为那个世界里的第二次机会和大器晚成的人会更少吗？
You get scored too early, so to speak, and then you’re tracked.
可以这么说，你太早得分了，然后你就被追踪了。
It’s a bit more like how European schooling systems can differ from American.
这有点像欧洲教育体系与美国教育体系的不同。
I think there will be a lot of second chances, and the reason that there will be is that oftentimes they’re effective.
我认为会有很多第二次机会，之所以会出现，是因为它们通常是有效的。
The models will identify that, and realize that maybe someone wasn’t the right fit for that first role.
模型会识别出这一点，并意识到某人可能不适合第一个角色。
There’s another role that they could be a really good fit for.
他们可能非常适合另一个角色。
I do think that there are jobs in the economy that almost everyone would excel at, and it’s really just this matching problem of finding the intersection of something that they’re excited about where they’ll also add an immense amount of economic value.
我确实认为经济中有一些工作几乎每个人都会擅长，而这实际上只是一个匹配问题，即找到他们感兴趣的事物的交集，从而增加巨大的经济价值。
As you know, there are AI services now.
如您所知，现在有人工智能服务。
You’re doing an interview across the top or the bottom of your screen.
您正在屏幕的顶部或底部进行采访。
The AI can give you advice, answers.
人工智能可以给你建议、答案。
Does that work at all?
这真的有用吗？
What do you think of those?
你觉得这些怎么样？
We run up against a lot of those.
我们遇到了很多这样的情况。
One thing I’ve found in talent assessment is that, initially, people tried to work against AI, similar to what we do in academic settings, where people would try to say, “We’re going to have you write the essay on paper so that you’re not able to use ChatGPT to help you with the essay.” Really, the right way of approaching it is seeing what people can do when using all of those tools.
我在人才评估中发现的一件事是，最初，人们试图反对人工智能，类似于我们在学术环境中所做的事情，人们会试图说，“我们将让你在纸上写论文，这样你就无法使用 ChatGPT 来帮助你完成论文。”事实上，处理这个问题的正确方法是看看人们在使用所有这些工具时可以做什么。
If we tell them, “Hey, use all of these phenomenal Codegen tools, and record your screen in building a product to see what you’re able to do over the course of an hour,” that’s a far better predictor of this person’s ability to actually deliver impact than it is to say, “Don’t use the tools at all.” So, I think that’s one shift that we’re going to see, and will likely frame the relevancy of a lot of these AI cheating tools over the coming few years.
如果我们告诉他们，“嘿，使用所有这些出色的 Codegen 工具，并在构建产品时录制您的屏幕，看看您在一小时内能够做什么”，这比说“根本不使用这些工具”更能预测此人实际产生影响的能力。因此，我认为这是我们将看到的一个转变，并且可能会在未来几年内确定许多人工智能作弊工具的相关性。
Can someone fool you by using an AI cheating tool, or do you feel you more or less always know?
有人可以通过使用人工智能作弊工具来愚弄你吗？还是你觉得你或多或少总是知道？
I think that there were cases where people could fool us, but now we’re quite good at figuring it out.
我认为在某些情况下人们可能会愚弄我们，但现在我们很擅长弄清楚这一点。
We’re quite good at figuring it out and also moving towards assessments where we almost encourage it, and are comfortable with the fact that they’re using these tools because we want to see what they’re able to do with them.
我们非常擅长弄清楚这一点，也很擅长进行评估，我们几乎鼓励这样做，并且对他们使用这些工具的事实感到满意，因为我们想看看他们能够用这些工具做什么。
On scaling the Thiel Fellowship
关于扩大蒂尔奖学金规模
You were a Thiel Fellow, and you dropped out.
你是泰尔研究员，但你退学了。
How could they improve their methods?
他们如何改进他们的方法？
Well, this is something we’ve talked with them a lot about because the Thiel Fellowship is constrained by that exact matching problem that we were talking about earlier, where they can only consider, and interview a fraction of a percent of the people in the world that they think would be a good fit for the fellowship.
嗯，这是我们和他们讨论过很多次的事情，因为泰尔奖学金受到我们之前讨论的精确匹配问题的限制，他们只能考虑并采访世界上他们认为最适合该奖学金的一小部分人。
So, we’ve worked with them on building out AI interviews that are able to better assess Thiel Fellows, and using models to analyze the transcripts of those recordings to see what are the signals to better select Thiel Fellows, and all of that, which I find very interesting.
因此，我们与他们合作构建了人工智能访谈，能够更好地评估泰尔研究员，并使用模型分析这些录音的笔录，看看有哪些信号可以更好地选择泰尔研究员，所有这些我觉得非常有趣。
But isn’t it part of their strength?
但这难道不是他们力量的一部分吗？
Say, Peter — he’s quite controversial, politically and otherwise.
比如说，彼得——他在政治上和其他方面都很有争议。
Being a Thiel Fellow has a certain brand that’s distinct from anything political, but it’s a very particular thing.
作为泰尔研究员有一个与任何政治不同的品牌，但这是一件非常特别的事情。
Not everyone wants to do it.
不是每个人都愿意这样做。
Doesn’t it work well because it’s an extremely local market, and you get people with a certain orneriness, and selecting from that pool just goes pretty well, and you don’t want to be in the bigger pool of people?
这不是很有效吗，因为这是一个非常本地化的市场，你会遇到一些脾气暴躁的人，从这个人群中进行选择就很顺利，而你不想加入更大的人群中？
I think that you’re right in the element that referrals are very important.
我认为您的观点是正确的，推荐非常重要。
Oftentimes, great people know great people, and so they’ll always need to leverage referrals.
通常，优秀的人认识优秀的人，因此他们总是需要利用推荐。
But at the same time, I think they rightfully care a lot about people that think unconventionally and come from unconventional backgrounds — the people from every part of the world that might otherwise not get a meeting with a venture capitalist or some of these more traditional institutions, and so, ensuring that they’re able to consider those candidates, and give them this opportunity, and incorporate them into the fellowship is incredibly important, and part of the mission.
但与此同时，我认为他们理所当然地非常关心那些非传统思维和来自非传统背景的人——来自世界各地的人，否则他们可能无法与风险投资家或一些更传统的机构会面，因此，确保他们能够考虑这些候选人，给他们这个机会，并将他们纳入奖学金是非常重要的，也是使命的一部分。
Could it be scaled 10X?
可以放大 10 倍吗？
Absolutely.
绝对地。
100X?
100 倍？
I think so.
我想是的。
100,000X?
100,000X？
FOODY: I think it ties to what we were talking about earlier, of the elasticity of demand for better investors, because in some ways, hiring has so much overlap with investing.
FOODY：我认为这与我们之前讨论的关于更好的投资者的需求弹性有关，因为在某些方面，招聘与投资有很多重叠。
Imagine if we could have Peter interview everyone in the world when they’re 18 or 20, or whatever the age is, and make a decision around whether he wants to give them a 100K check.
想象一下，如果我们可以让 Peter 采访世界上 18 岁或 20 岁或任何年龄的每个人，并决定是否要给他们一张 10 万美元的支票。
That would probably be very powerful with respect to economic mobility and how many companies we’re able to create.
对于经济流动性以及我们能够创建多少家公司而言，这可能会非常强大。
I think that will happen.
我认为这将会发生。
It’s just a matter of time of building the right technology and the right focus to enable it.
构建正确的技术和正确的关注点来实现它只是时间问题。
Is the following possible?
以下情况可能吗？
Let’s say Peter is just a tremendous interviewer — that’s easy to believe — but he’s really a great interviewer for the subset of people attracted to him.
假设彼得只是一位出色的面试官——“这很容易相信”——但对于那些被他吸引的人来说，他确实是一位出色的面试官。
If you just put him out in the broader pool, who’s going to be a lifeguard at the swimming pool, or something?
如果你把他放在更宽的游泳池里，谁来当游泳池的救生员之类的？
Maybe he’s just not that good an interviewer for that.
也许他只是不太适合做这方面的采访者。
I agree with that.
我同意这一点。
I think that’s certainly the case.
我认为情况确实如此。
Imagine if you had a panel of domain experts across every industry that were able to perform these interviews.
想象一下，如果您拥有一个由各个行业的领域专家组成的小组，能够进行这些采访。
Certainly, the best models will be better than any single best individual, but I would expect that the aggregate sum of all experts in each domain will likely remain better than the models for a long time.
当然，最好的模型会比任何一个最好的个体更好，但我预计每个领域所有专家的总和可能会在很长一段时间内保持比模型更好。
On a hypothetical gap year
在假设的间隔年
Now, you dropped out of school.
现在，你退学了。
Now, you’re doing the company.
现在，你在做公司。
Obviously, you’re very busy, but imagine, as an act of magic, you could have a free year just inserted between today and tomorrow, and you come back and nothing has changed.
显然，你很忙，但想象一下，就像魔术一样，你可以在今天和明天之间插入一年的空闲时间，然后你回来，一切都没有改变。
To go off and do anything you want with literature, with art, with travel, with music, climbing the Alps, I don’t know.
去从事任何你想做的事情，比如文学、艺术、旅行、音乐、攀登阿尔卑斯山，我不知道。
What would you do with the year?
过年你会做什么？
That’s a fascinating question.
这是一个有趣的问题。
Can it be AI-related?
它可以与人工智能相关吗？
No.
不。
Cannot be company-related.
不能与公司相关。
Let’s see.
让我们来看看。
I would love to travel.
我很想去旅行。
I think that sounds like it would be a lot of fun because, as you can imagine, in running the company, I’ve worked a hundred hours a week for the last three years, and I love doing it, and I’ll continue doing that.
我认为这听起来会很有趣，因为正如你可以想象的那样，在经营公司的过程中，过去三年我每周工作一百个小时，我喜欢这样做，而且我会继续这样做。
But I do think that seeing the world, and getting more of this understanding of how do perspectives vary by country and geography, how are people thinking about AI differently elsewhere is really interesting.
但我确实认为，看看世界，更多地了解不同国家和地区的观点如何不同，以及其他地方的人们如何以不同的方式思考人工智能，真的很有趣。
I really like that.
我真的很喜欢这样。
I remember after ChatGPT came out, Sam did this world tour of going to all the different places, seeing what they thought about AI, how they viewed it impacting their world.
我记得 ChatGPT 推出后，Sam 进行了一次世界巡演，去了所有不同的地方，了解他们对人工智能的看法，以及他们如何看待人工智能对他们的世界的影响。
I think that global perspective is incredibly valuable, and informative.
我认为全球视角非常有价值且信息丰富。
Where do you want to go the most?
你最想去哪里？
I want to go to Japan a lot.
我很想去日本。
I’ve never been to Japan, so I’ll have to make it out there.
我从来没有去过日本，所以我必须去那里。
That’s probably my top pick.
这可能是我的首选。
It’s a great visit.
这是一次很棒的访问。
One thing I’ve found since I have traveled a lot — obviously, I’m older, and in some ways, less busy than you are — that it helps me interview quite a bit, because people more and more come from all over.
自从我经常旅行以来，我发现一件事 ——显然，我年纪大了，而且在某些方面，比你不那么忙—— 这对我的采访很有帮助，因为越来越多的人来自世界各地。
It’s like if your model has the poetic taste of different eras — of John Milton, Wordsworth, Shakespeare, whatever — traveling is an individual’s version to get some version of that.
这就好像如果你的模型具有不同时代的诗意品味——约翰·弥尔顿、华兹华斯、莎士比亚等等——那么旅行就是个人的版本，以获得某种版本的版本。
Yes.
是的。
Say you hire a lot of people from India.
假设你雇佣了很多来自印度的人。
I suspect you do.
我怀疑你是这样的。
It’s a populous country, a lot in the Bay Area.
这是一个人口大国，很多都在湾区。
Going to India then becomes very important because you get a better sense of just where they’re coming from.
那么去印度就变得非常重要，因为你可以更好地了解他们来自哪里。
Yes, I completely agree.
是的，我完全同意。
I think also being able to connect with those individuals very quickly around “Hey, I’ve been to this place.
我认为也能够非常快速地与这些人建立联系“嘿，我去过这个地方。
I’m very familiar with India, and all these different things” is really helpful in building relationships and setting up trust across all the different people that we work with and interact with.
我对印度非常熟悉，所有这些不同的事情”对于与我们合作和互动的所有不同人之间建立关系和建立信任确实很有帮助。
On donuts, debates, and dyslexia
关于甜甜圈、辩论和阅读障碍
How did your 8th-grade donut company go?
你的八年级甜甜圈公司进展如何？
One of my favorite topics.
我最喜欢的主题之一。
I can tell the story, which is, I initially realized that Safeway donuts were selling for $5 a dozen.
我可以讲述这个故事，也就是说，我最初意识到 Safeway 甜甜圈的售价为 5 美元一打。
My 8th-grade mind was thinking that is such a deal.
我八年级的心在想，这就是一笔交易。
I would pay $2 a donut, and I bet my friends would as well.
我愿意花 2 美元买一个甜甜圈，我打赌我的朋友们也会这么做。
So, I would bike down to Safeway, and I would buy Safeway donuts for $5 a dozen, go to my middle school, and sell them for $2 each.
所以，我会骑自行车去西夫韦，我会以每打 5 美元的价格购买西夫韦甜甜圈，然后去我的中学，然后以每个 2 美元的价格出售。
Eventually, my middle school called me into the principal’s office to shut me down, because I was scaling up my operations.
最终，我的中学把我叫到校长办公室，要求关闭我，因为我正在扩大业务规模。
Then I moved my donut stand about 50 feet off of school campus so they couldn’t police me.
然后我把甜甜圈摊搬到了离校园约 50 英尺的地方，这样他们就无法警察来监视我了。
I paid my mom $20 a week to drive me in her minivan to be able to bring more donuts from Safeway.
我每周付给妈妈 20 美元，让她用她的小型货车载我，以便能够从西夫韦带来更多甜甜圈。
She charged you $20, your mom?
她收了你20美元，你妈妈？
She charged me $20, exactly.
她一共收了我 20 美元。
Was that underpriced or overpriced?
是定价过低还是定价过高？
I think it was about right.
我认为这是正确的。
I anchored it on the cost of an Uber.
我把它固定在 Uber 的费用上。
I was like, I’m not going to pay more than an Uber, but I need the car to wait long enough that I’m able to load up 10 or 20 dozen donuts, so I did this.
我当时想，我不会支付比 Uber 更高的费用，但我需要汽车等待足够长的时间，以便我能够装载 10 或 20 打甜甜圈，所以我这样做了。
I’d pay my friends in donuts because I perceived the cost of the donuts as my cost basis versus they perceived it as $2 each.
我会用甜甜圈付钱给朋友，因为我认为甜甜圈的成本是我的成本基础，而他们认为每个甜甜圈是 2 美元。
I had a little bit of arbitrage in the salaries.
我在工资方面有一点套利。
I had competition pop up, where they would sell Chuck’s Donuts, which are higher-end donuts, but they had a $1 cost basis.
我遇到了竞争对手，他们会在那里销售 Chuck’s Donuts，这是一种高端甜甜圈，但他们的成本基础是 1 美元。
I dropped my prices to $1 for two weeks to drive them out of business before I had learned anything about anti-competitive laws.
在我了解有关反竞争法的任何信息之前，我将价格降至 1 美元，持续了两周，以将他们赶出市场。
Those were just a few of the stories from my 8th-grade “donut dynasty” is what we called it.
这些只是我八年级的“甜甜圈王朝”（我们称之为“甜甜圈王朝”）的一些故事。
Other than just intelligence, what makes a person good at extemporaneous speaking?
一个人善于即兴演讲，除了智力之外，还有什么因素呢？
You won awards for this, right?
你因此获奖了，对吗？
I did.
我做到了。
Actually, I won awards for it, but I wasn’t nearly as good as my co-founders.
事实上，我因此获得了奖项，但我的表现还不如我的联合创始人。
In high school, we all did speech and debate together.
高中时，我们一起演讲和辩论。
You knew each other from high school.
你们从高中就认识了。
We knew each other from high school.
我们从高中就认识了。
Like age 14, right?
就像14岁一样，对吧？
Age 14, exactly.
确切地说，14 岁。
We were on the policy debate team together.
我们一起参加政策辩论小组。
We also did national extemporaneous speaking.
我们还进行了全国即兴演讲。
They were the winningest speech and debate team of all time in policy debate, the most competitive event, where they won the Tournament of Champions, the National Speech and Debate Association, and NDCA — the three largest national tournaments, which no other team has ever done.
他们是政策辩论中有史以来获胜最多的演讲和辩论队，也是竞争最激烈的赛事，他们赢得了冠军锦标赛、全国演讲和辩论协会和 NDCA 三大全国锦标赛的冠军，这是其他球队从未做到过的。
I did okay, but I’m dyslexic, and so I would always stumble over words, or mix things up.
我做得还不错，但我有诵读困难，所以我总是会结结巴巴，或者把事情搞混。
I wasn’t quite the same level as them.
我和他们水平不太一样。
I think there’re a few things that go into the answer of what makes one phenomenal at it.
我认为有几件事可以解释一个人在这方面的非凡表现。
I think that high clarity of thought often correlates very strongly with people that speak very well, and so, as you mentioned, intelligence plays another role.
我认为高度清晰的思维往往与善于表达的人密切相关，因此，正如您所提到的，智力发挥着另一个作用。
I think a second thing is confidence.
我认为第二件事是信心。
Someone that’s willing to speak and improve and iterate on it, because oftentimes, it’s just doing more of that activity that allows you to improve on it.
愿意发言、改进和迭代的人，因为很多时候，只是做更多的活动，就能让你改进它。
Then maybe a third one is, more than just intelligence, it’s also the speed of thought.
那么第三个可能不仅仅是智力，还有思维速度。
I think about those as different dimensions.
我认为这些是不同的维度。
There are certain people I think of as having very high aptitude, but thinking very deeply and slowly about a given thing.
我认为有些人具有非常高的才能，但对特定事物的思考非常深入且缓慢。
Other people that I think of as having reasonably high aptitude, or medium aptitude, but being able to be quick on their feet.
我认为其他人具有相当高的能力或中等的能力，但能够快速行动。
I definitely think there’s some innate element of that.
我绝对认为其中有一些与生俱来的因素。
Which are you?
你是哪一个？
I tend to think I’m more in this slower, deeper thinking bucket, but it depends a little bit on how much coffee I’ve had.
我倾向于认为我更适合这种更慢、更深入的思考，但这在一定程度上取决于我喝了多少咖啡。
COWEN: You started the company when you were 19.
COWEN：您 19 岁时创办了这家公司。
Why is it there’s a positive statistical correlation between being dyslexic and entrepreneurship?
为什么阅读障碍和创业精神之间存在正统计相关性？
There is one in some published papers.
一些发表的论文中有一篇。
What’s the mechanism?
机制是什么？
It’s shockingly strong, actually.
事实上，它的强度令人震惊。
I’m not sure exactly, but I find that one unique thing is that it feels like my brain works a little bit differently, and that there are certain things that people are so much better at than I am, where they’re reading through evidence in a debate round very quickly, and I could never do that.
我不太确定，但我发现一个独特的事情是，感觉我的大脑工作方式有点不同，而且在某些事情上人们比我擅长得多，他们在辩论中很快地阅读证据，而我永远无法做到这一点。
But there are certain ideas or ways of approaching a problem that are just different, that enable more creativity, potentially being unconventional in doing so.
但解决问题的某些想法或方法只是不同的，可以激发更多创造力，而且可能是非常规的。
I think that that is one advantage I’ve had.
我认为这是我所拥有的优势之一。
One of our early investors, actually, Scott Sandell, is dyslexic, and has backed a lot of dyslexic entrepreneurs, and so we’ve talked about this a little bit.
实际上，我们的早期投资者之一斯科特·桑德尔（Scott Sandell）患有诵读困难症，并且支持了很多患有诵读困难症的企业家，因此我们对此进行了一些讨论。
One of my hypotheses is that quite early on, you have to learn how to delegate.
我的假设之一是，你必须很早就学会如何授权。
That’s a skill that when people are not forced to learn, often very competent people don’t become good at it until much later, but a dyslexic person is good at it right away.
当人们不被迫学习这项技能时，通常非常有能力的人要等到很久以后才会擅长它，但阅读困难的人很快就会擅长它。
Totally.
完全。
Yes, asking people to help read something for them.
是的，请人们帮忙读一些东西。
COWEN: That’s right.
考恩：没错。
“Could you please do this for me?”
“你能帮我做这个吗？”
That certainly could be the case.
情况肯定可能是这样。
And focusing on bigger picture in some useful ways, at least for being a founder.
并以一些有用的方式关注更大的前景，至少对于作为创始人来说是这样。
Not good for every job, of course.
当然，并不适合所有工作。
Totally.
完全。
One thing I really came to appreciate, especially during high school, is that there are certain things that some people are phenomenal at, and others are horrible at.
我真正开始欣赏的一件事是，尤其是在高中期间，有些人在某些事情上表现出色，而另一些人则很糟糕。
I felt areas of debate in reading through evidence quickly where I felt extremely unintelligent, and it was super humbling.
在快速阅读证据时，我感受到了争论的领域，我觉得自己非常不聪明，这让我感到非常羞愧。
So much of finding success in your career is just understanding what are your strengths, and how do you leverage those, and much less about what are your weaknesses.
要在职业生涯中取得成功，很大程度上只是了解自己的优势是什么，以及如何利用这些优势，而不是了解自己的劣势是什么。
That’s something that I’ve taken with how I approach Mercor, but also how I encourage our employees to think about their roles within the business.
这是我对待美科的方式，也是我鼓励员工思考他们在企业中的角色的方式。
What are the things where they have these comparative advantages and phenomenal strengths, and how do they leverage those most effectively.
他们在哪些方面拥有这些比较优势和显着优势，以及他们如何最有效地利用这些优势。
On dating and dining out
关于约会和外出就餐
How much do you feel you’re in touch with the general culture of intelligent 22-year-old men in the United States?
您觉得自己与美国 22 岁聪明男性的普遍文化有多少联系？
Or are you just so in the company you have no idea what’s going on?
或者你在公司里根本不知道发生了什么？
I’m so in the company, I don’t know.
我在公司，我不知道。
I obviously was in college for a couple of years before I dropped out, and so I had some people around me.
显然，我在大学读了几年才退学，所以我周围有一些人。
So much of our company is 22 plus or minus a couple of years, so I guess I have that heuristic.
我们公司的大部分员工都已经工作了 22 年左右，所以我想我有这样的启发。
I certainly don’t think that I have spent as much time with people my age as if I had stayed in school, as another comparison.
作为另一个比较，我当然不认为我和同龄人在一起的时间比我在学校呆的时间多。
This is not a question about you, because we don’t ask personal questions, but a good tech friend of mine — you’ve probably heard of him — he says men in that age bracket, 22, 23, that there truly is a dating crisis, that something has gone wrong.
这不是关于你的问题，因为我们不问个人问题，而是我的一位好科技朋友 ——你可能听说过他 ——他说，22、23 岁这个年龄段的男性确实存在约会危机，有些事情出了问题。
Not about you, but just America in general, the very smart, possibly nerdy person in that age group.
不是关于你，而是关于整个美国，那个年龄段中非常聪明，可能是书呆子的人。
Is there a dating crisis?
是否存在约会危机？
I think certainly in San Francisco.
我想当然是在旧金山。
Not in New York, but certainly in San Francisco.
不是在纽约，但肯定是在旧金山。
You think it’s just gender imbalance?
你认为这只是性别失衡吗？
Or the country is screwed up more generally?
或者说这个国家普遍都搞砸了？
I haven’t thought too much about this.
我还没有想太多。
I think it’s probably gender imbalance in San Francisco, especially in certain industries.
我认为这可能是旧金山的性别失衡，尤其是在某些行业。
I think that dating apps are probably, generally in society — I don’t use dating apps — but are, generally, in society helping to drive a lot more efficiency in solving this matching problem.
我认为约会应用程序可能在社会中普遍存在 — 我不使用约会应用程序 — 但在社会中通常有助于提高解决匹配问题的效率。
So, you’re pro dating app.
所以，你是专业的约会应用程序。
Most of the people I know are against them.
我认识的大多数人都反对他们。
No, I think they’re good.
不，我认为他们很好。
I’m very much a proponent of better technology to solve these matching problems and enable people to be happy in their lives.
我非常支持更好的技术来解决这些匹配问题并让人们生活幸福。
Your last name is Foody.
你的姓氏是美食。
Should I believe in nominative determinism?
我应该相信主格决定论吗？
Are you a foodie?
你是吃货吗？
I certainly love my diet and good food.
我当然喜欢我的饮食和美食。
It’s funny.
这很有趣。
My dad always loved cooking growing up, and was certainly much more of a foodie than I am, but a little bit of it rubbed off on me.
我父亲从小就喜欢烹饪，他当然比我更热爱美食，但这一点也影响了我。
While I’m not as much into cooking, I love eating good food.
虽然我不太热衷于烹饪，但我喜欢吃美食。
Where in San Francisco should people eat?
人们应该在旧金山哪里吃饭？
Or nearby?
或者附近？
Lots of good restaurants.
有很多不错的餐馆。
There are the everyday restaurants that I think are very good, and then higher-end.
有我觉得很好的日常餐厅，还有更高档的。
Every day I love Mexican food.
我每天都喜欢墨西哥菜。
El Metate is a great Mexican restaurant.
El Metate 是一家很棒的墨西哥餐厅。
For higher end food, I like Cotogna in California, Quince, lots of good restaurants like that.
对于高端食品，我喜欢加利福尼亚州的科托尼亚 (Cotogna)、昆斯 (Quince)，有很多类似的好餐厅。
At the meta level, what’s the thing people should know about eating out here?
在元层面上，人们应该知道关于外出就餐的哪些事情？
Where I live, I would just say you need to know to go to the suburbs.
在我住的地方，我只想说你需要知道去郊区。
It may or may not be true here, but here, what do they need to know, other than particular names of restaurants?
在这里可能是真的，也可能不是，但是在这里，除了餐馆的特定名称之外，他们还需要知道什么？
I find that Beli is really accurate, the app for food ratings in San Francisco, because there’s a high density of users.
我发现旧金山的 Beli 食品评级应用程序非常准确，因为用户密度很高。
So, if you use Beli as your guide, you’ll generally find good spots.
因此，如果您以贝利为向导，通常会找到不错的景点。
On Mercor’s next steps
关于 Mercor 的下一步计划
Why is the company called Mercor?
为什么这家公司叫 Mercor？
Mercor means marketplace in Latin, and we want to build the largest marketplace in the world, so we named it Mercor.
Mercor在拉丁语中是市场的意思，我们希望建立世界上最大的市场，所以我们将其命名为Mercor。
We’re from Mercatus.
我们来自莫卡图斯。
Do you know what Mercatus means in Latin?
你知道Mercatus在拉丁语中的意思吗？
It’s a variant.
这是一个变体。
It means market.
就是市场的意思。
Okay, there you go.
好的，就这样吧。
Yes, we’re from the same named institution, in that sense.
是的，从这个意义上说，我们来自同一个机构。
Exactly.
确切地。
Well, it’s funny.
嗯，这很有趣。
In high school, my co-founders and I went to a Jesuit school, and my co-founder, Surya, studied Latin, and so we’ve always certainly thought a lot about Latin roots and Latin words.
高中时，我和我的联合创始人去了一所耶稣会学校，我的联合创始人苏里亚学习了拉丁语，所以我们总是对拉丁语词根和拉丁语单词思考很多。
Your family wasn’t Catholic, I believe, right?
我相信你的家人不是天主教徒，对吗？
That’s correct.
这是正确的。
Did going to a Jesuit school help you think?
去耶稣会学校上学对你的思考有帮助吗？
Or what did that add to the mix?
或者这又增加了什么？
Well, none of the three of us were Catholic despite going to Catholic school, which was a little bit funny.
好吧，尽管我们上的是天主教学校，但我们三个人都不是天主教徒，这有点有趣。
One interesting story is that my mom was concerned about whether I would start selling drugs when I was doing my donut stand in 8th grade, because it’s an easy step.
一个有趣的故事是，当我八年级摆甜甜圈摊时，我妈妈担心我是否会开始卖毒品，因为这是一个简单的步骤。
I like to think that Catholic school helped instill good values in what I should care about, and being very focused on school at the time, on speech and debate, on building companies, and so, very grateful for that education.
我喜欢认为天主教学校帮助向我应该关心的事情灌输了良好的价值观，并且当时非常专注于学校、演讲和辩论、创建公司，因此，非常感谢这种教育。
Last two questions.
最后两个问题。
First, what’s the next goal you have for the company?
首先，您对公司的下一个目标是什么？
The next goal for the company is really in scaling up a lot of these super-realistic evaluations that I’ve talked about, of how do we measure the ways that models use all sorts of different tools on trajectories that would take someone days or weeks to do is a big focus for us, and especially how that impacts enterprise, right?
公司的下一个目标实际上是扩大我谈到的许多超现实评估的规模，我们如何衡量模型在轨迹上使用各种不同工具的方式，这对我们来说是一个重点，尤其是这对企业有何影响，对吗？
For the last two years, people have been very focused on the idea of intelligence rather than the idea of models being useful.
在过去的两年里，人们非常关注智能的概念，而不是有用的模型的概念。
Bridging the gap between what do enterprises actually want to use, how do we measure that, and how do we get those capabilities and models is, to me, the most exciting thing that I could work on.
对我来说，弥合企业实际想要使用的东西、我们如何衡量它以及我们如何获得这些功能和模型之间的差距是我能做的最令人兴奋的事情。
What do you want to learn next, work-related or otherwise?
接下来你想学什么，与工作相关的还是其他方面的？
That’s an interesting question.
这是一个有趣的问题。
I feel like Mercor is at the intersection of labor markets and AI research.
我觉得 Mercor 正处于劳动力市场和人工智能研究的交叉点。
We grew up with the DNA in labor markets, of thinking all about how do we aggregate all these people on our platform?
我们在劳动力市场的 DNA 中长大，一直在思考如何将所有这些人聚合到我们的平台上？
How do we match them?
我们如何匹配它们？
We hired people that are deep domain experts in labor markets like Sundeep Jain, who was the chief product officer, chief technology officer at Uber.
我们聘请了劳动力市场的深度领域专家，比如 Sundeep Jain，他是 Uber 的首席产品官和首席技术官。
But I am most fascinated by all the advancements in AI research, of how do we apply human talent and human labor to all of these problems at the frontier, and more efficient ways to train models, and what are the specific rubrics or data types that are driving the most model improvement.
但我最着迷的是人工智能研究的所有进步，我们如何将人才和人力应用于所有这些前沿问题，以及更有效的模型训练方法，以及推动模型改进的具体规则或数据类型是什么。
I’ve been most interested in how to learn that.
我最感兴趣的是如何学习这一点。
Brendan Foody, thank you very much.
布伦丹·福迪，非常感谢你。
Thank you so much for having me, Tyler.
非常感谢你邀请我，泰勒。