Mark Zuckerberg on Muse, Meta's biggest AI bet yet

我成功拿到了那天的许可证。

I was able to get a permit for this day.

然后我就想，好吧，看来我得请假去跟我女儿爬山了。

And I was like, all right, well, guess I'm taking that day off from work to go climb a mountain with my daughter.

所以还挺酷的。

So it's kind of cool.

对。

Yeah.

对。

Yeah.

是吧？

Right?

所以它做这些事，但同时也帮我保持健康，帮助我训练。

So it does that, but it also helps keep me healthy, helps with my training.

我在我的MMA拳馆里装了摄像头，然后让它盯着摄像头，给我反馈。

I put cameras up in my in my MMA gym, and I tell it to watch the cameras and send me feedback.

而且，这还挺好玩的。

And it's like, it's pretty fun.

你知道，挺好的。

You know, it's good.

反馈不错？

It's good feedback?

有时候吧。

Sometimes.

对。

Yeah.

你知道，就是……

You know, it's...

其实，就是……

Mean, it's...

有时候反馈还挺搞笑的。

Sometimes it's funny feedback.

它就像，就像，它找到我……

It, like, it, like, finds me in...

就像，看起来你真的放弃了。

It's just like, it looks like you really gave up.

然后我就说，对。

And and I was like, yeah.

我确实放弃了。

I did.

我当时真的很累。

I was really tired

就在那里。

right there.

我就想，为什么你偏偏要指出这一点？

I like, why is that the thing that you're pointing out to me?

但是……

But...

不。

No.

挺好的。

It's good.

而且挺搞笑的。

And it's funny.

就像教练们，他们也会拿这个开玩笑。

Like, the coaches, like, laugh about it.

对。

Yeah.

不。

No.

这就像

This is like

这就是我们觉得有意思的地方。

This is what we Interesting.

感觉我们可以告诉你，但你的经纪人也在告诉你。

Feel like we could tell you, but your agent's telling you.

对。

Yeah.

嗯，这个嘛，

Well, it's,

是的。

yeah.

那挺好的。

That's good.

我之前没想过这个，直到听你说起，你身边显然有人能帮你搞定这些事。

I didn't think about this until hearing you talk about it, but you you have people that could obviously do all this for you.

但比如说，你怎么用像Muse agent这样的工具，去真正测试它作为助手的极限在哪里。

But, like, how do you use something like a Muse agent to, like, really test the limits of, like, how it can be helpful as an assistant.

对吧？

Right?

对。

Yeah.

比如说，你是不是……

Like like, are you...

你有没有在挑战它的极限？

Were you pushing it?

你是怎么用它，让团队觉得，哦，好吧。

Like, how have you pushed it in a way that the team is like, oh, okay.

我们得赶紧修一下这个。

We gotta gotta fix this.

或者……

Or...

我相信他们肯定……我是说，想想看。

I'm sure they're I mean, think

有意思的地方在于，每个人想用它做的事情都不一样。

part of what's interesting about it is that everyone just has such different things that they want to do with it.

所以在早期测试阶段，我们把它交给了很多人。

So in the early beta period, we handed it to a bunch of people.

然后我把它给了某个人。

And it's like I gave it to someone.

结果一天之内，他们就用它来帮忙管理家庭教育了。

And within a day, they're using it to help run their homeschool.

然后我就想，哇，这也太厉害了。

And then it's like, okay, wow.

你一天之内就上手了。

You just started this within a day.

然后另一个人，一天之内，甚至十二小时内，就说，我刚计划了一次旅行。

And then another person within a day or within twelve hours, they're like, I just planned a trip.

对。

Yeah.

它就像帮我规划好了所有事情。

It's like it just planned this whole thing for me.

对。

Yeah.

我是说，我认识的另一个人，她一般对技术挺怀疑的，我把这个给了她。

I mean, someone else I know who's generally pretty skeptical about technology, I I gave it to her.

然后她……

And then she was...

她好几天都没说什么。

She didn't say anything for, like, a few days.

后来她发短信问我，正式发布的时候，我的音乐还能保留吗？

Then she texted me and was like, so when you do, like, the general release, do I get to keep my music?

还是你们会重置？

Or are you gonna reset it?

我当时就想，行吧。

And I was like, alright.

这挺好的。

This is good.

这真是……

This is...

所以我觉得，嗯，我觉得这个效果挺好的。

So I think I think it's, like, I think this is working well.

我认识的每个参与测试的人，你知道，都对它评价很高。

Everyone I know who's been on the beta has very, you know, high things to say about it.

评价非常高。

High praise.

对。

Yeah.

但是每个人用它的方式都不一样。

But people do different things with it.

对。

Yeah.

而且我觉得我们应该更直白地告诉大家这是什么，这样他们才能理解，因为我觉得人们想到AI时，就会想到Meta AI或者ChatGPT那种。

And I think we should just also say what it is more plainly for people so they understand because I think people think of AI as like, you know, Meta AI or ChatGPT.

就像是来回提示那样。

It's like back and forth prompting.

对。

Yeah.

这里真正的突破在于，整个行业都在发生这种情况，不管是Grock、Bot、Town还是Instinct。

The the real unlock here, and this is happening in the industry more broadly, whether it's Grock, Bot, Town, Instinct.

我是说，有很多产品都在做这个，但关键是在后台加了一个虚拟机，让代理可以帮你控制电脑，登录进去，然后做各种事情。

I mean, there's many products doing this, but it's it's adding a virtual machine behind the scenes where, like, the agent can control a computer for you and log in and do things.

我觉得，对于那些只知道AI是那种……的人来说，这是一个巨大的变化。

And that's a that's a huge change, I think, for people who only know AI from It's the

它是长期存在的。

it's long lived.

所以它是……

So it's...

所以基本上，不像Meta AI、ChatGPT或Gemini那样，你发一个提示，它就给你一个答案。

So basically, instead of the model with, like, Meta AI or ChatGPT or Gemini or whatever you use where you send one prompt and then it gives an answer

对。

Mhmm.

在这种情况下，你基本上就是给它项目或者目标。

In this case, what you basically do is you give it projects or you give it goals.

是的。

Yeah.

然后它就会自己干活。

And then it just works.

而且它全天候工作。

And it works twenty four seven.

它会一直干，直到帮你达成目标为止。

And it, like, doesn't stop until it's helped achieve the goals.

你的团队跟我说，你们管这个叫“夜间学习”，是吗？

Your team was telling me it studies overnight, is what you guys call it?

哦，学习啊，它会把反思整理到记忆里。

Oh, Studies, it consolidates its reflections into memory.

它基本上就是专注于项目。

It basically just works on projects.

而且它还能提出新项目。

And it can also suggest new projects.

对。

Mhmm.

所以……

So...

是的。

Yeah.

我的意思是，我当时就像……

I mean, I was like...

我喜欢和我一个女儿玩电脑游戏《文明》。

I like play the computer game civilization with one of my daughters.

而且我喜欢……

And I like...

我就说，嘿。

I was like, hey.

你想不想给她做一份策略指南？

Do you wanna make like a strategy guide for her?

我就说，好啊。

I was like, yeah.

当然可以。

Sure.

所以我就说，行。

So I was like, okay.

现在我们有策略指南了，你想不想让我把它扩展一下，让它还能教一些关于不同文明的历史知识？

Now that we have the strategy guide, do you want me to, like, expand the strategy guide so it can also teach historical lessons about different civilizations?

我就说，好啊。

I was like, yeah.

当然可以。

Sure.

为什么不呢？

Why not?

所以它就，嗯，像是新建了一个标签页，然后它做的那个应用，就会……

So it just so it just kinda, like, built a new tab and and the app that it made, and that would...

那真的很酷。

That was that was very cool.

所以它就能这样不断扩展，然后……

So it can kind of just, like, expand and

而且它是主动的。

and It's proactive.

对。

Yeah.

它会主动建议一些东西。

It suggests things.

我觉得有意思的一点是，它将来真的能帮人们赚钱，也能帮人们省钱。

And one of the things that I think is interesting is that it, I think, is just going to be able to, like, make people money and save people money.

你觉得呢？

You think?

对。

Yeah.

我是说，部分……

I mean, part of the...

这里有意思的一点是，我们定价的经济模式。

Part of what's interesting here is the economic model for how we're pricing it.

就是说，如果你想的话，可以付费订阅……

Mean, you can pay for a subscription if you wanna...

如果你基本上想要那种模式的话。

If you basically kind of wanna have that that model.

但我们也让它免费使用量很大。

But we're also just making it so that you can get a very large amount of usage for free.

我觉得我们提供的东西，我觉得，开始……

I think we're we're offering something, I think, start.

大概每周免费提供一亿个令牌，就像……

It's like a 100,000,000 tokens a week for for free, like...

而且你还能得到一个虚拟机。

And and you get this virtual machine.

所以这就像是一台很强大的电脑。

So it's like a a lot of kind of computer.

这个梗不错。

That's a good meme.

确实是一台很强大的电脑。

It's a lot of computer.

确实是一台很强大的电脑。

It's a lot of computer.

对。

Yeah.

对。

Yeah.

对。

Yeah.

但我们这么做的原因，基本上是因为我们有信心支持这样一个事实：我们认为这对那些打算用它来经营小生意、赚钱、做交易或者某种商业活动的人来说，会非常有效。

But the reason for why we're doing this is we basically are confident in standing behind the fact that we think that this is going to effectively for people who are going to use it for running a small business or making money or transactions or commerce in some way.

我们其实只是觉得这会为人们赚很多钱，所以长期来看，我们预期的商业模式基本上就是只从每笔交易中抽取很小的一部分。

We actually just think it's gonna make so much money for people that that the business model over time that we expect is to effectively just take a very small cut of whatever the transaction is.

好生意。

Take great business.

小。

Small.

对。

Yeah.

对。

Yeah.

而且甚至不一定是从付款人那里抽成。

And not even necessarily the person paying for it.

这些费用会来自他们合作的那些企业。

It'll come from the businesses that they're working with.

但是，而且你正在

But And you're

和Stripe合作处理支付。

working with Stripe on payments.

对。

Yeah.

而且……

And...

对。

Yeah.

对。

Yeah.

所以……

So...

但我的看法是，我们应该能提供一个服务，让绝大多数人免费使用，这再次是至关重要的，如果你想为所有人构建这样一个未来，让每个人都拥有这些强大的超级智能代理。

But my view is, like, we should be able to have a service that you make free for the vast, vast majority of people, which again is critical if you want to build up this future for everyone where everyone has these powerful super intelligence agents.

我认为让所有人都能使用某样东西的一个重要部分，就是让它负担得起。

I think an important part of making something available to everyone is making it affordable.

所以我们想让它免费。

So we want to make it so that this is free.

所以你会得到大量的使用量，我们基本上就是在背后支持这一点，说我们认为这个东西真的会帮你赚钱、省钱，这就是你...

So there's just this huge amount of usage that you get, and we're basically just standing behind that and saying, we think that this thing is actually going to make you money and save you money, and that is how you're...

它会自己回本的。

It's going to pay for itself.

而且Meta的服务可以接入进来。

And Meta services can connect into it.

对吧？

Right?

所以理论上你可以管理你在Instagram上的广告支出，所有这些。

So you could theoretically manage your ad spend on Instagram, all that stuff.

嗯，你得……

Well, you have to...

你把它连上。

You connect it.

你可以把它连接到任何你想要的东西。

You you can connect it to whatever you want.

你知道，如果你愿意，它确实可以和Meta的服务配合使用。

You know, it does work with Meta services if you want.

当然，如果你不想连，也可以不连。

You obviously don't have to connect it if you don't want to.

抱歉，如果你用它来经营生意的话。

Sorry if you're using it to run a business.

它基本上可以直接连接到我们的广告系统。

It can basically just connect to our ad systems.

你可以让它帮你做点东西。

You can ask it to make something for you.

它们能帮你做产品，然后还能帮你打理生意。

They can help you make the product, and then it can help run the business.

所以所有这些事情，它们可以循环着做，一直做下去，全天候不停歇。

So all this stuff, they could just do that in a loop and just do it forever for 20 fourseven.

而且每次我们发布新模型，我们一直保持每个月都有重大更新的节奏，它就会变得更聪明，或者只是……

And every time we release a new model, which we've been on this cadence of shipping a meaningful update every month, it's just going to get smarter or just...

能力更强，能做越来越多的事情。

And get more capable and and able to do more and more stuff.

还有，你的团队告诉我，我在别的地方没听过这种用法，就是这种“舰队”概念，你让一群Muse代理一起学习，对。

And something your team was telling me that I haven't heard this approach used elsewhere is this fleet concept where you're letting the fleet of Muse agents learn Yeah.

一起学习。

Together.

所以这就是那个“想法和建议”的功能。

So that was the ideas and suggestions thing.

你打开应用。

So you open the app.

主标签页基本上就是你和你的Muse聊天的地方。

The main tab is basically your chat with with your Muse.

还有一个标签页，主要是根据你告诉它的内容，它如何扩展这些想法。

There's a tab for basically ideas from the things that you've told it, how it can expand those.

这就是我之前说的，首先，它帮我制定了和我女儿玩文明游戏的策略指南，然后把它扩展成了历史课程。

So that's the thing I was saying, which is, you know, first, it helped make the strategy guide for playing civilization with my daughter, then it helped expand that into historical lessons.

那是……

That was...

它想出了那个主意。

It came up with that idea.

对。

Right.

然后我就说，好的。

And then I was just like, yeah.

当然。

Sure.

那就做吧。

Do it.

对。

Right.

然后它就会找到各种方式来提升自己。

Then it it finds all these ways to basically augment itself.

我是说，像MMA教练那种，它会想出办法来改进。

I mean, the, like, MMA, like, coaching thing, it comes up with ideas for how to make it better.

它会说，你想不想让我更好地找到合适的画面发给你？

It's like, would you like me to get better at finding the right frame to send you?

然后你就说，好啊，去做吧。

It's like, yeah, go do that.

所以我觉得想法这个功能很重要，因为这样你基本上可以在整个团队里找到对不同事情感兴趣的人。

So the ideas thing I think is important because then you basically across the fleet, can find people who are interested in different things.

我觉得……

I I think...

我觉得，退一步来看。

I I I guess, taking a step back.

我认为AI存在的一个大问题是，很多人不知道该怎么用它。

One of the big issues that I think exists with AI is a lot of people don't know what to do with it.

对。

Yeah.

所以我觉得，如果智能体本身能帮你建议它能做些什么来帮助你，

So I think if if the if the agent can itself help you suggest things that it can do to be helpful for you

明白。

Mhmm.

那就能解决很大一部分问题，让你能充分利用它。

Then that solves a huge part of this problem of making it so that you can get the most out of it.

当你引入智能体的网络效应学习时，这真的没人做过，对。

When you're introducing, like, network effect learning for agents, which no one's really done where Yeah.

基本上，智能体从整个网络中学习匿名的洞察。

Basically, the agents are learning anonymized insights from the rest of the fleet.

明白。

Mhmm.

我的意思是，你可是网络效应的王者。

And I mean, you're, the king of network effects.

我对这个想法真的很感兴趣，因为我觉得没人这么做过。

Like, I I'm I'm really interested in this idea because I I don't think anyone's doing this.

对。

Yeah.

确实。

No.

我觉得现在，大多数行业都把智能体看作单机游戏，就是你有一个智能体，然后你自己用。

I think that right now, I think most of the industry is thinking about agents as a single player game where it's like you have your agent and you use it.

而通过让智能体互相交互，会有很多有趣的事情可以做。

And there are going to be all these interesting things that basically you can do by having agents interact with each other.

我们内部已经有很多有趣的例子了，人们让他们的智能体互相交互。

And we already have all these interesting examples internally where people have their agents interacting with each other.

这次发布大部分还不会推出这个功能。

This isn't, for the most part, rolling out in this release.

但随着时间的推移，这会是它运作方式的一个重要部分，就是嗯。

But it's gonna be an important part of how I think this works over time is just Mhmm.

比如，随着越来越多的人开始使用Muse，它会变得更好。

Like as more of the people who you know start using Muse, it just gets better

那这是不是就是差异化所在？你知道，你可以认为模型在前沿会不断商品化，或者产品都会变得相似，类似的框架。

and And is that the differentiator as, you know, you could buy the idea of models continue to, like, commodify at the frontier essentially, or, you know, the products all start to look similar, similar kinds of harnesses?

这种学习的网络效应是不是真正的优势？

Is the network effects of that learning the real edge?

嗯，我觉得我们做的有几件事是比较独特的。

Well, I think that there's a few things that are that are kind of unique that we're doing.

一个是我们从头开始设计模型。

One is we're designing the models from the ground up

对。

Right.

基本上就是为了这个用途而设计的，我觉得这真的很重要。

To to basically be good for this use case, which I think really matters.

第二点，我确实认为我们公司有这种社交基因。

Two is is basically, I do think we have this social DNA as a company.

我们帮助人们利用AI和智能体来增进人际关系，加强联系，从生活中那些柔软但非常重要的部分中获得更多。

We're helping people use the AI and agent to enhance their relationships and strengthen your relationships and get more out of the soft but very important parts of your life.

我觉得作为一家公司，我们可能比任何其他实验室都更关注这一点。

I think that that's something that we're probably just gonna be more attentive to as a company than any of the other labs.

第三点，我认为这将是我们的一个主要差异化优势，可能会让一些人感到惊讶，那就是隐私和安全。

The third thing that I would say is actually going to be a major differentiator for us that I think might be surprising to some people is privacy and security.

我们在这方面投入了巨资，非常巨大的投入。

And we're investing in this just a huge, huge amount.

我们的观点是，要让这个有用，它不仅要拥有最先进的智能。

And part of the view that we have on this is that in order for this to be useful, it needs to not just have state of the art intelligence.

它还需要真正了解你。

It needs to really understand you.

所以，为了能理解你的目标，你最终得把它连接到所有那些东西上。

So in order to be able to understand your goals, you end up connecting it to all the stuff.

对吧？

Right?

你提到把它连接到你的广告系统，但人们还会连接到消息和

You talked about connecting it to your ad system, but people connected to messaging and

邮件。

Email.

邮件，还有各种东西，健康信息之类的。

Email and all stuff, health information, whatever.

而要做到这一点，人们需要对系统有非常高的信任度。

And in order to do that, people need to have a very high degree of confidence in the system.

是的。

Yeah.

现在好消息是，Meta 到目前为止已经花了十多年时间，把 WhatsApp 打造成全球最大的端到端加密系统。

Now the good news here is that Meta has spent, at this point, more than ten years focusing on building WhatsApp into, I think, like, it is the largest, like, global end to end encrypted system.

我们设计的方式是，连Meta自己都看不到人们发送的消息，这真的带来了巨大的变革。

We've designed it in a way where even Meta can't see the messages that people send, and that's been this just really transformative thing.

我觉得这让人们信任WhatsApp。

I think it makes it so that people trust WhatsApp.

对Meta来说，这也是一个非常重要的教训，那就是这对我们在WhatsApp上的成功至关重要。

It's also been a very important lesson for Meta to learn that like that has been really important to our success with WhatsApp.

我们设计的系统连我们自己都看不到内容。

That we've designed the systems that even we can't see the content.

这意味着，不管人们担心什么，比如担心政府获取数据，或者黑客入侵，对吧。

That means that whatever people are worried about, if they're worried about, you know, government getting access to it, a hacker getting access to it Right.

或者Meta内部有人做坏事，这些都是他们不想看到的。

Someone at Meta doing something bad with it that they don't want.

如果你把系统设计成自己都看不到内容，那所有这些担忧都可以排除掉。

All that stuff you can kind of take off the table if you design the system so that you can't see it.

所以我们把这一点作为基础教训之一。

So we took that as one of the foundational lessons.

我们刚开始做这个的时候，我和Nat其实亲自去招募了Moxie Marlinspipe。

When we were getting started with this, Nat and I actually, you know, personally recruited Moxie Marlinspipe.

Signal的创始人。

The founder of Signal.

对。

Yeah.

而且他是其中一个...

And one of the person...

帮助我们构建这个的人之一

One of the people who helped us build the

WhatsApp标准加密

WhatsApp standard encryption

当年，2014年的时候。

back in the day and back in 2014.

他加入就是为了专门做这个机密虚拟机项目，这样我们就能...

He joined to specifically work on this confidential VM project, which makes it so that we can...

你可以拥有自己的虚拟机，把所有信息都放在你的Muse里，而且我们可以承诺，即使是Meta也无法看到里面的内容。

You can have your virtual machine and have all this information in your Muse, and we can make the commitment that even Meta cannot see the content that is in there.

这在技术上是可行的。

You can do this technically.

这真是令人难以置信的

It's it's a it's like an incredible

这个承诺在技术上是可以验证的。

of The commitment can be technically verified.

是的。

Yeah.

而且我们会在接下来的几周里发布更多相关信息，因为我们基本上会更广泛地推出这个功能。

And that's something that we're going to publish more about in the coming weeks as we basically get closer to rolling this out a lot more widely.

在这些实验室用这些代理做的所有早期虚拟机工作中，你觉得你们做的是独一无二的吗？

And out of all these early VM efforts that these labs are doing with these agents, you think this is unique, what you're doing?

我觉得没有人做过这个。

I don't think anyone is doing it.

我觉得没人这么做。

I don't think anyone's doing it.

我的意思是，我觉得我们还采取了其他很多安全措施，我觉得我们应该好好聊聊这些，因为即使这个还没准备好，那些措施也已经存在了。

I mean, I think that there's a lot of other security measures that we're putting in place that I think we should talk through because even before this is ready, there's that.

而且即使有人不想用这个，它也非常安全，因为我们从一开始就专注于这一点。

And even people who don't want to use this, it's incredibly secure because we focused on this from the beginning.

但还有自动批准功能。

But There's also the auto approve.

你得看看它是什么。对。

You have to see what it's Yeah.

那我们一会儿再聊这些。

So let's get into all that stuff in a second.

但据我所知，没有人有像Muse那样的机密虚拟机系统。

But I'm not aware of anyone having anything close to the confidential VM system that Muse has.

我觉得这非常根本，因为你想知道这是你的代理。

And I think it's a very fundamental thing because you want to know that this is your agent.

对。

Yeah.

而且，如果你把内容放进去，基本上可以放心，别人是拿不到这些内容的。

And that that if you put content in there that that basically you can trust that no one else is gonna get access to it.

那有哪两种方式呢？

So what are the two ways to do it?

今年早些时候，像Open Claw这样的东西出来的时候，很多人开始买Mac Studio。

Well, lot of people earlier in the year when stuff like Open Claw came out, they start getting Mac Studios.

对。

And Yeah.

你知道，一种让人放心的方式就是，你实际上把设备放在家里运行。

You know, one way to feel good about it is where you literally, like, you physically have your device running in your home.

但另一种方式是，你构建一种……

But the other way to do it is you build a a kind of...

那会有点棘手，因为那里有……

That's gonna be tricky because there there...

我觉得不会有好几十亿人去买Mac Studio，然后自己配置好放在家里用。

I don't think there are gonna be billions of people who are gonna buy Mac Studio and configure it and run it at their home.

尤其是现在内存价格这么高。

Especially with RAM prices right now.

对。

Yeah.

但是别——

But don't

但这也确实，技术上很难。

but it's also just it's, technically difficult.

是的。

Yeah.

对吧？

Right?

就像……

It's like what...

我们做Muse的一部分目标，就是打造一个真正好用的个人助理体验，能让我家里那些技术水平参差不齐的每个人都能用上。

Mean, part of what we're trying to do with Muse is build a version of that personal agent experience that just works that I can like give to everyone in my family of various levels of technical literacy.

对。

Yeah.

而且它真的很好用。

And like it just works.

一天之内，它就能帮他们处理生活中所有想做的事情。

And within a day, it's like doing kind of all the stuff that they that they want in their life.

所以，这其中的一部分就是，你不想让用户自己去配置电脑或虚拟机。

So part of that is like you don't want someone to have to set up their own computer or VM.

你只想在云端轻松部署，但同时又希望它具备像把设备放在家里桌子底下那样的安全性和保密性。

You just kind of like wanna be able to provision it in the cloud, but you want it to have the security and confidentiality that you'd have if you had the box sitting under your desk, like in your house.

当然。

Sure.

所以我觉得这是非常根本的一点。

So I think that's a very fundamental thing.

就像你说的，还有其他部分，因为不是每个人都会用那个。

Like you said, there's other pieces too because I mean, not everyone is gonna use that.

我是说，建一个安全的凭证存储。

I mean, built a secure credential store.

对吧？

Right?

你没有理由把信用卡什么的都公开存放。

There's no reason for you to just like store out in the open, like all your your kind of your credit card and your It

它生成一次性卡号，还有

creates one time card numbers and all

那些东西。

that stuff.

所以你的代理不应该知道那些东西。

So your agent, like, shouldn't know that stuff.

它只需要在需要的时候能访问，因为你让它登录某个东西，而且不...

It should just be able to kind of access it when it needs to because you've asked it to log into a thing and and no...

不然的话就不行。

Not otherwise.

其实这不是单一的东西。

It's actually not a single thing.

你有你的核心代理，但我们也构建了所有这些哨兵代理，它们基本上监控你的代理发送的进出流量和数据，目的是在你可能需要审查某些内容时提醒你。

You have your kind of core agent, but we also built all these sentinel agents that basically monitor the incoming and outgoing traffic and data that your agent is sending for the purpose of flagging to you when you might want to review something.

所以哨兵代理做的就是，它们实际上查看，试图看看是否有人在尝试进行提示注入。

So that's all that the sentinels do is effectively they of look at, they try to see if someone's trying to do a prompt injection.

它们会检查，好的，你的Muse代理是否分享了某些即将发送出去的内容，而这些内容可能让你感到不舒服？

They look at, okay, did your Muse agent share something that is going out that is not something that you might be comfortable with?

如果是这样，那么哨兵代理基本上有权触发这种人工介入的审查。

If so, then the sentinel agent is basically empowered to trigger this human in the loop review.

所以如果你要登录某个系统，如果你要进行支付，如果你要传输敏感信息，基本上每次都需要你批准。

So if you're going to log into something, if you're going to do a payment, if you're going to transfer sensitive information, you basically each time need to approve it.

你可以告诉它，我一般对这类事情没问题，比如这类事情总是允许的。

And you can tell it, I'm good with stuff like this in general, like always allowed this kind of thing.

但总的来说，Muse代理自己无法做出这些判断。

But in general, the Muse agent can't make those judgments itself.

这一点已经深深融入了系统和架构之中。

And that's built into the system and the architecture in a pretty deep way.

而且即使我们做连接器这类东西，比如你把它连接到你的邮箱。

And then even when we do things like the connectors, you connect it to your email.

我觉得有些人在设计这个的时候，他们就直接做成这样：好，你一连上，就什么都能访问了。

Think some people when they've designed this, they just make it so, Okay, you connect and now you have access to everything.

但我们采取的方法是，比如你连接邮箱，应该先设为只读。

But the approach that we've taken is like, alright, if you connect to your email, like, should start read only.

对吧？

Right?

然后如果你想要它能发邮件，那也行。

And then if you wanna, like, be able to have it send an email, then fine.

那就专门去跟它申请这个权限。

Like, go go ask it for that specifically.

尤其是大多数尝试这个产品的人，可能以前从没试过类似的东西。

Especially most people trying this have probably never tried a product like this.

所以，是的。

So it's Yeah.

所以，对我们来说，一个核心设计原则就是最小权限。

So this is like a core design principle for us is basically least privilege.

对。

Yeah.

对吧？

Right?

所以就是……

So just...

是的，你会让它做很多事情。

Yes, you're gonna ask it to do a lot of things.

每一步，都只给它完成当前任务所需的最小权限，只有在必要时才增加权限。

And each step along the way, get access to the least privilege that you need and only add to that as necessary.

所以这在产品设计中是非常核心的部分。

So this is very fundamental in the design of the product.

如果你看看市面上其他的智能体，我觉得没人能接近我们在这个产品里构建的复杂度和深度。

And if you look at all the other agents that are out there, I think like no one else is anywhere close to the level of kind of sophistication or depth that we've built into this.

这同样也受到了我们打造WhatsApp的经验影响，当时我们构建了全球领先的端到端加密系统，并且深知建立一个连Meta都无法查看内容的系统有多重要。

And again, it's sort of informed by our experience building WhatsApp into this like, state of the art end to end encrypted system around the world and the importance of building a system where even Meta can't see the content.

所以，让Moxie回归团队，是的。

And, so kind of getting the band back together and having Moxie Yeah.

像架构设计这块，我觉得是基础中的基础。从某些角度来说，可能大家没想到Meta会把重心放在这儿，但你知道的，我们确实……

Like, architect this has been, I think, one of the foundational things that in some ways it may not be what what some people would would think Meta would focus on, but but, you know, we're we're kind of...

挺有意思的。

You know, it's interesting.

我们是两件事并行。

We're two things.

我是说，社交媒体嘛……

I mean, social media is...

其实也不是说……

It's like not...

你知道，它本质上就是关于分享的。

You know, it's inherently about sharing.

对。

Mhmm.

但还有其他一些东西本质上关乎隐私和敏感内容，而这两方面我们都做得不错。

But then there's all these other things that are inherently about kind of privacy and sensitive context, and we've done well at both of those.

是的。

Yeah.

所以我觉得这更像是后者。

So I think that this is this is more the latter.

所以，专注于如何处理这些内容将会非常重要。

It's gonna be very important to to just, be extremely focused on how we handle that that content.

Mercury 是一种为像我这样的初创公司打造的现代银行服务。

Mercury is a modern take on banking built for startups like mine.

当我决定创办我的媒体公司时，Mercury 是让我能快速上手的最直接、功能最全的银行解决方案。

When I decided to start my media business, Mercury was by far the most straightforward full featured banking solution for me to set up quickly.

界面直观又简洁，每天都能帮我省下宝贵的时间。

The interface is intuitive and simple, saving me valuable time every day.

我用 Mercury 来跟踪支出、账单和发票。

I use Mercury to track my spending, bills and invoicing.

我喜欢它能让我把权限分配给团队，这样他们就能完全按照我想要的方式帮我打理事务。

I love that I can delegate permissions to my team so they can keep things running for me in exactly the way I want them to.

我最喜欢的是 Mercury 在 AI 方面的前瞻性。

My favorite part is how forward looking Mercury is with AI.

传统银行还停留在过去，而 Mercury 是为现代软件的工作方式而设计的。

Legacy banks are stuck in the past, but Mercury is built for how modern software works today.

我用它内置的指令助手来分析现金流，帮我转账。

I use its built in command assistant to analyze cash flow and help me move money.

而且 Mercury 还能连接其他 AI 工具，比如 ChatGPT 和 Claude。

And Mercury also connects to other AI tools like ChatGPT and Claude.

我经常用这个功能，而且Mercury的人还告诉我，我是这个功能用得最多的用户之一。

I use this feature all the time and the folks at Mercury actually let me know that I'm one of the top users of it.

所以相信我，现在终于可以随时随地轻松获取你业务的实时财务数据了。

So trust me, it's finally easy to get real time financial data about your business wherever you need it.

访问mercury.com了解更多信息，几分钟内即可在线申请。

Visit mercury.com to learn more and apply online in minutes.

Mercury是一家金融科技公司，不是FDIC保险银行，银行服务由Choice Financial Group和column NA members FDIC提供。

Mercury is a FinTech company, not an FDIC insured bank, banking services provided through Choice Financial Group and column NA members FDIC.

我经常在会议之间切换，常常还没来得及消化上一个会议的内容，下一个就开始了。

I spend a lot of time context switching between meetings, often with no time to process one before the next starts.

幸好，Granola一直在后台运行。

Thankfully, Granola runs in the background the whole time.

这是一个易于使用的AI会议记事本，适用于各种场合。

It's an easy to use AI notepad for meetings that works everywhere.

即使在电话会议中，我也用Granola来回忆会议内容并生成有用的摘要。

Even on phone calls, I use Granola to recall what was said in meetings and create helpful summaries.

我每天都用它来掌握团队里需要完成的事情。

I use it every day to stay on top of what I need to get done with my team.

它会连接我的邮箱，并建议后续跟进，让我快速查看和发送，节省了很多宝贵时间。

It connects to my email and suggests follow ups for me to quickly review and send saving me valuable time.

Granola 不仅仅是我工作流程中的核心部分，它基本上就是我的第二大脑。

Granola isn't just a core part of my workflow, it's basically my second brain.

去 granola.aisources 试试 Granola，并使用促销代码 sources 可以享受三个月的免费服务。

Try granola at granola.aisources and use the promo code sources for three months off.

AI 的用处取决于它拥有的上下文。

AI is only as useful as the context it has.

但当这些上下文分散在各种工具、讨论串和私信中时，你的团队和 AI 代理就像在盲飞。

But when that context is scattered across tools, threads, and DMs, your team and your AI agents are flying blind.

这正是 Atlassian 的 Jira 要解决的问题。

That's the problem Jira by Atlassian solves.

你的项目关联的目标是什么？

What's the goal tied to your project?

上周在Slack私信里决定了什么？

What got decided last week in Slack DMs?

Atlassian的团队协作图谱把Jira、Confluence、GitHub、Slack等平台上的所有重要信息整合在一起，确保没有任何遗漏。

Atlassian's teamwork graph pulls all of the valuable pieces together from Jira, Confluence, GitHub, Slack, and more, so nothing falls through the cracks.

结果准确率提高了44%，而token使用量减少了48%。

You get 44% more accurate results with 48% less token usage.

使用Jira，你可以轻松地将工作上下文分享给你已经喜欢的AI代理，比如Claude、Cursor和GitHub Copilot。

With Jira, you can easily share your work context with the AI agents you already love, like Claude, Cursor and GitHub Copilot.

直接给他们分配工作，或者通过MCP连接你的工具。

Assign them work directly or connect your tools through MCP.

所有这些让你不必再花时间翻找无尽的链接和消息，追查谁做了什么决定，而是把更多时间花在真正交付上。

All of this lets you spend less time digging through endless links and messages, chasing down what got decided and by who, and spend more time actually shipping.

了解更多请访问jira.com，就是jira.com。

Learn more at jira.com, that's jira.com.

Framer是一款AI网站构建器，它将代理带入你设计、管理和发布网站的同一画布中。

Framer is the AI website builder that brings agents into the same canvas where your website is designed, managed, and published.

这样你就能在不牺牲品味或控制权的情况下，更快地推进。

So you can move faster without giving up your taste or control.

Framer 为 Sources 播客网站提供技术支持，网址是 podcast.sources.news，你可以在那里找到新剧集、文字稿以及更多内容。

Framer powers the sources podcast website at podcast.sources.news, where you can find new episodes, transcripts, and a lot more.

想了解如何从你的网站中获得更多价值，可以咨询 Framer 专家，或者今天就免费开始建站，访问 framer.com/sources 即可享受 Framer 专业版年付计划 30% 的折扣。

Learn how you can get more out of your site from a Framer specialist or get started building for free today at framer.com/sources for 30% off a Framer pro annual plan.

网址是 framer.com/sources，享受 30% 折扣。

That's framer.com/sources for 30% off.

framer.com/sources。

Framer.com/sources.

可能适用相关条款和限制。

Rules and restrictions may apply.

你觉得这是一个赢家通吃的市场吗？

Do you think this is a winner take all market?

这个个人代理市场？

This personal agent market?

我觉得会有不少。

I I think that there's gonna be quite a bit.

我的意思是，人们认为赢家通吃的市场通常并不是这样。

I mean, things that people think are winner take all usually aren't.

所以我觉得……

So I think it's...

我觉得那是……

I think that's

你在网络效应这个领域里经历过，它们确实非常持久，虽然不是赢家通吃，但最终会有几家做到真正的规模。

As you've dealt in this business of network effects that they're incredibly durable, it's not winner take all, but it ends up being several at real scale.

你，你……

You you...

数量不多。

There's not a ton.

嗯，有……

Well, there's

很多。

a lot.

我的意思是，有……

I mean, there's...

对。

Yeah.

对。

Yeah.

你可能用两只手就能数过来有多少产品拥有超过20亿用户。

You could probably count on two hands how many products have over 2,000,000,000 users.

对吧？

Right?

就像，这……

Like, it's...

规模带来规模。

The scale begets scale.

我在想，

And I'm wondering if...

你是怎么看待个人代理这类东西的。

How you're thinking about, like, personal agents.

就是说，这是不是一种完全不同的模式，会有很多……

Like, is this a totally different paradigm where it's gonna be many...

每个人都会有各种各样的

Everyone has all kinds

代理？

of agents?

这是一个非常深奥的领域。

It's a very deep area to work in.

对。

Yeah.

所以我猜，可能不会有超过十几家公司有足够的技术水平去在这个领域做最前沿的工作。

So my guess is that there probably aren't going to be more than a dozen companies that have the sophistication to go do state of the art work in that.

所以我觉得，不管有没有网络效应，通常都会有一种幂律分布，就是如果你在某方面做得最好，通常你会获得大量的使用量。

So I think whether there are network effects or not, there's usually some kind of power law distribution around if you're the best at something, usually you end up getting a lot of usage.

而且我觉得很多细微差别最终来自于，其实人们关心的用途有很多种。

And I think a lot of the nuance ends up coming from, well, it turns out there there are all these different uses that people care about.

所以你可以擅长不同的事情。

So you can be the best at different things.

我们会尽量在尽可能多的方面做到最好，但是，你知道，帮你处理人际关系的东西，和那个最擅长帮你创业的个人代理，最终会不会是两回事呢？

And we will try to be the best at as many of these things as possible, but, you know, does building the thing that helps you with your relationships end up being a somewhat different thing than the personal agent that is the best at helping you build a small business?

也许吧。

Maybe.

对。

Mhmm.

我的意思是，我觉得我可以争辩说，也许Meta在赢得这两方面都很有优势。

I mean, I think that that I could argue, maybe Meta's very well positioned to win at both of those.

我的意思是，我们服务了数亿的小企业，也服务了数十亿的用户。

I mean, we serve hundreds of millions of small of small businesses, and we serve billions of people.

所以也许我们能在这些事情上都做到最好，但可能有些类别Meta不会是最强的。

So so maybe we can be the best at both of those things, but there are probably categories that Meta isn't going to be the best at.

那问题就是这些类别有多大。

And then the question is just how big are those?

对。

Right.

我猜即使能在云端拥有这种非常安全的机密虚拟机，可能还是有人想要家里的Mac Studio。

I would guess that even with the ability to have this, like, very secure confidential VM in the cloud, there are probably gonna be some people who still want the the Mac Studio at home.

是的。

Yeah.

但那会有多少人呢？

But but how many is that gonna be?

对吧？

Right?

那会是……

It'll be...

可能也就几百万吧，但我怀疑不会有几十亿。

Like, maybe it's millions, but I doubt it's billions.

对。

Yeah.

所以我觉得问题就在于不同的人追求什么，我们会尽力把它做到最好。

So it's I think there's just the question of, like, what different people optimize for, and we'll try to make this as good as possible.

但我确实认为，如果我们能做出一个对大家日常生活都非常有用的东西，我觉得Meta最擅长的就是把一个对消费者有效的产品推广给很多人。

But I do think that if we build something that ends up being just very useful for people generally in their day to day lives, One of the things that I think Meta is the best at is taking a product that works for consumers and distributing it to a lot of people.

当然。

Sure.

所以我觉得这就是...

So that is one that I think is...

一旦我们把它做顺了，我觉得我们就能把它带给几亿人，最终达到几十亿人。

Once we get this humming, I think we will be able to get this in front of, you know, many hundreds of millions of people and eventually billions of people.

而且我觉得这件事我们能做得很好。

And I think that that's something that that that we can do quite well.

那它是和Meta AI共存的，还是你觉得它们是分开的？

And it coexists with Meta AI, or do you see those as separate?

我

I

觉得是这样。

think so.

我们得看以后的发展。

We'll see over time.

我觉得现在它们风格上有点不一样。

I mean, I think right now they have somewhat different flavors.

我自己两个都用。

I mean, I use both of them.

Muse更偏向对话式，它会把你的问题理解成想了解你，以及你长期可能想要什么。

I mean, Muse is, know, it's more conversational, and it kind of interprets the questions that you ask it more as trying to understand you and what you might want over the long term.

所以如果你问它什么，它更可能会根据你说的话，自己去长时间地做一件事。

So it's more likely if you ask it something for it to just go off and work on a thing for a long time based on a thing that you said.

有时候你，你知道，你……

Where sometimes you, you know, you...

有时候你只是问个问题。

Sometimes you're just asking a question.

对。

Yeah.

没错。

Right.

然后你想要一个非常……

And you want like a very...

它给出了一个答案。

That gave an answer to it.

所以我觉得那才是我用Meta AI更多的情况。

And so I think that's more the type of thing that I use Meta AI for.

好的。

Okay.

不过我们走着瞧吧。

But we'll see.

也许那……

Maybe that...

也许它们最终会趋同，但我不太确定。

Maybe they'll they'll converge over time, but but I'm I'm not sure.

Semi分析，我不确定你有没有看到。

Semi analysis, I'm not sure if you saw it.

他们七月份发了一篇挺看好你的文章。

They had a pretty bullish piece about you in July.

他们说Meta在模型进展方面最有希望赶上OpenAI和Anthropic，处于前沿水平。

They said that Meta has the best shot at catching open AI and Anthropic on the frontier in terms of model progress.

而且，我觉得有一句话挺有意思的。

And, interesting quote, I thought.

大意是说，对MSL来说，重要的是斜率，而不是截距。

It was like, matters for MSL is the slope, not the intercept.

然后我看到了人工智能分析最近的一张图表，显示你们最新的模型Muse Spark，仅次于Claude。

And then I saw there was this recent chart by artificial analysis, which was showing the latest model you guys have, Muse Spark, behind only Claude.

我想是Fable 5.1和Opus五。

I think it was Fable 5.1 and Opus five.

这是最近的。

This was very recent.

所以你们在模型上取得的进展正在加速。

So the progress you guys are making on the models is picking up.

过去一年我们一直在讨论这个。

And we've been talking about this over the last year.

去年你重启了实验室。

And you rebooted the lab last year.

这在内部实际上是怎么发生的？

How has that practically happened internally?

你认为这些进步归功于什么？

What would you attribute the gains you're seeing to?

是文化的原因吗？

Has it been culture?

比如说，什么……

Like, what...

是的。

What's Yeah.

我的意思是，当我们创建Sure的时候，我们重组了团队。

I mean, well, we rebooted the team when we created Sure.

Meta当时非常公开。

Meta was very public.

你招了那么多人。

You were hiring all those people.

我的意思是，我对此的看法是，你知道，Meta是……

I mean, the way I've I've thought about this is, you know, Meta is...

它是一家在机器学习领域长期领先的公司。

It's a company that is a leader in machine learning for a long time.

你想想看，像Facebook或Instagram上的信息流、我们的广告系统，还有那个需要找出所有不适合放在网上的内容的审核系统，

If you think about, like, the feeds on Facebook or Instagram or our ad system or the integrity system that, like, needs to find all this content that is...

那些不适合出现在互联网上的内容，

That's, like, unfit to be on the Internet.

基本上全都是机器学习系统，而且我们在这些领域已经打造出了顶尖的领先系统。

I mean, are basically all machine learning systems, and we've built kind of state of the art leading systems in those areas.

所以当大语言模型开始火起来的时候，我们已经有FARE，那个早期研究Llama的实验室，但我们需要把它产品化，把它变成一种更工业化的流程，以便按照扩展定律的预测，把它扩展到更大的规模，从而获得这些成果。

So when Llms started gaining traction, we had FARE, there's a lab that did the early work on Llama, but we needed to kind of productionize that and build it into these this more kind of industrial process for scaling it to be larger as as the scaling laws predicted would would yield all these results.

而且我觉得当时我犯了个错误，就是以为因为我们在其他类型的机器学习上都很擅长，所以构建和扩展大语言模型的方法也会差不多。

And I think at the time, I made this mistake of just kind of assuming that because we were good at all these other types of machine learning, the approach of building and scaling LLMs would be kind of similar to that.

但实际上，这里面有很多非常不同的动态。

And in practice, there are a lot of very different dynamics.

所以我们最初通过Llama 4采取的方法，也就只能走到那一步了。

So the first approach that we took through Llama four, it got us so far.

我的意思是，Llama 3是个不错的模型。

I mean, Llama three was a good model.

我对Llama 4的发展更乐观一些，但结果它却……

I was more optimistic about where Llama four would go and then it just it...

你知道，当我们发布那个版本的时候，我觉得……

You know, when we launched that, I think it's...

我们已经偏离了应该走的轨道。

We were off the trajectory that we needed to be on.

所以就想，好吧，我们得改变一些东西。

So it's like, okay, we need to we need to change something.

但那时候我更加坚信人才密度的重要性。

But that's when I kind of got more religion around talent density.

对吧？

Right?

这不仅仅是一个可以让上千人同时工作、跑实验的系统。

It's like this isn't just a system where you can have, you know, like a thousand people working on it, running experiments.

你其实更希望，在某些方面，用尽可能少的人，他们能把整个事情记在脑子里，像小组科学项目一样一起合作。

Like you really just kind of want, in some ways, almost the smallest group of people that you can, who can keep the thing in their head, who can work together as sort of like a group science project.

如果团队里只有少数几个位置，那么每个位置都找到最合适的人就变得极其重要。

And if there's only a small number of seats on the team, then each seat getting the very best person is incredibly important.

所以我最后花了很多自己的时间在这上面。

So I ended up spending a huge amount of my own personal time doing that.

同时，我也想更贴近技术工作，这样我才能更好地理解并引导公司去做我们更广泛需要做的事情。

And I also, I wanted to be closer technically to the work, so that way I could understand and help guide the company to do the things that we need to do more broadly.

所以我们建了实验室，就建在我办公室座位周围。

So we built out the lab, we built it out literally around where I sit in the office.

所以团队基本上就在那附近。

So it's like the group is kind of around that.

随着我们对工作质量越来越有信心，我们也大幅增加了算力投资。

And we've significantly ramped up the compute investments as we've gained confidence in the quality of the work that we're doing.

我们正在建设很多很多吉瓦的算力，我们期望在这方面领先，也应该领先。

So we're building out the many, many gigawatts of of of compute, you know, we expect to be leaders on on that front, and we should be.

毕竟，我们有很多年、几十年的数据中心建设经验。

I mean, we have many years of experience, decades of experience building out data centers.