Inside the expert network training every frontier AI model | Garrett Lord
Garrett Lord: There will never be a time like this.
加勒特·洛德：永远不会有这样的时刻。
I've never seen anything like it.
我从来没有见过这样的事情。
I doubt I'll ever feel anything like this in business again where there's unlimited demand.
我怀疑我是否会在需求无限的商业环境中再次感受到这样的感觉。
How do you make sure that three months from now, six months, you have no regrets?
你如何确保三个月后、六个月后你不会后悔？
Get on the plane to go talk to a customer, make the late night push, check the data six times over again.
上飞机去和客户交谈，深夜推送，检查数据六遍。
Lenny Rachitsky: Your company creates new data to continue advancing the intelligence of models.
Lenny Rachitsky：贵公司创建新数据以继续推进模型的智能性。
This is a business that you built on top of a business you've already had.
这是您在已有业务基础上建立的业务。
Garrett Lord: We're the largest expert network in the world.
Garrett Lord：我们是世界上最大的专家网络。
We have this massive strategic advantage, which is like no customer acquisition costs.
我们拥有巨大的战略优势，就像没有客户获取成本一样。
The only moat in human data is access to an audience.
人类数据中唯一的护城河是接触受众。
Lenny Rachitsky: You guys come in after the model's trained to tweak the weights based on additional data that you you've created.
Lenny Rachitsky：你们在模型经过训练后进来，根据您创建的附加数据调整权重。
Garrett Lord: The models have gotten so good that the generalists are no longer needed.
加勒特·洛德（Garrett Lord）：模型已经变得如此优秀，以至于不再需要通才。
What they really need is experts.
他们真正需要的是专家。
Lenny Rachitsky: There's this tension between all these students training models to become smarter, and then there's that they will have harder time potentially finding jobs.
Lenny Rachitsky：所有这些训练模型以变得更聪明的学生之间存在着这种紧张关系，而且他们可能会更难找到工作。
Garrett Lord: That's not what we're hearing from our employers, this is just enabling human beings to be even more productive.
加勒特·洛德：这不是我们从雇主那里听到的，这只是让人类变得更有生产力。
You used to put a Google Search on a skill on your resume because you grew up with Google.
你过去常常在简历中的一项技能上进行谷歌搜索，因为你是伴随着谷歌长大的。
Being AI native, young people are at a huge advantage.
作为人工智能原生者，年轻人拥有巨大的优势。
Lenny Rachitsky: Today my guest is Garrett Lord.
莱尼·拉奇茨基：今天我的客人是加勒特·洛德。
Garrett is the co-founder and CEO of Handshake, which is one of the most interesting and incredible AI success stories that you probably haven't heard of.
Garrett 是 Handshake 的联合创始人兼首席执行官，这是您可能没有听说过的最有趣、最令人难以置信的人工智能成功故事之一。
Handshake has been around for over 10 years, they're essentially LinkedIn for college students, it's a place for students to connect with companies to find a job.
Handshake 已经存在了 10 多年，它们本质上是大学生的 LinkedIn，是学生与公司联系寻找工作的地方。
They are the platform of choice for every single Fortune 500 company.
它们是每家财富 500 强公司的首选平台。
Over 1,500 colleges, over 20 million students and alumni, and over 1 million companies use them to hire graduates.
超过 1,500 所大学、超过 2000 万学生和校友以及超过 100 万家公司使用它们来招聘毕业生。
At the start of this year, Garrett and his team realized that their huge proprietary network of students, including tens of thousands of PhDs and master's students, is extremely valuable to AI labs to help them create and label high quality training data.
今年年初，Garrett 和他的团队意识到，他们庞大的专有学生网络（包括数万名博士生和硕士生）对于人工智能实验室来说非常有价值，可以帮助他们创建和标记高质量的训练数据。
So, they launched a new business from zero to one in January.
于是，他们在一月份推出了从零到一的新业务。
Four months later, they hit 50 million ARR.
四个月后，他们的 ARR 达到了 5000 万。
They're now on pace to blow past 100 million ARR within just 12 months.
他们现在有望在短短 12 个月内突破 1 亿 ARR。
They'll exceed the revenue that they're making with their decade old business in under two years.
他们将在不到两年的时间里超过他们已有十年历史的业务所赚取的收入。
Lenny Rachitsky: This is a truly incredible and rare story, and one that I think a lot of teams can learn from because AI is creating a lot of opportunity but also a lot of potential disruption, and this is an amazing story where the company basically disrupted themselves.
Lenny Rachitsky：这是一个真正令人难以置信且罕见的故事，我认为很多团队都可以从中学习，因为人工智能创造了很多机会，但也创造了很多潜在的颠覆，这是一个令人惊奇的故事，公司基本上颠覆了自己。
This episode is packed with insights, including a primer on what the heck are people actually doing when they're labeling and creating data to train models?
这一集充满了见解，包括关于人们在标记和创建数据来训练模型时到底在做什么的入门知识？
A huge thank you to Garrett for making time for this, his wife just had a baby this week.
非常感谢加勒特抽出时间来做这件事，他的妻子本周刚刚生了孩子。
He's also in the middle of scaling this insane new business.
他还正在扩大这项疯狂的新业务。
So thank you, Garrett.
谢谢你，加勒特。
If you enjoy this podcast, don't forget to subscribe and follow it in your favorite podcasting app or YouTube.
如果您喜欢这个播客，请不要忘记订阅并在您最喜欢的播客应用程序或 YouTube 中关注它。
Lenny Rachitsky: Also, if you become an annual subscriber of my newsletter, you get a year free of a bunch of incredible products, including Lovable, Replit, Bolt, n8n, Linear, Superhuman, Descript, Wispr Flow, Gamma, Perplexity, Warp, Granola, Magic Patterns, Raycast, ChatPRD, and Mobbin.
Lenny Rachitsky：此外，如果您成为我的时事通讯的年度订阅者，您将获得一年免费的一堆令人难以置信的产品，包括 Lovable、Replit、Bolt、n8n、Linear、Super human、Descript、Wispr Flow、Gamma、Perplexity、Warp、Granola、Magic Patterns、Raycast、ChatPRD 和 Mobbin。
Check it out at lenny'snewsletter.com and click bundle.
请访问 lenny'snewsletter.com 查看并单击“捆绑”。
With that, I bring you Garrett Lord.
接下来，我为您带来加勒特·洛德。
Lenny Rachitsky: This episode is brought to you by CodeRabbit, the AI code review platform transforming how engineering teams ship faster with AI without sacrificing code quality.
Lenny Rachitsky：本集由 CodeRabbit 为您带来，CodeRabbit 是一个人工智能代码审查平台，它改变了工程团队在不牺牲代码质量的情况下更快地利用人工智能交付的方式。
Code reviews are critical, but time-consuming.
代码审查很重要，但也很耗时。
CodeRabbit acts as your AI co-pilot, providing instant code review comments and potential impacts of every pull request.
CodeRabbit 充当您的人工智能副驾驶，提供即时代码审查评论和每个拉取请求的潜在影响。
Beyond just flagging issues, CodeRabbit provides one click fix suggestions and lets you define custom code quality rules using AST GREP patterns, catching subtle issues that traditional static analysis tools might miss.
除了标记问题之外，CodeRabbit 还提供一键修复建议，并允许您使用 AST GREP 模式定义自定义代码质量规则，捕获传统静态分析工具可能错过的微妙问题。
CodeRabbit also provides free AI code reviews directly in the IDE.
CodeRabbit 还直接在 IDE 中提供免费的 AI 代码审查。
It's available in VS Code, Cursor and Windsurf.
它在 VS Code、Cursor 和 Windsurf 中可用。
CodeRabbit has so far reviewed more than 10 million PRs installed on 1 million repositories, and is used by over 70,000 open source projects.
迄今为止，CodeRabbit 已审查了安装在 100 万个存储库上的超过 1000 万个 PR，并被超过 70,000 个开源项目使用。
Get CodeRabbit for free for an entire year at coderabbit.ai using code Lenny.
使用代码 Lenny 在 coderabbit.ai 免费获取 CodeRabbit 一整年。
That's coderabbit.ai.
那是 coderabbit.ai。
Lenny Rachitsky: This episode is brought to you by Orkes, the company behind Open Source Conductor, the orchestration platform powering modern enterprise apps and agentic workflows.
Lenny Rachitsky：本集由 Orkes 为您带来，Orkes 是 Open Source Conductor 背后的公司，Open Source Conductor 是为现代企业应用程序和代理工作流程提供支持的编排平台。
Legacy automation tools can't keep pace, siloed low-code platforms, outdated process management and disconnected API tooling fall short in today's event-driven AI-powered agentic landscape.
传统的自动化工具无法跟上步伐，孤立的低代码平台、过时的流程管理和断开连接的 API 工具在当今事件驱动的人工智能驱动的代理环境中表现不佳。
Orkes changes this.
奥克斯改变了这一点。
With Orkes Conductor, you gain an agentic orchestration layer that seamlessly connects humans, AI agents, APIs, microservices, and data pipelines in real time at enterprise scale.
借助 Orkes Conductor，您可以获得一个代理编排层，可以在企业规模上实时无缝连接人类、AI 代理、API、微服务和数据管道。
Visual and code-first development, built-in compliance, observability and rock solid reliability ensure workflows evolve dynamically with your needs.
可视化和代码优先开发、内置合规性、可观察性和坚如磐石的可靠性确保工作流程根据您的需求动态发展。
It's not just about automating tasks, it's orchestrating autonomous agents and complex workflows to deliver smarter outcomes faster.
它不仅仅是自动化任务，它还协调自主代理和复杂的工作流程，以更快地交付更智能的结果。
Whether modernizing legacy systems or scaling next-gen AI-driven apps, Orkes accelerates your journey from idea to production.
无论是对遗留系统进行现代化改造还是扩展下一代人工智能驱动的应用程序，Orkes 都能加速您从创意到生产的旅程。
Learn more and start building at orkes.io/Lenny.
了解更多信息并开始构建，请访问 orkes.io/Lenny。
That's O-R-K-E-S.io/Lenny.
那是 O-R-K-E-S.io/Lenny。
Lenny Rachitsky: Garrett, thank you so much for being here.
Lenny Rachitsky：Garrett，非常感谢你来到这里。
Welcome to the podcast.
欢迎来到播客。
Garrett Lord: Yeah, thanks for having me.
加勒特·洛德：是的，谢谢你邀请我。
A long-time subscriber.
长期订阅者。
Lenny Rachitsky: I appreciate that.
莱尼·拉奇茨基：我很欣赏这一点。
Okay, so before we get into the insane trajectory that your data labeling business is on, which is just an amazing story that I think a lot of founders and product teams that are trying to navigate this AI disruption that's happening will have a lot to learn from.
好吧，在我们进入数据标签业务的疯狂轨迹之前，这是一个令人惊奇的故事，我认为许多试图驾驭正在发生的人工智能颠覆的创始人和产品团队将有很多值得学习的地方。
I want to first help people understand what the hell data labeling actually is.
我想首先帮助人们理解数据标签到底是什么。
Just like, what are people actually doing?
就像，人们实际上在做什么？
Why is this so valuable?
为什么这如此有价值？
Some of the most, I don't know, fastest-growing companies in the world today, including you guys are just, this is what you do.
我不知道，当今世界上一些增长最快的公司，包括你们，只是，这就是你们所做的。
Clearly there's something really important here.
显然这里有一些非常重要的东西。
I sort of understand it, probably not really.
我大概明白了，也许不是很明白。
I think a lot of listeners feel the same way.
我想很多听众都有同感。
So let me just ask you this, what is data labeling actually?
那么我想问一下，数据标签到底是什么？
What are people actually doing?
人们实际上在做什么？
And then, just why is this so valuable to frontier AI labs?
那么，为什么这对前沿人工智能实验室如此有价值呢？
Garrett Lord: Yeah.
加勒特·洛德：是的。
So, I think it's helpful to take a step back of what does training a model look like?
所以，我认为退一步思考训练模型是什么样子是有帮助的？
So, there's really two primary functions.
所以，实际上有两个主要功能。
There's a pre-training and a post-training process in training a model, and for a long time these AI providers, or LLMs, or Frontier Labs we're focused on basically sucking up more and more information on the pre-training side of the house.
训练模型有一个预训练和训练后的过程，长期以来，这些人工智能提供商、法学硕士或前沿实验室，我们基本上专注于吸收越来越多的预训练方面的信息。
And that's basically the entire corpus of written human knowledge.
这基本上就是人类书面知识的全部语料库。
So, that's not just written, but every YouTube video, every book, basically the pursuit of sucking up everything that was on the internet, and that was the pre-training side.
所以，这不仅仅是写出来的，而是每个 YouTube 视频、每本书，基本上都是在追求吸收互联网上的所有内容，这就是预训练方面。
And there was a lot of gains from pre-training, like models continue to get better.
预训练带来了很多收获，比如模型不断变得更好。
And about 18 months ago, 24 months ago, we started to really see an asymptoting of gains coming from, because they had essentially sucked up all of the knowledge on the internet.
大约 18 个月前，24 个月前，我们开始真正看到收益的渐近，因为他们基本上吸收了互联网上的所有知识。
And so, labs really shifted towards most of the gains now coming from the post-training side of the house.
因此，实验室确实转向了现在大部分收益来自于培训后的方面。
Garrett Lord: And what post-training is, is it's augmenting and improving the data they have across every discipline or capability area that they care about.
Garrett Lord：培训后的内容是增强和改进他们所关心的每个学科或能力领域的数据。
So take coding, or mathematics, or law or finance, they are focused on collecting high quality data that really improves the state of our capabilities, their models, and you can see a lot of these popular benchmarks on what are called model parts.
因此，以编码、数学、法律或金融为例，他们专注于收集真正提高我们能力和模型状态的高质量数据，你可以看到很多关于所谓模型部分的流行基准。
When Llama IV is released, you'll see the benchmarks across various domains, and each one of the research teams inside of the labs have different use cases.
当 Llama IV 发布时，您将看到各个领域的基准测试，并且实验室内的每个研究团队都有不同的用例。
Basically they're running experiments, almost think like the scientific process.
基本上他们正在进行实验，几乎像科学过程一样思考。
They have a hypothesis around how to improve the model.
他们对如何改进模型有一个假设。
They're trying to collect small pieces of data to see if that hypothesis works out.
他们正在尝试收集小块数据来看看这个假设是否成立。
If that hypothesis is proving true, then they expand the overall collection of the data in that advert.
如果该假设被证明是正确的，那么他们就会扩大该广告中数据的总体收集范围。
And it could look like reinforcement learning environments, it could look like trajectories, it could be audio and multimodal, it can be text-based like prompt-response pairs.
它可能看起来像强化学习环境，它可能看起来像轨迹，它可以是音频和多模式，它可以是基于文本的，如提示响应对。
Garrett Lord: It can also be reinforcement learning with human feedback, which is like preference ranking data.
Garrett Lord：它也可以是利用人类反馈的强化学习，就像偏好排名数据一样。
And so, that's the state of the art of models.
这就是模型的最新技术水平。
And most of the gains that are happening from models right now are coming from the post-training side of the house.
目前模型带来的大部分收益都来自训练后的部分。
And there's just an incredible amount of demand to stay at the absolute frontier of where models are going.
为了保持模型发展的绝对前沿，存在着令人难以置信的需求量。
Lenny Rachitsky: So training, pre-training is feeding it, say the entire internet.
Lenny Rachitsky：整个互联网都说，训练、预训练正在为其提供营养。
Here's like all the data that the humans have ever created, figure out knowledge and facts, and how to reason and all these things.
这就像人类创造的所有数据一样，找出知识和事实，以及如何推理等等。
Post-training, is it correct to say there's essentially two buckets of things to do?
培训后，说基本上有两件事要做，是否正确？
There's reinforcement learning, human feedback RLHF, and then there's kind of this bucket of fine-tuning?
有强化学习、人类反馈 RLHF，然后还有这种微调？
Garrett Lord: I mean, yes and no because take for example trajectories, or you want to be able to do, people use flight search or an accounting end-to-end process, or you want to be able to conduct biological experiments, you need actual trajectory data.
加勒特·洛德：我的意思是，是和否，因为以轨迹为例，或者你希望能够做到这一点，人们使用航班搜索或会计端到端流程，或者你希望能够进行生物实验，你需要实际的轨迹数据。
There's still very much, a lot of the labs are still, they have points of view on what data collect.
仍然有很多实验室仍然存在，他们对收集的数据有自己的观点。
It's evolving very quickly.
它的发展非常快。
But I think reinforcement learning is really preference ranking, like which question do you like more, question A or question B?
但我认为强化学习实际上是偏好排名，比如你更喜欢哪个问题，问题A还是问题B？
SFT data is a prompt and a response, and obviously the labs are very focused on these thinking or reasoning models.
SFT数据是一种提示，也是一种回应，显然实验室非常关注这些思维或推理模型。
So, in order to improve a reasoning model you need to actually have the step-by-step instructions, of which when you interact with a lot of these frontier models they struggle in very advanced domains.
因此，为了改进推理模型，您实际上需要有分步说明，当您与许多这些前沿模型交互时，它们会在非常高级的领域中遇到困难。
And so, I think there's a variety of data that they're working with to improve capabilities in their models.
因此，我认为他们正在使用各种数据来提高模型的功能。
Lenny Rachitsky: What I'm hearing is there's other ways to post-train.
Lenny Rachitsky：我听说还有其他方式进行后期训练。
Which of these are you guys focused on?
各位，你们关注的是哪一个呢？
Where do you help models most of these three-ish buckets?
您在哪里帮助对这三个左右的桶进行建模？
Garrett Lord: Our real unique proposition as a business is the fact that we have an engaged audience.
Garrett Lord：作为一家企业，我们真正独特的主张是我们拥有积极参与的受众。
We have 18 million professionals across, we have 500,000 PhDs, we have 3 million master students, we're a global platform.
我们拥有 1800 万专业人士、50 万博士、300 万硕士生，我们是一个全球平台。
And so, depending on what you're looking for across any area, academic knowledge, what is the definition of a PhD?
因此，根据您在任何领域寻找的学术知识，博士学位的定义是什么？
How do you get your PhD?
你如何获得博士学位？
You defend your thesis.
你为你的论文辩护。
Defending your thesis means, generally speaking, you have proven that you have extended the world's knowledge in a particular domain.
一般来说，论文答辩意味着你已经证明你已经在特定领域扩展了世界知识。
And so, the ability to hyper-target this audience into chemistry, math, physics, biology, coding and really touch parts of human knowledge that have never before made it to the internet is really where we excel.
因此，能够将这些受众定向到化学、数学、物理、生物、编码领域，并真正接触到以前从未在互联网上出现过的人类知识，这才是我们真正擅长的地方。
And I would say that when you talk about the labeling market, something to make it more abstract is like it used to be generalists' work.
我想说，当你谈论标签市场时，一些让它变得更加抽象的东西就像它曾经是通才的工作一样。
Garrett Lord: A lot of the market before the model started to get better was leveraging talented international lower cost labor to do basic generalist tasks.
加勒特·洛德（Garrett Lord）：在该模式开始变得更好之前，许多市场都在利用有才华的国际低成本劳动力来完成基本的通才任务。
But really what's happened is the models have gotten so good that the generalists are no longer needed.
但实际上，模型已经变得如此优秀，以至于不再需要通才。
What they really need is experts, experts across every area that the models are focused on.
他们真正需要的是专家，模型所关注的每个领域的专家。
And really, you could think about these model builders as they're focused on the most economically valuable capability areas in the economy.
实际上，您可以考虑这些模型构建者，因为他们专注于经济中最具经济价值的能力领域。
And so that, generally speaking, right now is focused on advanced STEM domains, advanced science and math domains, and then the derivative functions of accounting, law, medicine, finance, where they want to make the models more capable.
因此，一般来说，现在的重点是先进的 STEM 领域、先进的科学和数学领域，然后是会计、法律、医学、金融的衍生功能，他们希望使模型更加强大。
And then the work that we're doing, I think to come full circle to your question, we're doing work across so many domains.
然后我们正在做的工作，我想回到你的问题，我们正在跨多个领域开展工作。
I mean, we have millions of bachelor students that are being used for work in audio, work in customizing a model depending on the voice and tone, where you are geographically in the country, what do women versus men prefer?
我的意思是，我们有数百万学士学生从事音频工作，根据声音和语气定制模型，您在该国的地理位置，女性与男性更喜欢什么？
All the way to the most advanced PhD STEM domains out there.
一直到最先进的博士 STEM 领域。
Lenny Rachitsky: Okay.
莱尼·拉奇茨基：好的。
So, is it fair to say essentially all the data that is available has been trained on, and your company creates new data, new knowledge to continue advancing the intelligence of models?
那么，可以公平地说，基本上所有可用的数据都已经过训练，并且贵公司创建新数据、新知识来继续推进模型的智能吗？
Garrett Lord: Yep.
加勒特·洛德：是的。
And I also say we help point out where the models are weak.
我还说我们可以帮助指出模型的薄弱环节。
So, in order to break a model, it's pretty tough for the average person to break a model and get an incorrect response.
因此，为了打破一个模型，对于普通人来说，打破模型并得到错误的响应是相当困难的。
But if you are a PhD in physics, you can go in multiple subdomains of physics and prove where the model's actually breaking, either breaking in its reasoning steps or it's where it's broken in its ground truth right answer, or we start throwing tools in there or needing to follow some step-by-step process.
但如果你是物理学博士，你可以进入物理学的多个子领域并证明模型实际上在哪里出现了问题，要么是其推理步骤出现了问题，要么是其基本事实正确答案出现了问题，或者我们开始在那里投入工具或需要遵循一些逐步的过程。
And I wouldn't say it's easy for them, but the average person cannot break the models and that's where we really come in.
我不会说这对他们来说很容易，但普通人无法打破模型，这就是我们真正的切入点。
Lenny Rachitsky: So, essentially it's just catching mistakes that the model has made.
Lenny Rachitsky：所以，本质上它只是发现模型所犯的错误。
Okay.
好的。
So, what are these people actually doing?
那么，这些人实际上在做什么呢？
I know there's all kinds of different types, you described all the ways that data's generated, what kind of data is useful?
我知道有各种不同的类型，您描述了数据生成的所有方式，什么样的数据是有用的？
So, maybe just the most common examples.
所以，也许只是最常见的例子。
Let's say a PhD person is sitting there doing stuff, what are they actually doing?
假设一个博士坐在那里做一些事情，他们实际上在做什么？
Garrett Lord: A great example is a public paper called GPQA.
Garrett Lord：一个很好的例子是名为 GPQA 的公开论文。
So, for the engineers out there that want to read about it, essentially the crux of the paper is you break the model, you provide a ground truth, the right answer to the question, you provide the step-by-step a reasoning steps.
因此，对于想要阅读相关内容的工程师来说，本质上论文的关键是打破模型，提供基本事实，问题的正确答案，提供逐步的推理步骤。
So, you might imagine because models are non-deterministic, the model can get the answer right once, but it might not get the answer right three out of five times.
因此，您可能会想象，由于模型是不确定的，模型可以一次得到正确的答案，但五分之三可能无法得到正确的答案。
So, you actually prove where the model's failing.
所以，你实际上证明了模型的失败之处。
You actually break down into where is it failing?
你实际上会分解到哪里失败了？
Maybe it can get, it knows the question, it can get the right answer, but the actual steps to get there are wrong and they really focus on the steps to get there.
也许它可以得到，它知道问题，它可以得到正确的答案，但到达那里的实际步骤是错误的，他们真正关注的是到达那里的步骤。
Say there's 10 steps in a math problem, step 6 through 10 is wrong.
假设一道数学题有 10 个步骤，第 6 步到第 10 步是错误的。
So, how do you fix the actual steps?
那么，如何修复实际步骤呢？
Garrett Lord: And what are they doing?
加勒特·洛德：他们在做什么？
So they're going in, we're really focused on calling this branding the experience and treating people like experts.
所以他们进去了，我们真的专注于将这种品牌称为体验，并像专家一样对待人们。
PhD students expect to be treated different than a lower cost international labor with a different work expectation.
博士生期望获得与具有不同工作期望的低成本国际劳动力不同的待遇。
And so, these PhDs come into a community, we have a instructional design team and an assessments team that's going through and basically iteratively helping them understand how to use the tools that we built, and how to interact with the latest models.
因此，这些博士进入一个社区，我们有一个教学设计团队和一个评估团队，他们正在经历并基本上迭代地帮助他们了解如何使用我们构建的工具，以及如何与最新模型进行交互。
Then they go in and start actually creating data.
然后他们开始实际创建数据。
And that process is, on our side the model builders, they want to know that the data we're producing is high quality.
这个过程是，对于我们模型构建者来说，他们想知道我们生成的数据是高质量的。
Som we have our own research team, our own post-training team.
所以我们有自己的研究团队，自己的后期培训团队。
Garrett Lord: I hired a gentleman from Meta that went along on the post-training over there, and I-
加勒特·洛德：我从 Meta 聘请了一位绅士，他参加了那边的后期培训，然后我-
Garrett Lord: Yeah.
加勒特·洛德：是的。
So, war for AI talent is very expensive, but super, super privileged and proud to be working with him.
因此，人工智能人才的争夺是非常昂贵的，但与他合作是非常非常荣幸和自豪的。
And so, each unit of data, we have to build it an environment for them to actually create the data.
所以，每一个数据单元，我们都要给它搭建一个环境，让他们真正创建数据。
Then we have to understand in a unit level we're trying to approximate the actual gain from that piece of data and whether it can improve in a particular capability area.
然后，我们必须在单元级别上了解我们试图从该数据中近似获得实际收益，以及它是否可以在特定的能力领域得到改进。
And then, we're also focused on evolving the use cases to also follow what the model builders want, which is they want more more real world tool use and trajectory based data as well.
然后，我们还专注于发展用例，以遵循模型构建者的需求，即他们想要更多现实世界的工具使用和基于轨迹的数据。
Lenny Rachitsky: Okay.
莱尼·拉奇茨基：好的。
There's so much here, and we could go infinitely down here but I think that this is really interesting because just like people hear so much about all of this and they barely understand what the hell it actually is.
这里有太多东西，我们可以无限地深入这里，但我认为这真的很有趣，因为就像人们听到了很多关于这一切的事情，但他们几乎不明白它到底是什么。
So, this is for me really interesting.
所以，这对我来说真的很有趣。
I think it's going to help a lot of people.
我认为这会对很多人有帮助。
So essentially a PhD, say a biologist, biology PhD is just their job is find flaws in what, say ChatGPT is producing, and then come up with here's the correct answer.
因此，生物学家说，生物学博士本质上是博士学位，他们的工作就是找到 ChatGPT 所产生的缺陷，然后得出正确的答案。
And that is used to fine tune the mode, here's something you are doing incorrectly, here's the correct answer and that improves the model.
这是用来微调模式的，这是你做错的事情，这是正确的答案，可以改进模型。
Lenny Rachitsky: Is that a simple way to think about it?
Lenny Rachitsky：这是一个简单的思考方式吗？
And please correct anything I'm saying that is incorrect because I don't want people to misunderstand it.
请纠正我所说的任何不正确的地方，因为我不想让人们误解。
Garrett Lord: Yeah.
加勒特·洛德：是的。
I mean, a great example, let's take a non-verifiable domain like education.
我的意思是，一个很好的例子，让我们以教育这样不可验证的领域为例。
So there's a PhD student, Rachel on the network, she got her PhD from the University of Miami, spent two decades as a teacher teaching students in the eighth grade.
网络上有一位博士生，Rachel，她从迈阿密大学获得博士学位，当了二十年的老师，教八年级的学生。
And she was an adjunct professor at a local community college in the field of education.
她是当地一所社区大学教育领域的兼职教授。
And so, she is interacting with the state-of-the-art models in educational design.
因此，她正在与教育设计中最先进的模型进行互动。
So, actually trying to understand what is the best way to teach people, and how do you spot incorrect issues in a model in the way that they're training people, and help the models understand the forefront of educational design with the hands-on experience of being an eighth grade teacher for 10 plus years and having a PhD in education?
那么，真正尝试了解什么是最好的教人方式，如何在训练人的方式中发现模型中的错误问题，并通过十多年八年级教师和教育学博士学位的实践经验帮助模型了解教育设计的前沿？
So, that's an example of you can have that all the way down to a verifiable engineering problem that you're seeing the latest models fail on.
因此，这是一个例子，您可以将其一直归结为可验证的工程问题，而您会看到最新模型无法解决该问题。
Garrett Lord: Yeah, I think that gives you the gamut.
加勒特·洛德：是的，我认为这给了你全方位的信息。
You also have, we talk about professional domains like these reinforcement learning environments, there's a bunch of papers out there that basically speak to people narrating over their step-by-step tool use.
你也有，我们谈论像这些强化学习环境这样的专业领域，有很多论文基本上是在告诉人们如何逐步使用工具。
So, as they go to solve a problem from start to finish, interact with multiple different service areas, interact with multiple different tools, they're like, there's papers that talk about this, talking over what they're doing, actually following and screen recording where their mouse is going, how they're problem solving.
因此，当他们从头到尾解决问题，与多个不同的服务领域交互，与多个不同的工具交互时，他们就像有论文谈论这个，谈论他们正在做什么，实际上跟踪并屏幕记录他们的鼠标移动的位置，他们如何解决问题。
When they run into a roadblock, what do they do?
当他们遇到障碍时，他们会做什么？
So, they really want to understand how humans think.
所以，他们真的很想了解人类是如何思考的。
Lenny Rachitsky: You mentioned this term trajectory.
Lenny Rachitsky：您提到了轨迹这个术语。
Can you just explain what that actually means?
您能解释一下这实际上意味着什么吗？
Because it feels like you've mentioned that a few times and that feels important to all this.
因为感觉你已经提到过几次了，这对所有这一切都很重要。
Garrett Lord: Yeah.
加勒特·洛德：是的。
A trajectory is basically just like the entire environment that is collecting what you're doing.
轨迹基本上就像收集您正在做的事情的整个环境。
So it's your screen, it's your mouse-
所以这是你的屏幕，这是你的鼠标-
Lenny Rachitsky: Including this voiceover, okay.
Lenny Rachitsky：包括这个画外音，好吧。
And then, this might be too technical, but what is the output of all this work?
然后，这可能太技术性了，但是所有这些工作的输出是什么？
This, say teacher, is it just like a JSON file, an XML file, like a text file?
这个，老师说，是不是就像一个JSON文件，一个XML文件，一个文本文件一样？
Garrett Lord: Yeah, it can be managed JSON data.
Garrett Lord：是的，它可以管理 JSON 数据。
Garrett Lord: And then, we also have multimodal work like audio, like classifying music and understanding ...
加勒特·洛德：然后，我们还有音频等多模式工作，比如音乐分类和理解……
We're engaging thousands, or not thousands, probably hundreds of top music students at the music schools in the country who are improving models of understanding of music.
我们正在吸引该国音乐学校的数千名，或者不是数千名，可能是数百名顶尖音乐学生，他们正在改进对音乐的理解模型。
And you also have the thing called, which we haven't talked about here, a rubrics, and rubric models are, you can put a model in as a judge.
还有一种东西，我们在这里没有讨论过，即评分标准，而评分标准模型就是，你可以将模型作为法官放入其中。
What is a good educational design, or what's a good MRI result?
什么是好的教育设计，或者什么是好的 MRI 结果？
In some of these domains, you actually don't have a guaranteed correct right answer.
在其中一些领域，您实际上没有保证正确的答案。
And so, models can sit in the middle as a judge and actually understand what is ...
因此，模型可以作为法官坐在中间，真正理解什么是......
Think back on your school days.
回想一下你的学生时代。
How do you get A on your 5,000 word paper?
你如何在 5,000 字的论文上获得 A ？
Well, there's a great introductory statement and there's scientific proof.
嗯，有一个很好的介绍性陈述，并且有科学证据。
So, you can build a rubric that allows a model to sit in the middle and actually auto-evaluate responses.
因此，您可以构建一个评分标准，允许模型位于中间并实际自动评估响应。
We're seeing a lot of rubrics work as well.
我们看到很多规则也发挥了作用。
Lenny Rachitsky: And you would think, why would you trust this one teacher's opinion that this is the right way to do it?
Lenny Rachitsky：你会想，为什么你会相信这位老师的观点，即这是正确的做法？
But what's cool is the market speaks for itself.
但最酷的是市场不言而喻。
If these models are being used more and more, and people love them and value them, I imagine steps in between to verify this is good and other people think this is a good idea.
如果这些模型被越来越多地使用，并且人们喜欢它们并重视它们，我想在这之间采取步骤来验证这是好的并且其他人认为这是一个好主意。
It feels like the market dynamics will tell you if the data you're providing is correct at what people want.
感觉市场动态会告诉你你提供的数据是否符合人们的需求。
Is there something more there?
还有更多的东西吗？
Garrett Lord: I didn't get a PhD in AI, or math, or physics, and I haven't trained myself, we have frontier models, but there is a lot to each unit of data whether it's improving.
Garrett Lord：我没有获得人工智能、数学或物理方面的博士学位，我也没有训练过自己，我们有前沿模型，但每个数据单元是否在改进都有很多东西。
There's a ton of science and research out right now around how do you make sure that the data that you're producing is improving the model?
目前有大量的科学研究围绕如何确保您生成的数据正在改进模型？
And it's very hard for model builders to understand.
模型构建者很难理解。
They really care about, to zoom out, they care about three things.
他们真正关心的是，缩小范围，他们关心三件事。
They care about quality first and foremost.
他们首先关心的是质量。
You have to have high quality data.
您必须拥有高质量的数据。
And if you imagine you're training a model, like teaching a student and you're giving it the wrong data, it's extremely challenging to overcome that.
如果你想象你正在训练一个模型，就像教一个学生一样，但你给了它错误的数据，那么克服这个问题就非常具有挑战性。
So, quality is first and foremost.
所以，质量是第一位的。
And then, the other huge problems you have is volume.
然后，你遇到的另一个大问题是数量。
How do you generate thousands of pieces of data in the most advanced domains of chemistry, and mathematics, and physics, and how do you ensure that it's high quality?
如何在最先进的化学、数学和物理领域生成数千条数据，以及如何确保其高质量？
Garrett Lord: Well for us, say in physics, we just reach out to students at Stanford, and Berkeley, and MIT, and they're at the top GPA at the best physics schools in the country.
加勒特·洛德（Garrett Lord）：对我们来说，比如说物理学，我们只接触斯坦福大学、伯克利分校和麻省理工学院的学生，他们在全国最好的物理学校的平均绩点（GPA）名列前茅。
And so, our ability to get to scale or volumes of data, to produce very high quality data, is something they care deeply about.
因此，我们获得规模或数据量、生成高质量数据的能力是他们非常关心的。
And then, the other thing I would say model builders care about is speed, because they have all these hypotheses and they're constantly testing different pipelines.
然后，我想说模型构建者关心的另一件事是速度，因为他们有所有这些假设，并且他们不断测试不同的管道。
And so, you might have three or four bets going at once, and then as soon as one is actually showing a gain, imagine you're a researcher or you're assigned to the processes, once you're running a gain then you're trying to grow that pipeline and grow that piece of data that's actually improving it, and you're maybe ditching two or three other projects you had that weren't showing improvement.
因此，你可能会同时进行三到四个赌注，然后一旦一个赌注真正显示出收益，想象你是一名研究人员或者你被分配到流程中，一旦你运行了收益，那么你就会尝试扩展该管道并增长实际上正在改进它的那条数据，并且你可能会放弃其他两三个没有显示出改进的项目。
Garrett Lord: So, your ability to quickly turn around for them in a period of days, and then get to high volumes of data that are high quality is the number one thing they care about.
Garrett Lord：因此，他们最关心的第一件事是你能否在几天内快速为他们提供周转服务，然后获取大量高质量数据。
And so, there's quite a bit of technology we built on our side to assess each unit of data.
因此，我们开发了相当多的技术来评估每个数据单元。
We have our own post-training teams, we're renting our own GPUs, and we're trying to make sure that we can sit directly with these researchers and help share what we're seeing with the data that we're creating and how it could improve their model, how they could best train with it.
我们有自己的后期培训团队，我们正在租用自己的 GPU，我们正在努力确保我们可以直接与这些研究人员坐在一起，帮助分享我们所看到的数据以及我们正在创建的数据，以及如何改进他们的模型，以及他们如何最好地利用这些数据进行训练。
So, hopefully that helps.
所以，希望这会有所帮助。
Lenny Rachitsky: Going back to the types of post-training, just because I think this might be helpful, at least for me the mental model of there's pre-training, there's post-training, within post-training there's reinforcement learning, human feedback, there's this concept of fine-tuning.
Lenny Rachitsky：回到训练后的类型，只是因为我认为这可能会有所帮助，至少对我来说，有预训练、训练后的心理模型，训练后有强化学习、人类反馈，还有微调的概念。
There's also evals and stuff like-
还有评估和诸如此类的东西
Lenny Rachitsky: SFT, which is supervised fine-tuning?
Lenny Rachitsky：SFT，即监督微调？
Is that-
难道是——
Lenny Rachitsky: Yeah.
莱尼·拉奇茨基：是的。
So, the stuff you've been describing, would you mostly describe that as supervised fine-tuning?
那么，您所描述的内容，您主要会将其描述为监督微调吗？
Garrett Lord: Yes, and we're doing all of the above.
Garrett Lord：是的，我们正在做上述所有事情。
We don't do the auto eval, we produce rubrics which are used auto evals.
我们不进行自动评估，我们生成用于自动评估的量规。
But yeah.
但是是的。
Lenny Rachitsky: Okay, awesome.
莱尼·拉奇茨基：好的，太棒了。
So essentially there's a model, it's trained on all this amazing data.
所以本质上有一个模型，它是根据所有这些惊人的数据进行训练的。
You guys come in after the model's trained to tweak the weights based on additional data that you create.
在模型经过训练后，你们就可以根据您创建的附加数据来调整权重。
What's interesting is that this is a scalable system.
有趣的是，这是一个可扩展的系统。
I want to talk about just the supply of amazing people that you have producing this, but it's amazing that humans can do this.
我想谈谈你们所拥有的令人惊叹的人才的供应，但人类能够做到这一点真是令人惊奇。
You would think it needs to be this infinitely scalable thing, but humans sitting there creating data is working and improving the intelligence of models significantly.
你可能会认为它需要是无限可扩展的东西，但坐在那里创建数据的人类正在发挥作用并显着提高模型的智能。
Garrett Lord: Oh, yeah.
加勒特·洛德：哦，是的。
I mean, I think maybe a funny joke is all the MBAs think this is all just going to go away.
我的意思是，我认为也许有一个有趣的笑话是，所有 MBA 都认为这一切都会消失。
And I think for as long as models are improving, humans will be needed in this process.
我认为只要模型不断改进，这个过程就需要人类。
And when you talk to the lead scientists and researchers at these labs, it's like the data types will evolve and what they're trying to capture and collect, but there will be humans needed in the space for the next decade until we reach full ASI.
当你与这些实验室的主要科学家和研究人员交谈时，你会发现数据类型以及他们试图捕获和收集的内容将会不断发展，但在未来十年，直到我们达到完全的 ASI 之前，太空中仍需要人类。
So yeah, I mean, you think about in a lot of them will struggle to do basic trajectories right now.
所以，是的，我的意思是，你认为他们中的很多人现在都很难做到基本的轨迹。
So, right now people are very focused on academic domains, and I think they'll continue to be focused on academic domains, but there will also be far, far more demand for professional domains as well across basically every trajectory or step-by-step problem that a knowledge worker solves in the workplace, it's the pursuit of these labs to make sure that they're trying to collect the data to help add as much value in that process for humans as possible.
因此，现在人们非常关注学术领域，我认为他们将继续关注学术领域，但对专业领域的需求也将大大增加，以及知识工作者在工作场所解决的基本上每一个轨迹或逐步问题，这些实验室的追求是确保他们正在努力收集数据，以帮助在这个过程中为人类增加尽可能多的价值。
Lenny Rachitsky: So, let me ask you about this.
Lenny Rachitsky：所以，让我问你这个问题。
There's this tension, I imagine, people might feel between all these students training models to become smarter, and smarter, and smarter, and then there's that they will have harder time potentially finding jobs if models are so smart that people at entry level aren't being hired as much.
我想，人们可能会在所有这些训练模型变得越来越聪明的学生之间感到紧张，然后，如果模型太聪明，以至于入门级的人不会被雇用，那么他们将更难找到工作。
How do you think about just that tension?
您如何看待这种紧张感？
Do you think this is a real problem or not, or do you think this goes-
你认为这是否是一个真正的问题，或者你认为这是-
Garrett Lord: I'm probably in the camp of like GDP growth over universal basic income.
加勒特·洛德（Garrett Lord）：我可能属于GDP增长超过全民基本收入的阵营。
I very much believe that this is going to improve and accelerate every human's ability to create an impact in the economy in the world, and that we're hearing from, there's like a million companies that use Handshake.
我非常相信，这将提高和加速每个人对世界经济产生影响的能力，而且我们听说，有大约一百万家公司在使用 Handshake。
100% of the Fortune 500 uses Handshake, so we basically power the vast majority of how young people find jobs, and a lot of people are hyperbolic at saying that all young people won't have jobs, and that's not what we're hearing from our employers.
财富 500 强企业 100% 使用 Handshake，因此我们基本上为绝大多数年轻人寻找工作提供了动力，而且很多人夸张地说所有年轻人都不会找到工作，而这并不是我们从雇主那里听到的。
What we're hearing is pick social media marketing, before you needed somebody that could do Photoshop, and take pictures, and create the videos.
我们听到的是选择社交媒体营销，然后你需要有人会做 Photoshop、拍照和制作视频。
Then you needed somebody that understood marketing analytics platforms to track your posting on different social media forms.
然后，您需要了解营销分析平台的人来跟踪您在不同社交媒体表单上的发布。
It's like now one person, one young, talented, AI native, Iron Man suit enabled young person can get on and they can build their own videos, produce their own creative assets, post across multiple social media platforms, run all of their own analytics.
就像现在一个人，一个年轻的，有才华的，人工智能原生的，启用钢铁侠套装的年轻人可以上手，他们可以制作自己的视频，制作自己的创意资产，在多个社交媒体平台上发布，运行自己的所有分析。
They don't need a data science degree to be able to do that.
他们不需要数据科学学位就能做到这一点。
Garrett Lord: Or take an intern in our company, he had his first PR up I think the afternoon he started.
加勒特·洛德（Garrett Lord）：或者在我们公司找个实习生，我想他上班的那个下午就完成了他的第一份公关。
You were a PM, you realize how challenging that would've been historically to get your dev environment set up and figure out where to add value.
您是一名产品经理，您意识到设置开发环境并找出在哪里增加价值在历史上是多么具有挑战性。
You just took a bug and squashed it.
你只是抓住了一个错误并把它压扁了。
And so, I'm really a believer this is just enabling human beings to be even more productive and create more impact.
因此，我坚信这只会让人类变得更有生产力并创造更大的影响力。
And yeah, of course, hundreds of millions of jobs, the jobs will evolve.
是的，当然，数以亿计的工作岗位将会不断发展。
People will become displaced, they'll have to upscale and rescale, and I think Handshake has a huge role to play in helping knowledge workers evolve.
人们将会流离失所，他们将不得不升级和重新调整规模，我认为握手在帮助知识工作者发展方面可以发挥巨大的作用。
Lenny Rachitsky: This has come up a couple of times this point that I think is really good, that younger people coming out of school are actually going to be much more likely to be successful because they're growing up with these tools, and are much more native to all these advanced tools and so they just come in as beasts just doing so much more.
Lenny Rachitsky：这一点已经出现过好几次了，我认为这真的很好，从学校出来的年轻人实际上更有可能成功，因为他们是在这些工具的陪伴下长大的，并且对所有这些先进的工具更加熟悉，所以他们只是像野兽一样做更多的事情。
Garrett Lord: Well, do you remember when, this a little bit predates me, but you used to put Google search on as a skill on your resume.
加勒特·洛德：嗯，你还记得吗，这有点早于我，但你曾经把谷歌搜索作为你简历上的一项技能。
You were person, you were good at Googling, because you grew up with Google.
你是一个人，你擅长谷歌搜索，因为你是和谷歌一起长大的。
It's like I think being AI native and having your Iron Man suit on, and understanding how to leverage these tools is like young people are at a huge advantage.
我认为，作为人工智能原生者，穿上钢铁侠战衣，了解如何利用这些工具，年轻人就拥有巨大的优势。
Lenny Rachitsky: Yeah.
莱尼·拉奇茨基：是的。
And especially if they're involved in training these models, I imagine there's some other cool advantage there.
特别是如果他们参与训练这些模型，我想还有一些其他很酷的优势。
Garrett Lord: Yeah.
加勒特·洛德：是的。
Well I mean, just to hit on that, what we're hearing from our thousands of fellows is they're in the classroom, they're actually producing research.
好吧，我的意思是，顺便说一句，我们从数千名同事那里听到的是，他们在课堂上，实际上正在进行研究。
We're talking about PhDs at the top institutions in the country.
我们谈论的是该国顶尖机构的博士学位。
They can make 100, 150, $200 an hour in their area, in their field of expertise.
他们在自己的专业领域每小时可以赚 100、150、200 美元。
It's pretty sweet.
非常甜蜜。
You can make 25 bucks an hour being a teacher's assistant, or you can actually make $150 an hour breaking the latest models, and what we're hearing from our fellows is they're bringing a lot of those insights into the classroom to help them be more effective at teaching.
作为一名助教你可以每小时赚 25 美元，或者你实际上可以通过打破最新模型每小时赚 150 美元，我们从同事那里听到的是，他们正在将很多这些见解带入课堂，以帮助他们更有效地教学。
More importantly, they're starting to learn how to leverage these tools to actually advance their area of research.
更重要的是，他们开始学习如何利用这些工具来实际推进他们的研究领域。
So, they believe that these tools can help them advance their area of research by helping them be more effective with their time.
因此，他们相信这些工具可以帮助他们更有效地利用时间，从而帮助他们推进研究领域。
And so, it is quite cool to get paid to learn a skill.
因此，通过学习一项技能获得报酬是非常酷的。
Lenny Rachitsky: Before I get to the story of how this all emerged, because that is an incredible story, is there anything else about this whole field of labeling, of reinforcement learning that you think people just don't fully understand or you think that is really important?
Lenny Rachitsky：在我开始讲述这一切是如何出现的故事之前，因为这是一个令人难以置信的故事，关于整个标签领域，强化学习，还有什么你认为人们不完全理解或者你认为这真的很重要的吗？
There's just so much happening.
发生了很多事情。
Like I said, some of the fastest growing companies in the world are in the space, Scale was just acquired for 30, sort of acquired for $30 billion.
就像我说的，世界上一些发展最快的公司都在这个领域，Scale 刚刚被 30 美元收购，相当于 300 亿美元。
Just what else is there, if there's anything, that you think people need to understand?
如果有的话，您认为人们还需要了解什么？
Garrett Lord: Generally speaking, anytime that you're interacting with a model and you're asking it to do really advanced things, and it's not performing your expectations, like somewhere there's probably an expert that is the top mind in that domain working directly for the best researchers in the world at the Frontier Labs trying to understand and go through the scientific iteration process of how to make that better.
Garrett Lord：一般来说，任何时候当你与模型交互时，你要求它做真正先进的事情，但它并没有达到你的预期，就像某个地方可能有一位专家，他是该领域的顶尖专家，直接为前沿实验室世界上最好的研究人员工作，试图理解并经历如何改进模型的科学迭代过程。
And that the assumption there is that they already have the entirety of human knowledge that's written and recorded.
假设他们已经拥有书面和记录的全部人类知识。
And so, for as long as there are problems in solving any problem with AI, any human problem, there will need to be humans in the loop helping advance that.
因此，只要人工智能在解决任何问题、任何人类问题时存在问题，就需要有人参与其中，帮助推进这一进程。
And models don't generalize.
而且模型不能概括。
I mean, obviously the field will advance a lot and the type of data they'll collect will evolve a lot, but it's pretty exciting at the frontier.
我的意思是，显然这个领域将会取得很大的进步，他们收集的数据类型也会发生很大的变化，但它在前沿是非常令人兴奋的。
Lenny Rachitsky: Kevin Wheel was on the podcast, the CBO at OpenAI, and he made this point that really stuck with me that the model of today is the worst model you will ever use.
Lenny Rachitsky：Kevin Wheel 是 OpenAI 的 CBO，他在播客中提出了这一点，这一点让我印象深刻，即今天的模型是你使用过的最糟糕的模型。
Lenny Rachitsky: Will only get better, just boggles the mind, and now we know why things are getting better because all the work you guys are doing.
Lenny Rachitsky：只会变得更好，令人难以置信，现在我们知道为什么事情会变得更好，因为你们所做的所有工作。
Just one quick question on this whole scale thing, I guess they were, I don't know, the main company doing this, now they're swallowed up and Alex is running superintelligence in Meta.
关于整个规模的事情，我想问一个简单的问题，我猜他们是，我不知道，做这件事的主要公司，现在他们被吞并了，亚历克斯正在 Meta 中运行超级智能。
Are they still a big player in this labeling space or are they out of it and that's a big opportunity?
他们仍然是这个标签领域的重要参与者，还是已经退出这个领域，这是一个巨大的机会？
Garrett Lord: Yeah.
加勒特·洛德：是的。
I mean, kudos to the whole Scale team, a lot of respect for what they built, just many great companies operating the space.
我的意思是，对整个 Scale 团队表示敬意，对他们所创造的东西非常尊重，还有许多伟大的公司在运营这个领域。
I think to the core of your question, I think if you viewed your research team and your model building team, and the experiments they're running to be really the cornerstone of how you're improving, you probably wouldn't want the latest research of what you're trying to work on being invested in by a peer.
我认为你问题的核心是，如果你认为你的研究团队和模型构建团队以及他们正在进行的实验确实是你如何改进的基石，你可能不希望你正在努力的最新研究被同行投资。
I mean, that's just generally what we hear in this space.
我的意思是，这就是我们在这个领域听到的普遍情况。
And so, we have seen an incredible search and demand, and are I think extraordinarily well positioned.
因此，我们看到了令人难以置信的搜索和需求，而且我认为处于非常有利的位置。
We like to say the only moat in human data is access to an audience.
我们想说，人类数据中唯一的护城河是接触受众。
Basically, there are many, many small players in this space, some midsize players in the space, and they're basically running TikTok ads, running Instagram ads, paying money for Google Search display ads, YouTube ads, and they will be like, "Can you get me 200 physics PhDs?"
基本上，这个领域有很多很多小型玩家，还有一些中型玩家，他们基本上都在投放 TikTok 广告、投放 Instagram 广告、付费购买 Google 搜索展示广告、YouTube 广告，他们会说，“你能给我 200 个物理学博士学位吗？”
Garrett Lord: What do they do?
加勒特·洛德：他们做什么？
They only can do one thing.
他们只能做一件事。
They have 100 recruiters on staff, they all get on LinkedIn, they all send messages, they spend a couple million bucks on performance advertising campaigns.
他们有 100 名招聘人员，他们都在 LinkedIn 上，他们都发送消息，他们在绩效广告活动上花费了几百万美元。
Somebody's scrolling their Instagram feed that's a physics PhD of which you can't target them that well and they like see, "Come train a model." It's like, "I've never heard of this brand before." The huge advantage that we've had and why we've resonated so fast in the marketplace is we built a decade of trust with 18 million people, and they trust us, and we built a ton of brand affinity, and they use Handshake, and they have an active profile, and we have a ton of information around their academic performance and what they've done in school.
有人正在滚动他们的 Instagram 动态，内容是一位物理学博士，你无法很好地瞄准他们，他们喜欢看到“来训练一个模型”。就像“我以前从未听说过这个品牌”。我们拥有的巨大优势以及我们在市场上如此迅速引起共鸣的原因是，我们与 1800 万人建立了十年的信任，他们信任我们，我们建立了大量的品牌亲和力，他们使用 Handshake，他们拥有活跃的个人资料，我们拥有大量有关他们的学业表现和他们在学校所做的事情的信息。
And so, we're able to really target people really effectively, and get to scale and volume of high quality data faster than anyone else.
因此，我们能够真正有效地瞄准人群，并比其他任何人更快地获得高质量数据的规模和数量。
And I think that competitive advantage of access to an audience is really resonating in the marketplace.
我认为吸引观众的竞争优势确实在市场上引起了共鸣。
Lenny Rachitsky: Today's episode is brought to you by Anthropic, the team behind Claude.
Lenny Rachitsky：今天的节目是由 Claude 背后的团队 Anthropic 为您带来的。
I use Claude at least 10 times a day.
我每天至少使用 Claude 10 次。
I use it for researching my podcast guests, for brainstorming title ideas for both my podcast and my newsletter, for getting feedback on my writing and all kinds of stuff.
我用它来研究我的播客嘉宾，集思广益地讨论我的播客和时事通讯的标题想法，获取关于我的写作和各种东西的反馈。
Just last week I was preparing for an interview with a very fancy guest, and I had Claude tell me what are all the questions that other podcast hosts have asked this guest so that I don't ask them these questions?
就在上周，我正准备采访一位非常出色的嘉宾，我让克劳德告诉我其他播客主持人问过这位嘉宾的所有问题是什么，这样我就不会问他们这些问题了？
How much time do you spend every week trying to synthesize all of your user research insights, support tickets, sales calls, experiment results and competitive intel?
您每周花费多少时间来综合所有用户研究见解、支持请求、销售电话、实验结果和竞争情报？
Claude can handle incredibly complex multistep work.
克劳德可以处理极其复杂的多步骤工作。
You can throw a 100-page strategy document at it and ask it for insights, or you can dump all your user research and ask it to find patterns.
你可以向它扔一份 100 页的策略文档并要求它提供见解，或者你可以转储所有用户研究并要求它寻找模式。
With Claude 4 and the new integrations including Claude 4 Opus, the world's best coding model, you get voice conversations, advanced research capabilities, direct Google Workspace integration, and now MCP connections to your custom tools and data sources.
借助 Claude 4 和包括世界上最好的编码模型 Claude 4 Opus 在内的新集成，您可以获得语音对话、高级研究功能、直接 Google Workspace 集成，以及现在与自定义工具和数据源的 MCP 连接。
Claude just becomes part of your workflow.
克劳德只是成为您工作流程的一部分。
Lenny Rachitsky: If you want to try it out, get started at Claude.ai/Lenny, and using this link you get an incredible 50% off your first three months of the pro plan.
Lenny Rachitsky：如果您想尝试一下，请从 Claude.ai/Lenny 开始，使用此链接您可以在专业计划的前三个月享受令人难以置信的 50% 折扣。
That's Claude.ai/Lenny.
这是 Claude.ai/Lenny。
Lenny Rachitsky: Okay, this is an awesome segue to where I wanted to go, which is just how this business emerged.
Lenny Rachitsky：好的，这是我想去的地方的一个很棒的延续，这就是这个业务的兴起。
This is a business that you built on top of a business you've already had.
这是您在已有业务基础上建立的业务。
From what I understand, you were at like $150 million in revenue, you've been at this for a long time.
据我了解，你们的收入约为 1.5 亿美元，你们已经这样做很长时间了。
You found this opportunity, and now that looking back it's like obviously this is an amazing idea, labs need data, you guys have the supply of incredible experts.
你发现了这个机会，现在回想起来，这显然是一个了不起的想法，实验室需要数据，你们拥有令人难以置信的专家。
What an opportunity.
多么好的机会啊。
Talk about just how you first realized this was something that you could be doing, and should be doing, and then how you started to execute down this path.
谈谈你是如何第一次意识到这是你可以做、应该做的事情，然后你是如何开始沿着这条路执行的。
Garrett Lord: Yeah, I think it's been a pretty natural extension from helping people jumpstart, restart or start their career.
加勒特·洛德：是的，我认为这是帮助人们快速启动、重新启动或开始他们的职业生涯的一个非常自然的延伸。
Monetizing your skills and this new employment ecosystem is going to look very different in the future, and to zoom into how we discovered it, it's like because we have such a large access to this audience, and as the world shifted from generalists to experts, we're the largest expert network in the world.
将你的技能货币化和这个新的就业生态系统在未来看起来会非常不同，并且放大我们是如何发现它的，就像因为我们有如此多的受众群体，随着世界从通才转向专家，我们是世界上最大的专家网络。
We have more PhDs, 500,000 of them use Handshake than any other platform.
我们有更多的博士，其中 500,000 名博士使用 Handshake 比任何其他平台都多。
We have three million master students who are in school or alumni.
我们有三百万在校硕士生或校友。
And so, we started to see all what I would call middleman companies reaching out to us saying, "Can we recruit your PhDs and master's students?" And like any great marketplace we started sending them to these different platforms, and started to really realize that from hearing from our users that the experience was really frustrating.
因此，我们开始看到所有我称之为中间商的公司向我们伸出援手，说：“我们可以招募你们的博士生和硕士生吗？”就像任何伟大的市场一样，我们开始将他们发送到这些不同的平台，并开始真正意识到，从我们的用户那里听到的体验确实令人沮丧。
Garrett Lord: Training was very transactional, it was very amorphous how you could get paid.
加勒特·洛德（Garrett Lord）：培训非常具有交易性，如何获得报酬非常不明确。
There was immense amount of drop-off in the process to actual project like completion on these other platforms.
在其他平台上完成实际项目的过程中存在大量的下降。
So, we started to think the company was making tens of millions of dollars from helping these other platforms, and we started to realize what really kicked it off was hearing also from the Frontier Labs, they started to reach out to us and started to go direct and trying to almost cut out the middleman.
因此，我们开始认为该公司通过帮助这些其他平台赚取了数千万美元，并且我们开始意识到真正启动它的是来自 Frontier Labs 的消息，他们开始与我们联系并开始直接并试图几乎消除中间商。
And we started to realize, well, we could really serve our fellows, our PhDs, our experts, we could treat them.
我们开始意识到，我们可以真正为我们的同事、我们的博士、我们的专家服务，我们可以治疗他们。
We just believe there will need to be a platform, an experts first platform in the pursuit of ASI and advancing AI, and there will need to be a place that everyone in the world could go to, to monetize their skills and their knowledge as these labs are focused on improving in all these multidisciplinary.
我们只是相信，需要一个平台，一个专家优先的平台来追求 ASI 和推进人工智能，并且需要一个世界上每个人都可以去的地方，将他们的技能和知识货币化，因为这些实验室专注于在所有这些多学科领域进行改进。
And yeah, we entered the business in, really I started doing it over Christmas and New Year's.
是的，我们进入这个行业，实际上我是在圣诞节和新年期间开始做的。
That's when I started flying around.
就在那时，我开始四处飞翔。
Garrett Lord: My family thought it was a little wild that I was on planes trying to chase different leaders, but we built an incredible team of people that came from the human data world, and really started building out our platform in January, and then started really monetizing the relationships about five months ago.
加勒特·洛德：我的家人认为我在飞机上试图追逐不同的领导者有点疯狂，但我们建立了一支来自人类数据世界的令人难以置信的团队，并在一月份真正开始构建我们的平台，然后大约五个月前开始真正将关系货币化。
Fast-forward to today, we're working with seven of the Frontier Labs, basically every lab that's doing work and building the best large language models, and the team has exploded and revenue's exploded, and it's been really a incredible ride running back new company inside of a company for the second time over again.
快进到今天，我们正在与七个前沿实验室合作，基本上每个实验室都在开展工作并构建最好的大型语言模型，并且团队已经爆炸式增长，收入也呈爆炸性增长，这确实是一次令人难以置信的旅程，再次在公司内部运行新公司。
Lenny Rachitsky: And just to share some numbers, tell me if this is correct or if you're sharing these, but I heard that you hit 50 million in revenue just four months into this?
Lenny Rachitsky：只是分享一些数字，请告诉我这是否正确，或者您是否正在分享这些数据，但我听说您仅四个月就达到了 5000 万的收入？
Today we're at eight months in and you're on track to hit $100 million in revenue in the first year.
现在已经过去 8 个月了，您有望在第一年达到 1 亿美元的收入。
Garrett Lord: I think we'll blow through that number, but yeah.
加勒特·洛德：我想我们会突破这个数字，但是是的。
Lenny Rachitsky: Okay.
莱尼·拉奇茨基：好的。
Incredible.
极好的。
And I didn't even know there were seven Frontier Labs, that's-
我什至不知道有七个前沿实验室，那就是——
Garrett Lord: Zero to 50 is pretty good in four months, I think.
Garrett Lord：我认为四个月内从零到 50 已经相当不错了。
Lenny Rachitsky: Think zero to 50 million in four months, that's something.
Lenny Rachitsky：四个月内销量从 0 到 5000 万，这已经很了不起了。
It's like the bar has been shifting constantly.
就好像酒吧一直在不断变化。
A year ago that'd be legendary.
一年前，这将是一个传奇。
Now it's like, all right, well another one of these.
现在，好吧，又是其中之一。
50 million in four months, no big deal.
四个月5000万，没什么大不了的。
It's truly insane.
这真是太疯狂了。
Just to zoom out one second, for people that don't know a ton about Handshake, the original business, what was that?
缩小一秒钟，对于那些对 Handshake 最初的业务不太了解的人来说，那是什么？
What was actually this network that you had, that you sat on top of?
您拥有的、位于其之上的这个网络实际上是什么？
Garrett Lord: Yeah, that network does about 200 million.
Garrett Lord：是的，该网络的处理量约为 2 亿。
This will do about
这将做关于
Garrett Lord: So, we have 600-ish super passionate teammates that work on the core business, which I would separate those.
Garrett Lord：所以，我们有 600 名左右超级热情的团队成员，他们致力于核心业务，我会将他们分开。
These aren't two businesses, I think it's one business, but what is that business?
这不是两项业务，我认为这是一项业务，但那项业务是什么？
If you're a young person in America that's graduated in the last five, six, seven, eight years, you probably have Handshake on your phone.
如果您是最近五、六年、七、八年毕业的美国年轻人，您的手机上可能装有 Handshake。
You definitely know what Handshake is.
你肯定知道什么是握手。
It's a verb with young people in America, it's a verb with people that are in college in their PhD or master's program, and it is, I call it an unconnected graph, meaning you don't need to ...
这是一个与美国年轻人有关的动词，这是一个与在大学读博士或硕士课程的人有关的动词，我称之为不连通图，这意味着你不需要......
LinkedIn is very focused on who you know and what your experience is.
LinkedIn 非常注重您认识的人以及您的经历。
The first question on LinkedIn is what's your job?
LinkedIn 上的第一个问题是你的工作是什么？
And a lot of young people start off, they've never had a job before.
很多年轻人刚开始工作时，他们以前从未有过工作。
They don't have 500 connections to add to their graph.
他们没有 500 个连接可以添加到图表中。
Garrett Lord: Whereas on Handshake, you start off trying to discover, and explore, and figure out how to navigate through school and figure out, "Oh, I'm an engineer.
加勒特·洛德：而在握手中，你开始尝试发现、探索、弄清楚如何在学校中导航并弄清楚，“哦，我是一名工程师。
Maybe I want to be a PM, maybe I want to work at a startup, maybe I want to go to a larger company." What are the pros and cons you want to learn from your peers and young alumni?
也许我想成为一名产品经理，也许我想在一家初创公司工作，也许我想去一家更大的公司。”您想向同行和年轻校友学习哪些优点和缺点？
And so, Handshake's this I call a very social platform with groups, and messaging, and profiles, and short-form video and feed, all focused on your interests and helping really build your confidence in your early career to find your first job, your second job, and to manage 18 to 30, I would say.
因此，Handshake 是我所说的一个非常社交化的平台，其中包含群组、消息传递、个人资料、短视频和提要，所有这些都专注于您的兴趣，并帮助您真正建立对您早期职业生涯的信心，找到您的第一份工作，您的第二份工作，并管理 18 到 30 岁的工作。
Lenny Rachitsky: And how long has that business been around?
Lenny Rachitsky：这项业务已经存在多久了？
Lenny Rachitsky: 10 years, okay.
莱尼·拉奇茨基：10 年，好吧。
So it's just again, it just feels like such a holy shit, you guys are in the right place in the right time with the right network that is extremely valuable now.
所以，这又是一次，感觉真是太棒了，你们在正确的时间出现在正确的地点，拥有正确的网络，这现在非常有价值。
What an interesting story.
多么有趣的故事啊。
I feel like it's just another interesting example of you've been doing something for a long time and then all of a sudden AI is just, opens up a whole new way of leveraging something that you have been doing for a long time.
我觉得这只是另一个有趣的例子，你已经做了很长时间的某件事，然后突然间人工智能开启了一种全新的方式来利用你已经做了很长时间的事情。
It makes me think a little better about Bolt, and StackBlitz, which was building for seven years this browser based OS where you could run an OS in the browser.
这让我对 Bolt 和 StackBlitz 有了更好的认识，他们花了七年时间构建了这个基于浏览器的操作系统，你可以在浏览器中运行操作系统。
And they're like, "I don't know, no one needs this.
他们会说，“我不知道，没有人需要这个。
What are we doing?" And then, all of a sudden AI and they're like, "Oh, what if we build AI apps in the browser and just generate products for you with AI?" And now it's, I don't know, one of the fastest growing companies in the world.
我们在做什么？”然后，突然出现人工智能，他们会说，“哦，如果我们在浏览器中构建人工智能应用程序，然后用人工智能为你生成产品，会怎么样？”现在，我不知道，它是世界上增长最快的公司之一。
Lenny Rachitsky: So interesting.
莱尼·拉奇茨基：太有趣了。
And so, I think this is just an interesting time for our people to think about what have we done that may give us a new opportunity to build something huge based on this unfair advantage that we have?
因此，我认为这对我们的人民来说是一个有趣的时刻，让我们思考我们做了什么，可能会给我们一个新的机会，基于我们拥有的这种不公平优势来建立巨大的东西？
Garrett Lord: I think also as your company grows in size and headcount, and maturity, it's also hard to incubate something new inside of a business.
Garrett Lord：我认为，随着公司规模、员工数量和成熟度的增长，在企业内部孵化新的东西也很困难。
It's hard in so many ways.
从很多方面来说这都很困难。
The way that you build zero to one, and find product market fit, and scale a team very quickly and is very different than the way that you run a more mature business that has been around for 10 years with hundreds, and hundreds, and hundreds of people.
从零到一的构建、找到适合市场的产品以及快速扩展团队的方式，与运营一家已经存在了 10 年、拥有数百人、数百人的更成熟企业的方式有很大不同。
So, I've really had a ton of fun and found a ton of passion in running it back again for the second time inside the business.
所以，我真的得到了很多乐趣，并在第二次在公司内部再次运行它时找到了巨大的热情。
And then yeah, we have this massive strategic advantage, which is no cost or acquisition costs, and we have much higher conversion rates and retention than any of the other platforms by a large margin because we have such consumer affinity.
然后，是的，我们拥有巨大的战略优势，这是没有成本或收购成本的，而且我们的转化率和保留率比任何其他平台都要高得多，因为我们有这样的消费者亲和力。
Lenny Rachitsky: There's actually two threads here I want to follow, I'm going to follow the second one first, this idea of where this data labeling work can come from.
Lenny Rachitsky：实际上我想关注两条线索，我将首先关注第二条线索，即数据标记工作的来源。
This isn't a really clear, simple, understandable one, which is just experts sitting there creating data.
这并不是一个真正清晰、简单、易于理解的问题，只是专家坐在那里创建数据。
Another one that I know a lot of other companies in this space use Scale, I know especially with just like low-cost labor internationally.
另一家公司，我知道这个领域的很多其他公司都在使用规模，我尤其知道国际上的低成本劳动力。
Are there other methods for doing this that isn't one of those two?
除了这两种方法之外，还有其他方法可以做到这一点吗？
How are other companies doing this?
其他公司是如何做到这一点的？
Garrett Lord: I think if you care about building a really high quality business, and having good gross margin and high quality growth, the ecosystem here is, one of the leading players, they have 200 recruiters.
Garrett Lord：我认为如果你关心建立一个真正高质量的业务，并拥有良好的毛利率和高质量的增长，这里的生态系统是领先的参与者之一，他们有 200 名招聘人员。
It's unsustainable.
这是不可持续的。
There are like 200 people on LinkedIn sending individual messages to acquire these people, because there's no brand, there's no trust.
LinkedIn 上大约有 200 个人发送个人消息来获取这些人，因为没有品牌，没有信任。
They're spending tens of millions of dollars a month on performance advertising, Google Ads-
他们每个月在效果广告、Google Ads 上花费数千万美元-
Lenny Rachitsky: To find experts and to find folks.
Lenny Rachitsky：寻找专家和人员。
Lenny Rachitsky: And it's experts mostly at this point.
Lenny Rachitsky：目前主要是专家。
Garrett Lord: Yeah.
加勒特·洛德：是的。
And then they put them onto an experience that is treating them like they're drawing boundary boxes around stop signs in the Philippines.
然后他们将他们置于一种体验中，就像他们在菲律宾的停车标志周围画出边界框一样。
The frontier tax accountants don't want to be treated like low cost international labor, and I don't think anyone enjoys that process.
边境税务会计师不想被视为低成本国际劳动力，我认为没有人喜欢这个过程。
And so, the ability to build a experience that's rooted in community, that's rooted in high quality training.
因此，能够建立植根于社区、植根于高质量培训的体验。
If you're getting your PhD at MIT, chances are you're just not being taught well enough on how to use the tools.
如果您在麻省理工学院获得博士学位，很可能您在如何使用这些工具方面没有得到足够好的指导。
Garrett Lord: It's not you can't break the models, it's just like the other platforms, they're spending thousands of hours to acquire an individual user and they're put right into a project with no training.
Garrett Lord：并不是你不能打破模型，而是像其他平台一样，他们花费数千小时来获取个人用户，并且在没有经过培训的情况下将他们直接投入到一个项目中。
So, we just started from day one at building this expert ...
所以，我们从第一天就开始构建这个专家......
We believe there'd be a deep network effect here that's very connected to our core business of starting, jumpstarting or restarting your career.
我们相信这里会产生深刻的网络效应，这与我们开始、快速启动或重新启动您的职业生涯的核心业务密切相关。
And you come in, you build a profile, you see the community, there's groups and a feed of here's how people are learning.
你进来，建立一个个人资料，你看到社区，有小组和人们如何学习的提要。
You come into actual individual cohort with peers that look like you and have your similar background.
你进入了真正的个人群体，与那些看起来像你并且有相似背景的同龄人一起。
You're being taught on how to interact, and there's a trial and error, and we have an instructional design piece so you can't do it.
你正在被教导如何互动，并且有一个尝试和错误，我们有一个教学设计作品，所以你不能这样做。
Then you're put on the projects where building ...
然后你就被投入到建设......的项目中。
There's certain swim lanes where we're actually pre-building data and selling that data to all the labs.
在某些泳道上，我们实际上正在预先构建数据并将这些数据出售给所有实验室。
Garrett Lord: So, we can do this thing where we produce one unit of data ourselves.
Garrett Lord：所以，我们可以自己生成一个数据单位。
We pay for it, almost like a movie production.
我们为此付费，就像电影制作一样。
We pay for a unit of data, and then we make sure it's very high quality.
我们为一个数据单位付费，然后确保它的质量非常高。
We run our own post-training on it, and then we produce a bunch of specifications of the data, and we actually sell that individual package of data to many different labs.
我们对其进行自己的后期培训，然后生成一系列数据规范，并且实际上将单独的数据包出售给许多不同的实验室。
And so, you get put on a project like that.
所以，你就被安排去做这样的一个项目。
Once you're doing a really, really good job on our projects, oftentimes then we'll put you on customer projects where they only want the best of the best people in machine learning.
一旦您在我们的项目中做得非常非常好，我们通常会将您安排到客户项目中，他们只想要机器学习领域最优秀的人才。
And then they go from our projects to their projects.
然后他们从我们的项目转到他们的项目。
And so, there's a huge customer acquisition.
因此，有大量的客户获取。
You love going deep on your podcast, so just to talk about it, it's like you really have a couple of things that matter.
你喜欢深入研究你的播客，所以只要谈论它，就好像你真的有一些重要的事情。
Garrett Lord: You have a cost to customer acquisition, your CAC, and then you have your LTV, like the lifetime value of a user.
Garrett Lord：你有客户获取成本，你的 CAC，然后你有你的 LTV，比如用户的终身价值。
And an LTV is calculated pretty simply in this business.
在这个行业中，LTV 的计算非常简单。
It is based on the retention of a person, and how many projects they can participate in.
它基于一个人的保留率以及他们可以参与的项目数量。
So, if you treat people really well, you train them really well, well, A, we have no customer acquisition costs because we partner with 1,600 universities, power 92% of the top 500 schools in the country.
所以，如果你对待员工真的很好，你就对他们进行很好的培训，那么，A，我们没有客户获取成本，因为我们与 1,600 所大学合作，为全国 500 强学校中的 92% 提供支持。
We power almost every institution and community college in the country.
我们为该国几乎所有机构和社区大学提供电力。
We have no customer acquisition cost to acquire the people.
我们没有获取客户的成本来获取人才。
We have a ton of brand and trust with them built up, so they convert at really, really high rates.
我们与他们建立了大量的品牌和信任，因此他们的转化率非常非常高。
And then, if you treat them really well, because what they expect from us, they know Handshake, their school buys Handshake, we care about treating these people well but the universities would not tolerate our partnership with these fellows unless we treated them well.
然后，如果你对待他们真的很好，因为他们对我们的期望是，他们知道握手，他们的学校购买握手，我们关心善待这些人，但大学不会容忍我们与这些人的合作关系，除非我们善待他们。
Garrett Lord: So, you put them into this process where our LTVs and repeat engagement rate and retention rate on different projects is really high.
Garrett Lord：所以，您将它们放入这个流程中，我们的生命周期价值以及不同项目的重复参与率和保留率非常高。
And so, these structural advantages are quite significant when you contrast a leading provider that has 200 individual contributing recruiters, and are spending tens of millions of dollars a month on performance marketing.
因此，当您对比一家拥有 200 名个人招聘人员、每月在绩效营销上花费数千万美元的领先供应商时，这些结构性优势就显得相当显着。
So, that's I think why we've seen so much success.
所以，我认为这就是我们取得如此巨大成功的原因。
Lenny Rachitsky: That's extremely interesting.
Lenny Rachitsky：这非常有趣。
And it feels like, as you said, there used to be a big focus on generalists, which is people anywhere in the world for low-cost can do the work, like draw bounding boxes around things.
正如你所说，感觉过去人们非常关注通才，即世界上任何地方的人都可以以低成本完成这项工作，比如在事物周围绘制边界框。
And essentially the market has shifted from low-cost generalists to experts.
本质上，市场已经从低成本的通才转向了专家。
And a lot of these companies like Scale, we're optimizing for general work model training data, and you guys are set up to be extremely good at expert based data.
很多像 Scale 这样的公司，我们正在针对一般工作模型训练数据进行优化，而你们则非常擅长基于专家的数据。
And so, you're in the right place at the right time, at the right supply.
因此，您可以在正确的时间、正确的地点获得正确的供应。
What a business.
真是一桩生意啊。
Garrett Lord: I would say it's not been easy building business two inside of business one, but-
加勒特·洛德：我想说，在第一家企业内部建立第二家企业并不容易，但是——
Lenny Rachitsky: Yeah.
莱尼·拉奇茨基：是的。
So, let me follow that thread.
那么，让我跟随这个线索。
That's where I wanted to go.
那就是我想去的地方。
What was just that like?
那是什么样子的？
So, you started noticing that model companies were coming to your people, that people were having hard times with some of these other companies in this space and you're like, "Oh, maybe we should be doing this sort of thing"?
所以，你开始注意到模范公司正在向你的员工走来，人们在这个领域与其他一些公司相处得很艰难，你会想，“哦，也许我们应该做这种事情”？
How did that just initial inception start, and how did you start to explore that idea and to see if it was a real thing?
最初的想法是如何开始的？您是如何开始探索这个想法并看看它是否是真实的？
Garrett Lord: Tactically we were working with many of the middleman companies doing work.
Garrett Lord：从战术上讲，我们正在与许多中间商公司合作。
We started to see the demand, as I talked about earlier.
正如我之前提到的，我们开始看到需求。
We started to see direct outreach from the Frontier Labs reaching out to us, trying to cut out the middleman in their pursuit of getting higher-quality data.
我们开始看到 Frontier Labs 直接向我们伸出援手，试图在追求更高质量数据的过程中剔除中间人。
When we started to put together the dots on we could build a way better experience for our fellows, we could serve them directly to the labs and build a direct customer relationship with the labs, and basically cut out the middleman.
当我们开始将这些点放在一起时，我们可以为我们的同事建立更好的体验，我们可以直接为他们提供实验室服务，并与实验室建立直接的客户关系，基本上消除了中间商。
And provide a better experience to the labs, provide a better experience to our fellows and provide a better experience to our million companies in the network.
为实验室提供更好的体验，为我们的同事提供更好的体验，为我们网络中的数百万家公司提供更好的体验。
Garrett Lord: And you might think about just upskilling and reskilling, what's going to happen there.
加勒特·洛德：你可能会想到技能提升和再培训，那里会发生什么。
So, we walked into this space.
于是，我们就走进了这个空间。
We started in really December, exploring and learning more about it, like on expert calls and hammering down.
我们从 12 月份开始，探索并了解更多相关信息，例如专家电话会议和敲定。
I hired three expert firms, AlphaSights and GLG, and started doing a bunch of calls with the latest researchers, because we had resources.
我聘请了三个专业公司 AlphaSights 和 GLG，并开始与最新的研究人员进行大量通话，因为我们有资源。
One of the cool things about being a larger company is our core business is $200 million ARR, so it's like we had resources to be able to accelerate the learning curve here.
作为一家更大的公司，最酷的事情之一是我们的核心业务年收入为 2 亿美元，因此我们拥有能够加快学习曲线的资源。
And then, we started working with arguably the number one lab about five months ago.
然后，大约五个月前我们开始与可以说是第一的实验室合作。
Garrett Lord: different answers working with the number one lab, and have just now we're working with Devin on the Frontier Labs and the number one thing we're trying to do is just focus on scaling up.
Garrett Lord：与第一实验室合作有不同的答案，现在我们正在与德文在前沿实验室合作，我们要做的第一件事就是专注于扩大规模。
And we've gone from four or five people working on this to 75 plus people working on it.
我们的工作人员数量从四五个人增加到了 75 多人。
I think we had 12 people start last Monday.
我想我们上周一开始有 12 个人。
It's like we are so bottlenecked on just meeting this opportunity, because in this market there's essentially unlimited demand.
就好像我们在抓住这个机会方面遇到了瓶颈，因为这个市场本质上有无限的需求。
If you can produce high quality volumes of data, you most likely will be able to sell whatever you produce.
如果您能够产生大量高质量的数据，那么您很可能能够出售您生产的任何产品。
And so on our side, it's like we're really focused on making sure that we pick the right longer term strategy, making sure that we don't grow too fast as to erode the trust that we've built up with these Frontier labs.
因此，就我们而言，我们确实专注于确保选择正确的长期战略，确保我们不会增长得太快，从而削弱我们与这些前沿实验室建立的信任。
Yeah, but it's been fun.
是的，但这很有趣。
Lenny Rachitsky: You said it's also been really hard to start those business within an existing business.
Lenny Rachitsky：您说过在现有业务中开展这些业务也非常困难。
What's been hard?
有什么难的？
What's been hardest?
最难的是什么？
You touched on a couple of these elements already, but what else?
您已经谈到了其中的几个要素，但是还有什么？
Garrett Lord: I think I just followed a lot more of my intuition around this, doing this.
加勒特·洛德（Garrett Lord）：我想我只是更多地遵循我的直觉来做这件事。
The story of Handshake was we had to sign up 1,600 universities, so I had to learn how to be the best ...
Handshake 的故事是我们必须报名 1,600 所大学，所以我必须学习如何成为最好的......
We are the fastest growing higher education company in history.
我们是历史上发展最快的高等教育公司。
So, we signed up six 1,600 schools.
因此，我们签约了 6 所 1,600 所学校。
Then we had to build an employer business, where we had to figure out how to sell the 100% ...
然后我们必须建立一个雇主业务，我们必须弄清楚如何出售 100% 的……
All these Fortune 500 companies use it and 70% of it pay for it, so I had to learn about upmarket sales to Goldman Sachs, and General Motors, and Google and the biggest companies in the world, which is totally different than selling universities.
所有这些财富 500 强公司都使用它，其中 70% 是付费的，所以我必须了解向高盛、通用汽车、谷歌和世界上最大的公司进行高端销售，这与销售大学完全不同。
And then we had to learn how to build an incredible student social network.
然后我们必须学习如何建立一个令人难以置信的学生社交网络。
What does the best feed look like?
最好的饲料是什么样的？
What does group messaging look like?
群组消息传递是什么样的？
So, I felt a little bit of familiarity in those zero to ones.
所以，我对那些零到一感到有点熟悉。
Garrett Lord: Oftentimes marketplaces are like many zero to ones.
Garrett Lord：市场通常就像许多零到一。
Sometimes I dream that we just, I actually don't dream, but I make a joke that I just wish we were a cybersecurity company and we had one buyer and just one product, and it was just like we had to ...
有时我梦想我们只是，实际上我没有梦想，但我开玩笑说我只是希望我们是一家网络安全公司，我们有一个买家和一种产品，就像我们必须......
In a marketplace, you have to serve three different sides, you know from your time at Airbnb.
从您在 Airbnb 的经历中就知道，在市场中，您必须为三个不同的方面提供服务。
And so, one of my learnings in spinning up these three different businesses in starting Handshake was I was pretty hands-on.
因此，在启动 Handshake 时，我在开展这三项不同业务的过程中学到的东西之一就是我非常亲力亲为。
So, everyone reported directly to me.
所以，大家都直接向我汇报。
I really said in a lot of meetings, "I'm not trying to be the boss, I'm just trying to get another smart guy in the room." We've hired an incredible team of people that have spent a lot of time in the space and have been big leaders at a lot of the human data companies in the space.
我确实在很多会议上说过，“我并不是想成为老板，我只是想在房间里再找一个聪明的人。”我们聘请了一支令人难以置信的团队，他们在该领域花费了大量时间，并且是该领域许多人力数据公司的重要领导者。
Garrett Lord: And so, everyone saw very clearly the structural advantages that we had, and a lot of the focus was on making sure that we could deliver high-quality data to one customer before we expand to anyone else.
Garrett Lord：因此，每个人都非常清楚地看到我们拥有的结构性优势，并且重点是确保我们能够在扩展到其他客户之前向一个客户提供高质量的数据。
You had to say no to a lot of things.
你必须对很多事情说不。
And then, you also had a lot of people in the core part of the business that, rightfully so, there's just checks and balances that there's a lot of people that try to get involved.
然后，你也有很多人在业务的核心部分，理所当然，只是有很多人试图参与其中的制衡。
Everyone wants to say, not everyone, this is a stretch, but it's easy to say no. It's easy to be like, "I can't prioritize that this week or this month.
每个人都想说，不是每个人，这有点牵强，但很容易说不。人们很容易会说，“我不能在本周或本月优先考虑这件事。
I have an existing set of priorities." So essentially, with the exception of a few things, everyone just came straight into this new org that I built, everyone did not have any responsibilities in the existing part of the business.
我有一套现有的优先事项。”所以本质上，除了一些事情之外，每个人都直接进入我建立的这个新组织，每个人在现有的业务部分中都没有任何责任。
It was extremely clear who was the directly responsible individual across each area of the new co. And now we've got deeper coupling and integration points across the rest of the business, but we sat in a separate part of the office.
新公司各个领域的直接负责人是谁，一目了然。现在，我们在业务的其他部分有了更深入的耦合和集成点，但我们坐在办公室的一个单独的部分。
Garrett Lord: Everyone's in the office five days a week, a lot of weekends.
加勒特·洛德（Garrett Lord）：每个人每周五天都在办公室，还有很多周末。
There's a totally different expectation in hiring talent too, where it's like, "Hey, this is a 24/7 job.
在招聘人才时也有完全不同的期望，就像，“嘿，这是一份 24/7 的工作。
This is an early-stage company." The compensation was also different too, and based on hurdles in this new business so people felt owners creating the new co. And yeah, it's still extremely nimble, very, very flat.
这是一家早期公司。”薪酬也不同，并且基于这个新业务的障碍，所以人们觉得创建新公司的所有者。是的，它仍然非常灵活，非常非常平坦。
Just because you run one function doesn't mean you're the directly responsible individual on a project.
仅仅因为您运行一项功能并不意味着您是项目的直接负责人。
We pick the best person who's most capable of driving an initiative forward, regardless of the function to be the DRI.
我们会挑选最有能力推动计划向前发展的最佳人选，无论 DRI 的职能如何。
We're a lot more metrics-oriented.
我们更加注重指标。
When I built Handshake, we resisted this operating cadence for a long time, this weekly, monthly, quarterly operating cadence.
当我建立Handshake的时候，我们很长一段时间都抵制这种运营节奏，这种每周、每月、每季度的运营节奏。
With Handshake AI, we've been way more focused on operating with data, and metrics, and rigor from an early stage.
借助 Handshake AI，我们从早期阶段就更加专注于数据、指标和严谨性的运营。
There's a gentleman named Sahil on our team who's been doing an incredible job with that.
我们团队中有一位名叫萨希尔的绅士，他在这方面做得非常出色。
Shout out Sahil, shout out young, shout out Paco.
大声喊出萨希尔，大声喊出年轻，大声喊出帕科。
Yeah.
是的。
Lenny Rachitsky: Okay, this is incredible.
Lenny Rachitsky：好吧，这太不可思议了。
So, a few elements of what allowed this to succeed within a decade-old company.
那么，这家拥有十年历史的公司能够取得成功的几个因素。
And by the way, so you're at 200 million a year in revenue with the traditional business.
顺便说一句，传统业务每年的收入为 2 亿美元。
You're going to, as you said, blow past 100 million in the first year of this new business.
正如您所说，您将在这项新业务的第一年突破 1 亿美元。
So, it's wild that in the first couple years, if things continue to go this way, you'll exceed the run rate of a business that took you 10 years to build.
所以，在最初的几年里，如果事情继续这样发展，你将超过你花了 10 年建立的企业的运行率，这是疯狂的。
Incredible.
极好的。
To make this successful, a few of the things I noted as you were talking, one is clearly you were just in founder mode.
为了使这一成功，我在你谈话时注意到的一些事情，其中之一显然是你只是处于创始人模式。
You're the lead of this new business.
你是这项新业务的负责人。
You weren't delegating it to someone, "Hey, go start this thing." You dedicated people, "Here, we're going to pick people.
你不是把它委托给某人，“嘿，开始这件事吧。”你们敬业的人，“在这里，我们要挑选人。
You have nothing else going on, this is your new job.
你没有其他事情要做，这是你的新工作。
You're going to work on this stuff."
你将致力于这些事情。”
Lenny Rachitsky: You worked in different part of the office.
Lenny Rachitsky：您在办公室的不同部分工作。
There's a metrics-based cadence.
有一个基于指标的节奏。
It's just like, let's stay really diligent about here's how it's going, here's where we're going, here's our track, here's our KPIs, things like that.
就像，让我们对事情的进展、我们要去的地方、我们的轨迹、我们的 KPI 等等保持真正的勤奋。
Anything else there that you felt really important to making this work?
您认为还有什么其他事情对于完成这项工作非常重要吗？
Because a lot of companies are going to try to do this, I imagine, and so I'm curious what else you found important to make this work.
我想，因为很多公司都会尝试这样做，所以我很好奇您认为还有什么对实现这项工作很重要。
Garrett Lord: Yeah.
加勒特·洛德：是的。
I mean, I just really believe in separate and everything.
我的意思是，我真的相信分离和一切。
Separate engineering team, separate design team, separate accounts and operations team, separate finance team.
独立的工程团队、独立的设计团队、独立的账户和运营团队、独立的财务团队。
Early on, everything was separate.
早期，一切都是分开的。
People only had one job and one job only, and that was making Handshake AI successful.
人们只有一份工作，而且只有一份工作，这使得 Handshake AI 取得了成功。
We had a couple integration points more, and I had an incredible executive team and a core part of business, and now there's becoming more and more involvement.
我们还有几个整合点，我有一个令人难以置信的执行团队和业务核心部分，现在参与度越来越高。
But our executives that have built Handshake for a long time ran the core business, and I focused 80 plus percent of my time and attention on just this.
但我们长期建立 Handshake 的高管负责核心业务，而我将 80% 以上的时间和注意力都集中在这一点上。
And we hired an incredible engineering leader like Avery, who ...
我们聘请了像艾弗里这样令人难以置信的工程领导者，他......
We have a lot of entrepreneurs, people that have started companies inside the company.
我们有很多企业家，在公司内部创办了公司的人。
Or pardon me, people that started companies before.
或者请原谅我，那些以前创办过公司的人。
That was huge.
那是巨大的。
A lot of familiarity with hiring talent that have only worked at early stage companies so that feels super comfortable with ambiguity.
对招聘只在早期公司工作过的人才非常熟悉，因此对模糊性感到非常舒服。
Garrett Lord: We were also way more upfront around this is going to be chaotic.
加勒特·洛德（Garrett Lord）：我们也更加提前地意识到这将会变得混乱。
Just owning that narrative in front of all hands at the core company, owning it directly with the team.
只需在核心公司的所有人面前拥有这种叙述，并直接与团队拥有它。
We have a separate all hands, we have separate onboarding, we have a separate recruiting team.
我们有一个单独的全体员工，我们有单独的入职团队，我们有一个单独的招聘团队。
I had some connection points, but mostly separate.
我有一些连接点，但大部分都是分开的。
And I think that was absolutely critical.
我认为这绝对至关重要。
We took some of the top people, I mean, we have great people in the core business, we took some great people from the core businesses and basically said, "Sorry, I know you love your old team.
我们聘请了一些顶尖人才，我的意思是，我们在核心业务中有一些优秀的人才，我们从核心业务中聘请了一些优秀的人才，基本上说，“抱歉，我知道你爱你的老团队。
I know you love what you're doing.
我知道你热爱你正在做的事情。
Will you join us in Handshake AI?" And they completely forego their historical responsibilities and came over.
你愿意加入我们握手人工智能吗？”他们完全放弃了自己的历史责任，过来了。
That became really critical with engineering when things started to scale and topple, and we're growing so quickly we took some of our top senior engineers, who were very entrepreneurial, and principal engineers, staff of engineers, parachute them in.
当事情开始规模化和颠覆时，这对于工程来说变得非常重要，而且我们发展得如此之快，我们聘请了一些非常有创业精神的顶级高级工程师，以及首席工程师和工程师团队，将他们空降进来。
It's been awesome to ask some of the most talented people in the core business like, "Hey, do you want to come over here and do this?"
向核心业务中一些最有才华的人询问“嘿，你想来这里做这件事吗？”真是太棒了。
Garrett Lord: And sometimes they say no. They're like, "I don't want to work most of the weekends." The number of 2:00 AM, 3:00 AM nights we done in this business, I mean, it's quite regular.
加勒特·洛德：有时他们会说不。他们会说，“我周末大部分时间都不想工作。”我的意思是，我们从事这项业务的凌晨 2:00、凌晨 3:00 的夜晚数量是相当规律的。
People sometimes don't want to commit to that, but we've been up front, like here are the expectations for this team.
人们有时不想承诺这一点，但我们一直在前面，就像这就是对这个团队的期望。
It's an insane pace.
这是一个疯狂的速度。
If you want to be a part of one of the fastest growing businesses in Silicon Valley, you can join it.
如果您想成为硅谷发展最快的企业之一的一部分，您可以加入。
The ownership too has also been huge, like owning this outcome, and we have this motto to leave nothing to chance.
所有权也很大，就像拥有这个结果一样，我们的座右铭是不留任何机会。
For a while there we drew the number of days in the year on the whiteboard and it was like, there will never be a time like this.
有一段时间，我们在白板上画了一年中的天数，感觉永远不会有这样的时刻。
I've never seen anything like it, I doubt I'll ever feel anything like this in business again where there's unlimited demand and it's just our ability to execute against it.
我从来没有见过这样的事情，我怀疑我是否会在商业中再次感受到这样的事情，那里有无限的需求，而这只是我们执行它的能力。
Garrett Lord: And so, we had this motto like leave nothing to chance.
加勒特·洛德：因此，我们的座右铭是“不留任何机会”。
How do you make sure that three months are not six months?
您如何确保三个月不是六个月？
You have no regrets.
你没有遗憾。
Get on the plane to go talk to a customer, make the late night push, check the data six times over again, ship the extra feature that helps.
登上飞机去与客户交谈，深夜推送，检查数据六次，交付有帮助的额外功能。
And really, a huge celebratory culture too.
事实上，这也是一种巨大的庆祝文化。
It's very flat so there really isn't this principle of ...
它非常平坦，所以确实没有这个原则......
There's so many people putting up points, directly calling out the people that are putting up points and creating a really fun environment around impact I think has been awesome.
有这么多人提出观点，直接呼唤提出观点的人，并围绕影响力创造一个非常有趣的环境，我认为这非常棒。
Lenny Rachitsky: The leave nothing to chance piece I imagine speaks partly to the value of trust in what you're doing.
Lenny Rachitsky：我认为“不留任何机会”部分说明了对自己正在做的事情的信任的价值。
You win if they can trust that your data's awesome, and great, and consistent, and I could see why that ends up being such an important part of what you're building.
如果他们能够相信你的数据很棒、很棒且一致，那么你就赢了，我可以理解为什么这最终会成为你正在构建的内容的重要组成部分。
And just listening to you describe this, I understand it's obviously a massive opportunity, obviously a massive advantage you guys have, and just the stress that comes with that burden also imagine is very high of just like we can't screw this up.
听你描述这一点，我知道这显然是一个巨大的机会，显然是你们拥有的巨大优势，而且可以想象，这种负担带来的压力也非常高，就像我们不能搞砸一样。
Garrett Lord: No. Yes, Handshake should be a ...
加勒特·洛德：不。是的，握手应该是......
Business does billions dollars revenue as a public company, we should be able to continue to ...
作为一家上市公司，企业收入数十亿美元，我们应该能够继续......
I mean, and it also helps our core business.
我的意思是，这也有助于我们的核心业务。
The longer term opportunity that we see is it's connecting, it's building the best job mashing marketplace on the internet.
我们看到的长期机会是它的连接，它正在互联网上建立最好的就业混搭市场。
It's probably one of the largest problems in the world like labor supply mashing.
这可能是世界上最大的问题之一，就像劳动力供应混乱一样。
It's where people spend most of their time and energy, just hours of their life they spend it at work.
这是人们花费大部分时间和精力的地方，他们一生中只有几个小时花在工作上。
The process of searching for a job, applying to a job is going to be completely reinvented with AI.
人工智能将彻底重塑寻找工作、申请工作的过程。
We've been leading the charge there.
我们一直是那里的领头羊。
An AI interviewer that's collecting skills and actually asking about your experiences, doing work simulation experiences that help employers find the best candidates.
人工智能面试官会收集技能并实际询问您的经历，进行工作模拟体验，帮助雇主找到最佳候选人。
I mean, I don't know the last time you've done this, but the hiring manager process, reviewing 200 resumes, are you kidding me?
我的意思是，我不知道你上次这样做是什么时候，但是招聘经理的流程，审查 200 份简历，你在开玩笑吗？
Garrett Lord: I'm going to sit there and review 200 resumes?
Garrett Lord：我要坐在那里审查 200 份简历？
Not a chance five years from now.
五年后就没有机会了。
Students manually making cover ...
学生手工制作封面...
Not a chance.
没有机会。
So, there will need to be a marketplace that wins in connecting supply and demand, and talent with opportunity, and we think and get psyched about the opportunity for impact here.
因此，需要有一个能够连接供需、人才与机会的市场，我们思考并为这里产生影响的机会感到兴奋。
That's my story, I went to community college, I paid my way through school.
这就是我的故事，我上了社区大学，我自己付了学费。
I went to a no name school in Upper Peninsula of Michigan.
我去了密歇根州上半岛的一所无名学校。
I worked at Palantir as an intern, it totally changed my life, and I started Handshake because I wanted to make it easier for anyone regardless of who you knew, what your parents did, what school you went to, to find a great opportunity.
我在 Palantir 担任实习生，这彻底改变了我的生活，我开始 Handshake 是因为我想让任何人更容易找到一个很好的机会，无论你认识谁，你的父母做什么，你上哪所学校。
And I think AI, totally step function improvement in matching.
我认为人工智能在匹配方面完全实现了功能改进。
And I think that our human data business is really serving as the foundation for improving matching.
我认为我们的人类数据业务确实是改善匹配的基础。
Garrett Lord: A lot of things that we're doing in the human data business are being integrated to our core business.
Garrett Lord：我们在人类数据业务中所做的很多事情正在整合到我们的核心业务中。
I think that's going to improve outcomes for employers, save them in the aggregate like billions of dollars over time.
我认为这将改善雇主的成果，随着时间的推移，总共可以节省数十亿美元。
And I think it makes the experience way better for students.
我认为这让学生的体验变得更好。
So, it's just like we have to meet the moment.
所以，就像我们必须见面的时刻一样。
We still have the stamina, and the excitement, and the passion internally in our core and in the new business to go charge after this.
我们仍然有耐力、兴奋和激情，在我们的核心和新业务中继续前进。
And that's a lot of the message we've been sharing internally is it's time to amp it up.
我们在内部分享的很多信息都是，是时候加强它了。
This is a once in a lifetime opportunity to be positioned as well, and we are going to make the moment as a team.
这也是一生一次的定位机会，我们将作为一个团队抓住这一时刻。
Lenny Rachitsky: It really is.
莱尼·拉奇茨基：确实如此。
This very much feels like a once in a lifetime opportunity.
这感觉就像是一生一次的机会。
Let me ask a few other questions along these lines that are something I've been thinking about, something that a lot of people think about, just while I have you, there's always this question of will we run out of data?
让我沿着这些思路问一些其他问题，这些问题是我一直在思考的，也是很多人思考的，就在我和你们在一起的时候，总是有这样一个问题：我们会用完数据吗？
Will model stop advancing?
模型会停止前进吗？
Are we going to hit some plateau and there's not actually going to be some AGI moment, SGI moment?
我们是否会达到某个平台期，实际上不会出现 AGI 时刻、SGI 时刻？
So first of all, do you think we'll run out of data?
首先，您认为我们的数据会耗尽吗？
There's a point at which we just can't produce more knowledge and data to feed these models?
到了某个时候，我们就无法产生更多的知识和数据来支持这些模型了吗？
And along those lines, what do you think is the biggest bottleneck to advancing models faster and further?
沿着这些思路，您认为更快更进一步推进模型的最大瓶颈是什么？
Garrett Lord: Yeah, I mean, it's just the type of data we're going to need is going to evolve.
加勒特·洛德：是的，我的意思是，我们需要的数据类型正在不断发展。
It's going to be CAD files, it's going to be scientific tool use data as they try to automate scientific discoveries and drug discovery.
这将是 CAD 文件，将是科学工具使用数据，因为他们试图自动化科学发现和药物发现。
It's going to be esoteric operating systems that exist on scientific tools.
它将是存在于科学工具上的深奥操作系统。
So, I love this trajectory and stitching together step-by-step instruction following.
所以，我喜欢这个轨迹和将遵循的分步说明拼接在一起。
The type of data we're going to need is going to evolve a lot.
我们需要的数据类型将会发生很大的变化。
And we haven't even talked about multimodal, and video, and text and audio.
我们甚至还没有讨论多模式、视频、文本和音频。
There's a huge demand for audio data right now.
目前对音频数据的需求巨大。
So, the type of data's going to evolve.
因此，数据类型将会不断发展。
Lenny Rachitsky: Yeah, I use voice mode all the time.
Lenny Rachitsky：是的，我一直使用语音模式。
That's on my default ChatGPT experience, just talking to-
这是我默认的 ChatGPT 体验，只是与-
Garrett Lord: It's amazing.
加勒特·洛德：太棒了。
It's amazing.
太棒了。
I just had a baby on, or my wife had a baby on Sunday, and voice mode's been incredible.
我刚刚生了孩子，或者我的妻子周日生了孩子，语音模式非常棒。
I mean, every night, every two hours it's like I have more questions.
我的意思是，每天晚上，每两个小时我就会有更多的问题。
Voice mode's been huge.
语音模式非常强大。
So, shot out voice mode.
于是，就拍出了语音模式。
Yeah, so the type of data is going to collect a lot, or change a lot.
是的，所以数据类型会收集很多，或者改变很多。
I think synthetic data has a role to play and in verifiable domains, but what we consistently hear from companies it's like synthetic data is not going to dominate.
我认为合成数据在可验证的领域可以发挥作用，但我们一直从公司那里听到的消息是，合成数据不会占据主导地位。
There's billions, and billions, and billions of dollars of value to extract as a company over the next decade and following the frontier of AI development.
作为一家公司，在未来十年内，追随人工智能发展的前沿，可以获取数十亿美元的价值。
Lenny Rachitsky: Let me first say just huge kudos to you for just having a kid, your wife just having a kid a few days ago, and building this business that is growing bananas and doing this podcast conversation.
Lenny Rachitsky：首先让我对您表示极大的敬意，因为您刚刚有了一个孩子，您的妻子几天前刚刚有了一个孩子，并且建立了种植香蕉和进行播客对话的业务。
I really appreciate you.
我真的很感激你。
Lenny Rachitsky: Of course.
莱尼·拉奇茨基：当然。
Is there anything else that we haven't covered that you think might be helpful for folks to hear, or a part of your story that you think might be helpful for folks to learn from, or something you may want to just double down on that we've talked about before we get to a very exciting lightning round?
还有哪些我们没有提到但您认为可能对人们听到有帮助的内容，或者您认为可能对人们学习有帮助的故事的一部分，或者您可能想在我们进入非常激动人心的闪电回合之前加倍强调我们已经讨论过的内容？
Garrett Lord: I mean, the thing I always love talking, I'm really passionate about people starting companies and helping them do so.
加勒特·洛德：我的意思是，我一直喜欢谈论的事情是，我真的很热衷于人们创办公司并帮助他们这样做。
I just think in this moment right now with AI, for young entrepreneurs that listen, that read this podcast, because I've been a reader since 2020.
我只是想，此时此刻，对于人工智能来说，对于那些聆听、阅读这个播客的年轻企业家来说，因为我自 2020 年以来一直是一名读者。
We looked.
我们看了看。
Lenny Rachitsky: Yeah, we did check.
Lenny Rachitsky：是的，我们确实检查过。
That's incredible.
那真是难以置信。
Garrett Lord: Yeah, been a long-term reader.
加勒特·洛德：是的，我是一名长期读者。
I'm just so curious and love sucking up-
我只是很好奇并且喜欢拍马屁-
Garrett Lord: ...
加勒特·洛德：……
your interviews.
你的采访。
But it's like you just focus on doing something of meaning that really helps people.
但这就像你只专注于做一些真正能帮助人们的有意义的事情。
And I think with AI, there's going to be so many opportunities to improve the way people learn.
我认为有了人工智能，将会有很多机会来改善人们的学习方式。
I'm just really passionate about trying to make Handshake a platform that is not only an incredible business, but is also something that really helps solve a societal problem that matters.
我真的很热衷于努力让 Handshake 成为一个平台，它不仅是一项令人难以置信的业务，而且还真正有助于解决重要的社会问题。
And yeah, that's be my one shout out here.
是的，这就是我在这里大声疾呼的。
If anyone wants advice on how to do that or wants to reach out, I'm happy to chat.
如果有人需要有关如何做到这一点的建议或想要联系，我很乐意与您聊天。
Lenny Rachitsky: Okay, so this is an offer to share advice on starting companies within AI.
Lenny Rachitsky：好的，这是一个分享关于人工智能领域初创公司的建议的提议。
Is that the offer here?
这是这里的优惠吗？
Just so folks-
就是这样，各位——
Lenny Rachitsky: Okay.
莱尼·拉奇茨基：好的。
I don't know how much time you'll have for the hundreds of thousands of people coming your way, but I appreciate the offer.
我不知道你能有多少时间来迎接成千上万的人，但我很感激这个提议。
That's very cool.
太酷了。
Anything else before we get to a very exciting lightning round?
在我们开始激动人心的闪电回合之前还有什么要说的吗？
Lenny Rachitsky: Well, with that Garrett, we reached our very exciting lightning round.
Lenny Rachitsky：嗯，通过加勒特，我们进入了非常激动人心的闪电回合。
We've got five questions for you.
我们有五个问题要问你。
Are you ready?
你准备好了吗？
Lenny Rachitsky: What are two or three books that you find yourself recommending most to other people?
Lenny Rachitsky：您最常向其他人推荐的两三本书是什么？
Garrett Lord: I'm a sucker for Peter Thiel's Zero to One.
加勒特·洛德（Garrett Lord）：我很喜欢彼得·泰尔（Peter Thiel）的《从零到一》。
I read it when I started the company, and watched Peter Thiel's startup school class at Stanford he taught back in the days where there wasn't everything written on the internet about how to start companies, and just think he was the coolest.
我在创办公司时读过这本书，并观看了 Peter Thiel 在斯坦福大学教授的创业学校课程，当时互联网上还没有关于如何创办公司的所有内容，只是认为他是最酷的。
Love Shoe Dog.
爱鞋狗。
I think it's the epitome of starting a company.
我认为这是创办公司的缩影。
Hard Things About Hard Things obviously, but these are all quite common books.
困难的事情显然是关于困难的事情，但这些都是很常见的书。
Lenny Rachitsky: But also classics.
Lenny Rachitsky：但也是经典。
Ben Horowitz is coming on the podcast, talk about Hard Things About Hard Things.
本·霍洛维茨 (Ben Horowitz) 即将参加播客，谈论艰难的事情。
Lenny Rachitsky: The Hard Thing About Hard Things.
莱尼·拉奇茨基：困难中的困难。
Yeah.
是的。
Okay, have you seen a recent movie or TV show you really enjoy it?
好吧，你最近看过一部你非常喜欢的电影或电视节目吗？
I imagine you don't have much time for this, but-
我想你没有太多时间来做这件事，但是-
Garrett Lord: I'm going to get blasted for this, but I did start Game of Thrones with my wife, and I cannot-
加勒特·洛德：我会为此受到谴责，但我确实和我的妻子一起开始了《权力的游戏》，而且我不能-
Garrett Lord: So, I got a lot of catching up to do.
加勒特·洛德：所以，我有很多事情要做。
Lenny Rachitsky: Why would you get ...
莱尼·拉奇茨基：你为什么要……
No, this is great.
不，这太棒了。
It's like people that have watched it-
好像看过的人一样——
Lenny Rachitsky: You've loved it so far?
Lenny Rachitsky：到目前为止你喜欢它吗？
Okay.
好的。
It's quite gruesome, that's the only downside of that show.
太可怕了，这是该剧唯一的缺点。
Don't watch it before you go to bed, I don't know how many gruesome scenes you've seen already.
睡觉前别看，我不知道你已经看过多少恐怖场面了。
Do you have a favorite product you've recently discovered that you really love?
您最近发现了您真正喜欢的最喜欢的产品吗？
Garrett Lord: The SNOO.
加勒特·洛德：SNOO。
The baby automated SNOO has really helped us a lot.
婴儿自动SNOO真的帮了我们很多。
So love the, shout-out SNOO team.
非常喜欢 SNOO 团队。
Lenny Rachitsky: Amazing.
莱尼·拉奇茨基：太棒了。
I had a SNOO as well.
我也有一个SNOO。
We never actually turned it on, we just ended up using it as a basinet the whole time.
我们从未真正打开过它，我们只是一直将它用作摇篮。
Garrett Lord: Yeah, most of the time it's not turned on, but a couple of cries it's been turned on, it's been very helpful.
Garrett Lord：是的，大多数时候它都没有打开，但是有几次它已经打开了，这非常有帮助。
Lenny Rachitsky: Do you have favorite life motto that you find yourself coming back to, sharing with other people?
Lenny Rachitsky：您有没有最喜欢的人生格言，并且会与其他人分享？
Garrett Lord: I love that leave nothing to chance, leave it all out on the field.
加勒特·洛德：我喜欢不留一丝侥幸，把一切都留在球场上。
Grew up in a really hardworking family, and dad worked really hard to provide, make it happen for us and it's like just give it your all.
在一个非常勤奋的家庭中长大，爸爸非常努力地工作来养活我们，让这一切发生，就像全力以赴一样。
Leave nothing a chance.
不给任何机会留下任何机会。
Lenny Rachitsky: Okay.
莱尼·拉奇茨基：好的。
So the last question, I was researching you in prep for this podcast and there's a story that I love about your hustle early on is when you were going from campus to campus pitching schools to join Handshake, and there's a story where you had to shower in the Princeton's pool to save money because you just didn't have a place to stay.
所以最后一个问题，我在为这个播客做准备时对你进行了研究，有一个关于你早期的忙碌的故事，我很喜欢你从一个校园到另一个校园推销学校加入握手的故事，有一个故事，你不得不在普林斯顿的游泳池里洗澡以省钱，因为你只是没有地方住。
Is there something there?
那里有什么东西吗？
Is there a story there you could share?
有什么故事可以分享吗？
Garrett Lord: Yeah, so it was a tough one.
加勒特·洛德：是的，所以这是一场艰难的比赛。
I mean, I almost got arrested at Princeton, because I mean, I guess for entrepreneurs that are traveling around all the time, we were sleeping out of our car.
我的意思是，我差点在普林斯顿被捕，因为我的意思是，我想对于一直在各地旅行的企业家来说，我们睡在车外。
We had this Ford focus, we would put 20, 30,000 miles on it, sleep in the back of like McDonald's parking lots because they're well lit and had good wifi back in the day.
我们有这辆福特焦点，我们会在上面行驶 20、30,000 英里，睡在麦当劳停车场的后面，因为那里光线充足，并且在白天有很好的 wifi。
And instead of staying in a hotel, a way to freshen up ahead of your meeting is every university has a pool and the pool's almost always, it is always open.
与住在酒店不同，在会议前梳洗一番的一种方法是每所大学都有游泳池，而且游泳池几乎总是开放的。
We never had a situation where it's always open for people to swim in the morning, like fitness.
我们从来没有遇到过早上总是开放供人们游泳的情况，就像健身一样。
Faculty, students.
教师、学生。
And every pool, what do they have?
每个泳池都有什么？
They have a shower.
他们有淋浴。
Garrett Lord: So, you could go to any pool at any university in the country, and you can get a free shower and freshen up.
加勒特·洛德：所以，你可以去全国任何一所大学的任何一个游泳池，你都可以免费淋浴并梳洗一番。
So, the Princeton campus security did not appreciate me showering as a non-student, but I think it meaningfully helped us because the Princeton campus security called the career service center director we were selling to, being like, "Who's Garrett Lord?
所以，普林斯顿校园的保安不喜欢我作为一个非学生洗澡，但我认为这对我们很有帮助，因为普林斯顿校园的保安打电话给我们要推销的职业服务中心主任，问道，“加勒特·洛德是谁？
Is he really here to pitch you software for your career center?" And it made the start of the meeting with the career center really stimulating and exciting, because they were like, "You showered in our pool, you drove here?" "Yeah, we drove here from Michigan." And so, I think that showed a level of commitment that was exciting for them.
他真的是来向你推销职业中心软件的吗？”这使得与职业中心的会议开始时非常刺激和令人兴奋，因为他们就像，“你在我们的游泳池里洗澡，你开车来这里？” “是的，我们从密歇根开车来这里。”因此，我认为这表明了他们的承诺程度，这让他们感到兴奋。
Lenny Rachitsky: Fast-forward to all these founders now starting to use this growth lever of getting in trouble with the campus police to get better meetings with the school leaders.
Lenny Rachitsky：快进到所有这些创始人现在开始利用与校园警察发生麻烦的增长杠杆来更好地与学校领导会面。
Incredible.
极好的。
Garrett, this is such an insane, amazing, inspiring story, just like what you're building and the opportunity here, and just how it's fast, it's going, and all the advantages you have.
加勒特，这是一个如此疯狂、令人惊叹、鼓舞人心的故事，就像你正在建设的东西和这里的机会一样，以及它的速度、进展以及你所拥有的所有优势。
If I was an investor in Handshake, I'd be like, "All right, 10 years, it's going great." And now it's like, "Whoa, holy shit.
如果我是 Handshake 的投资者，我会说，“好吧，10 年了，一切都进展顺利。”现在就像是，“哇，天哪。
Where did this come from?" Incredible.
这是从哪里来的？”极好的。
And it's just also really meaningful.
这也非常有意义。
So, I'm really happy that you made time for this in spite of the madness you are in right now.
所以，我真的很高兴你能抽出时间来做这件事，尽管你现在正处于疯狂之中。
Two final questions, where can folks find you if they want to maybe reach out or maybe if you're hiring, let us know.
最后两个问题，如果人们想要联系或者如果您正在招聘，可以在哪里找到您，请告诉我们。
And then, how can listeners be useful to you?
那么，听众如何对你有用呢？
Garrett Lord: I mean, sign up for Handshake.
加勒特·洛德：我的意思是，报名参加“握手”活动。
If you want to message me on there, it's the easiest way to reach me.
如果您想在那里给我发消息，这是联系我的最简单的方法。
Just find me at garrettlord@Handshake, and you find me on Twitter.
只需通过 garrettlord@Handshake 找到我，您就可以在 Twitter 上找到我。
Love X, huge X guy.
爱X，巨大的X家伙。
You can email me at garrett@joinhandshake.com and double R, double T.
您可以给我发电子邮件 garrett@joinhandshake.com 和双 R，双 T。
And how can you be helpful?
您可以如何提供帮助？
We are trying to hire so many people.
我们正在努力雇用这么多人。
We have offices in New York and in San Francisco, and London and Berlin.
我们在纽约、旧金山、伦敦和柏林设有办事处。
If you have friends that are maybe passionate about this, you let them know, or you're interested in the learning more, please reach out.
如果您有朋友对此充满热情，请让他们知道，或者您有兴趣了解更多信息，请与我们联系。
We'd love to talk to you.
我们很乐意与您交谈。
Hiring is like the number one problem we have right now to meet the demand.
招聘是我们目前满足需求的首要问题。
So, if you're talented and interested in learning more about Handshake, if you want to work on our consumer product, if you want to work on our employer products, cool PLG issues or the state-of-the-art consumer social experience, like reach out, or you want to work on the AI business we'd love to talk to you.
因此，如果您有才华并且有兴趣了解有关 Handshake 的更多信息，如果您想从事我们的消费者产品工作，如果您想从事我们的雇主产品、很酷的 PLG 问题或最先进的消费者社交体验（例如伸出援手），或者您想从事人工智能业务，我们很乐意与您交谈。
Lenny Rachitsky: To make it even more clear for folks, what roles are you most hiring for?
Lenny Rachitsky：为了让大家更清楚，你们最常招聘哪些职位？
Is it every role?
每个角色都是这样吗？
Is it engineering?
是工程学吗？
Lenny Rachitsky: Engineering, all right.
Lenny Rachitsky：工程学，好吧。
If you're an engineer and want to join one of the fastest growing AI companies in the world right now, here we go.
如果您是一名工程师并想加入目前世界上发展最快的人工智能公司之一，我们就来吧。
We'll link to your careers page in the show notes.
我们将在演出笔记中链接到您的职业页面。
Lenny Rachitsky: Yeah, of course.
莱尼·拉奇茨基：是的，当然。
Garrett, thank you so much for being here.
加勒特，非常感谢你来到这里。
This was incredible.
这太不可思议了。
Bye everyone.: Thank you so much for listening.
大家再见：非常感谢您的聆听。
If you found this valuable, you can subscribe to the show on Apple Podcasts, Spotify, or your favorite podcast app.
如果您发现这很有价值，您可以在 Apple Podcasts、Spotify 或您最喜欢的播客应用程序上订阅该节目。
Also, please consider giving us a rating or leaving a review, as that really helps other listeners find the podcast.
另外，请考虑给我们评分或留下评论，因为这确实有助于其他听众找到播客。
You can find all past episodes or learn more about the show at Lennyspodcast.com.
您可以在 Lennyspodcast.com 上找到所有过去的剧集或了解有关该节目的更多信息。
See you in the next episode.
下一集见。