看不见数据时如何构建法律智能体：Legora CTO Jacob Lauritzen

好的，我们正在演播室，特邀嘉宾是 Lora 的 CTO Jacob Lauritzen。我念对了吗？

Okay, we're here in the studio with special guest Jacob Lauritzen, CTO of Lora. Did I get it right?

Lorisen 就行。对，没问题。我知道丹麦语怎么念。对，没错。

Lorisen is fine. Yeah, that's okay. I know how it is in Danish. Yeah, exactly.

这个名字有什么特别的来历吗？

Does it have a special history?

有，Sen 就是儿子的意思，某某之子。你有 son，对，所以就是 Lords 之子。这就是我的名字。

Yeah, just know Sen is like son, son of. You have son, and yeah, so son of Lords. That's my name.

不错。我们之前也碰过几次面。你在几个活动上演讲过，而且显然 Lora 是史上增长最快的法律科技公司之一。

Nice. We've crossed paths here and there. You spoke at a few events, and obviously Lora is one of the fastest growing legal tech companies in history.

是的。

Yeah.

你在开场前说了一件我觉得很有意思的事。你说你们的起源非常欧洲，尽管你们显然正在纽约和旧金山扩张。你们的文化感觉有什么不同？

And you said something at the start in a pre-show that I thought was really interesting. You said that your origin is very European, even though obviously you're expanding in New York and SF. What feels different about your culture?

我觉得我们骨子里可能对所做的事非常谦逊，因为我们从一开始就一直是弱势的一方。没人相信斯德哥尔摩地下室里的小创业公司能打败美国那家远远领先于我们的巨头。所以我觉得我们一直是弱者，因此一直非常谦逊。我们一直都得为生存而战。我觉得随着我们越来越壮大、进入美国并在美国待得更多，我们不得不学会的一件事是稍微更大胆一些，告诉世界我们有多棒，因为这对我们中的一些人来说并不是自然而然的。

I think we are just at the core probably very humble about what we're doing because we've always been the underdog from the very very beginning. No one believed that small startup in a basement in Stockholm was going to outcompete this huge company that was way ahead of us in the US. So I think we've always been underdogs and so we've always been very humble. We've always sort of had to fight for our existence. I think it's something we've had to learn as we've also become more and more, as we've entered the US and been in the US a lot more, is to be a little bit more bold and tell the world how great we are, because it doesn't come naturally to I think some of us.

我想先聊聊你们的产品，因为我觉得很多这类法律产品都挺神秘的，你知道，不该展示出来。但你能不能讲讲产品长什么样，也许你们是怎么走到今天的？比如哪些产品迭代让你们达到了今天的位置？我觉得那会很有意思。

I kind of want to start by going over your guys' product because I feel like with a lot of these legal products, they're kind of under wraps, you know, you're not supposed to show it. But if you could just talk about what does the product look like and maybe how did you guys get there? Like what have been the product iterations that got you to where you are today? I think that would be really interesting.

是的，当然。一切的开端就是 ChatGPT 里加了一个瑞典法律搜索工具。那就是全部了。所以我们就从那里开始。我们从法律研究做起，这基本上是世界上一个巨大的市场，律师花大量时间弄清楚之前有哪些类似的案例、法律先例是什么样的、以前的人在这些情况下是怎么做的。这极其重要。所以这一直是我们产品的一部分。基本上我们从一个你可以提问、它会查询法律的智能体开始。那是非常简单的起点。然后从那里扩展。所以它扩展到了智能体能做什么。它现在能很好地起草很多东西。它能编辑 Word 文件。它能移动你的文件、整理一切。运行非常复杂的分析。我们还给它加了一大堆功能。所以，我们的一个功能叫表格视图，现在这个名称已经在行业里被使用了，我们很高兴看到 Claude for Legal 在用我们的术语，他们正在构建的东西。

Yeah, absolutely. The way everything started was ChatGPT with a Swedish laws search tool in it. That was literally the entire everything. And so that's where we started. We started with legal research which is like a huge market basically in the world where lawyers spend a lot of time figuring out what are previous cases that look like this, what's the legal precedent like, what have previous people done in these situations. That's extremely important. And so that has always been part of our product. Basically we started from an agent you could ask questions and it would query laws. That was the very simple beginning. And then from that it's expanded. And so it's expanded in what can the agent do. It will now draft a lot of stuff really well. It will edit Word files. It will move your files away, organize everything. Run really complex analysis. And we've added just a bunch more features to it. So, one of our features called tabular view, which is now a name that's being used in industry, and we're very happy to see Claude for Legal using our terminology, what they're building.

我们得聊聊那个。我们得聊聊那个。

We got to talk about that. We got to talk about that.

这实际上就是一种批量提取和批量审阅大量协议的方式，你在尽职调查或融资等场景中会这么做。所以这部分就是我们帮助律师更高效、更好地完成工作。然后除了做工作本身，我们现在还比一年前更多地帮助协调工作。所以我们有不同种类的——我们有一个叫 Lists 的东西，对我们来说有点像 Linear。我们还有 Intake，这是一种与客户协作的方式。我们还有 Portal，这也是一种协作方式，你知道，律师对律师、客户对律师、律师对智能体、也许智能体对智能体。我们花了很多时间在那里，因为坦率地说，10 年前甚至今天与律师合作的方式真的很痛苦，说实话，有很大的空间可以做得更好。所以我觉得就是这两件事：做工作，以及协调工作。

Which is actually a way to just bulk extract and bulk review many, many, many agreements which you do in due diligence or when you raise money or whatever. So that's part of it is like we help lawyers just do their work much more efficiently, much much better. And then outside of just doing the work, we're also now helping orchestrate the work a lot more than we were just even a year ago. So we have different sort of—we have a thing called Lists which is sort of like Linear for us. And we have Intake which is a way to collaborate with clients. And we have Portal which is also sort of a way to collaborate, you know, lawyer to lawyer, client to lawyer, lawyer to agent, agent to agent maybe. We're spending a lot of time there because frankly the way that you work with a lawyer 10 years ago or maybe even today is really painful and there's loads of room to do that in a much better way to be honest. And so those are like I think the two things. It's doing the work and it's also orchestrating the work.

你们什么时候开始做或构建智能体的？

When did you guys start doing or building the agent?

我的意思是，我们基本上经历了从助手到我们所谓的智能体的转变，对吧。

So I mean we've basically had this switch from the assistant to the agent we call it, right.

我猜智能体——我不知道——我想说大概一年前。

I guess the agent—I don't know—I want to say a year ago probably.

是的。

Yeah.

所以这是一个粗略的时间线，给了解 YC 2023 的人。

So this is a rough timeline for people who are understanding YC 2023.

是的。

Yes.

然后你是 YC 创始人。

And then you were a YC founder.

YC 24。

YC 24.

对。我是 YC 23 的。Goro 是 24。

Yeah. I was in YC 23. The Goro was 24.

好的。然后所以这有点像在智能体真正起飞之前一点，然后推理模型显然改变了一切。我的意思是，你知道，从某种意义上说今天——你知道我们预览了产品——它看起来确实不错——预览我们私下看的，但我们会把视频放到 YouTube 上——它基本上就是给律师用的 ChatGPT，对吧？这么说很公平。但侧边栏有很多不错的自定义工具，比如表格视图，比如我很喜欢的 Intake，还有你知道的一堆其他东西。

Okay. And then so there is this is like a little bit pre—like agents really taking off and then like reasoning models obviously change everything. I mean you know in one sense today—you know we had a preview of the product—it does look well—preview we had a private thing but we will put up a video on YouTube—it's mostly like ChatGPT for lawyers, right? Like very fair to say that. But then the sidebar has a lot of nice custom tools like the tabular view, like the Intake which I really liked, and you know a bunch of other things.

没错，就是这样。它就是同事版 ChatGPT，对法律工作真的非常非常好，对于像内部公司律师、寻找即将出台的法规的人，所有这些事情，我们可以多聊聊那里的产品。

Exactly, that's exactly what it is. It's coworker ChatGPT is like really really good for legal and the workflow for like in-house corporate lawyers, people looking for upcoming regulations, all these things, and we can talk about more of the products there.

对。你们底层用的是什么模型？能聊聊吗？

Yeah. What models are you guys using underneath? Like can you talk about that?

我的意思是，我们根据任务使用一堆不同的模型，对吧？那是——

I mean, we're using a bunch of different ones depending on task, right? That's um—

那律师看不到你们用的这些？

And then that's not available to the lawyer to see you guys?

对于——是的。嗯，我们在这件事上反复讨论过。是的。

For the—yeah. Well, we've been back and forth on this. Yeah.

和所有人一样，我认为智能体里有一个针对主模型的选择器，但有些工具我们会用不同的模型。比如表格视图，我们尝试——我觉得最有效的做法是把问题细分，在细分任务的同时做路由。不过我们现在确实在智能体里放了一个选择器。说实话，我不确定它会不会一直留在那里。所以是的。

Just like everyone else, I think in the agent there's a model selector for the primary model, but there are certain tools where we use different models. There's tabular view, and we try— I think we found it most effective to subdivide the problem and route as we subdivide the task. But we do right now have a picker in the agent. I'm not sure if that's going to stay there always, to be honest. So yeah.

是的。我补充一点背景，因为我是 Cognition 的顾问。

Yeah. I mean, I bring a little bit of context to this because I'm an adviser for Cognition.

是的。

Yeah.

我刚加入的时候，我对智能体实验室是什么的一部分论点是——智能体实验室理想上就是一个领域专用的 GPT 套壳，对吧？我们负责运营——我们是领域专家。我们会关注你，比前沿实验室更好地服务你的需求，并按照我们的判断把前沿实验室最好的东西给你。所以我们的工作是替你决定，因为你的工作不是——

And when I joined, you know, part of my thesis for what an agent lab is— an agent lab is, ideally, a domain-specific GPT wrapper, right? We run the— we are the domain experts. We'll pay attention to you. We'll serve your needs way, way more than the frontier labs, and we'll give you the best of the frontier labs according to our judgment. So our job is to decide for you, because your job is not to—

正是如此。我们会一直让你处在 PTO 前沿，确保它始终很棒，你不需要深入所有细节。因为有一点——说到这个，如果你去找律师，问他们：哦，你要根据这个先例起草这份合同，你觉得这是 Sonnet 高配还是 Opus 中配的任务？你知道，他们不会知道的。就给我这个——

Exactly. We'll always keep you on the PTO frontier and make sure that it's great all the time, and you don't have to dive into all the details. Because one of the things— to this point, if you go to a lawyer and you ask them, oh, you're going to draft this contract based on this precedent, do you think that's a Sonnet-high or an Opus-mid task? You know, they won't know. Just give me this—

就确保它能用。

Just like make sure it works.

时薪 1000 美元。

Hourly rate is $1,000 an hour.

没错，就让它能用就行。

Exactly, just like make it work.

所以是的，我最初是带着这个想法来的。我当时想，哦对，这就是 Devon 与众不同的地方。而现在我们的模型选择器里有五个选项，包括 Fable、Opus 和 Astra。人们想试试那些花哨的新模型。

So yes, I originally came in with that. I was like, oh yeah, this is what Devon does differently. And now we have five things in the model picker, including Fable and Opus and Astra. And people want to experiment with the fancy new model.

你觉得——情况一样吗？

Do you— is that the same?

我觉得是的。我觉得这——在法律领域肯定比工程领域少得多。

I think so. I think it's pro— it's definitely way less in legal than it is in engineering.

是的。

Yeah.

但确实有人就想马上用 Table 1。你知道，人们一直在问我们 Mythos 之类的东西。尽管我认为现实中很多法律工作并不需要 Mythos 级别的智能。

But there are definitely people that just want table one right away. And you know, people were asking us about Mythos and everything. Even though I think realistically a lot of legal work does not require Mythos-level intelligence.

是的。所以完全正确。我觉得基本上你就是说，好吧，我们给你这个选项。你要求了，但你的额度会消耗得快 10 倍——

Yeah. So, exactly right. Like I think basically you're like, okay, we'll give you the option. You asked for this, but you're going to run down your credits like 10x faster than—

是的。我本来很想——我们抗拒引入它很久，最后我们还是妥协了。我们说，好吧，如果你想要这种程度的控制，那我们就给你。

Yeah. I would have loved— we fought introducing it for a very long time, and then in the end we just kind of caved in. We're like, okay, if you want this level of control, then we'll give it to you.

是的。所以你觉得模型已经达到律师需要的水平了？就是没必要再变强了？

Yeah. So, you think the models are already like as good as lawyers need? Like there's no need for them to get better.

这是——我觉得这是一个能力分布。大概 60% 的工作我们现在就能做得很好，从现在起就是成本和延迟的问题。但剩下的部分，尤其是非常复杂的分析，我们仍然需要更多智能。然后我觉得还有工作范围的扩张——就像我们现在有这些工作，但律师的工作大概也会随着智能而扩张，我想。

It's a— I think it's a power distribution. So there's probably like 60% of the work we can do really well right now, and from now on it's a cost and latency thing. But then there's the rest of the thing where, especially for very complex analysis, we still need more intelligence. And then I think there's also just the work expansion— kind of like that's the work that we have right now, but also the work of a lord— thus probably expands with intelligence, I imagine.

早期，你们刚开始的时候，我猜是转向智能体或者构建智能体的时候，你们用的是什么模型？我猜大概是 Opus 4.6 之类的。

Early on, like when you guys first started, I guess transitioning to the agents or building the agents, like what models were you guys using? I guess it would have been like Opus 4.6 or something.

那应该是——是的。Sonnet 4.6。Opus 4.6 大概没错。

That would have been— yeah. Sonnet 4.6. Opus 4.6 is probably right.

那时候智能体不擅长、但现在强得多的是什么，尽管才过了大概 9 个月？

What were the agents not good at at that point that they're much better at now, even though it's only what, like 9 months later?

子智能体委派是其中之一，我们现在已经让它运转得很好了。对于扇出型检索查询，现在对我们来说效果极好。以前这很难。

Sub-agent delegation was one of them that we've gotten working really well now. For fan-out type retrieval queries it works extremely well for us now. That was difficult before.

我们基本上会追踪——我们有一个专门做评估的团队。所以我们追踪所有模型在不同执业领域的表现，也就是法律的不同子领域，以及这些领域里的不同任务。所以我们基本上有一张大的热力图，然后我们——

We basically track— so we have a team that's just doing evals. And so we track all of the models' performance in different practice areas, which is like subgroups of legal, and different tasks across all those. So we basically have a big heat map, and we're like—

然后我们能看到它们的变化，通常它们会改善大部分东西，但实际上有很多模型发布是改善了一些、在另一些上退步了。

And then we can see them move, and usually they improve most stuff, but there are actually a lot of model releases where they improve some and regress in others.

好。

Okay.

而这实际上比你想象的要常见得多。

And that actually happens a lot more than you'd think.

比如那些任务是什么？你熟悉你们怎么跑这些评估吗，比如你们看的那些不同子领域是什么？

Like what are some of those tasks? Are you familiar with how you guys are running a lot of the evals, like what are those different subsegments, I guess, that you look at?

是的。所以有不同种类的框架设定。我们基本上有不同的世界——本质上就是强化学习环境——就像，这里有一个——因为律师总是在做项目，或多或少。所以就像，这里有一个一万个文件的数据室。你在做尽职调查，最后一份 grit 报告长这样。它包含所有这些红旗警示，不包含那些奇怪的东西，我们希望它简洁，等等。我们大概就是这样构建的。所以有不同任务，其中一部分是起草报告：草稿看起来好吗？它包含所有正确的内容吗？措辞合理吗，分析合理吗？还有找红旗警示的部分，就是在那之前的分析。所以我们可以稍微细分一下，我们就是这样跑评估的。

Yeah. So there's different just like framing stuff. So we have different basically like worlds— like RL environments essentially— which is like, here's a— because lawyers work in projects always, more or less. So it's like, here's a 10,000-file data room. You're doing a due diligence, and in the end a grit report looks like this. It contains all these red flags, and it doesn't contain whatever weird stuff, and we want it to be concise, etc., etc. That's kind of how we've built it. So there's different tasks, which is like part of it is drafting a report: does the draft look good? Does it contain all the right stuff? Is the wording sound, the analysis sound? There's also the part of finding the red flags, like the analysis before that. So we can kind of subdivide it a little bit, and that's how we run the evals.

有意思。等等，你们的评估团队是什么样的？你们会雇律师来做这个吗？

Interesting. Wait, what does your eval team look like? Like do you hire lawyers to like do that?

是的，是混合的。我们有工程师，也有律师，是一个团队。现在大概有三个工程师和三个律师。然后我们有一个池子，我们称之为法律工程师，就是那些在大律所工作了大概三四年的人，通常他们可能有计算机学位，非常懂技术，然后他们来找我们，因为他们想帮助改变这个行业。

So yes, it's a mix. We have engineers and then we have lawyers, and it's one team. So there's like three engineers and like three lawyers right now. And then we have a pool of what we call legal engineers, which is people that have worked in big law for like three or four years, typically maybe they have a CS degree, and they're very tech forward, and then they come to us because they want to help change the industry.

他们是在大律所，还是大律所的律师？

Are they in big law or lawyers in big law?

他们是大型律所的律师。

They're lawyers in big law.

还有计算机学位。

With CS degrees.

是的。

Yes.

我的天。这是个维恩图。是的。没错。

Oh my god. It's a Venn diagram. Yeah. Exactly.

那大概是我们最难招的人。但他们也是巨大的力量倍增器，因为我们可以拉他们来做评估和产品开发，就像——你有一个 100 名专家的池子，他们每天使用产品、每天和客户交谈，而且他们亲身经历并一直经历着这些，我们就拉他们来做评估、帮我们塑造产品。这真是巨大的优势。

That's like our hardest person to hire for. But they're also such a force multiplier, because we can pull on them for the eval stuff and for the product development, which is like— you have a pool of 100 experts that use the product every day and go and talk to clients every day, and also like they live this and have lived this, and we just pull on them to do evals and help us shape the product. And it's such an advantage.

我很好奇，你们什么时候开始组建那个团队的？

And curious, like when did you guys start that team?

评估在某个时候是从助手智能体团队里长出来的。我觉得直到最近，有一段时间它老实说并不太值得做，因为所有模型都在变好。

Evals grew out of the assistant agent team at some point. I think there was a point up until recently where it was like honestly just not really worth it, because all models just got better.

所以当时的问题是：我们是花大量时间做非常复杂的评估，还是直接去构建？因为构建更好的测试框架能带来很多收益，而且我们相信一切都会变好。结果一切确实都变好了。

So it was like, do we want to spend a lot of our time doing really complex evals, or do we just want to build? Because there was so much to be gained from building a better harness and just leaning into everything's going to improve. And everything kind of just improved.

但没错，就像每两三个月你就能直接通过。对吧。

But yeah, like every 2-3 months you get just pass it through. Right.

没错。我的意思是，我们曾有一个问题：它引用不正确，导致难以跟进。结果下一个模型版本就解决了。从大约 70% 提升到 99%。我们什么都没做。真是太好了。

Exactly. I mean, we had one problem where it wouldn't cite properly, so it would be difficult to follow. And it was like the next model release solved it. It went from like 70% to like 99%. We did nothing. And like, great.

是啊，也许真正的工作是保持所有数据私密，使其可标记、可排序，添加新功能比如表格视图。那是……

Yeah, maybe the real work is keeping all the data private, making it taggable, sortable, adding new features like the tabular view. Is that the...

你是指除了这些之外？

You mean over and above?

是的，但显然现在你也在做评估。我猜你对开源模型的使用也在增加。我们可以谈谈这个。呃，所以，我不知道我在这里说什么，除了，你知道，工作是什么？工作是……

Like yes, but obviously now you're also doing evals. I assume your use of open models is also increasing. We can talk about that. Uh so yeah, I don't know what I'm saying here apart from like, you know, what is the job? The job is...

也许不是。早期并不太需要做评估，因为主要是向律师展示模型能力。很大一部分工作是弄清楚，有了这些技术，法律行业在五年、十年后会是什么样子。编程领域可能领先一两年，所以我们可以稍微参考一下。但基本要素是什么？人们如何工作？如何协作，如果不是通过 200 个线程的邮件和来回发送 v5 Word 文件？那会是什么样子？这甚至不一定是纯粹的人工智能。那只是构建更好的产品体验。是的，并确保在受监管的行业中做到审计日志、数据保留等所有真正重要的事情。然后是人工智能部分，让工作更快更好，以及主动智能体在事情进来时做出反应，这样你就不必启动一切。所以我认为很多工作只是弄清楚我们认为五年后会是什么样子。如果模型能力持续提升，如果价格持续下降，我们能做什么？你能登录后看到一个指挥中心，就像你需要对智能体最近一直在做的五件事做出反应，并审查、检查和改进系统。也许那就是我们要去的地方，或者正在去的地方。

Maybe not. In the early days it was not so much to do evals because it was mostly exposed model capability to lawyers. Yeah, the huge part of it is just figuring out what does the legal profession look like in five years, 10 years with all of this stuff. Like coding is maybe a year or two ahead. So we can kind of look to that a little bit, but like what are the primitives? How do people work? How do you collaborate if it's not a 200-thread email with like v5 Word files back and forth? Like what does that look like? And that's not even necessarily purely AI. That's just like build a much better product experience. Yeah, and make sure you do it with audit logs and data retention and all of the really important things in a regulated industry. And then there's the AI part which is like make the work faster and better and proactive agents that react to stuff when it comes in so you don't have to kick off everything. So I think a lot of the work is just figuring out what do we think it's going to look like in 5 years. If model capabilities keep increasing, if price keeps coming down, what could we do? Can you log in and it's like a command center where it's just like you need to react to these five things that the agents have been doing lately and review and check and improve the system. Maybe that's where we're going, or is where we're going.

是啊。是啊。是啊。我正想问，你能再深入一点吗？比如，几年后法律行业会是什么样子？

Yeah. Yeah. Yeah. I was going to ask, like could you go into that in a little more depth? Like yeah, what is the law profession going to look like in a few years?

我得先声明，我不知道。我的意思是，谁知道呢，但我认为我们可以多参考编程领域。我认为所有律师都会比之前高效得多。他们能够以更低的成本完成更多高质量的工作。我认为这个职业本身、工作本身会大大改善。会更多……我认为律师会把更多时间花在他们成为律师时想做的事情上。你知道，弄清楚正确的风险状况、可管理的风险是什么，以及他们所有的判断、战略判断，还有与客户合作。他们会做更多这些，而少做那些翻查大文件寻找确凿证据之类的工作，或者少做许多律师一直在做的那些琐碎、超级烦人的工作。

I should preface this by saying I don't know. I mean, who knows really, but I think we can look to coding a lot. I think all lawyers are going to be way more productive than they were before. They're going to be able to do a lot more stuff at a higher quality for lower cost. I think the profession itself, the work itself, is going to improve a lot. It's going to be a lot more... I think lawyers are going to spend more of their time doing the stuff that they became lawyers to do. You know, figuring out what's the right risk profile, what's manageable risk, what's like the all of their judgment, strategic judgment, and like working with clients. They're going to do that a lot more and much less of the like trolling through big documents to like find the smoking gun or like do all this menial, super annoying work that a lot of lawyers have been doing.

你认为合伙模式会发生什么？你觉得它会保留还是怎样？

What do you think is going to happen in the partnership model? Like do you think that stays or like what happens?

我实际上上周刚好好思考过这个问题。是的，

I gave this I actually gave this a lot of thought just like last week. Yeah,

因为合伙模式……我不想涉及太多细节，但合伙模式的存在是因为这些公司唯一拥有的资本类型是人力资本。

Because the partnership model... I don't want to get into too much of the details, but like the partnership model kind of exists because the only type of capital that these companies have is human capital.

所以不像我们或软件工程公司或制造商，你有实物资产或某种无形资产，随着时间的推移积累，并向股东支付股息。但如果你在专业服务这样的行业工作，咨询公司等也一样，你所有的价值都在你的人身上，你无法真正拥有人力资本。你知道，你不能，你知道，奴隶制已废除。你不能拥有 people，但你可以租用他们。你可以付钱给他们。所以合伙制本质上是一种结构，让人力资本成为公司的所有者，以便他们留下来。

So like it's not like you know us or like a software engineering firm or like a manufacturer whatever you you have physical assets or like some sort of intangible assets that that you acrew over time and and they pay dividends to the shareholders. But if you work in an industry like professional services and it's the same for like consultancies and and whatever else, all of your value is in your people and you can't really own human capital. You know, you can't, you know, slavery is abolished. You can't own people, but you can rent them. You can pay them money. And so partnerships essentially is a structure where you make the human capital the owners of the company so that they stay.

是的。

Yeah.

那会改变吗？我觉得不一定。但我认为我们真正想用 Lor 做的是，我们实际上希望律师、律所能够建立某种无形资本，某种持久资本，比如你可以建立一个学习你工作方式的系统。现在很多律所认为，嗯，有前律所的工作方式，我们可以将其编码化，我认为我们可以建立一个系统，无论是自改进的知识图谱还是什么，以及先例文件和文档，其中有一种工作方式能产生非常好的输出，那可以成为他们的知识产权。所以这是一种将人类专业知识和判断编码到系统中的方式，我觉得现在我只是在胡言乱语，但嗯，这是我们想要帮助赋能的事情。

Will that change? Like I don't it doesn't have to. But I think what we really want to do with Lor is we actually want lawyers to be able law firms to be able to build some sort of intangible capital, some durable capital in like you can build a system that learns of how you work. A lot of firms right now think about like well there's the ex law firm way of doing things and we can codify that I think and we can build a system and whether that's like a self-improving knowledge graph or whatever it is and like president files and documents where there's like a way of working that produces a really good out output and that can be their IP. So it's like a way to codify the human expertise and judgment into a system that I think is now I'm just rambling completely but like um that's something that we want to help empower

那这会改变合伙模式吗？因为合伙模式中的激励不同，你只需要带来更多客户

And does that change the partnership model then cuz the incentives are different in the partnership model you just got to bring more clients

而且你不想从带来客户的时间中抽走时间去编码你的信息，基本上是这样

and you don't want to take away time from bringing in clients to codify your information basically

因为就像你得到的奖励来自，你知道，eat what you kill 或 lock step 之类的，嗯，那需要稍微改变，我们需要弄清楚一种激励结构，或者合伙伙伴被激励去编码东西，为公司建立资产，但现在他们不是，因为一旦他们离开合伙关系，他们就不会从他们建立的资产中获得任何股息。

Cuz like you get rewarded from you know eat what you kill or make this lock step whatever um that needs to change a little we need to figure out an incentive structure or partnerships partners are incentivized to codify things for to build an asset for the company but right now they aren't cuz as soon as they leave the partnership they don't um get any dividends from the assets that they built.

是的。

Yeah.

所以就像，是的。我认为整个商业模式是，呃，我们拭目以待。

So it's like a Yeah. I think the the the whole business model is is uh we'll we'll see.

是的。是的。现在一切都在疑问中。

Yeah. Yeah. It's all it's all in question right now.

我实际的另一个观点，你知道，在合伙层面上，那是行业的高端。实际上，我的观点更多在低端，对吧？会发生什么？

My my actual other take you know so on the partnership level that's like the top end of the the industry. Actually, my take is more on the bottom end, right? What happens?

律师助理没了。

Paralegals gone.

是的。

Yeah.

对。就像初级律师没了。我不需要他们。就像

Right. Like junior lawyers gone. I don't need them. Like

嗯，实际上我们看到的不是这样。所以是的，我认为，我知道。你会说我们在工程领域看到了这种情况吗？

um that's actually not what we're seeing. So yeah, it's I think it's I know. Would you would you say that we're seeing that in engineering?

呃，是的。所以初级工程师有点完蛋了。就像你 kind of 必须从大学出来，呃，辍学，完全成熟，否则你无法成功。

Uh yeah. So junior engineers are kind of cooked. Like you kind of have to come out of uh drop out of university fully baked or else you're not going to make it.

我觉得为了喜剧效果我有点夸张了，但我认识很多律师，他们从律师助理做起，读了法学院，经历了所有这些，对吧？而现在他们不被需要了。他们真的就是不被需要了。

I feel like I'm being a bit dramatic for comedic purpose, but I know a lot of lawyers who started as paralegals, went through law school, all these things, right? And now they're not needed. They're seriously just not needed.

是啊。

Yeah.

合伙模式，也就是那种金字塔结构，会发生什么很有意思。我觉得总体而言，很多律师，你仍然需要下一代。至少对工程来说，我的想法是，今天你有一个好工程师，他们比两年前值钱得多，因为他们能产出多得多的东西。前提是他们足够优秀。我觉得法律行业也是一样，如果他们优秀，他们实际上能做出很多非常好的工作。所以我觉得更多是，真正优秀的律师或初级律师值钱得多。

It's interesting what happens to the partnership model, like the pyramid. I think generally a lot of lawyers, you still need the next generation. And at least for engineering, the way I think about it is you have a good engineer today, they're worth a lot more even than they were 2 years ago because they can produce so much more stuff. If they're good. And I think it's the same thing in legal, where they can actually do a lot of really good work if they're good. So I think it's more that really good lawyers or junior lawyers are worth a lot more.

是啊。一直都是这样，对吧？即便如此，它变得更加两极分化了。

Yeah. It's always been the case, right? Even so, it's polarizing even more so.

我其实对这一点很好奇。整个法律生意基本上是按小时收费。那么如果现在每个项目都能快很多完成，会不会变得便宜很多？

I'm actually curious about that. So the whole law business is basically charging per hour. So if each project now can be done a lot faster, is it going to be way cheaper?

这其实是个有趣的故事。我当时在读 Max 的 YC 申请，帮他进 YC。把当时叫 Leia 的公司弄进 YC。我当时唯一的一点就是——

This was actually a fun story. I was reading Max's YC application and helping him get into YC. Get Leia, as it was called back then, into YC. And this was my one thing I was like—

我很好奇，Leia 在丹麦语里是某个东西的意思吗？

And curious, Leia is a Danish thing for something?

不，是瑞典语。不，它也没什么意思。有个特别好笑的故事，我记得是一家西班牙报纸在早期报道了我们，Leia，然后他们解释说这个缩写是“法律高效尤达助手”。

No, Swedish. No, also it means nothing. There was this really funny story in, I think it was a Spanish newspaper that wrote about us in the early days, Leia, and then they explained that the abbreviation was legally efficient Yoda assistant.

为什么是尤达？

Why Yoda?

是啊，我不知道。我当时就想，为什么？为什么会这样？

Yeah, I don't know. I was like, why? Why is that?

你看，幻觉不只是大语言模型才有。

See, hallucinations are not just from LLMs.

不，我觉得记者才会产生幻觉。不过不，它没有任何意思。所以他写了申请。我给他的反馈是，人们已经尝试过一百万次了。法律科技从来就没成过，因为律师没有动力去——他们为什么要更高效？那样他们赚得更少。这完全说不通。但我觉得律师确实想采用这个的原因，是他们明白形势的严峻。他们知道商业模式和一切都面临压力。他们需要改变，因为竞争压力太大了。

No, I think journalists hallucinate. But no, it doesn't mean anything. So he wrote the application. And my feedback to him was, people have tried this a million times. Legal tech has never been a thing because there's no incentive for lawyers to—why would they be more efficient? They earn less money. It doesn't make any sense. But I think the reason that lawyers do want to adopt this is that they understand the gravity of the situation. They know that the business model and everything is under pressure. They need to change because the competitive pressure is so high.

计费模式本身我觉得也会改变。我们已经看到了，尤其是在美国，很多工作实际上按项目收费，而不是按小时收费。尤其是这个谱系里偏低端的东西，比如小型审查之类的。说实话有些甚至是免费做的，有些就按固定费率定价。

The billing model itself I think is also going to change. We're already seeing that, particularly in the US, that a bunch of stuff is actually being priced on a project fee and not on an hour fee. Particularly the lower end stuff on the spectrum, like small reviews and whatever. Some of it honestly is being done for free and some of it is just being priced on a flat rate.

那钱从哪里赚呢？是因为——还有战略性工作？

So where is the money being made then? Is it because—and the strategic work?

是啊，我觉得你们聊的很多都是并购，具体来说，那才是所有钱所在的地方。

Yeah, I feel like a lot of what you guys talk about is M&A specifically, like that's where all the money is.

这是个很简单的例子。诉讼也是，大活儿仍然在那里，而且非常有价值，所以律师会把更多时间花在那上面。客户关系也超级重要。

This is a very easy example. Litigation is also the big work still is there and is very valuable, and so lawyers will spend more of their time there. Client relationships also super important.

像客户关系管理那样？

Like a CRM?

不，就是花更多时间和客户在一起。很多律所会，或者说历史上基本上会免费做一些事，只是为了和客户保持那种关系。

No, just like spending more time with the client. A lot of law firms will, or historically have, basically done stuff for free just to have that relationship with the client.

是啊。

Yeah.

所以现在他们可以这么做，但不必为此损失很多钱。

And so now they can do that but they don't have to lose a lot of money on doing it.

我还以为我只是——我得付律师钱就为了让他做我的朋友。那不就是预付聘金那回事吗？

I thought I just—I have to pay the lawyer money just to be my friend. Like, isn't that the retainer thing?

是啊。

Yeah.

是啊。你就这么告诉我可以免费做，就像——

Yeah. That's how you tell me I can do it for free, like—

是啊。你只需要找到对的律师。

Yeah. You just have to find the right lawyers.

好，酷。

Okay cool.

或者 L space 得做大得多。我不知道。

Or L space has to grow much bigger. I don't know.

是啊。我们在努力。我们在努力。

Yeah. We're trying. We're trying.

好，所以你是说，显然战略性工作和所有这些高附加值的东西仍然在，但我们的律师费并没有下降。可你也说律师助理没受影响，或者说他们只是在转移。这也是我们在 Cognition 担心过的事，人们担心“我们会取代你的初级工程师”这种说法，因为没人喜欢那种话。但实际上，是的，其中一部分确实发生了。另一些初级工程师拥抱它，变得更有生产力，就像你刚才说的。所以我猜，这个角色或这份工作有没有什么新叫法，在你的客户中开始流行？

So okay, so you're saying obviously the strategic work and all these high value add is still there, but our legal fees are not going down. But you also said paralegals are not affected, or like they're just shifting. And this is something that we also did with, we worried about with Cognition, like people worry about the phrasing of like we will replace your junior engineer, because nobody likes that kind of statement. But in actual fact, yes, some part of it does happen. Other junior engineers embrace it and they become more productive, exactly like you just said. So I guess is there a new term for this role or this job that is having traction with your customers?

你是说超级 AI 初级法律？我还没找到什么。也许我们该造一个。

You mean the super AI junior legal? Nothing I found yet. Maybe we should coin one.

不就是法律——法律工程师吗？

Isn't it legal—legal engineer?

法律工程师？那是我们的叫法，现在也成了一个行业术语。

Legal engineer? That's ours and also now an industry thing.

是啊。CSJD 们。

Yeah. CSJDs.

没错。是啊。也许那就是——那些人是做这类事情最有优势的人。我觉得那个贬义词“肉身代理”，就是你坐在机器前，输入别人让你输入的东西，然后重复机器让你说的话。

Exactly. Yeah. Maybe that's the—those are the people that are best positioned to do this kind of stuff. I think the derogatory term of meat proxy, where you just sit in front of the machine and type in something that someone else tells you to type and then you repeat what the machine tells you to say.

但我确实觉得一个非常好的“肉身代理”就是，你非常了解智能体，你非常了解大语言模型，你知道怎么——这是我的工具箱，每个行业都有工具箱，我可以用我的工具去做你需要我做的事。

But I do think a very good meat proxy is just like, well, you know agents very well and you know LLMs very well and you know how to—this is my toolkit, and you know every industry has a toolkit, and I can use my tools to do the thing that you need me to do.

但我猜那你就不是真正的代理了，对吧？你实际上就是在做有意义的工作。

But I guess then you're not really a proxy, right? You're actually just doing meaningful work.

是啊。就像我不再——好吧，我来说说表格审查，因为那是经常发生的事。我有几百、几千、几百万——几千。

Yeah. It's just like I no longer—well, I'll talk about tabular review because that's a thing that happens a lot. I have hundreds, thousands, millions—thousands.

我们单个好像能做到最多一百万个单元格——那很厉害。

We do up to a million cells I think in a single—that's impressive.

我们该聊聊表格和窗口化。是啊。你知道，我在处理一个大案子。我有一百万份文件。我把它输进去。通常我会扔一个律师助理团队——锁——坐在那里到凌晨 4 点。

We should talk about the table and the windowing. Yeah. You know, I'm working on a big case. I have a million documents. I type it in. Usually I'll throw a team of paralegals—lock—sit there till like 4 a.m.

那种事不需要发生了，对吧？因为现在他们可以把它提取出来，你可以去重新提示一切。

Like that doesn't need to happen, right? Because now they can extract it and you can go and reprompt everything.

所以我觉得现在我不需要律师助理了，但一个好的超级 AI 律师助理会做别的事情，你知道吧。

So like yeah I think now I don't need a paralegal for that but like a good super AI paralegal would do something else you know.

对，没错，它们能做更多事情，或者把更多时间花在以前没时间做的事情上。

Yeah exactly what they can do more stuff or spend more time on the things that they didn't have time for before.

或者真的去睡觉，谁知道呢。

Or actually sleep who knows.

我有个问题，你们收购了一家公司叫 Cura，是这么念吗？你们用自家产品做了这笔并购。

I have a question so you guys bought a company Cura is that how you say it and you guys used your own product to do the M&A.

是的，我们经常这么做。

Yes we do that all the time.

那具体是怎么运作的？能聊聊吃自己的狗粮吗？

So how did that work like can you talk about dogfooding?

嗯，效果很好，我们经常这么做。我们的法务团队，我是说 intake 就是和法务团队一起构建的，因为现在我们有 Slack 工作流，大家把东西扔进去。他们觉得这很烦。

Yeah no that works great we're doing that a lot. Our legal team is I mean intake has been built with our legal team cuz right now we have a slack workflow where everyone throws their stuff in. They're like this is annoying.

嗯。嗯。

Yeah. Yeah.

对，而且效率高。所以就觉得很好，我们应该这么做。

Right. And it efficient. So it's like great we should do for that.

嗯。嗯。能稍微讲讲那个过程是什么样的吗？

Yeah. Yeah. Can you talk a little bit about that like you know what that process look like?

当然可以。我们自己做了很多尽职调查，因为可以用表格审查和其他工具来做。我们仍然会让律师来评估某些风险，比如在澳大利亚有 payroll，所以 Zakur 那笔很简单，因为全在瑞典，但我们在加拿大和澳大利亚也做过，所以澳大利亚有 payroll，好吧，我们是不是想让人更战略性地看看这个？所以那部分我们还是会用律师来做战略部分，他们可以真正成为 sparring partner，帮我们很多，比如如何以正确的方式构建这个。但很多更琐碎的事情我们其实可以自己做。

Yeah absolutely. So it's we do a lot of the due diligence ourselves because we can do it with tabular review and other tools. We do still have counsel coming in to like assess certain risk or like you know there's payroll in so Zakur was simple because I was all in Sweden but you know we've done one in Canada and one in Australia and so there's like payroll in Australia okay do we want someone to look at this like a bit more strategically so there we will still use lawyers for the strategic part where they could actually be sparring partners and help us a lot on like how to figure out how to structure this the right way but a lot of the more menial stuff we could actually just do ourselves.

所以很多早期的东西我们会发现，而且我们有过这样的情况：我们发现了一些红旗警告，然后我们就踩了刹车。

So like a lot of uh yeah a lot of the early things we would find And we have had cases where we found something uh that's been like a red flag and then we we've pulled the brakes.

能稍微讲讲其中任何一个吗？

Can you talk a little bit about any of that?

不，我觉得不行。

No, I don't think so.

嗯。

Yeah.

好吧。

All right.

你之前说早期客户是瑞典的，比如瑞典律所。你有没有看到和瑞典律所合作与和美国律所合作之间的区别，诸如此类？

Early on you said you your early customers were Swedish, like Swedish law firms. Have you seen a difference between working with like Swedish law versus like American law firms, that sort of thing?

而且你们主要和大型律所合作，对吧？

And you guys are mostly working with like big big law firms, right?

嗯，我们也去较小的律所。不，我们肯定是从非常大的律所开始的，但 50 名律师的律所也用我们，并且获得很多价值。

Well, we're also go smaller law firms also. No, we definitely started really really big law firms, but 50 lawyers also use us and get a lot of value.

嗯。那么，你有没有看到和瑞典律所合作与和美国律所合作之间的区别？也许甚至，我不知道，亚洲的？

Yeah. So, have you seen differences between working with like the Swedish ones versus American ones? Maybe like even I don't know Asian ones.

嗯。嗯。我觉得欧洲的律所行动稍慢一些。美国的律所非常前卫，行动非常快。我觉得在美国公司那里，这件事在议程上的优先级也更高，比欧洲公司高。他们认为这是战略上超级重要的事情，我们需要在这里做很多。美国公司，无论是企业还是律所，通常也更愿意定制构建东西，他们都想做，比如“嘿，我们想和你们一起构建这个巨大的东西，以某种方式结构化”，而欧洲人则更像“我们就用现成的”。

Yeah. Yeah. I think the European ones have been a bit slower to move. The American ones are very forward move really quickly. I think it's also very high on their agenda of the American companies more so than the European ones. I think it's like a strategic super important thing we need to just like do a lot here. The American companies generally both corporates and law firms also have a much higher appetite to like custom build stuff as like a thing that they all want to do like hey we want to build this like huge thing with you guys and like structured in whatever way where Europeans are much more like we'll just take the off-the-shelf.

他们想自己构建的东西有什么例子？

What's an example of something they would want to build themselves?

集成到他们自己的系统，比如有些客户构建自己的 CRM 和文档管理系统以及各种东西，他们想构建这些，然后想构建触发器，比如你在 Salesforce 里做审查，由于某种原因它转到某个人那里，他们做了些事情，然后进入 Lor，然后再回来，这是简单的一端；还有一些极其雄心勃勃的，比如他们想和我们一起构建一个全新的产品，你知道我们真的做不到。

Integrations into their own like some of these customers build their own CRM and their own document management system and their own all sorts of stuff and they want to build that and then they want to build triggers that like you do a review in Salesforce and it goes to this one person for whatever reason and they do something and then it goes into Lor and then it goes back in and there's that's like the simple end and there are some that are extremely ambitious where it's like a whole new product that they want to build with us and I you know we can't really do that.

但这确实是美国公司比欧洲公司更想自己构建的一点。

But that's definitely a thing where like American companies really want to build their own more than European ones do.

而欧洲的则像“嘿，你知道，他们给你产品建议或者？”

And the European ones are like hey like you know they give you product suggestions or?

嗯，当然。我的意思是，我们和所有客户都有很好的合作。所以他们会说“嘿，这个很烂。这个很好。我们想要更多这个”，然后我们就构建一个能做到的产品。

Yeah, for sure. I mean, we have great collaboration with all of them. So, they'll they'll, hey, this sucks. This is great. We want more of this, and we'll just build a product that does that.

好的。所以，他们不觉得自己的业务面临生存威胁。

Okay. So, they don't see like an existential threat here in their business.

我是说，

I mean,

而美国公司则觉得有，或者

Whereas the American ones do, or

我不知道这是否是他们想要定制构建与否的原因。我认为他们都知道这对律所成功和企业成功极其重要和关键。所以，对他们来说都极其重要。我认为只是实现方式不同。

I don't know if that's the reason that they want to custom build versus not custom build. I think they both know that this is like extremely important and key to their firm success and their corporate success. So, they it's both extremely important. I think it's just a different way to achieve it.

好的。另一个问题，我想也许你想提这个，就是产品指标。比如你怎么知道我们实际上在这里做了好事？

Okay. Another question I think maybe you want to bring this up is like product metrics. Like how do you know that like you know we're actually doing something good here?

嗯。你怎么考虑这个？我猜这也和评估有关。

Yeah. How do you think about that? I'm assuming that's also related to evals.

所以我们有我们非常关心的常规产品指标。人们是否在我们的产品上花很多时间？他们似乎从中获得很多价值吗？他们续订吗？他们经常回来吗？对于企业软件来说，我们的数据非常好，比如我们的 DAO 超过 MAO，以及你知道我们在产品上的每月支出非常高。然后我们还做一些在线评估，就是试图大致弄清楚我们是否解决了这里的问题，就像用一个超级便宜的 LM 消息，试图有一个在线指标，然后我们也可以在新东西推出时使用它，我们只是按功能标志比较，好吧，它是否改进了，或者是否灾难性下降了。因为实际上有一件事可能很有趣，对我们的业务来说很困难，就是我们基本上不能看任何追踪。我们不能看任何数据，因为都是超级机密的，这工作起来极其令人沮丧，因为你基本上不知道人们在做什么。所以就像这里有一个智能体，它可以做很多惊人的事情，但我不知道人们实际在做什么。

So we have the regular product metrics that we care a lot about. Are people spending a lot of time in our product? Do they seem to get a lot of value out of it? Do they renew it? Do they come back a lot? We have really good numbers for an enterprise software of like our DAO over MAO and our like you know our spent per month in the product are extremely high and then we do some sort of we do online evals also where it's like try to figure out generally like did we solve the problem here just like with like a super cheap LM message try to just like have an online metric and then we can use that also when we do roll outs of new things we just like compare feature flag-wise okay does it did it improve or like did it catastrophically decline because one thing actually that's might be interesting that's difficult for our business is we can't look at any traces basically. We can't look at any data cuz it's all super confidential and that's like extremely frustrating to work with cuz you basically don't know what people are doing. So it's like here's an agent and it can do so many amazing things and I have no idea what people are actually doing.

我本以为你们和律所签了某种协议，他们会说“哦，好的”。

I would have assumed you guys like signed some sort of agreement with the law firm and they're like oh yeah.

我希望如此。是啊，我希望如此。

I wish Yeah, I wish.

不，但问题是他们拥有的很多数据甚至不是他们自己的，对吧？那是他们客户的数据，而且超级机密，所以我们不能看。所以基本上我们有一个 EVA 团队，试图近似我们认为每个人都在做的事情，并把它构建到 EVA 集中。所以这只是业务中一个超级烦人的小问题。

No, but the thing is like a lot of the data that they have is not even theirs, right? It's their clients' and it's super confidential so we can't look at it. And so basically we have an EVA team that's like trying to approximate what we think everyone is doing and like build that into EVA sets. So that's just like a super annoying kink of the business.

你能做转述吗？比如 ChatGPT 有一个 /feedback 功能，直接完整地说出整个轨迹。但显然这对你们来说是不可能的。但你可以用语言模型来转述，比如从这个真实的东西出发，做一个足够接近的合成东西，但改变所有名字，改变所有细节，但仍然……是的。

Can you do paraphrase so for example ChatGPT has like a slash feedback thing that directly fully says the full trajectory. But obviously that is not possible for you guys. But you can use an LM to paraphrase like from this real thing, make a synthetic thing that is like close enough but change all the names, change all the details but like still yeah.

我们真的不能。我的意思是总有一个范围，对吧？比如如果你一直转述到只有三个值，然后你选其中一个，那显然没问题。如果你只是去掉名字，那可能就不行。我们通常选择最安全的方式，尽可能限制自己，因为你知道应该完全没有风险。但有些情况是，好吧，我们知道失败模式可能是这 10 或 20 种。所以我们可以把它分类到其中一种，这是一个有用的信号。我们也可以把它分类到任务类型和实践领域。然后我们就有想法了。这是我们一直试图以聪明的方式弄清楚如何做的事情。

We can't really. I mean there's always a range, right? Of like if you paraphrase all the way down to like three values and you pick one of them then obviously that's fine. If you just remove the names then that's maybe maybe not fine. As like we are generally pick the most safe way like we restrict ourselves as most possible because we you know there should be no risk at all. But there are things where it's like okay we know that the failure modes are probably these 10 or 20. So we can classify it into one of those and that's a helpful signal. We can classify it into the types of tasks and the practice areas also. And then like we have an idea. It's something that we're always trying to figure out how to do in a smart way.

另外，在隐私方面，我猜，我假设甚至两年前，让律师事务所同意这件事一定非常困难。但他们怎么看待他们的数据被发送到 Anthropic？这一定是个大问题。我们所有的基础设施都在 Azure 上。是的。主要就是这个原因，因为当我们开始的时候，律师事务所说 AWS。不。绝对不行。

Also, like on the privacy front, I guess, I'm assuming even two years ago, it must have been like insanely difficult to get the law firms to be okay with this. But like how do they think about their data being sent over to Anthropic or it must be like a huge thing. All of our infrastructure is on Azure. Yeah. Primarily for this reason because like when we started law firms were like AWS. No. Like no way.

我们不信任 AWS。

We don't trust AWS.

是的。是的。现在不同了。不。但他们会说亚马逊、谷歌，我们不能把任何数据托付给他们。所以我们基本上不得不把所有东西都托管在微软上，因为他们已经有 365。

Yeah. Yeah. It's different now. No. But they were like Amazon, Google, we can't trust them with any of our data. So we basically we had to host everything on Microsoft cuz they already have 365.

哦，所以他们接受微软。

Oh, so they're okay with Microsoft.

没错。因为他们已经对所有文档和 Word 文件等做了安全审查。所以微软在他们心中是经过批准的供应商，因此我们不得不把所有东西托管在 Azure 上，这在过去两年发生了变化。所以现在人们明白，好吧，我们需要上云，我们需要做所有这些事情。当时还有更多人在谈论本地部署，这也完全消失了。人们明白你应该直接在云上做，而且很安全，可能比你正在做的更安全。所以变化很大，但才三年，所以实际上就像

Exactly. Cuz they've already done like whatever security review etc. of all of their documents and all of their Word and whatever. So like Microsoft is an approved vendor in their mind and so we had to host everything on Azure and that's changed over the past 2 years. So like now people understand okay we need to get on cloud we need to do all this kind of stuff. Um there was also a lot more people talking about onrem back then that's also disappeared completely. People understand that like you should just do it in the cloud and and it's safe and it's probably safer than what you're doing. Um so that's changed a lot but it's only been three years so like it's actually like

是的。是的。这是一个巨大的转变。这很棒。我的意思是，对你们来说是好事。但是的，我的意思是，我不指望律师事务所，法律行业在过去 10 年里以某种方式忽略了这一点。是的。我认为我们需要 AI 来踢每个人一脚。所以，理解我们需要稍微改变我们的方式。

Yeah. Yeah. It's a huge shift. It's it's it's great. I mean, good for you guys. But yeah, I mean, I don't expect the like, you know, the legal firm, legal industry has somehow ignored this for the last like 10 years. Yeah. I I think we needed the AI uh kick in the butt for everyone. So, understand that we need to change our ways a little bit.

好的。好的。那么，现在开源模型也变得非常好了。所以，你是怎么考虑这个问题的？

Okay. Okay. So, also then on like open source models now are getting really good. So yeah, how are you thinking about that with the

是的，我们正在考虑，就像你说的，带着我们的客户一起，沿着前沿，以及所有模型的地图，它们擅长什么，它们的延迟、成本、治理和区域可用性。我们如何在这所有方面进行优化？所以一些开放权重模型在某些任务上很棒，我们就会在那里使用它们。一些会进行后训练，一些就只用前沿模型。

Yeah, we're just thinking about to to your point um just bring our clients along on like the paralo frontier and like the of the the map of all the models and what they're good at and their latency and their cost and their uh governance and and and their um regional availability. How do we just optimize across all of that? So some open weights models are great for certain tasks and then we'll use them there. Um some will post- train and and some will just use the frontier models.

是的，你称之为模型套利。你有不同的偏好。你在这方面做了一些研究，年初时你坚定地使用 Anthropic。Grok 进步很大。嗯，是的，你想举一个例子，一个整体排名不高，但你想大声称赞它出奇地好的模型吗？

Yeah, you call uh you call it model arbitrage. Uh you have different uh favors. You did some research on this where you were firmly anthropic at the start of the year. Uh Grok is improving a lot. Um yeah, you want to give an example of a model that wasn't ranked highly overall, but like you want to shout it out as uh surprisingly good.

是的，我认为不是整体好。Gemini，早期的一个 Gemini 版本，对于某些类型的转录和说话人归属非常出色。所以我们用它来做转录。

Yeah, I think not overall good. Gemini, one of the early Geminis is uh yeah, extremely good for certain types of um uh transcription and speaker attribution. So we use that for transl transcription.

哦，所以你有像 Granola 那样的转录？

Oh, so you have a transcript like a granola.

我们没有，我们实际上可能会构建一个会议相关的东西，但这只是像语音转文字、文字转语音，就像任何其他有这功能的应用一样。我们实际上也会转录音频文件，如果你上传听证会或其他内容。我们会转录它们，把它们变成文本，这样我们就可以对它进行所有分析。

We don't have we actually probably will build like a meeting thing, but this is just like talk to speech speech to text, text to speech just like in in um any any other app that has this. And we do actually also transcribe files if you upload hearings or whatever. We'll we'll then like transcribe them, put them into text and so we can do all of our analysis on it.

不用 ElevenLabs？

No 11 labs.

不用 ElevenLabs，因为 DPA 的原因。同样，DPA 是指数据处理协议。所以我们的数据处理协议上只有很少的供应商，因为我们需要保持紧密。没错。我们需要对任何数据的去向保持超级严格。所以我认为，不，这是 Gemini 和 OpenAI 的好处之一，他们说你来我们这里，我们有你所需的大部分东西，这实际上非常有价值。即使它不是最先进的，只要它对所有任务来说基本足够好，那谁在乎呢？是的。就像是的。不，绝对。我实际上应该称赞微软的转录模型。他们做了一个，我的转录。

No 11 lab for DPA reasons. Same thing is like a data processing agreement I should say. So like our data processing agreement has very few vendors on it cuz we need to keep it first. Exactly. We need to keep it super tight on where any data goes. And so um I think no that that is one of the benefits of like a Gemini and OpenAI where they're like you come to us we have most of the things that you need and that is actually very valuable. Even if it's like it doesn't have to be like state-of-the-art as long as it's like mostly good enough for all tasks then who cares? Yeah. Like yeah. No, absolutely. I should actually shout out also Microsoft's transcription model. They made one my transcribe.

是的。好的。它很棒。好的。它比旧的 Gemini 好。是的。是的。他们取得了一些好结果。我不知道。甚至不知道，我觉得通常它们都在前二、前三、前四，就竞技场分数而言，那就足够了。

Yeah. Okay. It's great. Okay. It's better than the old Gemini. Yeah. Yeah. They they they did some good results. I don't know. Don't even know like I think like usually all of them are like top two, top three, top four in terms of like the the arena scores and that's good enough.

是的。没错。就像你拥有所有治理，而且超级可扩展，那可能就足够了。

Yeah. Exactly. It's like you have all the governance and it's super scalable and it's like then that's probably good enough.

是的。我想说，在这一点上，语音转文字模型是不是已经足够好，好到即使这里那里错一个词，如果错了，那也

Yeah. I was going to say like at this point are pretty much the uh speech to text models like good enough to where you're like one word here there if it's off that's

不，当然，我做过很多次，特别是在法律领域，有些……我们有过这种情况，我想是在澳大利亚，那里对法庭上的某个人有特定称呼，而它总是把它转写成 condom。

no for sure like I've done this a lot particularly in legal there's like some we we had this thing where I think it's I forget if it's like in Australia where there's like a certain name for like a a person in court and it would consistently prescribe it as condom.

哦，所以那里……是的，有些东西你不想在那里出现。

Oh so there Yeah, there's there's certain things that you don't want there.

是的。是的，但通常你会在上面运行一个语言模型层。然后你提取一些关键词，然后用关键词再次运行转录模型。就像这样，这是我目前为止的工作流程。

Yeah. Yeah, but so usually you run an LM layer on top of it. Uh and then you you extract some keywords and then you run the transcription model again with the keywords. Like this that's kind of my my workflow so far.

但没错，如果你只是大规模地做这件事，在任何不是日常普通对话的领域，你都会遇到大量的转录错误。

But yeah, if you've just done this on any large scale, you will pick up tons and tons of transcription errors on any domain that is not like average day-to-day conversation.

是的。

Yeah.

是的。绝对如此。如果里面有行话，那就难办了。

Yeah. Absolutely. If there's any sort of jargon in there, it's gonna be tough.

是的。我觉得 Whisper 和 11 Labs 仍然把 Claude Code 转录成 Claude Code。

Yeah. Like I think Whisper and 11 Labs still transcribe Claude Code as Claude Code.

是的。这实际上非常非常常见。所以这就是你知道还有很多需要改进的地方。

Yeah. And like that's actually very very common. So like that's how you know there's like a lot more to improve.

是的。绝对如此。

Yeah. Absolutely.

我确实想稍微谈谈智能体产品方面。我们已经涵盖了一些顶层内容，对吧？我们已经提到了 job dropped 和 tableau review。我们提到了 intake，你还有一些我们甚至还没触及的东西。但我只想涵盖两大产品。我们称之为两大产品，这样说公平吗？

I do want to cover a little bit on the agent product side. So we've covered a few of the top level stuff, right? We've already named job dropped tableau review. We name drops intake and you've got some other stuff that we haven't even touched on yet. But I just want to cover the big two. Let's call it like is that fair to say do the big two

智能体和表格审查，可能还有我们的 Word 插件，因为很多人在 Word 里工作，所以我们的 Word 插件很重要，还有 Outlook 插件。

Agent and tabular review, and probably yeah, and our Word add-in is also something because a lot of people work inside of Word and so our Word add-in is huge, and Outlook add-in also.

是的。所以我只想涵盖那些每个人必须知道的重要产品。所以 Tableau review 基本上就像是对每个文档进行大量分发。你可以在表格中收集的所有文档上问同一个问题，然后来回操作。

Yeah. So I just want to cover like the big ones that everyone must know. So Tableau review is basically like fan out of like a lot across each document. You can ask the same question across all documents collected on the table and then work back and forth.

一百万很令人印象深刻。还有其他工程考虑或性能方面的事情，或者这有什么难处？

A million is impressive. What's any other engineering considerations or performance things or like what's hard about this?

就像在我们做那之前，你能先带我们过一遍律师的用户体验和流程吗？是的。

Like before we do that, can you just like walk through like what is the user experience for a lawyer and then like flow? Yeah.

是的。是的。就像他们用表格审查做什么，然后我们可以深入讨论。

Yeah. Yeah. Like so like what are they doing like with tabular review and then we can go into that.

是的。嗯，所以他们通常——我的意思是，一个很好的例子就是尽职调查，对吧？就像我们需要审查所有雇佣合同、所有供应合同等等。所以雇佣合同，你把 10,000 份雇佣合同放进去。然后你嗯

Yeah. Um so they it's typically I mean a good example is just a due diligence, right? where it's like we need to review all employment contracts, all the supply contracts, all of whatever. And so employment contracts, you put 10,000 employment contracts in there. And then you um

就像每个单元格是一份雇佣合同。

Like each cell is one employment contract.

每一行是一份雇佣合同。然后对于单元格，你基本上就是提问。所以就像第一个问题，第一个——呃抱歉，是按列。所以第一列可能是终止期是多久？

Each row is one employment contract. And then for cells, you ask questions basically. So like the first question, the first uh sorry, per column. So the first column might be what's the termination period?

然后它会为每个人填写。是 30 天或其他。下一个问题是是否有知识产权转让条款之类的。下一个问题是我们解雇他们需要支付多少，或者他们的工资是多少，然后它会填写整个内容。你也可以直接让智能体做这件事。就像如果你把 10,000 个文件扔给智能体，说对这个做尽职调查，它会启动一个表格审查并运行所有文件，等待完成，然后分析结果，然后它可以突出显示危险信号。比如，这个 CEO 的薪酬高得离谱，或者有这种奇怪的奖金结构。然后显然我们需要关注那个，然后智能体可以据此起草一份备忘录发送给客户，或者它可以直接做审查，然后让律师用户说好的让我们看看这个，这里是危险信号。还有协作功能，所以你可以评论供其他人查看等等。

And then it's going to like fill out for everyone. It's 30 days or whatever it is. And the next question is like is there an IP assignment clause or whatever. Next one is like how much you know how much do we have to pay them if we fire them or whatever it is like how much are they getting paid and then it's going to fill out the entire thing. You can also just ask the agent to do this. Like if you just drop 10,000 files into the agent and say like do a due diligence on this it will spin up a tabular review and run all of them and wait for it to be done and then analyze the results and it can then highlight like red flags. Okay, this the CEO is getting paid an outrageous amount or there's like this weird bonus structure. Okay, then like obviously we need to look at that and then the agent from that can draft a memo to send to the client or it can just do the review and then have you know a lawyer the user just be like okay let's look at this here are the red flags. There's also collaborative so that you can comment on it for other people to look at etc.

你为什么选择那种表格格式？

And why do you choose that like spreadsheet format?

是的，它只是好得多，它超级容易使用，很容易快速了解正在发生的事情并进行审查。所以我们生产的所有东西都有引用回溯到源文件，所以你可以基本上超级容易地审查，如果你只是想确保一切正确。你也可以高亮这些已审查或其他。这些是被标记的，你可以评论和协作。这比我们尝试过的任何其他形式都容易得多。

Yeah, it's just a lot better to it's super easy to work with work in h it's really easy to get an overview super quickly of what's going on and to review it. So all of the stuff that we produce also has citations back to source documents and so you can basically review it super easily if you want to just make sure that everything's right. You can also like highlights these are reviewed or whatever. These are the ones that are flagged and you can comment and collaborate. That's much easier than any other form factor that we've tried.

是的。我会说表格形式在某些领域有很大的吸引力，但不知何故并没有真正传播开来。

Yeah. I will say that the spreadsheet form factor has a lot of pull in some areas but somehow hasn't really spread.

我觉得很奇怪。就像 Clay 是 GTM 方面的另一个，你知道以表格闻名的

I find it very weird. Like Clay is another one on the GTM side that you know very known for the spreadsheets

你们，我真的不知道是否有其他人做大型表格 AI 的东西。不，不。

You guys I don't really know if anyone else does a large spreadsheet AI thing. No, no.

曾经有像 V7 这样的尝试推出类似表格的东西。甚至在金融领域也有。我真的没有像做表格的顶级名字

There was like V7 that like try to launch there's like sheets. There's even in finance domain. I don't really have like a top name that like does spreadsheets

这个真的很好。是的。不，我同意。我认为如果你有分发并且做完全相同的事情并且提取东西，这是一个很好的形式来理解正在发生的事情。

This really well. Yeah. No, I agree. I think it's like a great if you have fan out and you do the exact same thing and you're extracting stuff, it's such a great form factor to like understand what's going on.

你觉得人们必须导出到 Excel 还是喜欢住在你的表格里？

Do you find like you must export to Excel or like people are happy to live in your spreadsheet?

人们喜欢住在那里。

People are happy to live there.

你们用 ag grid 还是自己的？grid 似乎是领先的。我的意思是，他们做这个大概有 15 年了。

Do you use ag grid or your own use? grid that seems to be like the leading one. I mean, they've done this for like I don't know 15 years.

是的。是的。他们知道怎么做。是的。

Yeah. Yeah. They know how to Yeah.

是的。所以，是的。我的意思是，我对表格角度很满意。我的意思是，我认为工程方面唯一其他有趣的事情是分发。

Yeah. So, yeah. I mean, I'm happy on that spreadsheet angle. I mean, I think the only other interesting thing for me engineering wise is the fan outs.

工程师也许能很好地理解权衡，就像，

Engineers maybe can are well equipped to understand the trade-offs of like,

好吧，在你把这个东西分发到一百万份文档之前，也许先在两个上测试一下。

Well, okay, before you fan this thing out on a million documents, maybe test it on like two.

是的。是的。是的。在你发疯之前，那行得通吗？是的。

Yeah. Yeah. Yeah. Does that work before you go nuts? Yeah.

他们这样做吗？

Do they do that?

是的，他们会的。所以我们的表格视图实际上会先运行前 10 行。

Yeah, they will. So our type of view actually will run the first 10 rows.

是的。只是检查一下，

Yeah. Just check,

对吧？是的。这有意义吗？它是否显示了我们认为它会显示的内容？

right? Yeah. Does this make sense? Did it show us what we thought it was going to show us?

那真的很好。哦，但还有对于人们，当人们构建这个时，对他们来说，快速迭代前 10 行直到他们得到他们想要的结果，然后他们可以全力以赴，这真的很好。

That's really good. Oh, but also for people when people build this, it's just really good for them to iterate quickly on the first 10 rows until like they have a they're getting the results that they want and then they can go ham.

是的。我很乐意继续下一个。Intake，这是你们的 Slack 机器人。嗯，每个人都有 Slack 机器人。呃，你们的有什么不同？

Yeah. I'm happy to move on to the next thing. Intake, which is your Slackbot. Um, everyone has a Slack bot. Uh, what's different about yours?

所以，我们的只是法律特定的。我们发生的情况是，它实际上不是——可能我们也会构建一个只是智能体接口的，但它实际上是专门用来启动一个智能体，然后向我们的列表添加任务。

So, ours is just legal specific. And what happens with ours is it's not actually it might also we'll probably also build one that's just an interface to the agent but it's actually specifically to basically kick off an agent that then adds a task to our list.

所以我们有这个 lists 功能，对你来说就像 Linear，你知道，那是最接近的等价物。所以它会把它添加到一个特定的 intake 列表，然后律师和智能体可以进入 Lore，就像这样：好吧，这些是人们发送给我们的东西，作为我们的律所，或者作为大公司里的法务团队。我们可以为它们启动智能体，我们可以协作处理它们，然后我们可以把它们标记为完成，然后它就会在 Slack 中发回。所以它更像是一个——不仅仅是启动一个智能体。它更像是一个店面，就像：嘿，这是工单，我想让这个东西被审查。好的，现在它回来了，不需要任何人改变任何东西。法务团队就坐在 Lore 里管理智能体和一切。

So we have this lists thing, which is like Linear for you, you know, that's the closest equivalent. And so it will add it to a specific intake list, and then lawyers and agents can go into Lore and just be like, okay, these are the things that people are sending to us as our law firm or as our legal team in a big corporate. We can kick off agents for them, we can collaborate on them, and then we can just mark them as done, and then it gets sent back in Slack. So it's more of a—it's not just like kicking off an agent. It's more of a storefront thing where it's like, hey, here's the ticket, I want this thing reviewed. Okay, now it's back, no one needs to change anything. The legal team just sits in Lore and manages the agents and everything there.

我觉得这类东西的数据结构真的很有趣，因为它基本上就是收件箱，对吧，对于所有这些查询，它们可能转化为一个东西，再转化为另一个东西，再转化为另一个东西。

I think that the data structure for this kind of thing is really interesting because it's basically the inbox, right, for all these kinds of queries that can potentially translate into a thing, into something else, into something else.

而且，我不知道，把所有这些与其他所有功能集成起来难吗，还是你根本不担心？

And like, I don't know, is it hard to integrate all these with all the other features, or do you just not worry about it?

不，不难。我认为我们有一些非常好用的原语。比如 lists 就是一个很好的原语，因为它本质上——它非常灵活。它就像一个 Notion 数据库。

No, it's not. I think we have some primitives that work really well. So like lists is a great primitive because it's essentially—it's like extremely flexible. It's like a Notion database.

是的，一个 Notion 数据库。

Yeah, a Notion database.

它就像一个 Notion 数据库。所以你可以基本上根据你的需要配置它。我们当然投入了很多来使它超级可扩展并且运行良好。你可以在它们上面设置触发器、自动化。它就像一个很棒的胶水，把所有东西粘在一起，因为这样你可以基于列表在发生某事时启动一个智能体运行，然后它回来并填充一个字段或其他什么，添加新文档或其他什么。所以就像你可能用 Linear 或某种任务管理工具协调一堆编码智能体一样。我们可以用 lists 类似地做到。所以智能体可以创建列表，列表可以创建智能体，智能体可以读取列表。所以就像智能体集群活起来了。

It's like a Notion database. And so you can basically configure it for whatever you want. And we of course invest a lot into making this super scalable and work really well. And you can have triggers on them, automations on them. And it's like a great glue to glue everything together because then you can kick off an agent run based on a list if something happens, and then that comes back and populates a field or whatever, adds new documents or whatever it is. And so it's like just like you might coordinate a bunch of coding agents with Linear or some sort of task management tool. We can do it similarly with lists. And so like agents can create lists and lists can create agents and agents can read lists. And so it's like the agent swarm comes alive.

是的。人们对此感到兴奋还是不知所措？你知道，我觉得这基本上就是终端用户编程。你可以让人们设置疯狂的自动化。

Yeah. Do people get excited or overwhelmed by that? You know, like I feel like this is basically end user programming. You can let people set up crazy automations.

有些人对此超级着迷，他们变得像影响者，你知道，有像玩影响者会很酷。嗯，这值得吗？

Some people get super obsessed by it and they become like influencers, you know, it'd be cool to have like play influencers. Um is that worth it?

太过于复杂了。

Too over complicated.

我们需要灵活性，即使我们实际上不向最终用户暴露它。假设这完全是内部的。那么这对我们来说是一种非常快速的方式，利用像我们的法律工程师这样的人来基本上极快地构建许多产品。所以你可以构建自定义的基金管理仪表板。你基本上可以在 lists 上用你自己的生成式 UI 构建它，然后你基本上就有了一个新产品，有些人会喜欢，我们可以有一个市场，里面有一堆专门的产品，我们可以超级快速地扩展，因为我们所有的法律工程师都可以直接构建新的。如果他们和客户在一起，哦你想要你自己的，我们就给你修一个。所以这有点像 Lore 内部的 Lovable。如果我们不想暴露它，那也是我们构建自己自定义功能的方式。

We need the flexibility even if we don't actually expose it to the end user. Let's say this was just all internal. Then it's a really fast way for us to utilize people like our legal engineers to basically build a lot of products extremely fast. So like you can build custom fund management dashboards. You can basically build that on lists with your own generative UI, and then you basically have a new product that some people are going to love, and we can have a marketplace of just a bunch of specialized products that we can scale super quickly because all of our legal engineers can just build new ones. If they're out with a client, oh you want your own, we'll just fix one for you. So it's like a lovable inside Lore a little bit. And if we don't want to expose it, that's also how we build our own custom features.

所以我们在英国又做了一次收购，叫 Wexler，他们做事实智能，主要用于诉讼，就像你为一场诉讼上传一百万份文件之类的，然后我们用命名实体和日期等从中提取事实，然后当你拥有所有事实时，你基本上可以把它们相互对照，看看哪些被证实，哪些是你所说的分歧。这非常强大，因为然后你也可以转录法庭，然后说，哦，他说了这个，但实际上这里的电子邮件——如果你现场看到，那太酷了——而且我们也可以在这个原语上构建。所以基本上我们可以在这一个原语上构建许多非常酷的东西。这就是我们的思考方式。就像我们必须——随着公司扩展，我们只需要解决很多事情，所以我经常思考如何构建一个核心原语，我们可以将其扩展到许多不同的最终产品，并创建非常可定制的体验。

So we made another acquisition in the UK called Wexler, and they're doing fact intelligence, which is for litigation mostly, which is like you upload a million documents for a lawsuit or whatever, and then we extract facts from those with like named entities and dates and whatever, and then when you have all the facts, you can basically hold them up against each other and see which ones are corroborated and which ones are what you call disagreement. And that's extremely powerful because then you can also transcribe a court and be like, oh, he said this thing, but actually the email over here—and it's if you see it live is like so cool—and that we can also build on this primitive. And so like we can build a lot of really cool things on this one primitive basically. That's how we think about it. Like we have to—as we scale the company, we just have to solve so many things, and so I think a lot about how do we build a core primitive that we can scale to many different end products and create very customizable experiences.

好的，那么你有——你有哪些核心原语？让我们为大家重新陈述一下。

Okay, so you have—what core primitives do you have? Let's just restate for people.

List 是一个核心的、可配置的。是的，项目基本上是文件的容器、文件系统。嗯，智能体当然你可以启动它并以非常不同的配置运行它。

List is a core one, configurable one. Yeah, projects basically containers of files, file systems. Um the agent of course you can kick that off and run that in very different configurations.

顺便问一下，你们给智能体起名字了吗？

Do you have a name for the agent, by the way?

没有，我们 intake。没有。

No we intake. No.

好的。

Okay.

是的。不，你的意思是我们叫它 layer 之类的，或者我们给它起了个名字。

Yeah. No, you mean like we called it like layer or like we gave it a name.

是的。是的。Layer 会很酷。是的。嗯，人们过去常把 layer 当作一个命名实体来谈论。

Yeah. Yeah. Layer would be cool. Yeah. Um people used to talk about layer as like a named entity.

嗯，但是是的，我们停止了那样做。

Um but yeah, we stopped that.

很可能把它带回来。

May well bring it back.

可能会令人困惑。我不知道。

It might be confusing. I don't know.

所以是 list 智能体。

So list agent.

嗯，表格视图，我们正在考虑那实际上是不是一个实例化，你可以把这个深入到你想要的任何程度，对吧？但那些可能是大的。是的。

Uh tabular view which like we're considering is that actually an instantiation of like you can take this as far down as you want, right? But those are probably the big ones. Yeah.

是的。所以我实际上想稍微谈谈像 FTE 这样的东西，因为这听起来有点像那个，而且我认为 Max 已经谈过一点，就像，好吧，当你引入一个新的大客户时，你想尽快让他们获得价值，这样你知道他们就会留在 Gora 上。

Yeah. So I actually wanted to talk a little bit about like FTEs because this sounds kind of like that and I think Max has talked about that a little bit that like okay when you onboard a new like big customer you want to get them value like ASAP so that you know they stay on stay on the Gora.

是的。那是如何演变的，比如你们什么时候开始那个前置部署团队的？

Yeah. How has that evolved like when did you guys start that forward deployed team?

所以我们雇了我们的第一个前置部署工程师，那是我们在纽约的第一个员工，也是 YC 创始人，大概九个月前，我其实不确定。

So we hired our first forward deployed engineer that was our first hire in New York also YC founder nine months ago maybe I'm not actually sure.

他们被认为是法律工程师吗，还是不？

Are they considered as legal engineers or no?

不，工程师，但我们做这个。我思考的一件事是如何在规模上实现这一点。嗯，我认为一个真正核心的事情是如何赋能我们的法律工程师，如何赋能我们的高级用户和法律工程师以强大的方式定制产品。然后当然我们仍然会有前置部署工程师来帮忙，你知道，做所有的自定义集成和范围界定，弄清楚什么是可能的，并把所有这些反馈到核心产品开发中。所以这就是我的思考方式。这几乎像第一线第二线防御，而法律工程师应该能够定制很多东西并构建出色的体验，然后在他们不能的地方，或者我们需要进行核心产品更改或自定义集成或其他什么。我们有四个部署工程师，他们也会在后面稍微轮换，有时他们只是做产品，有时他们与客户合作。

No engineers but we do it. It is something that I think about is like how do we exactly enable that at scale. Um and I think a really core thing is like how can we empower our legal engineers like how can we empower our power users and our legal engineers to customize the product in a powerful way. And then of course we we'll still have forward deploy engineers like help out you know do all the custom integrations and the scoping and figuring out what's possible and feed all that back into core product development. So that's kind of how I think about it. It's like almost a first line second line defense whereas like legal engineers should be able to customize a lot of stuff and build a great experience and then where they can't or we need to make core product changes or custom integrations or whatever. We have four deployed engineers who also then cycle in a little bit behind like sometimes they'll just work on the product and sometimes they'll work with clients.

你能走一遍一个例子吗？比如，好，你刚接到这个新客户。第一步是什么？你们是不是就说，嘿，我们给你派一位前驻工程师。他会去为你构建某种产品。这具体是怎么运作的？

Can you walk through an example? Like, okay, you just got this new client. What's step number one? Are you guys like, okay, hey, we have a forward deployed engineer for you guys. They're gonna go build some kind of product for you. Like, how does that work?

每个客户的情况差别很大，对吧？但如果是个大客户，而且他们非常投入，脑子里有一堆想做的事，或者通常他们已经尝试过构建某个东西但失败了，然后他们说，这是我们要解决的问题。那我们基本上会给他们配法律工程师，视情况可能再加一位前驻工程师。然后我们就说，嘿，这些人会过来，他们会让你成功。他们会和你的所有律师坐在一起，教他们怎么用，梳理出你现有的所有工作流，并把它们构建进 Lora。其中一部分只需要法律工程师，另一部分则需要前驻工程师。

So it's very different client to client, right? But if it's a big client and they're super engaged and they have all these dreams of things they would want to do, or typically they've already tried to build something and it's failed, and they're like, this is a problem that we'd like to solve. Then we'll basically staff them with legal engineers and a forward deployed engineer, potentially, depending on what the thing is. And then we'll just be like, hey, these guys will come, they'll make you successful. They will sit with all your lawyers and teach them how to use it and figure out all the workflows that you have right now and build them into Lora. And some of that will just be a legal engineer, and some of that will require a forward deployed engineer.

这个分工是什么？能不能稍微讲讲？

What is that separation? I guess if you could talk about that a little bit.

法律工程师像高级用户那样配置产品。

Legal engineers configure the product as a power user would.

他们不写代码。

They don't write code.

对，对。

Yeah. Yeah.

而前驻工程师基本上是把代码写进生产环境。我想这就是——这就是——

And a forward deployed engineer writes code into production, basically. I guess that's the—that's the—

前驻工程师通常是什么背景？你刚才说创始人。

What kind of background do the forward deployed engineers normally have? You said founder.

对，创始人、普通工程师，他们资历都比较老。

Yeah, founders, regular engineers, they're old.

对。D 是你的——

Yeah. D was your—

对，没错。D 是我们的第一位。对。做过两家 YC 公司，在 Meta 做过搜索。

Yeah, exactly. D was our first. Yeah. Two YC companies and worked on search at Meta.

对。她很厉害。对，对。

Yeah. She's awesome. Yeah. Yeah.

她很棒。对，对，对，对，对。是我们在美国的第一批落地人员。

She's great. Yeah. Yeah. Yeah. Yeah. Yeah. First boots on the ground in the US.

好。这个门槛很高。

Okay. That's a high bar.

那能举些例子吗？比如这些人为客户构建的工作流或东西？你们是不是有那种，哦，这几个用例是所有律所都会有的——你知道，这是我们的前驻工程师进去后要做的第一件事？

So what are some examples of, I don't know, workflows or things that these people are building for the clients? Do you guys have like, oh, these are a couple use cases that all the law firms are—you know, this is the first thing that our FDEs are going to do when they go in?

差别很大。有些核心工作流每个人都有，我们可以直接构建进 Lora，可能是，比如，我们是这样处理这类保密协议的，诸如此类，你激活它就行。对至少大多数前驻工程师来说，这更像是技能构建。那更像是——

It's very different. Some core workflows everyone has and we can just build in Lora, and that could be anything from, like, this is how we handle these types of NDAs or whatever, and it's just like you activate it. And it's more like skill building, really, for at least most of the FDEs. That's more like—

有些事我们还做不到，比如我们需要在某些基础组件上有更多灵活性，或者随便什么情况，或者我们有个非常复杂的系统需要集成进去。那他们就会出去，基于我们的 API 构建集成，也许还会扩展我们的 API 或别的什么，让这件事跑通。

There's something that we can't do yet, like we need some more flexibility in some of the primitives or whatever it might be, or we just have this really complex system that we need to integrate into. And then they'll go out there and they'll build the integration against our APIs and maybe expand our APIs or whatever to make that work.

那法律结构上是怎么运作的？因为你说过你们自己都不能处理交易。对。那他们怎么参与客户项目？

And how does the legal structure work there? Because you said you guys yourselves can't even work on deals. Yes. So like how do they work on client projects?

你是说律所，他们怎么参与客户项目？

You mean the law firms, how do they work on client projects?

你们的前驻工程师怎么参与客户项目，如果他们——

How do your FDEs work on client projects if they—

他们仍然不碰数据。所以他们只是构建系统。但有时候他们也会像员工一样被入职。比如他们可能会拿到一台笔记本电脑之类的——对。对。如果需要的话。

So they don't touch the data still. So they'll just build the system. But they will also be onboarded like an employee sometimes. So like they might get a laptop and everything to—yeah. Yeah. If they need to.

有意思。这不太常发生，因为通常他们只是构建一个集成，差不多就能用了，但他们——对，如果需要的话可以。

Interesting. That doesn't happen very often because usually they'll just build an integration and it kind of works, but they—yeah, they can if they need to.

我猜只有当客户达到一定规模时，你们才会把全职员工入职到客户那边。

And you only onboard FTEs into a customer, I'm assuming, if they're a certain size.

对。对。这得说得通，因为前驻工程师也可以做核心产品，对吧？所以这是个持续的优先级排序问题。

Yeah. Yeah. It needs to make sense, 'cause FDEs can also work on core product, right? So it's like it's a constant prioritization thing.

有没有什么规则，比如——我猜是不是每个——Palantir famously 有过那种，你知道，四天待在客户现场，一天回来给产品做贡献。有类似这样的吗？

Is there a rule like—I guess does every—so Palantir famously had like, you know, four days you spend on site, one day you come back and contribute back to products. Anything like that?

没有，我们没有。说实话，很多这方面都还处在早期。

No, we don't. I mean, to be fully honest with you guys, it's still early days in a lot of this.

你觉得它会往哪走？也许未来一年。

Where do you see it headed? Maybe over the next year.

我确实认为会更——也许不完全像 Palantir，但我认为我们会有一个更大的、持续轮转的前驻工程师团队，也许我们在一个客户那里待两个月左右来构建这个东西，然后他们去下一个，等等等等。对，但我确实设想我们会带着法律工程师和前驻工程师进驻，去做特定项目。

I do think that it's going to be more—maybe not exactly like Palantir, but I think we will have a larger in-constant-circulation group of FDEs that maybe we sit with a client for like two months or whatever to build out this one thing, and then they go to the next one, etc., etc. Yeah, but I do imagine us rolling in with legal engineers and forward deployed engineers to work on specific projects.

对，API 业务其实很有意思，就像你说的，因为你知道那是你可以有的另一个角色，其中——

Yeah, the API business is actually interesting, like you said, because you know that's another role that you can have where—

你是 Lora 平台的首席技术官兼维护者。

You are CTO maintainer of the Lora platform.

对。比如我们会把 Lora 网页应用里的所有东西都作为 API 提供，然后你们想用什么就用什么。

Right. Like we will offer everything that we have in the Lora web app as an API and then you guys can use whatever.

那是——那是你们做出的承诺吗，还是——

Is that—is that a promise that you make or—

那不是我们现在承诺的事。我们确实有可以基于其构建的 API，但不是产品的全部界面。

That's not something we're promising now. We do have APIs that you can build against, but not the entire surface of the product.

这就是有意思的地方。比如，那是策略吗？显然有些人会喜欢，因为我不用学新工具。我可以直接用我现有的工具，对吧？然后我就——

That's the curious thing. Like, is that the strategy? And obviously some people will love it because I don't have to learn a new tool. I can just use the existing tool I have, right? And then I just like—

比如把 Lora 装进 Word 文档那套东西里。

Install Lora, for example, with the Word document stuff.

除此之外，你知道，我确实认为那你就得维护它。

Other than that, you know, I do think like then you have to maintain it.

没错。这就是——这就是我一直的想法。这就是为什么我们尽量不做。我们一直没——我们一直没有 API。我是说，我们大概六个月前才开始做 API。我们抗拒了很久。主要因为，对，我就是不想——我们在产品上迭代太快了，我不想因为要为我们已经改过的数据模型维护旧版 API 而被拖慢。我是说，现在我们更大了，情况不同了。所以现在这是我们可以承担的成本。但当我们只有 20 个工程师时，我们迭代太快了，做不了这个。

Exactly. That's so—that's been my thing. That's why we've tried not to do it. We haven't—we haven't had APIs. I mean, we started our APIs like six months ago or something. We fought it for a very long time. Mostly because, yeah, I just don't want to—we move so quickly in the product that I don't want to be slowed down by having to maintain old API versions for a data model that we've changed. I mean, now we're bigger and things are different. So now it's a cost that we can take on. But like when we're 20 engineers, it's just like we move so quickly that we can't do this.

我琢磨过这个，对吧？比如一般来说，所有公司都有平台团队和产品团队是好事，因为产品团队你有一个，但只要平台和产品不是绑得太死，你很容易再开第二个。

I've toyed with this, right? Like it's just generally good for all companies to have a platform team and a product team, because the product team you have one, but then you can easily spin up a second one as long as the platform is not super tight to product.

对。

Yeah.

你知道，很简单的一个例子就是网页应用、移动应用，对吧？或者网页应用、移动应用、API，然后随便什么其他集成。那你的前驻工程师反正也只是整个平台的又一个客户而已。

You know, like a very simple one is web app, mobile app, right? Or like web app, mobile app, API, and then like whatever other integrations. Then your FDEs just are one more client to the overall platform anyway.

对。对。而且我觉得连智能体也可以这么看。有一个类似运行框架的团队，或者说运行智能体的东西，然后你甚至可以围绕它构建产品，加上技能、自动化之类的。

Yes. Yeah. And I think like even the agent you can think about it this way. There's like a harness team or like a—that runs the agent, and then you can even build a product around it with like skills and automations and whatever else.

对，这就是非常著名的亚马逊云服务（AWS）指令。贝索斯说，不能再有后门访问。你只能向其他团队提供你的 API，也就是你向客户提供的同一套 API。如果没有，你就自己建。

Yeah, this is very famously the Amazon Web Services mandate. Bezos was like, no more backdoor access. You only offer other teams your APIs, the same APIs you offer customers. If you don't have it, you build it.

对。

Yeah.

然后客户也能拿到。

And then the customers also get it.

对。我觉得这一点也在发生一些变化。现在很多 API——我们正在琢磨的一件事是：智能体能不能就是 API？我知道这听起来很奇怪，但对某些东西来说，他们真正想要的是触发某些事情发生。对其中一部分来说，你可以直接——我不是说我们在这么做，但这是我脑子里在琢磨的事——能不能就是，好，你触发智能体，让它去做这件事？显然，这比有一个你总是调用的 API 效率低得多。但我在琢磨的是：我们的产品里有多少可以直接就是智能体，而不是其他那些东西。

Yeah. I think that's also changing a little bit. A lot of APIs now — one thing that we're toying with is, can the agent be the API? I know it sounds weird, but for some of this stuff, what they actually want to do is they want to trigger certain things to happen. For some part of this, you could just be — I'm not saying that we're doing this, but it's a thing that I'm toying with in my head — could that just be like, okay, you trigger the agent and ask it to do this thing? Obviously it's way less efficient than if you just had an API that you always hit. But it's something that I'm toying with: how much of our product can just be the agent versus all the other stuff.

嗯，你知道，无限制的执行、成本和延迟。

Well, you know, unbounded execution and cost and latency.

这就是——这是唯一的阻碍吗？

That's the — is that the only holdback?

这是——对。然后我觉得还有确定性。也许 100 次里有 2 次，它会做一些我们不希望它做的离谱事情。

That's — yeah. And then I think determinism. Like maybe two out of 100 times it's going to do something whack that we didn't want it to do.

我们待会儿会聊到 harness 和模型微调。但我觉得有一个关于克制——也就是你不做什么——的产品问题，因为那些你都可以做。显然还有优先级问题。比如，你现在更倾向于对大多数事情说不，只做少数几件事？还是说，不，其实我们想做更多事，只是需要再招 100 个人？

We're going to go into harness and model fine-tuning in a bit. But I think there's a product question about restraint — what you don't do — because you could do all that. And obviously there's a prioritization issue. Like, are you more at a point of, let's say no to most things and only do a short number of stuff? Or no, actually we want to do more things and we just need to hire like 100 people?

很长一段时间里，我非常强烈地倾向于第一种。就对一切说不。保持工程团队——我想我以前说过——我们有一次演示，我在 Liora 待了半年后，翻出电影《300 勇士》的那张幻灯片，我说，公司剩下的历史里我们就只做 20 个工程师，我们要打败所有比我们大得多的公司。现在我们已经有 100 个工程师了，感觉还是太小。我本想那样做，但我觉得在现实世界里，我们面临激烈的竞争。人们想要太多东西。我可以做个清教徒，就说不行，我们必须保持 20 个人，我们只能做某某东西。但那样我们就输了，人们会去别处。所以我们必须做更多。所以我觉得我的工作是弄清楚，如何建立一个系统，让我们能做很多，但实际上只——让客户觉得我们做了很多东西，而我们实际上只做了少数几件事。这就是我现在对自己工作的理解。

I was very strongly the first thing for a very long time. Let's just say no to everything. Keep the eng team — I think I've said this before — one of our presentations, after I'd been at Liora for half a year, I pulled up this slide of the 300 Spartans from the movie and I was like, we're just going to be 20 engineers for the rest of the company's history and we're going to beat everyone who's much bigger. And we're 100 engineers now and it feels like we're way too small. I wanted to do that, but I think in the real world we're in fierce competition. People want so much stuff. I could be a Puritan and just be like, no, we have to stay 20 people and we can only build whatever. But then we just lose and people would go elsewhere. So we have to build more. And so I think my job is to figure out how do I build a system where we can build a lot but actually only — where the customer feels like we've built so much stuff but we've actually only built a few things. That's how I think about my job right now.

对。我是说，原语（primitives）是对的，你知道。

Yeah. I mean, primitives is right, you know.

对。没错。我觉得那会非常强大。

Yeah. Exactly. I think that's going to be really powerful.

你们会测试新产品，看看这里有没有需求，然后你说，不，我们还是专注主业？还是——

Do you test new products, just seeing if there's any pull here, and then you're like, no, let's just focus on our main thing? Or —

我们还没有——我是说，我们做过一些战略押注，但我们没试过任何我们本来就不太有信心的事。主要是因为我们有堆积如山的需求——比如，这些东西我们都想要——所以去试“嘿，也许这个东西能行”对我们来说并不太合理，因为有那么多人在为其他东西呼喊。所以我们主要做那些。然后有时我们会不同意客户，我们会说，其实我觉得这边这个东西在战略上会很有意义，然后我们还是会试着去做它。

We haven't made — I mean, we've made certain strategic bets, but we've not tried anything that we don't have a lot of confidence in already. Mostly because we just have a mountain of demand — like, we want all of these things — and so it hasn't really made sense for us to try to be like, hey, maybe this thing could work, because there's so many people screaming for the other things. So we mostly build those. And then sometimes we'll disagree with clients and we'll be like, actually I think this thing over here will make a lot of sense strategically, and then we'll try to build that anyway.

比如对你们来说有战略意义。

Like strategically for you.

对，没错。

Yeah, exactly.

那那些东西是什么？

So what are those things?

其中一件事其实就是智能体。其实没人要求——比如“我想让我用的助手有更多自主性，不用我问就做事”。从来没人提过。但我们就是觉得，这绝对是行业的走向。这会让你能做很多很酷的事。协作是另一个，我们认为行业正走向一个地方，让你可以在你工作的地方——也就是 Liora——里面协作，而不是在邮件和短信里。所以对，有那么几件。

So one of those things was the agent actually. No one was actually asking for — like, I want the assistant that I'm using to have more agency and do things without me asking it. No one ever. But we were just like, this is definitely where the industry is going. This is going to empower you to do so much cool stuff. Collaboration was another one, where we think the industry is going to a place where you can collaborate inside of where you're working, which is Liora, and not in emails and text messages. So yeah, there are a few of those.

然后怎么——一旦你说，好，我们希望你们用智能体，我猜在律所内部，他们要真正学会怎么做需要花很多功夫。所以这就是你说，嘿，FDE（前向部署工程师），进去教他们——

And then how — once you say, okay, we want you guys to use agents, I'm assuming internally at the law firm it's going to take a lot for them to actually learn how to do that. And so that's where you kind of say, hey, FDEs, go in there and teach them what they —

对。主要是法律工程师，但有人愿意的话叫他们前向部署法律工程师，但本质上就是那样。然后他们进去，帮助他们围绕智能体的能力来组织工作流。

Yeah. Legal engineers mostly, but some people call them forward-deployed legal engineers if you want, but that's essentially what they are. Then they go in and help them structure their workflows around the capabilities of the agent.

你们之前说过——我想是 Max 说的——早期具体来说，你们没有微调自己的模型。现在还是这样吗？

You guys said previously — I think Max said this — that early on, specifically, you guys weren't fine-tuning your own models. Is that still the case?

对，所以到目前为止我们没微调过任何模型。我觉得现在，过去几个月里，开放权重模型变得极其强大，所以对于某些任务，如果我们认为前沿模型在成本和延迟上能做到的，与我们认为后训练模型能做到的之间存在巨大差距，我们就会去做。法律研究就是个很好的例子。这是一个扇出检索问题，你可以让一个非常有效的模型以极低的延迟、极快极便宜地完成那些查询。

Yeah, so we haven't fine-tuned any models up until now. I think now, in the past few months, open weights models have become extremely capable, and so we will do it for certain tasks where we think there's a huge delta between what a frontier model can do at the cost and latency versus what we think a post-trained model can do. Legal research is a great example. It's a fan-out retrieval problem, and you can make a really effective model do those queries super quickly and super cheaply at low latency.

所以具体来说，你们是在后训练它们，让它们在非常具体的受限任务上表现更好。

So specifically, you're post-training them to be better at very specific bonded tasks.

对，没错。对。我们不是想训练自己的“寓言”模型来当主智能体。我不觉得那是——我们赢不了那个，你知道，对上 Anthropic。很可能还有很多，比如说，搜索和检索研究或 RAG 研究要做。

Yeah, exactly. Yeah. We're not trying to like train our own fable to like be the main agent. I don't think that's a — we won't win that, you know, versus Anthropic. Most likely there's still a lot of, let's say, search and retrieval research or RAG research to do.

那可能没那么光鲜。它甚至不一定是模型训练。有时候只是架构和流水线。你们和 Turbopuffer 做过一个案例研究，但你知道，还有——你们做后期交互（late interaction）吗？是的，你们绝对做那个？

That is like maybe less glamorous. It's not actually even model training. Sometimes it's just like architecture and pipelines. You did a case study with Turbopuffer, but like, you know, there's also like — do you do late interaction? Yes, absolutely you do that?

对。其实我们不做那个，但你说得完全对，这是——团队里有一个人，Son，他特别在意搜索，他基本上老来找我，说我们应该做这些事，还有这些新的嵌入模型，我们可以做这个后期交互。

Yeah. We actually don't do that, but you're completely right that this is — so there's one guy on the team, Son, who just really cares about search, and he basically comes to me all the time and he's like, we should do these things, and like there's these new embedding models, and like we can do this late interaction.

他实际上与 Turbo 团队沟通，让他们构建 multiffactor，以便我们能够做这类事情。但不，通常有很多低垂的果实，甚至在模型之前，甚至在 harness 中，老实说，如果你只是想为我们的客户创造更好的输出，有很多事情我们可以做，在训练模型之前就能产生即时影响。

He actually talked to Turbo teams to get them to build multiffactor so that we could do these types of things. But no, there's loads of low hanging fruit generally even before models and even in the harness honestly like if you just want to create a better output for our clients, there's like so much stuff that we can do that has immediate impact before training models.

比如，什么造就了法律行业的好 harness？我猜你们在打造，我假设是你们自己的定制 harness。

Like what makes a good harness for like the law industry? I guess like you guys are making I'm assuming like your own custom harness.

是的，当然。我们编辑文件的方式很重要。我们进行分析的方式很重要。呃，你进行法律研究的方式都融入其中。所以就像你可能遇到一个问题或看到某件事，然后它知道如何进行法律研究来找出答案是什么，先例是什么，这都非常重要。当然，还有所有围绕它的治理内容。

Yeah, of course. The way that we edit files is important. The way that we do an analysis is important. uh the way you do legal research is all baked into it. So it's like you might have a problem that comes up or see something then it knows how to go and do the legal research to figure out what the answer is, what the president is, that's all very important. And then of course there's all the governance stuff around it as well.

是的，看起来搜索可能是，你会说这是你们真正在定制的最重要工具吗？因为如果你不能获取正确的信息

Yeah, it seems like the search is probably like would you say the most important tool that you guys are really customizing cuz if you can't pull in the right information

是的。然后你就完蛋了。这是呃，我认为这是这是这是嗯

Yeah. and then you're screwed. It's uh I think it's it's it's um

你们在搜索大量文件

you guys are searching like a lot of documents

和分析。是的，在搜索中，是分析，也可能是起草，因为像起草文字，起草大合同实际上也不是超级简单

and analysis. Yeah, in search it's analysis and it's drafting probably cuz like drafting word drafting huge contract is actually also not super trivial

比如，哪部分不简单

like what about it is not

你必须，比如说你从头起草一份大合同，那么你必须找到先例，比如我们之前是怎么做的，我们是否需要这类条款

you have to let's say you're drafting a huge contract from scratch then you have to find precedent of like how do we do this previously do we need this type of clause or not

特定律所之前是怎么做的

specific law firm how did they do it previously

是的，所以我是律所，我即将起草一份大合同，随便什么，NDA 很简单，但有些合同要复杂得多，然后你需要先找到正确的先例，因为律师不会只是坐下来开始打字。他们会弄清楚我们在类似交易中使用了哪些条款，而类似可以意味着很多不同的事情，我相信你知道，在搜索中，可能是同一行业、同一类型的公司、同样的什么。嗯，所以你需要识别正确的先例，你可能有多份文件，然后你试图弄清楚应该从哪份文件中取哪部分，这实际上并不简单。所以本质上也是一个搜索问题。

yeah so I I am the law firm and I'm about to draft a huge whatever NDA is very simple but like there's there are contracts that are way more complex and then you need to find the right president first cuz like lawyers just Don't don't like sit and like start typing stuff. They figure out what clauses have we used in similar deals and similar can mean a lot of different things as I'm sure you know in search it could be like same industry, same type of companies, same whatever. Um and so you need to identify the right person and you might have like multiple documents that you then try to figure out which part should I take from what document and that's actually non-trivial. So it's essentially also a search problem.

哪个模型最擅长起草？

Which model's best at drafting?

所以，我会说，我昨天在比较这个。Fable 和绝对非常健谈。我相信你们已经知道了。呃，它们非常冗长。

So, I'll say I told I was faring this yesterday. Definitely fable and are extremely chatty. I'm sure you guys have known this. Uh, they're very verbose.

所以，那不好。那通常不好。

So, that's not good. That's generally not good.

我注意到 Astra 很擅长写作。

I've noticed Astra is pretty good at writing.

Astra，是的，很擅长写作，更简洁。

Astra, yeah, pretty good at writing much more concise.

嗯，你们是在运行 100 个不同的问题来像

Um, are you running like 100 different problems to like

是的，当然。我们可以，是的，我们可以，我们可以做，我们可以做某些事情，但我认为健谈冗长实际上对我们来说有点困难。所以，这是我认为

Yeah, sure. We can Yeah, we can we can do we can do certain things, but I think the chattiness robosity is actually a little bit difficult for us to do. So, this is I think

你知道，让我们看看。但可能 Anthrop 的下一个模型会在这方面好得多。Astro 实际上非常简洁。

you know, let's see. But probably the next model from Anthrop is going to be much better at this. Astro is actually very concise.

是的。而且我假设他们都和你有一个 Slack 频道，你在告诉他们这个，对吧？就像这对他们来说不是新闻。

Yeah. And I I assume they they all have a Slack channel with you and you're telling them this, right? Like this is not news to them.

不，不，不。他们知道这个。然后我真正关心的另一件事是幻觉。

No, no, no. They know this. Then the other thing that I really care about is hallucination.

他们会编造东西吗？

Do they make stuff up?

所以，呃，我们有一个引用系统，我们基本上检查他们写的内容是否有链接。所以，用户有一个链接可以点击，它会用边界框高亮，比如在 PDF 或其他地方，显示信息来自哪里。我们可以确保它所说的内容总是可以在某处引用，甚至多层向下。所以就像如果智能体基于表格审查写了什么，你知道，它可以引用表格视图中的单元格或表格视图中的行。所以你可以实际点击多层向下，找到，你知道，某种程度上一直验证上去。

So, uh, we have a system of a citation system where we basically check what they're writing has a link. So, the user has a link they can click and it highlights with a bounding box like in the PDF or whatever where where this information was taken from. And we can sort of make sure that what it says always is citable somewhere even multiple layers down. So like if the agent writes something based on a tabular review, you know, it can site the cell in the tabular view or the row in the tabular view. So you can actually like click multiple layers down and find, you know, sort of validate it all the way up.

是的。

Yeah.

你发现律师们会验证这些引用是否正确吗？

Are you finding that lawyers are like verifying that these citations are correct or not?

最初有效，但现在就像幻觉链接到最高法院。他们不在乎。

Initially work, but now exactly like hallucinated links to Supreme Court. They don't care.

那会发生。是的，没错。那会发生。是的。我认为这就像编程，你知道？你有工程师就是不在乎，只是发送智能体，你有人喜欢坐下来读每一个逗号。一样。

That happens. Yeah, exactly. That happens. Yeah. I I think it's like coding, you know? You have engineers that just don't give a and just like send agents and you have people that like sit and read every comma. Same.

有那么多，一个人怎么能做到，对吧？

There's just like so much that like how can one human do that, right?

是的。但我认为他们可能就像我们在代码审查中知道要寻找什么。系统是什么或 whatever。我认为他们也知道，好吧，这是事情。这些是红色，就像这是重要的事情。让我只是再检查这些事情。而且这些听起来合理，所以可能没问题。

Yeah. But I think they probably just like we know in code review what to look for. What is the system or whatever. I think they also know, okay, this is the the thing. These are the red like this is the thing that's important. Let me just like double check these things. And like these sound reasonable, so it's probably fine.

好的。有趣。

Okay. Interesting.

就像如果你从初级律师那里收到东西，你做同样的事情，对吧？就像我应该再检查这个，而且这 probably 没问题。

Just like if you if you receive something from a junior law, it's the same thing you do, right? like I should double check this and like this probably is fine.

还有其他元素，我们称之为法律推理，或者什么是法律前沿，呃，模型还不够好，你希望从下一代模型中得到什么？

Any other elements of let's call it legal reasoning or what's the the legal frontier that uh models are not good enough yet that you want from the next generation of models?

我现在具体想不出什么。我认为长文档推理仍然困难。嗯

Nothing I can think of specifically right now. I think long document reasoning is still difficult. Um

长像百万 token 上下文。

long like million token token context.

没错。你有一份巨大的合同，嗯，就像你在整个合同中做编辑。这对今天的模型来说也是一件困难的事情，就像记住你在这里改变了所有这些事情，定义变了，然后顺序变了等等。嗯，所以最简单的是一次性，你把所有东西喂进去，然后一次性输出。你不那样做。你把它分成块，然后协调编辑。

Exactly. You have a huge contract um and like you're making edits across it. It's a difficult thing to do for a model today even and like remember that you changed all these things up here and like the definitions changed and then you the ordering changed and whatever else. Um so the simplest would be one shot you put feed everything in and then one shot output. You don't do that. You break it up into chunks and then you orchestrate edits.

是的。没错。没错。这就像那些有趣的事情之一，参与这个行业很有趣。对吧。所以我们的说法是，我们已经做了，你知道，我们做了 map reduce，我们改变了，有一个大的，有像一个大的一次多次。就像我们不断改变我们围绕模型构建的脚手架，随着模型变化。

Yeah. Exactly. Exactly. And this is like one of those things that are funny like that are fun to be part of in this industry. Right. So our word at was like we we've done you know we've done map reduce and we've changed and there was one big one and there was like one big one in multiple times. It's like we keep changing the the scaffolding that we build around as models change.

是的。我我我会说，我我嗯，我最近，我做很多新闻报导，显然因为我的工作，我最近从 map reduce 转向把所有东西扔进上下文

Yes. I I I would say I I um I've recent I do a lot of like news reporting obviously because of what I do and I've recently shifted from map reduce to throw everything into context

而且更好。

and it's better.

是的，没错。但它也不同。我的意思是，我们在搜索中遇到这个问题，就像你在巨大的合同中搜索某些东西，你可以 map reduce 它，但然后它可能，如果你在总结任何东西，那么它可能会丢失块之间的一些重要细微差别，本质上如果你只是总结各个部分。

Yeah, exactly. But it's also it's different. I mean, we had this problem with search where it's like you're searching for certain things in huge contracts and you can map reduce it, but then it might if you're summarizing any sort of thing, then it might lose some important nuance in between the chunks essentially if you just summarize the individual parts up.

所以这一直是个问题：我们可以做两种检索。我们也能做另一种，就是需要在中间的不同步骤里进行推理。我正想找一个特别好的例子。有些文档会花很长篇幅讲一件事，比如讲两页之类的，然后每一个单独的片段未必和搜索词或他们想搞清楚的东西相关，但合在一起就极其相关。所以是的，这些就是一些问题，其中——

So this has been a problem where we can do two types of search. We can also do the other one where you need to have reasoning in the different steps in between. I'm trying to think of a really good example here. There are some documents where they talk about something at length over, let's say, two pages or whatever, and then each individual chunk is not necessarily relevant to the search term or what they're trying to figure out, but together it's extremely relevant. So yeah, these are some of the problems where it's—

而且这一定很难，就像你之前说的，要真正让模型在这方面变得更好，因为你甚至看不到那些合同，对吧？真实的合同。所以你们有没有问过他们，比如“嘿，你们能不能做一些和你们真实合同相似的示例合同”？

And it must be hard, like you said earlier, to actually make the models better at that because you can't even look at the contracts, right? Real contracts. So do you guys ever ask them for, like, hey, can you make example contracts that are similar to your real ones?

我们有反馈渠道，可以收到他们的反馈，如果他们愿意配合——没错——或者他们就说“嘿，这对我们特别重要，但现在效果不好”。然后我们就说“好，太好了，你们能不能把文档和提示词之类的做匿名化处理”，如果我们能复现，就能修好。

So we have feedback and we can receive feedback from them, and if they're willing to engage—exactly—or they're just like, hey, this is super important to us and it's not working well. And then we're like, okay great, can you anonymize the documents and the prompt and whatever, and if we can recreate it, then we can fix it.

是啊。我是说，这件事的一个版本就是你去买一家律所。

Yeah. I mean, so one version of this is you buy a law firm.

对。

Yeah.

这是有人在考虑的。

Which people are considering.

是的。但这里有些问题：他们的数据到底算他们的数据，还是客户的数据。所以如果客户都没了，那也许就没问题。我其实也不确定。

Yes. But there's some stuff there on whether their data is actually their data or if it's the client's data. So if the clients are all gone, then maybe it's fine. I actually don't know.

有意思。

Interesting.

对。但这是——是的。这是获取大量数据的一种方式。对。

Yeah. But it's—yes. It's a way to get a lot of data. Yeah.

是啊。就像 Google 收购 Spirit 那样。

Yeah. People like Google's buying Spirit.

没错。对。

Exactly. Right.

对。他们买了那个——对。大概花了一百万还是多少。对。

Yeah. They bought the—yeah. For like a million or whatever. Yeah.

你知道，如果你有一家失败的公司，你还有你的 Slack。你可以把 Slack 卖掉。我觉得有一家 YC 公司就是收购你失败的公司。

Like, you know, if you have a failed company, you still have your Slack. You can sell the Slack. I think there's a YC company that buys your failed company.

是的，有。对。对。

Yes. There is. Yeah. Yeah.

对。因为就像“哦，我就是想要真人在干活”。

Yeah. 'Cause it's just like, oh, I just want real people doing work.

没错。然后慢慢把东西抽出来。

Exactly. And then just slowly pull stuff out.

没错。买律所数据的问题在于，你能拿到所有输入，真实的输入，这很好。输出就稍微难一些，因为你只拿到最终结果，很难搞清楚——你基本上得逆向工程他们是怎么走到那一步的。所以在我们的产品里，当我们从真实用户那里拿到真实的轨迹时，价值极高。但如果我们只拿到这个输出，不知道它是好是坏、是怎么产生的，那就有点难办。

Exactly. The problem with buying law firm data is, well, you get all the inputs, the real inputs, which is great. The outputs are a little bit more difficult because you just get what the end result was, and it's difficult to figure out—you have to basically reverse engineer how they got there. So in our product, when we get real traces from real users, it's extremely high value. But if we just got this output and we don't know if it's good or bad or how it was created, it's a little bit difficult.

我们很快就要收尾了。我最后想给大家指出的另一点是，嵌入这块还有很多 alpha 空间。如果你的工作负载很重 RAG，你只看 MTEB 排行榜、直接拿排名最高的模型，那你其实浪费了很多。你应该看看 Voyage 或者一些自定义嵌入，你——

We're going to wrap up soon. The last thing I'll point people to is also there's a lot of alpha left on embeddings. If you're a RAG-heavy workload, you're just looking like MTEB and you're taking like the top model, you're leaving a lot on the table. You should look at like Voyage or some of the custom embeddings you—

对。

Yeah.

对。所以基本上别用默认嵌入，哪怕是开源嵌入、Gemini 嵌入，都无所谓。要定制你的嵌入。

Yeah. So basically don't take default embeddings, even open embeddings, Gemini embeddings, doesn't matter. Customize your embeddings.

是的。你们会支持这个。

Yes. You will support that.

对。我觉得说得对。我觉得说得对。唯一一点是你得确保支持多语言。对我们来说，我们服务那么多语言。所以那是——

Yes. I think that's right. I think that's right. The only thing is you need to make sure that you're multi-language. So for us, we serve so many languages. So that's—

欧洲是这方面的王者。

Europe is the king of this.

对，没错。

Yeah, exactly.

Mistral 在这方面很有帮助。

Mistral helpful there.

不，直到——嗯，直到现在吧。他们刚融了一大笔钱，对吧？所以也许——

No, not until—well, until now. They just raised a ton of money, right? So maybe—

对，让我们——你们。

Yeah, let's—you guys.

他们也融了一大笔钱，要当你们的捍卫者。

They also raised a ton of money to be your champions.

我觉得我们需要更多来自欧洲的模型公司。是的。

I think we need way more model companies out of Europe. Yes.

我就是想问问你对葡萄酒行业未来一年有什么预测？我们之前稍微问过一点。我把它改成你的第二个问题：两年后，律师会在 Legora 里完成工作吗？他们会用 Claude 或 Traceora 之类的其他工具，比如 API 吗？

I just wanted to ask you what predictions you have, maybe over the next year, for the wine industry? We asked a little bit of that earlier. I'll modify it to your second question, which is: in two years, will lawyers do their work inside Legora? Will they use other tools like Claude or Traceora, like the API?

我觉得两者都会有。这真的取决于情况。我觉得如果你是大企业，你可能会用 Claude 协作，然后你会——我觉得这类似于我通过 Claude 用 Linear，但我还是会去 Linear。我觉得企业会那样做。我觉得律所会直接用 Legora，因为那更简单也更好。

I think both. It really depends. I think if you're a big enterprise, you might have Claude co-work and then you'll—I think it's similar to how I use Linear through Claude but I still go to Linear. I think enterprise will do that. I think law firms will just use Legora because that's simpler and better.

这让我想起医学领域的 OpenEvidence。你基本上就是法律领域的那个。

It kind of reminds me of OpenEvidence for medicine. You're basically that for law.

没错。

Exactly.

在编程领域，事情已经到了实验室会因为你和他们竞争而切断你的地步，这已经发生过两次了。你觉得法律领域会发生这种事吗？

In coding, it's gotten to a point where the labs will cut you off for competing with them, and this has happened twice. Do you think that will happen to law?

不，我不这么认为。我觉得因为，你知道，他们也在推出法律服务，对吧？他们在——

No, I don't think so. I think because, you know, they are releasing legal services, right? They are—

我不——我是说，我们显然经常和他们聊，但也会问，嘿，你们会不会——

I don't—I mean, we obviously talk to them a lot, but it's also like, hey, are you going to—

你们的销售代表会说不会。他们不做决定。没错。那不是他们的职权。而且他们往往甚至不知道实际发生了什么。我不认为会那样。他们在知识工作，或者说企业知识工作里，有太多东西要覆盖，法律必然是其中一部分，就像数据分析必然是，销售、营销和财务也是。所以他们必须把那些做好，但他们永远不会像我们做得那么深。我觉得如果 Claude 拿下小客户、小企业客户，他们通过 Claude 处理一些法律事务，那很好，没问题。但如果你在做一笔 4 亿美元的并购交易，你会想要 Legora。

Your sales reps will say no. They don't make the call. Exactly. That's not their right. And they don't even know often what's actually going on. I don't think that's the case. They have so much stuff to capture in knowledge work or in enterprise knowledge work generally that legal has to be part of that, just like data analytics has to, and sales and marketing and finance. So they have to do things that are good there, but they're never going to go as deep as we are. And I think if Claude takes small client, small corporate clients, and they do some of their legal via Claude, like that's great. That's fine. But if you're doing a $400 million M&A transaction, you're going to want Legora.

我有个不太相关的问题，但我很好奇。你们会去法学院，让他们在课堂上教 Legora 吗？这具体是怎么运作的？

I had kind of an unrelated question, but I'm curious. Are you guys going to law schools and getting them to teach Legora in classes? Like how does that work?

这肯定是我们已经启动的一个项目。我其实不太清楚我们具体进展到哪了，但我们确实希望法学院的人学会用 Legora。

It's definitely a program that we have started. I actually don't know exactly where we are, but we do want people in law school to learn to use Legora.

比如，你看到法学院在大幅调整课程设置吗，还是并没有？

Like, are you seeing that law schools are changing their curriculums a lot or not really?

我不知道。我其实不太清楚。

I don't know. I wouldn't know actually.

有意思。

Interesting.

我其实会换个方式做。你们应该为计算机专业、为工程师设立法学院奖学金。我们会覆盖你的——在你读完计算机之后，我们把你送进法学院。计算机加上一些软件经验，我们把你送进法学院。

I would actually do it a different way. You should do scholarships for law school for CS majors, for engineers. We will cover your—after you've done CS, we'll put you in law school. CS plus some software experience, we'll put you in law school.

我们会覆盖。就是那种硕士加 JD 的项目。

We'll cover it. Where you do like the MA or like master's plus JD programs.

对。你不在乎那个硕士。

Yeah. You don't care about the master's.

嗯，意思是说，因为很多人本科没学计算机科学，所以就读个硕士，只是为了补上基础。

Well, meaning like because there's a lot of people who didn't do CS undergrad, so then they do like masters just to kind of get caught up.

对，对。我不知道，我不太常见到这种情况。

Yeah. Yeah. I don't know. I don't see that as often.

对，对。

Yeah. Yeah.

有人读法律博士加工商管理硕士，这是有的。

People do JD-MBA. That's a thing.

但法律博士加计算机科学，我没见过。对。

But JD-CS, I haven't seen it. Yeah.

也有。对。

Happens. Yeah.

对，对。好，总的来说，你知道，进展非常惊人。我是说，恭喜你取得的所有成功。你还会保持这个节奏吗？对。而且我记得你们在旧金山、纽约招人。

Yeah. Yeah. Okay. Look, overall, you know, super impressive progress. I mean, congrats on all your success. Like, are you still going to keep up the pace? Like, yeah. You know, and I think you're hiring in SF, New York.

我们在招工程岗，在纽约、斯德哥尔摩和伦敦招人。

We're hiring in engineering. We're hiring in New York, Stockholm, and London.

对。

Yeah.

其他所有岗位我们也在各地招人。

And all other roles we're hiring everywhere.

你们现在纽约有多少人？

How many people do you have in New York right now?

100 人。

100.

好。那你们计划在纽约招到多少人？

Okay. And you guys are planning on how many in New York?

在纽约？

In New York?

对，就大概整体来说，

Yeah. For just like generally,

就这么说吧，对，整体上

let's just say, yeah, generally in

嗯，很多。能招多少招多少。

um, a lot. As many as you can.

对，而且这个目标很激进。

Yeah. And it's aggressive.

对。我是说，就美国而言，整个公司法就是这样，对吧。所以，对。我是说，我很期待你们收购一家律所，然后，你知道，成为——

Yeah. I mean, well, that's the whole of corporate law, right, as far as America's concerned. So, like, yeah. I mean, I'm looking forward to you buying a law firm and, like, you know, becoming—

对，这非常令人兴奋。我觉得，你知道，每个行业，尤其是顶尖的——医生、律师、程序员——你知道，他们都得有，你知道，两三家智能体实验室完全主导这个领域。所以你们有一个很棒的市场。

Yeah. It's super exciting. I think, like, you know, every industry, especially like the top—like doctors, lawyers, coders—you know, they all have to have, like, you know, two or three agent labs that just completely dominate the space. So you guys have, like, a great market.

对，对，我也这么认为。而且我觉得随着法律变化、对更多人变得更可及，这个市场只会大得多。

Yeah. Yeah, I think so. And I think it's only going to get much bigger as, like, law changes and becomes more available to more people.

对。另外我觉得，我是说，我们其实已经超级——但你做的这些问题非常处于知识工作的前沿，你知道吗？

Yeah. Also, I think, I mean, we're already like super over—but like, the problems that you work on are very like the frontier of knowledge work, you know?

对。

Yeah.

就是这样。好。除非你真的在做科学，比如纳维-斯托克斯方程之类的科学，除此之外，大多数工作基本上都收敛到法律。

There. Okay. Like unless you're literally like doing like science, Navier-Stokes and science, other than that, like most work is like basically converges to legal.

对。就像，对。嗯对，最后都是法律。对。招聘、金融，一切。就像我只需要咨询我的律师。所以对，我知道，我知道。这就是为什么我觉得做这件事非常令人兴奋，因为现代社会在某种程度上是靠法律运转的。

Yeah. Like, yeah. Uh yeah, it's all legal in the end. Yeah. Hiring, finance, everything. Like I just have to consult my lawyer. So yeah, I know. I know. Which is why I think it's a really exciting thing to work on, cuz like modern society runs in some way on legal.

对，对。

Yeah. Yeah.

好。行，就到这里。谢谢。谢谢。谢谢。

Cool. All right, that's it. Thank you. Thank you. Thanks.