Anthropic 实验室负责人探讨 Fable 的能力与构建 AI 原生产品——专访迈克·克里格

Anthropic 不只是在做大型模型，它在产品开发上也同样强大。

Anthropic is not just a massive model builder, it's a massive product builder as well.

像 Cloud Code 和 Cowork 这样的产品，最近几个月简直火得不行。

Products like Cloud Code and Cowork that have taken off like crazy over the past few months.

Cloud Code 是从一个很少人了解的地方冒出来的，叫 Anthropic Labs。

Cloud Code came out of a place that few of us know much about called Anthropic Labs.

Anthropic Labs 是 Anthropic 内部的一个团队，专注于以 AI 为核心打造下一代前沿产品。

And Anthropic Labs is a organization within Anthropic that is working on building the next level of frontier products with AI at the center.

今天，我们很幸运请到了这个实验室的负责人——Mike Krieger，他是 Instagram 的联合创始人，现在担任 Anthropic Labs 的负责人。

And so today we are lucky to hear from the person running that lab, Mike Krieger, is the cofounder of Instagram and now the lead of of Anthropic Labs at Anthropic.

接下来，我们会请他上台，和《连线》杂志的 Lauren Goode 一起，和我共同进行这场对话。

We're gonna welcome him on stage along with Lauren Goode of Wired who will join me as a co interview.

Mike 和 Lauren，让我们为两位鼓掌欢迎。

Mike and Lauren, let's let's let's hear it for both of you guys.

你好。

Hey.

谢谢。

Thank you.

谢谢。

Thank you.

你来了。

You're here.

很好。

Great.

面对持续的变革与机遇，TMT领域的领导者需要交付切实的成果，而不仅仅是想法。

In the face of ongoing disruption and opportunity, TMT leaders need to deliver tangible results, not just ideas.

当速度和表现最为关键时，普华永道将市场洞察、行业专长与人工智能、云计算和新兴技术相结合，加速您的转型，从战略到执行实现可衡量的投资回报。

When pace and performance matter most, PwC combines market insights and deep sector experience with AI, cloud, and emerging tech to accelerate your transformation and drive measurable ROI from strategy to execution.

普华永道能帮助您预见未来，超越变革，赢得竞争。

PwC can help you anticipate what's next, outpace disruption, and compete.

如需更多信息，请访问 pwc.com。

For more information, visit pwc.com.

好的。

Alright.

好的，迈克。

Alright, Mike.

所以现在在Anthropic这边挺清闲的。

So chill times in Anthropic Land.

没什么事发生。

Nothing going on.

这周挺慢的。

Slow week.

这周挺慢的。

Slow week.

劳伦，你来开头吧？

You want to start, Lauren?

好。

Yeah.

首先，我们想聊聊你在实验室负责的工作，跟大家说说你的角色。

First of all, I mean, we want to talk about what you're working on at labs and explain your role to folks.

不过我想先问你一下，你现在跟白宫那边的联系有多紧密？

But I want to ask you first, how close are you right now to the situation with the White House?

比我在之前职位时要少多了。

Less than in my role.

我大概五个月前转到了现在的实验室岗位。

So I transitioned about like five months ago into this labs role.

要是还在那个位置上，我肯定深度参与其中。

I think in the role, I think would have been deep in it.

现在嘛，说白了，我们当然希望把事情重新推起来。

Now I'm more, you know, obviously, you know, we wanna restore it.

作为产品负责人，我得确保相关资源能到位，但跟以前当高管时比，离核心远多了。

And and as a, like, product person, I wanna make sure that that gets access, but less close to it than, you know, in that sort of c level role that I had before.

明白了。

Okay.

亚历克斯，你有后续问题吗？

Alex, you have a follow-up?

哦，我确实有

Oh, I have

足足八个呢。

like eight follow ups.

当然有。

Definitely.

有个叫本的人在X平台上说，只要Anthropic重新为我开启Fable功能，他就寄一份原始的长格式出生证明给我们。

Well, we there was a guy named Ben on x who said, I will mail Anthropic an original copy of my long form birth certificate if they will enable Fable for me again.

我听起来就像那些痴迷于Four O的人一样疯狂。

I sound like those lunatics who were obsessed with four o now.

你会接受本的那份长格式证明吗？

Will you take Ben's long form?

我不确定我们会收他的证明，但这件事确实挺有意思的。

I don't know that we'll take his long form, but it has been interesting.

我的意思是，Fable 只上线了几天，但自从那以后，我每发一条推文，他们都没看我具体在说什么，反而大部分都在喊‘把 Fable 带回来’，就像 Instagram 当年取消 Gotham 滤镜一样。

I mean, Fable was only available for a few days, but definitely every time I've tweeted since then, they have not read whatever I was tweeting, they've mostly been, like, bring back Fable, which, like, in Instagram, we got rid of Gotham.

你还记得那个 Gotham 滤镜吗？

Do you remember Gotham, the filter?

之后整整八年，我听到的全是‘把 Gotham 带回来’。

This is like and then for the rest, like, the next eight years, all I heard was bring back Gotham.

所以我们都没想到，Fable 居然比 Gotham 还早回来。

So it struck in our but Fable will come will back before Gotham did.

不过，确实那些用过它、并把它融入日常的人，感受很不一样。

But, yeah, it's it's clearly, the folks that have gotten to use it and started incorporating it.

这其实挺有意思的。

It's actually really interesting.

我现在已经不敢轻信模型刚上线那天，或者第一周的反馈了。

I've learned to not really trust day of or even week of model reactions.

你得真正让它经受住实际使用的考验，才能知道它到底怎么样。

Like, you don't really know until you've put it through its paces.

所以在新模型刚发布头几天，我几乎会主动屏蔽所有噪音，因为每个人都会拿自己那套小把戏去试新模型，但真正考验它好不好用，还得靠实际工作场景。

And so, like, I almost just completely block out the noise in the first couple of days of any new model release because, I don't know, everybody has maybe their, like, toy, like, example thing that they like to do with the new model, but it's hard to actually put it through its paces until you've actually had real work done with it.

我觉得大家这才刚开始真正用起来，结果我们就不得不把Fable撤回来了。

And I think people were just starting to do that, and then, you know, we had to sort of pull back Fable.

我记得去年十二月发布Opus 46的时候，正好赶上圣诞节假期，很多人都回家了，圣诞节到新年之间有一周空闲。

But I remember in December when we put out Opus four six, it was, like, this interesting time where everybody went home for the holidays, and a lot of people had that week off between Christmas and New Year's.

等他们回来后都说：‘我那几天花了不少时间用它，现在终于明白Opus为什么厉害了，我决定长期用下去。’

And then they came back and were like, oh, I spent a lot of time, and I really get why Opus is good, and I'm gonna do it.

所以我觉得Fable还没来得及经历这样的过程。

So I don't think Fable has had that opportunity yet.

但即便如此，特朗普政府那边的反应还是相当强烈的。

But despite that, though, I mean, this was a pretty big reaction from the Trump administration.

我相信在座各位，尤其是听过亚历克斯·斯塔莫斯那场对话的人，都清楚背后发生了什么。

And I think everyone here, especially if you were listening to the Alex Stamos session, understands what's going on.

但这一切都是在模型发布后短短几天内发生的。

But this happened within a few days of the model release.

如果要给《连线》杂志一点肯定的话，他们昨晚刚报道说，这件事是因为Anthropic向一家电信公司开放了模型访问权限，这才引起了政府方面的警觉。

If I can give Wired some credit, Wired just reported last night that it was due to Anthropic's relationship with their having giving access to the model to ask a telecom that could have raised flags within the administration.

你对这种 backlash 来得这么快感到意外吗？

How surprised were you by how immediate that that backlash was?

是的，我觉得这种迅速做出反应的决定确实出人意料。

Yeah, I think the sort of reaction decision was surprising.

真的挺意外的。

Surprising.

我们当时也立刻跟他们联系，试图尽快恢复访问权限。

And And we when were sort of immediately engaged with them to, you know, restore access as well.

同时，我们内部经常讨论的一件事是，以前Facebook墙上贴过一张海报，上面写着‘每一天都像过了一周’。

And so at the same time, you know, the one thing that we think a lot about internally is, you know, there used to be a poster on the Facebook wall when we were saying that was, like, every day feels like a week.

我觉得现在在AI领域，这句话正变得越来越真实，整个行业都该时刻提醒自己：我们正处在前所未有的时代。

And I think that's becoming true in AI, and I think a good thing to remind ourselves of in general in the industry is we're dealing with unprecedented times.

我们面对的都是全新的情况，而且它们的发展速度可能快得惊人。

We're dealing with new situations, and they can develop really quickly as well.

所以我觉得，提升能力、建立联系，确保这些对话能迅速展开，真的非常重要。

And so I think also developing the capabilities and the connections and and to make sure those conversations can happen quickly is really, really important.

我有个新标语建议，可以贴在墙上。

I have a new motto suggestion for you for the wall.

往前挪一挪。

Move that.

快点，绕过限制。

Fast and jailbreak things.

我觉得他们不会这么做。

I don't think they're going to

别用这个，Lauren。

use that Lauren.

好的。

Okay.

我们刚从Alex那里得知，有些能力其实早在之前的模型上就存在了，或者现在也能在其他模型上找到漏洞。

We just heard from Alex that some of these capabilities have been available on previous models or can can find bugs with other with other models that are available today.

那你觉得为什么Anthropic在这一点上被特别挑出来？

So why do you think Anthropic got singled out on this front?

我不清楚为什么是Fable，毕竟Fable本身也是一个非恶意意图的模型。

I don't know why, like Fable specifically, it was singled out as, you know, again, Fable being the non cyber intention model as well.

我觉得随着时间推移，真正变化的是：如果你知道该怎么提问、知道自己在找什么，能力就会不一样。

I think that the thing that does change over time is there's, you know, the capabilities if you knew what to prompt and knew what you're looking for, and then there's the capabilities.

我不太认同Alex说的那种‘榨取高中运动员’的比喻，但确实，我们特别关注模型带来的提升效果，尤其是在安全方面。

Think, you know, I'm not sure I resonate with the, like, juicing high school athletes metaphor from Alex, but, like, you know, the the uplift that you get like, uplift is a thing that we think about a lot when we think about model safety.

比如看我们的模型卡片，评估生物领域风险时，我们会对比普通人用模型和普通人只用互联网的差异，看看提升有多大。

So if you look at our model cards, for example, one of the ways that we look at risk in the bio domain is comparing the uplift from sort of lay a lay person using the model versus, you know, an expert or a lay person just using the Internet and and seeing what the comparison there is.

这种趋势随着模型能力增强一直在持续发展。

And so that is one trend that is you know, has been, you know, progressing as the models get more capable.

所以与其纠结为什么Fable被单独拎出来，不如看看整体的发展方向。

And so maybe less why Fable is singled out and maybe more like what the overall trajectory is.

有意思。

Interesting.

我们刚请了Ramp公司的首席经济学家卡拉齐安来播客做客。

So we just had our Karazian, the lead economist on ramp on the podcast.

他提到一件挺有意思的事：五角大楼事件期间，媒体都在报‘某公司不再使用Anthropic’，但Ramp的实际数据显示，他们对Anthropic模型的支出反而增加了。

And one of the interesting things that he said was, know we had during the Pentagon situation, there were these headlines, okay this company will not use Anthropic anymore, but actually the data from Ramp shows that spending actually increased to Anthropic models.

这事儿对Anthropic来说，反倒成了个不错的公关机会。

It was apparently a good publicity moment for the company.

这其实也印证了我们节目里常讨论的一个问题：到底有多少是真正对问题的担忧，又有多少是Anthropic的营销操作？

So I think it sort of plays into a debate that we have like here on the show about like how much of this is like real concern for the issues and how much of it is from Anthropic is marketing.

今天我们请来了Anthropic的人，正好能帮我们澄清这一点。

And like we have somebody from Anthropic here who can actually shed some light on that.

之前斯塔莫斯从安全角度谈了一些，但我们今天能请到你，真是很幸运。

We just said, Stamos talking a little bit about it from his perspective on the security side, but we're lucky to have you here today.

所以，这到底是实质性的变化，还是纯粹的营销，或者两者兼有？

So so is it material or is it marketing or some combination?

我的意思是，

I mean,

我觉得最难的，就是真正相信一件事是真实的，而不是营销手段。

I think it is one of the hardest things to like really deeply believe that something is true and not marketing.

不只是Anthropic，我觉得大家对任何公司说的话保持怀疑都是对的，你也应该用自己的判断去过滤这些信息。

And then due to not even just Anthropic, I think like, you know, I think people are generally right to be skeptical of any company saying anything, and you should, like, put it through your own filters as well.

但对我个人来说，我总是觉得：不是这样的。

But for me personally, I'm always like, no.

但这是真的。

But it's real.

我们真的非常重视安全，而且我们公开谈论这些事，只是为了帮助大家看清未来很可能发生什么、我们已经看到或发现的趋势。

And, like, we, like, both deeply care about safety and are, like to the extent that we are being vocal about anything, is to either sort of help paint the picture of what very likely is coming or we believe is coming or what we've already seen and spotted.

比如在Mythos项目里，我们主要做的是漏洞扫描和漏洞发现，而且是和那些参与过Project Glasswing初期公告的公司合作进行的。

You know, for example, in the in the Mythos case, really just looking at, like, vulnerability scanning and bug finding and doing it in partnership with with companies that were in that kind of Project Glasswing initial announcement.

这项技术真的很厉害，不是那种空洞的吹嘘。

And so the technology is awesome, not in the okay.

它确实在做着令人惊叹的事，所以当我们指出正在发生的事情时，听起来可能有点夸大其词。

Like, in the just you know, it is it is doing really incredible things, and therefore, even calling out what we see as what is happening, I think, can seem hype y.

真希望我能按个按钮，让所有人都相信我们真的没在吹牛。

I wish I could press a button and was and, like, I could make everybody believe that we are not being hype y.

我明白现实没那么简单，但至少从我的角度看，我们一直力求实事求是。

I realize that's not the reality that we operate in, but at least from my perspective, we try to call it like it is.

我了解到，特别是在Anthropic的实验室和研发部门，你们是用最先进的AI模型来实际开发新产品、做原型并验证你们的假设。

My understanding is that within labs in particular and research and development at Anthropic that you're using the best AI models to actually build new products, to prototype new products and sort of test out your thesis.

那这个禁令现在是不是在很大程度上限制了你们在实验室里的这些工作？

So is this is this is this ban essentially now limiting your ability to do that within labs?

是的，说实话，Fable是我用过最好的模型。

Yeah, I mean, definitely Fable is the best model I've ever used.

这并不是说工作完全停了，但确实比不上那些其他模型的效果。

And it's not to say that, like, work has stopped, but it's definitely like less good than the other models that that are.

我们也在用比它差一些的模型。

We are using models that are less good than that as well.

而且说实话，可能正相反——就像第一周那种不信任的反应，恰恰就是没有Fable时会发生的情况。

And it is also I mean, maybe the reverse of the distrusting the first week, you know, response is what happens when you don't have Fable.

而且很明显，那些早就开始用这个模型的人，对Twitter上的反应特别强烈。

And, like, obviously, the Twitter reaction for the people that had already kind of, like, gotten into the model and and were using it was strong.

但就我个人使用来说，我现在用的是Opus 48，感觉也还不错。

But I'd say even in my personal use, I'm like, oh, like, I'm on Opus four eight, and it's good.

我照样能高效工作。

Like, I'm still productive.

我确实在干活，至于具体工作方式因为Fable或者类似模型的变化有多大，我们回头可以细聊，但差别确实很明显。

I'm doing work, but and we can go into sort of, like, how my work changed with sort of these, Fable or, like, you know, the models of that sort of family, but it is noticeable for sure.

是的，我们也挺想了解这些的。

Yeah, I think we'd like to know that.

说真的，公众接触Fable的时间连半分钟都不到。

Mean, the public had access to Fable for like a half a minute.

像Project Glasswing这样的团队，倒是能用上Mythos。

Groups in this like Project Glasswing have had access to Mythos.

我们其实不太清楚，今天能用的模型和那些顶尖的Anthropic模型之间，到底差在哪。

We don't really know what the difference is between using a model that you can use today is and using one of these, you know, super anthropic models.

是的

Yeah.

那么，使用 Fable 或 Mythos 究竟能让你做些什么不一样的事呢？

So what actually could you do differently with a fable or a mythos?

对我来说，关键在于任务分配的范围和规模。

I think for me, it's the sort of scope and scale of delegation.

但这些工具都还很不完善。

And all these things are really imperfect.

比如，人们总会问，这现在算不算一个五级或六级的软件工程师？

Like, you know, people will say, like, oh, is this now a level five software engineer or a level six?

但任何用过这些模型的人都知道，它们的能力依然很不稳定。

But anybody who's used these models extensively knows that they're still spiky in capabilities.

能力，对吧？

Capabilities, Right?

没错。

Right?

在某些方面，它们的工程能力其实比我强多了。

In In some some ways, ways they're, they're, you you know, know, in in many many ways they're a better engineer than me.

但另一方面，比如今天我就在抱怨，它居然没把字母G的下延部分显示全，那个叫下延部。

And in other ways, like, you know, I was complaining today that it had missed a descender like the G, that bottom part is called the descender.

我心想，你们是怎么设计这个界面的，怎么把字给截断了？

I'm like, how did you put that in the UI and it's clipped?

当然了，视觉识别、调试能力，还有那种基本的人类常识，这些方面我们人类还是远胜过现在的模型。

And of course, like there's vision capabilities I need to improve and there's debugging capabilities and there's sometimes even just sort of human common sense that is, you know, we're way better at than when the models are.

但总的来说，我觉得我工作方式最大的转变，其实挺有意思的，正好跟我重新回到一线开发角色的时间吻合。

But overall, I think the big shift for me working and it's really interesting because it sort of coincided with me going back into a builder role.

所以我亲眼见证了，从以前作为高管使用这些模型，到现在亲自上手开发。

So I really got to see going from, you know, using these models as an executive.

当高管的时候，你只能尽量利用它们，但不可能指望它们帮你把每封邮件都写得完美。

You're trying to do the most of them, but, you know, like, not gonna have it right all your email.

我觉得战略方向还是得你自己定，模型的作用是帮你验证和打磨这些想法。

And, you know, I think strategy still needs to come from you, and then you can use the models to sort of pressure test it.

不过回到开发者的角色，从之前的状态转变过来，是这样的。

But going back into a builder role and going from, okay.

以前我会把任务分块委派出去，比如‘请修一下这个bug’，或者‘我在考虑实现这个功能’。

I am delegating chunks like, please fix this bug, or I'm thinking of implementing this feature.

现在我们会来回沟通，最终变成更直接的方式，比如‘我收到了用户的一个bug反馈’，或者‘我有一个想做的东西的想法’。

Let's go back and forth to something that ends up being much more sort of, all right, like, I got this bug report from one of our users, or I have this notion of something that I want to build.

比如，‘你能构思出两三种实现方案吗？’

Like, can you sketch out two or three ways in which we could do it?

对。

Right.

听起来挺合理的。

Like, that seems plausible.

其实我觉得这样挺好的。

Often, I find actually it's nice.

它会给我一些解释或者方案，然后我就说，行吧。

It'll give me the the sort of explanation or proposal and be like, okay.

这确实超出了我的能力范围。

That actually is over my head.

你明显比我聪明多了，能不能给我讲讲，别当我是五岁小孩，但至少别用你那种方式讲，有时候你这么一讲，我就懂了。

Like, you are clearly way smarter than me, like, explain it to me like I'm maybe not five, but at least, you know, not you, and it'll sometimes, you know, explain it that way.

然后你就去把它做出来，而且做得特别准，效率高得惊人。

But then go build it and get it right, getting it right, like, you know, at a very, very, very high rate.

我觉得这真的会彻底改变你的工作方式。

And I think that starts really changing how you operate.

比如我睡前会先确保给Fable安排好足够多的、能撑一整晚的任务。

Like, I moved much more to before going to bed, making sure that I had, like, queued up for Fable, like, enough chunky work to last.

我会整晚都让它跑着，第二天再检查，结果它一小时就搞定了，然后就闲着等我接下来七个小时再吩咐。

I would call it the whole night, and I would check-in later and it got it done in an hour, and it was, like, the guest hanging out for the next seven hours.

但关键是，把任务分派出去，这已经不只是手段，而成了目标。

But, like, really, like, delegating, like, much more of a goal than just a

比如说，一个具体任务。

So one one task, for example.

是的。

Yeah.

比如，举个你交给他做的具体任务？

Like, give us an example of one task you would hand

来吧。

to it.

我举个有点夸张的例子，对程序员来说——我们有个实验室项目是用 Python 写的。

I mean, here's a kind of crazy one, which is for the programmers in the audience, like, I I had written one of our labs projects in in Python.

那是我最熟的语言，而 Instagram 当时全用 Python。

That's like the language I know while Instagram was all Python.

但因为一些不太重要的原因，我们最终得把它改成 TypeScript 才能上线。

And for, like, some not super exciting reasons, we actually needed it to be in TypeScript to deploy it.

我当时就想，行吧。

And I was like, alright.

在 Instagram，我们好几年都在讨论要不要从 Python 换成 PHP，或者 Hack，就是 Facebook 被收购后用的那个语言。

That's gonna be, like, you know, in Instagram, we, for years, talked about moving from Python to PHP or hack or the the Facebook language after the acquisition.

但在我待那会儿，从来没实现过。

And then at least when I was there, never did.

不过我们有个功能叫动态工作流，可以把任务拆分成很多子任务。

But I basically like, we have a feature called dynamic workflows where you can have it, like, also break down the task into, like, a lot of subtasks.

我不只是让它完成单个操作，而是直接处理了数百万行、几十万行代码的整个语言转换。

And I trusted it to sort of not just do the individual action, but here's, like, a whole language conversion of millions of by hundreds of thousands of lines of code at that point.

去干吧。

Go off and do it.

去规划吧。

Go plan it.

去执行吧。

Go execute it.

去检查成果，再复查一遍。

Go verify the work, double verify the work.

等我再回来时，工作已经完成了。

And then I came back to the work being complete.

所以这个任务规模挺大的，属于那种比较复杂的块。

So that level of like, this is a big sort of chunky task.

所以你的意思是，这样更快了？

So that was so you're basically saying it was faster.

一个小时就完成了？

Did it in an hour?

你这是在拿Fable和以前的情况做比较，纯属猜测吧。

You're you're guessing compared to with Fable compared to what it would it have been before.

我觉得主要区别在于，过去做完就完了，觉得搞定就行了。

I think the main difference is in the past, you would it would be like, great, I did it.

但你其实心里清楚，是不是偷了懒？是不是哪里没做好？得再去核对一下，或者你干脆省了某个步骤。

And you'd like, did you you kind of took a shortcut here or this is not quite right or I need to go verify it or like, Oh, you cut this corner or

这就跟过去一年大家总在说的‘管实习生’那套说法一模一样。

it's like the managing interns thing that everyone's been saying for the past year.

确实如此，不过这么说对实习生其实挺不尊重的。

Yeah, it's which is very offensive to interns, by the way.

没错，就是这样。

But yes, yeah, exactly.

正是如此。

Exactly.

我不确定你有没有带过实习生。

I don't know if you managed an intern.

说得对。

Yes, true.

你意思是说，结果更准确了。

And it was in you're saying it was more correct.

确实是。

It was.

所以速度更快了。

So it's faster.

更准确，也更可靠。

It's more accurate, more reliable.

然后，根据美国政府的说法，这是危险的。

And then according to the U.

美国。

S.

政府，危险。

Administration, dangerous.

我觉得另一部分，用词可能不太准确，但大致意思是项目理论，这样就不会说，哦，我要做个改动，然后就会说，好，我改。

I think the other piece of it has like a greater of mind is the wrong word, but sort of like theory of project so that it's less, you know, oh, I'm going to make this change and it'll say, great, I'll make this change.

不过说真的，特别是如果你有过大规模软件工程的经验，最好的工程师会考虑到系统各个不同部分之间的关联，而且他们能预见未来：我知道可以做这个改动，但如果不用这种方式来做，那接下来的改动就会越来越难。

But really, especially if you've done like software engineering at scale, the best engineers kind of keep in mind all of the disparate parts of how this things and they also see around the corner is like, I can make this change, but if I don't do it in this way, then the next change is going to be incrementally harder.

我觉得这就是我在这一类模型中看到的显著差异。

And I think that's like been a significant difference I've seen in that kind of class of models.

所以

So

说到 Anthropic Labs，大家首先想到的肯定是 Claude Code，毕竟它是你们最出圈的产品。

think when we talk about Anthropic Labs, right, people think of Claude code because it is really your breakout product.

听起来你被委派的任务，就是找出下一个Claude Code该是什么样子。

And and it sounds like you've been tasked with basically figuring out what the next Claude code is.

你觉得这个说法准确描述了你在实验室的工作吗？

Would you say that's an accurate description of what you're doing at labs?

而且，Anthropic为什么需要设立实验室？

And also, why does Anthropic need labs?

实验室？

Labs?

是的。

Yeah.

对。

Yeah.

也许值得想一想，2024年我刚来的时候，为什么需要实验室，而今天为什么还需要它，因为我觉得答案其实已经变了。

It's also maybe worth thinking about why we needed labs in 2024 when I arrived and why we need labs today, because I think that the answer kind of shifts.

我刚到Anthropic第三周，就和联合创始人本·曼一起组建了最初的实验室团队，这其实是我早就想推动的事。

I started the labs, the original labs team with Ben Mann, who's one of the cofounders of Anthropic in my third week at Anthropic, and it had been something that I've been bubbling under.

那时候的原因完全不一样。

And at the time, the reason was really different.

当时我们的产品工程团队只有25个人，模型也没现在这么强。

It was all of our product engineering with team was 25 people, and we didn't have the models really like that.

我加入的时候，Claude还只是Sonnet 3，就像你们现在说的Opus 3那样。

We had Claude that when I joined the sonnet three, like you've been Opus three.

那些模型在当时算不错了，但根本算不上实习生的水平。

Like, those were for the time good models, but you weren't gonna they were not even interns.

对吧？

Right?

连IC3级别的工程师都算不上。

They weren't even IC three engineers.

所以如果你团队只有二三十个工程师，他们只能盯着下一个小小的改进。

So if you have a team of only 25 or 30 engineers, they are working on like the next, you know, incremental thing.

我们感觉模型确实在进步，但没什么产品能真正把这种进步展示出来。

And we were feeling like the models are starting to get better, but we don't have any products that sort of show that off.

对我来说，一个很好的检验标准是，当我们准备发布一个模型时，有没有一个产品、演示，或者别的什么能清楚展示出完全不同的东西？

Like a good litmus test for me is when we get ready to release a model, do we have either a product or a demo or, you know, some other illustration of something that is very different?

随着时间推移，这变得越来越难了，比如Fable，哪怕只是想清楚那个周末任务，或者这种需要长时间投入的工作，都不容易呈现出来。

And it gets harder over time, like, with with Fable, you know, even, like, illustrating, like, that weekend task or, you know, this longer amount of work.

所以当时Labs的核心目标就是，别让我们的产品落后于模型飞速发展的步伐。

So, really, Labs at the time was, let's make sure we don't like, our products don't fall behind the model exponential that's happening.

Claude Code就是从这个想法里诞生的，因为其他产品团队根本没人考虑过编程这件事，也没人有空间去想：如果我们彻底改变交互形式，完全接受模型会朝这个方向进化，会怎么样？

And, yeah, so Claude Code came out of that initial one because nobody in the rest of the product orgs people were thinking about coding, but nobody was sort of had the, like, space to go and think about, well, what if we totally change the form factor and we embrace the fact that the models were gonna evolve in this way?

我们在Labs里最常用的两个思维练习之一，就是想象当前模型的能力和大多数人实际使用方式之间的差距。

And a lot of the like, the two most useful thought exercises we do in labs, one is, like, visualize the gap between what the models can do today and how most people use it.

然后想办法缩小这个差距。

It can be closed that gap.

这是第一个。

So that's one.

另一个是，想想现在模型还做不好的事，但六个月后可能会做得很好，然后确保我们到时候已经有相应的产品准备好了。

And the other is, imagine what the models are bad at now that they're actually gonna be really good at in six months, and let's make sure we have a product ready for that by then.

我觉得这两点就是实验室的核心指导原则。

And I think those are, like, the two guiding questions for for labs.

从最初的版本里，还诞生了计算机操作功能。

And then also out of that first incarnation came computer use.

不过计算机操作这个功能不一样，我们做出来的时候简直太糟糕了。

Computer use was different though because when we built it, it was really bad.

我们试过好几个产品，那时候大概是Sonic 35左右。

Like, we tried a bunch of products with it, and this was around, you know, Sonic three five.

用户会说：‘Claude，你能帮我整理一下桌面吗？’

And they're like, Claude, can you help me, you know, clean up my desktop?

它就会点一下，直接把文件删了。

And it would, like, click the thing, just delete the file.

这根本没法发布，太不安全了。

Like, this is not not safe for release.

我们绝对不可能把它做出来或者推向市场。

We're definitely not gonna go and and and build this or to ship this.

但我们保留了这个产品，每次推出新版本，都会先在内部测试，看看电脑辅助功能有没有改进。

But we had that product so that every new model that we'd we'd release, we'd first check it internally and say, did computer use get better?

我们会把改进或退步的情况反馈给研究团队，直到觉得足够好为止。

And we'd tell the research team how it gotten better or worse until the moment where we it's good enough.

我们真的打算在这个基础上推出一个正式产品。

We're actually going to put a product out around this.

它还像一个标杆，帮我们衡量未来产品的进步方向。

It also gives you this sort of sort of beacon into the future that then you can kind of measure your your future products against.

但也要拿它和现在的情况做对比。

But then compare it to now.

我们现在有一个非常活跃的产品团队。

So we have a thriving product team.

有CoWork，也有Claude Co.，都发展得不错。

There's co work, there's, you know, Claude Co.

规模已经扩大了很多。

It has grown a lot.

我们有自己的平台。

We have our platform.

现在我觉得，这些产品团队其实已经不太会做这种思考了。

And now I think it's actually much, much less about none of these product teams are doing this sort of thinking.

更重要的是，模型进步得太快了，我们与它们互动的方式也得跟着升级。

And I think it's much more that the models are advancing really quickly and even our capability to interact with them needs to evolve.

所以我们和实验室、Cloud Code 合作，推出了今天上线的 Cloud Code 组件。

So one of the things we collaborated with Labs and Cloud Code that we ship today is Cloud Code artifacts.

让 Cloud Code 不只是能打字回复你，还能画画、给你出图。

Having Cloud Code not just be able to type back to you, but also sort of draw a picture or give you an illustration.

这个想法部分来自我们在实验室里花大量时间得出的结论：光靠一个文本框和一大段文字回复，已经不够用了。

And that partially came from spending a lot of time in labs saying, just a text box and a big text response is not gonna cut it anymore.

就像我之前说的，模型和我聊天时，有时候我觉得它们比我聪明多了，这时候我就会想：你能给我画张图吗？

Like, when I mentioned that the models feel like they're way smarter than me when they talk to me, sometimes I'm like, can you draw me a picture?

因为这才是我真正需要的，才能完全理解。

Because this is what I actually need to fully Yeah.

完全理解这一点。

Fully understand this.

但我们真正一直在想的是，没错，我们的产品确实多了不少。

But it's really what we've been thinking about is, you know, yes, we have a lot more products.

实际上，我们的产品还有很多需要整合的地方。

You know, we have we are actually have a lot of consolidation to do in our products.

这本身也是我们正在推进的另一个方向。

Like, that's another initiative that we have.

但即便如此，我们依然有机会让那些不整天琢磨提示词、不纠结于高、中、低投入差异的人，也能更轻松地使用。

But within that, we still have the opportunity to make things much more accessible to a person that does not spend all of their time thinking about prompting and the exponential on the difference between high, low and medium effort.

这里面能说的还有很多。

Like, there's a lot we can say.

不过，迈克，使用Anthropic模型的用户现在其实处在一个挺特别的位置。

But but Mike, so there's a it's it puts people using Anthropic Models in an interesting place.

对吧？

Right?

你知道吗？Cursor 最近被 SpaceX 以六百亿美金收购了，有人在推特上发了个梗，说要是没有这个 guy，Cursor 本来能卖到三千亿，照片上的人就是 Claude Code 的创始人 Boris Cherni。

There know, Cursor I think just sold for 60,000,000,000 to SpaceX and someone put this meme on Twitter that like, you know, Cursor would have sold for 300,000,000,000 if it wasn't for this guy and it's a picture of Boris Cherni, the person who created Claude Code.

对于那些打算基于 Anthropic 技术做产品的公司来说，他们肯定会犹豫：我到底是该和 Anthropic 合作，还是说 Anthropic 自己会直接推出我本来想做的那个产品，哪怕我已经和他们合作了？

And so for companies that are going to build on top of Anthropic technology, they're going to wonder, do I want to partner with Anthropic or is Anthropic going to go ahead and the product that I'm gonna wanna build, potentially even after partnering with them?

是的。

Yeah.

我们更愿意聚焦在智能编码这个方向上，而同时扮演平台和产品双重角色，这个思路其实特别有意思。

I mean, we'll we'll take the, like, agentic coding side, and I think the broader sort of aspect of of, you know, being both a platform and a product, I think, is is really interesting.

我们接项目的时候，目标通常是推动整个行业在这块往前走。

When we take on projects, the goal is often to sort of push that area of the industry forward.

以前也有 AI 编辑器，有些做得不错，但没人像我们做 Cloud Code 时那样，这么自由地去思考这个问题。

So, you know, there were AI coding editors and some of them were really good and, you know, but nobody was quite thinking about it in as sort of freeform a way as we got to think about it with Cloud Code.

现在，越来越多的产品都开始往这个方向走了，要不是我们，可能没这么快。

And now a lot more products have that flavor than I think would have otherwise.

所以如果你觉得我说得不对，随时可以指出来，Alex。

And so I think if we're ever you can call me out on this, Alex.

如果我们要进入一个行业，大家做的都是千篇一律的事，只是挂了个Anthropic的牌子，我觉得这纯粹是浪费我们的时间，也浪费了实验室或产品团队的精力。

Like, if we're ever entering an industry where, like, all you're doing is the same thing everybody else is doing, but, like, you've got the Anthropic brand, I feel like that's bad use of our time and a bad use of our either labs or product team time.

如果我们真要进某个领域，那应该是为了说：嘿，我们觉得未来的发展方向是这样的。

Like, if we're going in somewhere, it should hopefully be to say, all right, we think that the direction of travel is this way.

我们可以做出符合这个方向的产品。

We can build a product of that.

而且，根本不可能也不应该出现所有产品都变成Anthropic自家产品的局面。

And then, by the way, there's no world nor should there be a world where, like, all the products are Anthropic products.

那样的世界会很糟糕。

That'll be a bad world.

没错。

Right.

所以，我们希望要么为其他公司开辟出新的空间，要么指明方向，让别的产品也能借鉴融入。

So like that is hopefully either creating new space for for companies or sort of showing the way where other products can incorporate.

对。

Yeah.

这就有点像在一家同时做社交聊天、视频和后续功能的科技公司工作一样。

It would almost be like working for a tech company that has like social messaging, video, follow-up to.

对吧，迈克？

Right, Mike?

是的。

Yeah.

明白了。

Okay.

你确实离开了。

You did leave.

比如，当Anthropic推出一款被看作与Figma竞争的产品时，而你之前曾任职于Figma的董事会，当时就有人提出疑问。

Well, there was some question, for example, when, you know, Anthropic launched a product that was seen as competitive to Figma and you had been on the Figma board prior to that.

我想你后来退出了。

And I think you stepped down.

是的。

Is Yeah.

那

That

所以亚历克斯提了个好问题，关于硅谷为什么能拥有如此健康、充满活力且敢于冒险的创业生态。

And so it's a good question that Alex has brought up, think, where Silicon Valley is known for this really healthy, vibrant, risk tolerant startup ecosystem.

当大公司带着大量风险投资和丰富资源涌入时，人们就会担心：等等，他们是不是直接抄走我的点子？

And when the big start coming in with tons of venture capital and, you know, a lot of resources, people say, well, wait, are they just are they essentially just going to steal my idea?

对。

Yeah.

不，我认为我们这种双重身份是其他公司也必须面对的。

No, I think I think our dual existence and it's something that other companies have to navigate.

我们刚才在上一个环节聊了很多亚马逊，他们就得处理这种双重角色：既是基础设施提供方。

We talked talked about Amazon a lot in the in the previous panel, like they have to navigate this role where they are both the infrastructure provider.

他们显然拥有庞大的电商业务。

They obviously have a very large e commerce.

他们做视频，同时也分发视频。

They they do video, but they also serve video.

总的来说，客户其实能适应这种双重角色——他们知道，自己在用对方的基础设施，而对方也在用同样的基础设施做自己的事。

And then, you know, by and large, customers can live in that dual world of like, Okay, I'm using their infrastructure, also knowing that they they are also using their infrastructure to do that.

你可以去问问我们的客户，看看我们在这方面做得怎么样。

And I think the you can talk to our customers and see how well we're doing at this.

我一直以来都尽量保持透明，这是我的原则。

Like, the thing I always try to do is, like, at least approach it with a lot of transparency.

比如光标这个例子，我和迈克尔过去聊过很多次。

So the cursor example is an interesting one where, like, Michael and I talked a lot over the time around.

我们一直在讨论事情的发展方向。

Here's where things are heading.

同样地，对于我们其他产品的思考，我觉得有几点很重要。

And, you know, with similarly with with the other products that we think about, like, can we I think it's a couple of things.

一是透明，二是共享的基础模块。

It's transparency, and then it's shared building blocks.

说实话，我觉得根本就不存在这种情况。

Like, yeah, I think in general, I actually don't think there's any cases where this is even true.

我们是在别人已有的能力基础上继续搭建东西。

Like, we're trying to build on top of the same capabilities that are available elsewhere.

我上次在联邦俱乐部这个舞台上，是今年年初的医疗日，那时候我们还没推出Claude Healthcare。

The last time I was here in the Commonwealth Club on this stage was our health care day at the beginning of the year, and we didn't ship like Claude Healthcare.

只有我们有这个功能，别人都没有。

Only we have it like nobody else has it.

我们发布了一堆插件、技能、MCP，还有一些配套的能力。

We shipped a bunch of like plug ins and skills and MCP's and like complementary abilities.

所以我说这事儿不简单，也不好说一清二楚，但这就是我们应对这个复杂局面的方式。

So that's how I'm not claiming it's easy or that it's a straightforward thing, but it is how we're trying to navigate what is, like, admittedly a complicated Right.

有点复杂的情况。

Sort of situation.

说到初创公司，Anthropic严格来说还是个初创公司，但你们已经值不少钱了。

Speaking of startups, Anthropic is still technically a startup, but you're worth a lot of money.

我的意思是，现在估值是多少？

I mean, what what's the latest valuation?

是多少？

Is it

96.5亿。

$9.65.

9650亿左右吧。

$965,000,000,000 or something like that.

我当年把Instagram卖了一十亿。

I sold Instagram for a billion.

对吧？

Right?

初创公司。

Start up.

在2010年。

In 2010.

对。

Right.

从那以后，财务状况已经发生了不小的变化。

Financials have changed quite bit since then.

但Anthropic已经为自己定好了位置。

And yet, Anthropic has positioned itself.

它是一家公益公司，一直把自己定位为在开发AI时更注重伦理的公司。

It is, you know, a PBC and it's positioned itself as sort of a more ethical company around building AI.

我想知道，你如何看待这种定位，以及Anthropic在改变硅谷文化方面扮演的角色？

And I'm wondering if you could talk a little bit about how you see that positioning and Anthropic's role in particular changing the culture of the Valley.

就像我知道，回想起2000年代初，谷歌在很多方面彻底改变了硅谷的文化。

Like, I know I think back to how Google in the beginning of the 2000s really changed the culture of Silicon Valley in so many ways.

你觉得现在Anthropic的文化，会如何塑造下一个时代？

And how do you see Anthropic's culture now dictating this next era?

这是个很有意思的问题。

Yeah, that's a really interesting question.

我先从内部说起吧，我觉得外部也有影响。

Maybe I'll start like inside, and I think there's an external component, too.

我当初加入的原因，是因为我正准备结束我的第二家创业公司，当时我决定去一家前沿实验室工作，因为我已经开始用这些模型来写代码了——虽然它们那时候代码能力很差，但我看得出来，这已经是它们最差的状态了。

I think the reason I joined in the first place, so I was winding down my second start out and knew I wanted to go work at a frontier lab because I had started to use these models for coding, and they were bad at coding, but I could see that they were as bad as they were ever gonna be at coding.

以后只会越来越好。

They were going to improve.

那时候我已经开始基于这些 API 做开发了。

And I had started building on top of these APIs.

我当时的创业项目叫 Artifact，主要是做 AI 驱动的新闻推荐。

So the startup I was doing was called Artifact, and we did sort of AI powered sort of news recommendations.

她以前经常通过 Artifact 看 Big Technology 的内容。

And she read a lot of Big Technology via Artifact back in the day.

谢谢。

Thank you.

我们后来加了一些功能。

Things we added.

所以。

So.

但没连网。

But not wired.

不过说实话，你们的付费墙确实太严了。

But, you know, you guys had a really hard paywall, to be honest.

这我能理解。

Fair enough.

但我们不这么认为。

We didn't agree.

我们是这么做的。

We do.

你需要我给你订阅打折吗？

Do you need a discount on my subscription?

我可以帮你申请一个。

I can get one for you.

好的。

Okay.