7个月不足200人！扎克伯格与亚历山大·王如何打造出爆款Mu

7个月不足200人！扎克伯格与亚历山大·王如何打造出爆款Muse？

发布于 2026-10-13 · 时长 2:14:31

字幕摘录

时间	英文	中文
0:00	At the end of last year, we're like, all right, we are quite excited about the progress.	在去年年底,我们像,好吧, 我们对进展相当兴奋。
0:05	We haven't seen the models yet.	我们还没有看到模型。
0:06	We're going to write this check for, I don't know, $100 billion or $200 billion.	我们要写这张支票,我不知道,一千亿美元或两千亿美元。
0:11	I'm like, Alex, you sure this is going to work?	我很喜欢,亚历克斯, 你确定这会工作?
0:16	Cheers.	欢呼声。
0:21	Let's launch it.	让我们启动它。
0:25	So give me the update on Muse.	给我关于Muse的最新消息
0:27	If you were looking at the third-party reports, you would see that this might be the fastest-growing agent or AI app that I think has existed yet.	如果你在看第三方报告你会发现这可能是我认为已经存在的增长最快的代理或AI应用程序
0:41	What are your guys' personal favorite Muse use cases, either yours or other people's?	你的人最喜欢的是什么Muse的案子无论是你的还是别人的?
0:47	I was actually talking about this at dinner yesterday.	我昨天在吃饭时真的在说这个

展开字幕全文（1943 条）

序号	英文	中文
1	At the end of last year, we're like, all right, we are quite excited about the progress.	在去年年底,我们像,好吧, 我们对进展相当兴奋。
2	We haven't seen the models yet.	我们还没有看到模型。
3	We're going to write this check for, I don't know, $100 billion or $200 billion.	我们要写这张支票,我不知道,一千亿美元或两千亿美元。
4	I'm like, Alex, you sure this is going to work?	我很喜欢,亚历克斯, 你确定这会工作?
5	Cheers.	欢呼声。
6	Let's launch it.	让我们启动它。
7	So give me the update on Muse.	给我关于Muse的最新消息
8	If you were looking at the third-party reports, you would see that this might be the fastest-growing agent or AI app that I think has existed yet.	如果你在看第三方报告你会发现这可能是我认为已经存在的增长最快的代理或AI应用程序
9	What are your guys' personal favorite Muse use cases, either yours or other people's?	你的人最喜欢的是什么Muse的案子无论是你的还是别人的?
10	I was actually talking about this at dinner yesterday.	我昨天在吃饭时真的在说这个
11	My favorite thing is I have my Muse, whose name is Euler, constantly source increasingly obscure teas from increasingly sort of like obscure vendors of tea.	我最喜欢的是我的Muse,她的名字叫Euler, 不断从越来越隐蔽的茶叶中找到越来越隐蔽的茶叶。
12	So at this point, he's on a weekly basis, my Muse is buying some random tea from some farm in Taiwan or some farm in Japan and sourcing it over and bringing it back.	所以此时,他每周一次, 我的缪斯正在从台湾的某个农场或日本的某个农场购买一些随机茶, 并采购过来,带回来。
13	What's your favorite Muse use case?	你最喜欢的Muse用案是什么?
14	I don't know.	师曰.
15	I mean, I use it for a ton of stuff.	我的意思是,我用它做一吨的东西。
16	I still like, my favorite both use case and probably eval for the product is having it coach me in MMA.	我还是喜欢,我最喜欢的既使用箱子, 并且也许 eval产品是它让我在MMA教练。
17	So I put cameras in my gym and I basically, it watches the cameras when I'm in there and it gives me feedback in between the rounds, which is a very hard problem, it turns out.	所以我把摄像头放进我的健身房, 基本上,当我在那里的时候,它会看着摄像头, 它会给我反馈到中间, 这是一个非常困难的问题, 结果发现。
18	Because not only does it have to, it basically has to process all the video quickly enough because you only get one minute in between the rounds.	因为它不仅需要,它基本上必须快速地处理所有视频, 因为你之间只有一分钟的时间。
19	Oh, it coach you between the rounds?	哦,它指导你之间的回合?
20	Yeah, I want feedback on how to win.	是的,我要反馈如何赢
21	Did you have to instruct us?	你一定要教我们吗?
22	I've wondered about this too, because looking at videos is a really good AI use case.	我也想知道,因为看视频是一个非常好的AI使用案例。
23	It feels like you can't just give it a multi-gigabyte video file.	感觉你无法只给它一个多基加比特视频文件.
24	So do you tell us how to slice stills out of it?	那么,你告诉我们如何切出它的静物?
25	No, it basically, yeah, it pulls stills.	不,基本上,是的,它拉静态。
26	I think it kind of figured that out by itself.	我想这是自己想出来的
27	It turns out that Muse has pretty good native understanding of MMA.	原来缪斯对MMA有相当良好的本土理解.
28	So that's good.	所以这是很好的。
29	So like the multi-modal.	所以,像多模式。
30	It's in the model.	在模型内.
31	Someone on the team knew.	团队里有人知道
32	Now we're just kind of out there in the corpus.	现在,我们只是有点在那里的保镖。
33	I think it's a good test of the kind of coding ability.	我认为这是对编码能力的一种良好测试.
34	Because really basically what it's doing, it's like within the agent, I'm not seeing any code,	因为实际上基本上它在做什么, 它就像在特工内部, 我没有看到任何代码,
35	but it basically is writing this whole video processing image processing pipeline.	但它基本上写这整个视频处理图像处理管道。
36	And I'm constantly telling it, all right, you look like you took 30 seconds to give me that feedback.	我不停地说,好吧,你看起来像花了30秒给我的反馈。
37	I'm going to have a 60-second break.	得六十刹那间.
38	You've got to go faster.	你得快点
39	Just optimize yourself, optimize yourself.	只是优化自己,优化自己。
40	So it's I think a good example of how stuff is going to go in the future,	所以我想一个很好的例子说明将来会怎样
41	where basically everyone is actually going to be writing tons of code.	基本上每个人都会写很多代码
42	They're just not going to know that they're writing code.	他们不会知道他们在写密码
43	What kind of MMA steers does a model give you?	什么样的MMA导师会给你一个模型?
44	You want to see?	你想看吗?
45	Basically, let me see if I can pull some stuff out.	基本上,让我看看,我能不能拉出一些东西。
46	But it's basically it gives me a lot of stuff about balance and like hip position,	但它基本上它给了我很多的东西关于平衡和像臀部位置,
47	or it's like, oh, you're like when you're kind of inside control here,	或者它就像,哦,你就像当你有点内部控制在这里,
48	it's just like you're like you didn't get this underhook in this place,	就像你没有在这个地方得到这个底线一样
49	so you're vulnerable to this getting rolled over here.	所以你很容易被卷进来
50	I'll pull up some stuff.	我会拉一些东西。
51	It was definitely one of his first use cases actually of even the initial prototypes.	这绝对是他最先使用的案例之一甚至是最初的原型。
52	Yeah.	对
53	我和OpenClob合作过还没让Muse工作
54	It was just like and then that's what I was saying is it's kind of like become an eval for this too.	这就像,然后,我是说, 它有点像成为 eval的这一点。
55	I actually didn't know that it was an eval for the product, so noted.	我其实并不知道那是产品的精华所以注意到了
56	How is all this in the model's training data?	模型的训练数据如何?
57	Like the fine-grained, this is what an MMA fighter looks like when they're appropriately balanced,	和精致的战士一样这就是MMA战斗机在平衡时的样子
58	and this is when they're like too much on the back foot,	这是当他们像太在后脚,
59	or just I'm wondering how has it consumed so many images of appropriate versus inappropriate MMA posture?	或者我想知道它是如何消耗这么多的图像适当和不适当的MMA姿态?
60	Here you could say it's just like it just pulls out frames from it and it's like,	在这里你可以说,它就像它只是拉出框从它,它就像,
61	I'm like too much on my back leg there.	我像太多我的后腿在那里。
62	But is this the emergent intelligence like it does actually,	但是,这是新兴的智能就像它实际上,
63	again, does it know from seeing lots of images of MMA,	再次,它知道从看到许多MMA的图像,
64	or does it know because it understands the concept of where you're meant to be balancing its red books about MMA,	或者它知道,因为它理解的概念在哪里你应该平衡它的红色书关于MMA,
65	and then it sees that you're too much on the back foot or whatever?	然后它就会发现你脚后脚太重了?
66	Well, we did train the models with a lot of visual perception data.	我们用很多视觉感知数据训练模型。
67	I think that that's like kind of making good multimodal models.	我认为这就像一种制造良好的多式联运模式。
68	This gets into like I think a kind of core approach thing that we have,	这就像我觉得一种核心方法的东西我们有,
69	which is we're training the models end-to-end to be good personal agents.	这就是我们训练模型端到端成为好个人特工。
70	The question is like why would Meta want to build these models?	问题是,梅塔为什么要建造这些模型?
71	First of all, I think just super intelligence overall is going to allow us to run the company better	首先,我想只是超级智能整体将使我们能够更好地管理公司
72	and build all kinds of different things.	并造种种异物.
73	But for the products that we're building,	但对于我们制造的产品,
74	there are certain capabilities and certain mixes of skills that basically are just much more applicable to these cases.	有些能力和技能组合基本上更适用于这些案件。
75	This is one of the things that Alex focused on from the beginning was like,	这是亚历克斯从一开始就关注的问题之一
76	we really want to make sure that our multimodal program is good	我们真的想确保我们的多模式方案是好的
77	because that's disproportionately important for the types of things that people use all these services for.	因为这对人们使用所有这些服务的种类非常重要
78	How do you train a good multimodal model?	如何训练好多式联运模式?
79	Is this where Meta just has an advantage because you guys have so much image and video data?	这就是Meta有优势的地方因为你们有这么多的图像和视频数据?
80	It's very helpful.	这很有帮助
81	Yeah, and I think ultimately like a lot of this it comes down to how are you focused on it, right?	是啊,我想最终会像很多这样的问题... ... ......
82	Like I think because of the fact that we wanted to build personal super intelligence,	我想是因为我们想建立个人超级智能
83	and fundamentally I think most people in the world really experience the world in a highly multimodal way.	我基本上认为,世界上大多数人都以高度多式联运的方式体验世界。
84	They like talking to other people and listening to things and seeing videos and filming things of their own.	他们喜欢和别人交谈,倾听事物,看到视频,拍摄自己的事物.
85	I think most consumers have this highly multimodal experience,	我认为大多数消费者都有这种高度多式联运的经验,
86	and especially earlier this year I think there was actually a point of tension almost in the industry	特别是今年早些时候我觉得这行业里几乎有一点紧张
87	because I think so much of the consensus view within the AI industry was that it was just all about coding	因为我认为AI行业的共识在于它只是关于编码
88	and coding was all that mattered, and I think that for the success of a product like Muse	编码才是最重要的我想为了Muse这种产品的成功
89	and the success of a lot of these consumer use cases it really depends on the ability to be strong and multimodal.	以及许多消费者使用案例的成功真正取决于他们是否有能力强大和多式联运。
90	Yeah, I think we're just kind of scratching the surface of that.	是啊,我想我们只是那种刮的表面。
91	I mean a lot of the initial theory that we had was that most people don't actually like reading that much, right?	我的意思是,很多最初的理论我们曾经是,大多数人不喜欢阅读这么多,对不对?
92	So even the kind of text-based interaction model I think it's something that people do.	所以,即使是那种基于文本的交互模型, 我认为这是人们做的事情。
93	And when you think about stuff like the glasses that we're working on or even our other products that are very visual,	当你想到像眼镜一样的东西我们的工作,甚至我们的其他产品非常视觉,
94	the glasses will eventually just be a pure multimodal product.	眼镜最终会只是一种纯粹的多式联运产品。
95	It's like you can let it hear what you are listening to and it can talk to you and it can see what you see if you let it,	就像你可以让它听到你正在听的, 它可以和你说话, 它可以看到你看到的,如果你让它,
96	and then eventually, I don't even know if we're going to have an app store on the glasses,	最后,我甚至不知道我们是否要有一个应用商店在眼镜上,
97	it'll just be like a multimodal generation that kind of creates whatever you need right in your vision.	它只是像一个多模式的一代那种创造任何你需要的东西在你的视觉。
98	And I imagine something like that will probably end up powering a lot of the kind of apps that we build too.	我想象这样的东西可能最终会给很多我们所建造的应用程序提供动力。
99	But yeah, I mean we're still in the early days of that, but for now I mean the perception work has been really impressive.	但是,是的,我的意思是,我们仍然在早期, 但是现在,我的意思是, 感知工作真的很令人印象深刻。
100	Yeah.	对
101	I want to ask you guys about the product space here in AI and try something out on you.	我想问问你们在AI的产品空间并试一下你们。
102	If I was to describe to you, say with Stripe, what I think a winning product looks like,	如果我要跟你描述一下和史黛丽说说我觉得一个赢的商品看起来像什么
103	it would take me pages and pages to talk about because we have the developer experience	这需要我的页面和页面谈论因为我们有开发者的经验
104	and we have to talk about all the international heterogeneity of payment methods,	我们必须谈谈所有国际支付方法的异样性,
105	then we have to talk about stablecoin, it would take me a long time.	然后我们得谈谈马钱, 这会花我很长时间。
106	Whereas I notice a phenomenon in AI where it feels like the winning products collapse very quickly	我注意到AI里有个现象感觉胜出的产品很快就崩溃了
107	onto an agreed-upon product space that everyone is pursuing.	进入大家追求的商定产品空间。
108	And so Transformers worked and then like the chat agent UI really worked.	于是变形金刚工作,然后就像聊天代理UI真的工作.
109	And now we are seeing, you mentioned OpenClaw earlier, the kind of OpenClaw persistent agent,	现在我们看到,你刚才提到OpenClaw, 那种OpenClaw坚持代理,
110	has a computer, able to do things, computer use out there in the world view,	有一台电脑, 能够做一些事情, 电脑使用世界观,
111	like we're collapsing on that product space.	就像我们崩溃在产品空间。
112	But there's something about AI where the winning products tend to be very simple,	但AI有些东西获胜的产品往往很简单
113	pursued by all the companies that are kind of taking it seriously.	所有公司都认真对待
114	You could describe on the back of a postcard, like you could describe very concisely,	你可以在明信片背面描述就像你可以描述得非常简洁
115	and obviously implementing them is massively complicated, like we'll get into the new model.	很明显地,执行它们是巨大的复杂, 就像我们将进入新的模式。
116	There's huge amounts of frontier work that goes into making a good version,	有大量的前沿工作进入一个很好的版本,
117	but the winning, now you're talking about the glasses or the future versions,	但获胜者,现在你说的是眼镜或未来的版本,
118	but somehow there's a very small number of clearly correct products in AI,	但不知何故,AI中明显正确的产品数量很少,
119	as opposed to it being this very broad and complicated product space.	而不是它是一个非常广泛和复杂的产品空间。
120	I kind of disagree.	我有点不同意
121	Sure.	当然
122	I mean, well, first of all, I think part of the theory from the beginning was that this kind of agent	我的意思是,首先,我认为理论的一部分从一开始是,这种代理
123	or whatever form it took was going to be quite a different thing from this text-based assistant.	不管是什么形式都会和这个基于文字的助手大不相同
124	Every company has one of those.	每个公司都有一个
125	Ours is MetAI, it reaches a ton of people, but it's the same thing that all these companies have.	我们的是MetAI,它覆盖了一吨人, 但这是同样的东西所有这些公司有。
126	We, I think, specifically think that that is used for very informational purposes,	我认为,我们特别认为,这被用于非常信息化的目的,
127	but people don't think about it as a thing that gets stuff done for them.	但人们不认为这是一件事情得到的东西为他们做。
128	It's retrieval, not delegation or something.	这是回收,不是授权什么的。
129	Yeah, and even when you say it has a computer, yeah, but there's lots of different ways to do that.	是的,甚至当你说它有一台计算机, 是的,但有很多不同的方式做到这一点。
130	I think that it's only this most recent wave that gives the agent a dedicated VM,	我认为只有最近一波才给特工一个专注的VM
131	which I think is actually incredibly important for making you feel confident	我觉得这对让你感到自信很重要
132	that the security and privacy of your information is not connecting to some elastic compute that everyone uses.	您的信息的安全和隐私不是连接到某种弹性计算每个人都使用。
133	But my claim isn't that it's easy to implement, nor that it's obvious a priori.	但我的主张不是说它易于执行,也不是说它显然是先验的.
134	I think there are going to be a lot of things.	我想会有很多事情发生
135	Okay, you think there's going to be a lot of different things?	好吧,你认为会有很多不同的事情?
136	Because again, my claim is that now that people see Muse, everyone looks at that and says,	因为,我的要求是,现在人们看到穆斯, 每个人都看着它并说,
137	oh yeah, that's clearly how you want the AI agent product to work,	哦,是的,这很明显你希望AI代理产品的工作,
138	and that the future AI agent products that we'll see will be fairly Muse-shaped.	而我们将会看到的未来AI代理产品将相当的Muse形状。
139	So I think my take on this is I think we're just so early in the realm of what AI products will look like	所以,我想我的假设是,我认为我们只是这么早的领域 AI产品会是什么样子
140	that I think a lot of the interesting differentiation about the products kind of gets collapsed in how we view the products today.	我认为很多有趣的区别产品的类型崩溃在我们今天如何看待产品。
141	I think the traditional way in which we wrote down and think about product specs,	我认为传统的方式我们写下来思考产品规格,
142	those aren't the relevant dimensions and levers and axes to consider the important design decisions on AI-native products.	这些不是考虑AI本地产品重要设计决定的相关维度和杠杆和轴.
143	And so Muse, for example, I think so much attention was put in the design of the model for the product,	举例来说,Muse, 我认为在产品模型的设计中, 已经引起太多的注意,
144	and there were a lot of very intentional decisions around in all these situations,	在所有这些情况下有许多非常有意的决定
145	how exactly should the model behave and what's the exact personality of the model in these situations	模特儿应该怎样表现模特儿在这种情况下的个性是什么
146	and how should it handle these sorts of tool use scenarios.	以及它应该如何处理这些类型的工具使用情景。
147	And these are very intentional decisions, and probably in the same way that the full spec of how we wanted the model to behave	这些决定都是有意的, 可能和我们想要模型的行为一样
148	in all these scenarios would be an extremely long document.	在所有这些设想中,这将是一份极为冗长的文件。
149	But because of just how we think about products today, it's just AI.	但由于我们今天对产品的看法,只是AI而已.
150	It's kind of like how when mobile first started happening, it was like, oh, it's just an app and everything is just an app.	这有点像手机刚开始的时候, 它就像,哦,它只是一个应用程序,一切都只是一个应用程序。
151	Well, it turns out there's a lot of nuance in what exactly an app is,	嗯,它发现有很多细微的具体是一个应用程序,
152	and we're kind of in this phase where, oh, it's just AI or it's just an agent or it's just an agent with a VM or something like that.	我们正处于这个阶段, 哦,它只是AI 或者它只是一个代理或者它只是一个代理与VM或类似的东西。
153	But the interesting differentiations we have to tease out.	但是,有趣的区别,我们必须取笑。
154	OK, so you guys think I'm making the classic mistake, which is like my domain is complex and this domain is, again, from a product point of view, simple.	好了,所以你们认为我犯了一个经典的错误, 这就像我的域是复杂的, 这个域也是,从产品的角度来说,很简单。
155	It's like all payment processors seem the same.	就像所有付款处理器都一样
156	I really agree with what you're saying.	我非常同意你说的
157	One thing that we talked about from early on is that basically there is this pattern where I think the products are getting thinner	我们从早到晚谈的一件事是基本上有这种模式我认为产品越来越稀薄
158	and the model is getting to be more of the product.	模型越来越成为产品。
159	So one of the things that we say internally is like the model is the product.	所以,我们内部所说的一件事就是, 模型就是产品。
160	And when you're talking about an agent, a lot of the complex product work that went into Muse	当你谈论一个代理, 很多复杂的产品工作进入Muse
161	is basically around the privacy and security model for handling your data and the things that get connected to it.	基本上围绕着处理你的数据的隐私和安全模式以及与之相连的东西。
162	And that, I think, is something that we're just very proud of that we really focused on.	我认为,这是我们非常自豪的事情, 我们真正专注于它。
163	But that's a lot of the product surface that we need to innovate on.	但是,这是很多的产品表面我们需要创新。
164	But the capabilities, the personality, a lot of the stuff is like you're training into the model.	但是能力,个性,很多东西就像你正在训练进入模型。
165	And whether they're the people in the lab who are working on that or the people on Nat's team who are working on the product,	不管是实验室里的人在研究这个还是在纳特团队里的人在研究这个产品
166	I think when Nat is thinking about what is the key stuff that we need to make the product feel good,	我想当纳特想着什么是关键的东西我们需要让产品感觉良好,
167	he's talking about post-training the model, right?	他说的是训练后模特儿,对吧?
168	Not like, I mean, of course, the UI of the product matters too, but the most important part of the UI is the product.	当然,我并不是说产品的UI也很重要,但是UI最重要的部分是产品.
169	Yeah, exactly.	没错
170	I think there's still very viscerally, I think from a user, there's a feeling of using Muse over time.	我认为,从用户的角度看, 仍有一种长期使用Muse的感觉。
171	And maybe some meaningful portion of that is just the reliability of Muse to handle certain tasks.	也许其中的有意义的部分只是Muse处理某些任务的可靠性。
172	But then another meaningful portion is how is it proactive?	但另一个有意义的部分是它是如何积极主动的?
173	And in what context is it proactive?	在什么背景下是积极主动的?
174	And when it talks to you, how do you feel about how it talks to you?	当它和你说话时,你对它怎么跟你说话有什么感觉?
175	And so there's clearly a lot of, and I'm starting to see more and more of this actually show up on the X feed,	因此,显然有很多, 我开始看到越来越多的这个实际出现在X的种子,
176	which are people are starting to tease out these, they're all proactive agents,	他们都是主动的特工
177	but what are the actual meaningful differences between them?	但是它们之间有什么实际有意义的区别?
178	And these, I think, are the interesting product decisions, which the funny thing is, how do you specify these part decisions?	我认为,这些是有趣的产品决定, 有趣的是,你如何说明这些部分的决定?
179	Well, it's ultimately in model design decisions, which ultimately boil down to training methods or data or valves or all these things.	最终是由模型设计决定的, 最终归结为培训方法或数据或阀门或所有的东西。
180	Yes.	对
181	As we talk about product space, one notable Muse detail is that there is no model selector and no effort selector.	在我们谈论产品空间时,一个值得注意的Muse细节是,没有模型选择器,也没有努力选择器.
182	Do you guys think that's just obviously correct in a consumer AI product?	你们觉得消费者AI产品显然正确吗?
183	I just want to make it simple.	我只是想让事情变得简单。
184	Yeah, I think that we've been, I think somewhat infamously, a lot of the AI products have been designed by a highly technical group	是的,我认为我们一直, 我认为有些不光彩, 许多AI产品已经设计了一个高技术组
185	that do not have the sort of randomly sampled American in their mental model.	在他们的心理模型中没有随机抽样的美国人
186	And so the products, I mean, the naming of some of the models is so egregious.	所以产品,我的意思是, 一些模型的命名是如此惊人。
187	The numbers go backwards.	数字往后走。
188	Some of the times the numbers go backwards, and sometimes it's just like a totally different word, and you have another set of words.	有些时候数字倒着, 有时它就像一个完全不同的词, 你有另一组词。
189	And just like, how is somebody who's not carefully reading API design specs supposed to know what they're supposed to pick?	就像,一个不仔细阅读API设计规格的人怎么会知道应该选什么?
190	It's just far too confusing.	这太混乱了
191	OK, we're talking about models a lot.	我们经常谈模特儿
192	Tell me about the new model.	说说新模型.
193	It sounds awesome.	这听起来真棒。
194	Yeah, I mean, I think we're quite proud of this.	是啊,我的意思是,我觉得我们相当骄傲这一点。
195	Yeah, the new model, Muse Wonder, it's a new class of model that is at the frontier.	是啊,新的模型,Muse Wonder, 这是一个新的类型的模型在前沿。
196	I mean, we've basically been working on this since the beginning.	我们从一开始就一直在研究这个
197	Basically, when you're doing these pre-trains, you don't just go to whatever the biggest cluster is that you have.	基本上,当你做这些预赛的时候, 你不只是去最大的集群是你所拥有的。
198	You want to work your way up the ladder, training models that are maybe four or five times bigger, and then you just keep on repeating that process.	你想在梯子上努力, 训练模型, 可能大四五倍, 然后你继续重复这个过程。
199	So the first set of models that we've had, Muse Spark, was all based on this initial pre-train that we did in, what was it, December?	所以,我们拥有的第一套模型,Muse Spark, 都基于这个初步的预列车, 是什么,12月?
200	Yeah, the end of last year.	是的,去年的结束。
201	Yeah, and it's actually, I mean, I think we've all just been quite surprised and impressed with how much intelligence and capability the team has been able to kind of build into a model which is frankly not that big.	是啊,其实,我的意思是, 我想我们只是非常惊讶并给大家留下了深刻的印象, 团队能够建立多少智能和能力成为一个模型, 坦率地说,它不是那么大。
202	And then after that, we were working on this next model, and the pre-train has been done for a bit, and we've been working on the post-training, and now it is here.	之后,我们研究了这个下一个模型, 预训已经做了一点, 我们研究了后训练,现在它就在这里。
203	So it's the next kind of class of model that is significantly bigger, so I think you will kind of feel that in terms of the nuance in which it can think.	所以,这是下一类的模型, 显著的更大, 所以我想你会觉得, 从它能思考的细微差别来看。
204	And it really, a lot of the areas that we focused on were making it the best personal and professional, agentic model.	这真的,我们所关注的许多领域都使它成为最好的个人和专业的代理模式。
205	Some of the different things that you can optimize for, that's it for us.	一些不同的东西,你可以优化, 这就是我们。
206	And then, yeah, I mean, a lot of the basic capabilities that we focused on building in, I mean, very good memory, kind of very good ability to persist through complicated tasks.	然后,是的,我的意思是,很多基本的能力, 我们专注于建设, 我的意思是,非常好的记忆, 那种非常好的能力坚持完成复杂的任务。
207	But, I mean, a ton of work went into this across the lab and the product teams and all of this, and I think it's also just sort of a testament to the team being able to methodically figure out the process to scale these models.	但是,我的意思是,一吨工作进入了整个实验室和产品团队以及所有这一切, 我认为它也只是一种证明团队能够有条不紊地找出这些模型的规模。
208	So I think this is also just a milestone on the way towards the time between each subsequently bigger class of models, I think, will now start to kind of compress, and that's very exciting for the type of stuff that we can build.	因此,我认为这也只是一个里程碑, 走向每一个后来的更大的模型类别之间的时间, 我认为,现在将开始压缩, 这对我们能够制造的东西来说是非常刺激的。
209	At Meta or in the industry generally?	在梅塔还是整个行业?
210	At Meta.	在梅塔(英语:Meta).
211	I mean, I think you're saying something like this everywhere else, but I think part of it...	我意思是,我认为你在说类似的东西其他地方,但我认为这是一部分...
212	Is this an RSI effect or something else?	这是RSI效应还是别的?
213	Well, I think at Meta, part of this is just we needed to...	我想在梅塔,我们需要...
214	Well, I think quite famously, we were not happy with the trajectory that we were on with Lama 4.	我认为很有名的是我们对跟喇嘛4的轨迹不满意
215	We kind of rebuilt the whole team.	我们重建了整个团队
216	I mean, that's been this kind of crazy thing over the last year and a half that we should probably spend a bunch of time on.	我的意思是,这是这种疯狂的事情在过去的一年半, 我们也许应该花一堆时间。
217	We'll get to that, yeah, yeah.	我们会得到,是的,是的。
218	Some war stories.	一些战争故事。
219	But the team basically had to go build up different tiers of models to kind of get to this one.	但是,团队基本上必须去建立不同层次的模型来达到这个。
220	And now that we are kind of at the frontier, now everything that we do going forward will just kind of be pushing the frontier,	而现在,我们有点在边境, 现在,我们所做的一切前进将只是那种推动边境,
221	as opposed to there's a lot of kind of background work that we needed to do before that.	而不是有很多类型的背景工作我们需要做之前。
222	I don't know.	师曰.
223	What would you say?	你会说什么?
224	Yeah.	对
225	Well, I think we're really excited about it.	嗯,我觉得我们真的很兴奋。
226	We will have two sort of two models in this series that are going to be available.	我们将有两种模式在这个系列中,将会有两种模式。
227	因此,有Muse Wonder和Muse Wonder Full. 穆斯Wonder是一款游戏.
228	Laugh.	笑个甚么.
229	Yeah.	对
230	All right.	好吧,我们走。
231	There you go.	给你。
232	Yeah.	对
233	Very good.	很不错。
234	Which is basically a much higher end.	这基本上是一个更高端。
235	There's no studio band version.	没有录音室乐队版本.
236	Yeah.	对
237	No.	没有
238	A much higher end model.	一个更高端的模型。
239	这是Muse Wonder的完整版本.
240	I know.	我知道,我记得
241	I got it.	我来
242	Yeah.	对
243	It's not.	这不是。
244	It's two words.	谓二句.
245	Wonder Full 版本。
246	You're clearly a father.	你显然是一个父亲。
247	What's up?	这是怎么回事?
248	Excellent dad joke.	很棒的爸爸笑话。
249	There you go.	给你。
250	And the performance, I mean, we've seen like incredible performance across a slate of evaluations on personal, professional, agentic tasks.	表演,我的意思是,我们已经看到令人难以置信的表演跨越一系列关于个人, 专业,代理任务的评价。
251	It's state of the art based on our measurement.	根据我们的测量结果,这是最先进的
252	And then there's a number of, I think, specific highlights which we're really excited about, which we've designed into the model over time.	还有,我想,我们非常兴奋的一些具体的亮点, 随着时间的推移,我们设计了模型。
253	So active memory, like Mark mentioned, is one area where we think that to build personal agents, personal and professional agents,	就像马克提到的活跃的记忆是一个领域我们认为要建立个人特工个人和专业特工
254	like memory is just one of the clear requirements to build reliable agents and is one of the areas where we've invested, I think, disproportionately actually compared to many of the other efforts.	就像记忆只是建立可靠代理人的明确要求之一, 也是我们投入的领域之一, 我认为,实际上与许多其他努力相比, 是不成比例的。
255	And we have our own internal evals of memory bench, which demonstrate a lot of the capabilities there.	我们有我们内部的记忆长凳, 展示出很多的能力。
256	What does memory mean in a model to a layperson?	模特儿对外行人来说记忆是什么意思?
257	Is it just how much stuff it can, like the working memory of if you're working on a task, how much stuff the model can kind of keep in its, how is it different from context?	是不是它能保存多少东西, 比如工作记忆,如果你在做一项任务, 模型能保存多少东西, 和上下文有什么不同?
258	I think it's in the experience of using the model.	我想这是在使用模型的经验。
259	How well does it remember things that you mentioned to it over days, weeks or months and then leverage those in the process of doing future tasks?	它在多大程度上记得你在几天、几周或几个月里提到的事情,然后利用那些正在完成未来任务的人?
260	I see.	原来如此
261	这是OpenClaw的窃听器
262	Like, I think we're all extremely fond of OpenClaw, very happy OpenClaw user, but it ultimately does this compaction where it has like the long-term memory and the working short-term memory or whatever.	像,我认为我们都非常喜欢OpenClaw, 非常快乐的OpenClaw用户, 但最终它做这个紧凑的地方它有像长期记忆和工作短期记忆之类的。
263	And I think once people use OpenClaw for a while, they find the memory pretty buggy.	我想一旦人们使用OpenClaw一段时间, 他们就会发现记忆很模糊。
264	Yeah, so we've designed basically both the model and a system for the model to use to develop a much stronger form of memory.	所以我们设计了模型和系统用来开发更强大的记忆
265	We call it active memory.	我们称之为活动记忆。
266	In addition, as Mark mentioned on the persistence.	此外,正如马克在持久性方面提到的那样。
267	But so this is outside the model, like the model exists.	但是,这是在模型之外, 就像模型存在。
268	No, no, but some of the capabilities that you're talking about, like the compaction, are things that the model does.	不,不,但是你所说的一些能力, 比如紧凑,是模型做的东西。
269	Right, so basically being able to do this well both with the agent.	好吧,所以基本上能够做到这一点与代理。
270	It's not a harness, it's in the model itself.	它不是绳子,而是在模型本身。
271	Yes, it's in the model itself.	是的,它是在模型本身。
272	Exactly, yeah, yeah.	没错,是啊,是啊。
273	So it's aware of its memory as a core system.	因此它意识到它的记忆是一个核心系统.
274	And it's about how it's able to leverage various sources to be able to conduct tasks in the future, as well as understand its ability to do compaction, how it should do that compaction most effectively, et cetera.	它是如何利用各种来源来完成未来的任务的, 以及理解它做压缩的能力, 如何最有效地做压缩等等。
275	Okay, but this is a big deal, because I think a lot of people's early experience of AI agents has been buggy memory.	好吧,但这是个大问题, 因为我认为很多人的早期经验 AI的代理一直有错误的记忆。
276	And so, yeah, improvement that is.	所以,是的,这是改进。
277	And to be clear, there's still plenty of room to improve, like our evaluations show this.	说得很清楚,仍有很大的改进余地, 就像我们的评价所显示的。
278	But this is something that we're quite proud of, and we think this is a meaningful contribution.	但是,这是我们相当自豪的, 我们认为这是一个有意义的贡献。
279	Okay, so active memory is one thing.	好吧,所以积极的记忆是一回事。
280	Yeah, active memory, active focus is another, which is effectively just the ability for the model to, you know, when conducting tasks, be able to sort of carry them through.	是的,活性记忆,活性焦点是另一个, 这实际上只是模型的能力,你知道,当执行任务时, 能够排序地通过它们。
281	Especially as we've seen Muse users, as well as Muse code users, give sort of like higher and higher complexity and more and more ambiguous and more and more complex tasks, we want the model to be able to conduct those and do those very effectively.	特别是当我们看到Muse的用户,以及Muse的代码用户, 给出了类似更高和更高的复杂性, 以及越来越模糊和越来越复杂的任务时, 我们希望这个模型能够进行这些并且非常有效。
282	And this is one of the things actually in early dogfooding and testing that a lot of people have noticed about the model that is sort of special and stands out with respect to other models that they've tried or used.	在早期的狗食和测试中, 很多人都注意到了一种特殊的模型, 相对于他们尝试或使用过的其他模型来说,它非常突出。
283	The third one is domain expertise.	第三个是领域专门知识。
284	Is focus in AI terms?	用人工智能来说是焦点吗?
285	Is it like putting a lot of compute on the task?	这就像给任务做很多计算吗?
286	Is this getting distracted with other, like, kind of persistence?	这是否与其它的一样, 一种坚持?
287	Yeah, like persistence.	是啊,像坚持。
288	It's basically like the ability to conduct complete tasks over very long horizons.	基本上就像在很长的地平线上完成全部任务的能力.
289	Okay.	摆
290	And then when faced with various setbacks along the way, the ability to sort of identify other solutions to those problems in an aligned way.	然后,在一路遇到各种挫折时,能够以一致的方式找到解决这些问题的其他办法。
291	Okay.	摆
292	Then there's, it's actually a big step up in coding capabilities, so frontier coding capabilities.	接下来,它实际上在编码能力上迈出了一大步,所以是前沿编码能力.
293	So it is actually state of the art on DeepSui and Sui Marathon and does very well on other benchmarks like Terminal Bench 4 and many others.	因此它实际上在DeepSui和Sui Marathon上达到了最先进的水平,并且在诸如终端座椅4等其他许多基准上也做得非常好.
294	So we think that, you know, users who use it for coding, which will be a minority of all of our users, but the ones who do, will see a big improvement and a big step up.	所以,我们觉得,用它来编码的用户, 将是我们所有用户中少数的, 但那些这样做的用户, 将会看到一个很大的改进, 一个很大的步骤。
295	Okay, so we should also just be using MuseWonder full for coding at Stripe.	好吧,所以我们也应该使用MuseWonder 充电的编码在史黛芬。
296	Potentially, yeah.	很有可能,是的。
297	I mean, let's go out there and tell the...	我们出去告诉...
298	But like if it's state of the art on Sui Bench, yeah.	但就像苏伊法官的艺术水平一样
299	We'd be missing out.	我们会错过。
300	Yeah, yeah, no, it's a big improvement on coding.	是啊,是啊,不,这是一个很大的改进编码。
301	I mean, obviously, you know, we want people to try it and see how they experience it, but we've been very pleased in our own internal testing.	我的意思是,很明显,你知道, 我们希望人们去尝试,看看他们是如何体验的, 但我们在自己的内部测试中非常满意。
302	Okay, so this is...	好吧,所以这是...
303	Yeah, I think people will like it for a lot of things.	是啊,我想人们会喜欢它很多事情。
304	Okay, so this is both just your latest frontier model, like people can come, you know, developers can come directly to you, they can use it on an open router, like they can just consume it as a model.	好吧,所以这都是你最新的前沿模型, 就像人们可以来,你知道,开发者可以直接来找你, 他们可以用在开放的路由器上, 就像他们可以消费它作为一个模型一样。
305	And it's also a big upgrade inside of Muse that people will just feel as, you know, active users of Muse.	这也是Muse内部的一次大升级, 人们会觉得,你知道,Muse的活跃用户。
306	Yeah, well, what we're going to do in Muse, because it's a lot bigger and it's more expensive to serve, where we're going to start off primarily in paid subscriptions.	是啊,嗯,我们在Muse要做的, 因为它是大得多, 它更昂贵的服务, 我们在那里开始主要付费订阅。
307	So what we want to do with Muse as a product, we want it to be free for everyone to use, so we have this huge amount of free tokens that people can use with Muse Spark as the model.	所以,我们想用Muse作为产品, 我们希望它可以免费让所有人使用, 所以我们有这么多免费的代币, 人们可以用Muse Spark作为模型。
308	But if you want to use Wunder for most of the hard tasks, then at least for now, you'll use the paying version.	但如果你想用Wunder来完成大部分的艰难任务,那么至少现在,你会使用付费版本.
309	But, you know, I mean, eventually we'll, you know, quickly the field moves, and we hope to, you know, at some point, not too far from now, have a model that is as good as Wunder, but in the Spark form factor.	但是,你知道,我的意思是,我们最终会,你知道, 迅速的实地移动, 我们希望,你知道,在某个时候, 离现在不远的地方, 有一个和Wunder一样好的模型, 但是在Spark形式因素中。
310	And that will be a great thing to be able to serve hundreds of millions of people around the world.	这将是一个伟大的事情能够服务全世界亿万人民。
311	Okay, so you go to Muse, you subscribe to the, what's it called, like the premium or pro plan, or you pay money.	好吧,所以你去缪斯, 你订阅, 它叫什么, 像溢价或专业计划, 或者你支付钱。
312	Yeah.	对
313	And just don't stop until you've taken out your credit card.	在你拿出信用卡之前别停下来
314	Yeah.	对
315	And then you'll be able to, you'll select it or it'll just magically...	然后,你可以,你会选择它或者它只是神奇的...
316	Oh, it just gets better.	哦,它只是变得更好。
317	It just gets better for you.	它只是变得更好给你。
318	And...	还有...
319	然后也可以在API,金属模型API和OpenRouter上找到.
320	And then it'll also be available for Muse code users.	然后它也会提供给Muse代码用户.
321	So, and then Muse Wunder full will be available to subscribers of Muse code.	因此,然后Muse Wunder的完整将提供给Muse代码的订阅者.
322	So there's, it's pretty exciting, I think we want to, you know, as Mark mentioned, it is a much larger model.	所以,这是非常令人兴奋的, 我认为我们想, 你知道,正如马克提到的, 这是一个更大的模型。
323	And so we're working through how to serve that most effectively to as many people as possible.	所以,我们正在努力如何最有效地为尽可能多的人服务。
324	But we're excited to get it in the hands of as many people as possible.	但我们很兴奋地把它交给尽可能多的人手中.
325	It's a big step forward.	谓前大步.
326	I mean, this is like, and we really with Muse Spark, we were very happy with the performance that we were able to deliver as we were getting up to this point.	我的意思是,这就像,我们真的与Muse Spark, 我们非常满意的表演,我们做到了在这一点上。
327	But like, this is the beginning of like, MSL kind of being at the frontier.	但是,就像,这是开始像,MSL有点在前沿。
328	We feel like the models are quite competitive with the best models that are out there.	我们觉得模型相当有竞争力与最好的模型在那里。
329	And I think that...	我觉得...
330	Okay, so you think it's like Astrophable and Wunder all together at the frontier?	好吧,所以你认为这就像可以天体和Wunder 一起在边境?
331	Yeah, well, I mean, you can look at it.	是啊,我的意思是,你可以看看它。
332	Just empirically, by the benchmarks, you think...	从经验来看,根据基准,你认为...
333	By the benchmarks, one would conclude that, yes.	根据基准,可以得出这样的结论。
334	I see, yes.	原来如此
335	And I think actually, one of the things that we're especially proud of is, you know, we have taken the time, kind of as we've mentioned, to optimize models for the product form factor of Muse.	我认为,我们特别引以为傲的一件事是, 你知道,我们已经花时间,就像我们提到的, 优化了Muse产品形式因素的模型。
336	So this is the state of the art model.	这就是艺术模型的状态。
337	We think it really is the best model in the world for the use case of personal and professional agents.	我们认为,这确实是世界上最好的模式, 用于个人和专业代理。
338	People inside companies like complaining about Goodheart's law, you know, a metric that becomes a target ceases to become a good metric.	公司内部的人喜欢抱怨 Goodheart的律法, 你知道,一个成为目标的指标不再成为一个好的衡量标准。
339	You hear complaints that models are bench maxed and that at a certain point, kind of overfitting to the benchmarks actually makes the models slightly worse.	你听到的抱怨是,模型是最大级的, 在某一时刻, 有点过于符合基准实际使模型稍差。
340	As people who actually have to train AI models, is that true?	作为真正需要训练AI模型的人,这是真的吗?
341	Are the benchmarks pretty good at grading models?	这些基准是否对定级模型相当好?
342	Do you start to see like weird edge failure cases from the...	你开始看到像怪异的边缘故障案件从...
343	Benchmarks have to keep moving, right?	基准必须继续移动,对不对?
344	Because like every...	因为像每一个...
345	I mean, it happens very quickly, but they just get saturated so quickly.	我的意思是,它发生得很快, 但他们只是得到饱和如此之快。
346	So I mean, I do think that...	所以我觉得...
347	Do you think non-saturated benchmarks are fine?	你认为不饱和的基准很好吗?
348	I mean, as we've gone on, we steer the models more towards our internal benchmarks for the use cases that we're focused on.	我的意思是,随着我们继续前进, 我们引导模型更多向内部基准我们的用例,我们专注于。
349	So early on when we were kind of just doing the science of trying to climb the ladder, we cared more about what the external benchmarks were just to make sure we were on the right track.	所以,在我们刚刚开始尝试攀登梯子时, 我们更关心外部基准是什么, 以确保我们走上正确的道路。
350	And then I'd say, you know, each release that goes by the larger share of what we're looking at are our internal ones.	然后,我会说,你知道, 每一个发行由更大的份额我们所看到的是我们的内部。
351	And one of the things that we actually quantified in our research blog post as part of Me's Wonder is this metric for big model smell, which is kind of, you know, big model smell and feeling of bench maxing are kind of like inverses.	作为"我的神奇"的一部分,我们的研究博客文章中将一个数字化的事物是一个关于大模型气味的衡量标准, 这是一种,你知道的,大模型气味和长凳最大感有点像反面。
352	This idea that you can just like tell when you're using a really big model.	这个想法,你可以只是喜欢知道当你使用一个真正的大模型。
353	It just... you can tell.	它只是... 你可以告诉。
354	It just has nuance.	它只是有细微的。
355	It's got that genocer claw.	它的基因发爪。
356	And so the way we quantify that is actually almost explicitly what you're mentioning, which is the gap between in distribution performance and out of distribution performance.	我们的量化方法几乎是你们提到的, 就是分配性能与分配性能之间的差距。
357	So what is the... you know, we kind of measure how well does the model perform on tasks that are in distribution of our RL and post-training versus tasks that are out of distribution?	那么,什么是... 你知道,我们有种衡量这个模型是如何执行任务在分配我们的RL和培训后的任务相对于分配出的任务?
358	And what does that gap look like for... what does that gap look like between the Me's Spark line and the Me's Wonder line?	那间隙是什么样子的... 那间隙是什么样子的...
359	And we actually see this huge step up from the Me's Spark to the Me's Wonder line because of, as Mark mentioned, like it's just a much better base model powering this new line.	我们实际上看到这个巨大的步骤从"我的火花"到"我的奇妙线", 因为,正如马克所提到的, 它只是一个更好的基础模型,为这条新线提供动力。
360	And there's an entire genocer claw to that.	还有一个完整的基因组爪子
361	But I think we've been able to actually quantify this effect.	但是,我认为我们已经能够量化这种效果。
362	Yes.	对
363	And it's an obviously dimension that we're excited to continue improving on.	它显然是一个维度, 我们兴奋地继续改进。
364	But if you're saying that external benchmarks have ceased to become as important and you care much more about your internal benchmarks,	但如果你说外部基准不再重要了你更关心你的内部基准
365	is that kind of an argument against benchmarking that you think that just trying to get the top score on the external benchmarks that everyone talks about is maybe not going to lead you to develop a great model?	那种反对基准的争论,你认为,仅仅试图在大家谈论的外部基准上获得最高分也许不会导致你发展出一个伟大的模型?
366	Well, yeah, but also we're focused on specific things.	嗯,是的,但是我们也专注于具体的事情。
367	I mean, part of the whole strategy is personal super intelligence and just trying to build different types of products that other people are trying to build.	我的意思是,整个策略的一部分是个人的超级智能, 并且只是试图建立不同种类的产品, 其他人正在试图建造。
368	So even if we did have the exact same evals as everyone else, it just...	所以,即使我们有完全一样的 evals 和其他人一样,它只是...
369	Yes, and though I would say a personal super intelligence is not like a super niche interest.	是的,虽然我会说个人的超级智能不像超级特殊利益。
370	It's probably something that other models won't be able to do.	这可能是其他模型无法做到的事情.
371	I think that there are things that are very different about it.	我认为有些事情是完全不同的。
372	I mean, I think like the personality probably matters more.	我觉得性格可能更重要
373	I mean, one of the things that we talk about is this idea of discretion.	我的意思是,我们谈论的一件事就是这种谨慎的想法。
374	So if you want to use the model to get something done,	所以如果你想用模型来做点什么
375	it needs to be able to know a lot about you and get things done in the world while disclosing as little as possible.	它需要能够了解你很多并让事情在世界上完成同时尽可能少披露。
376	So if you have an allergy or you're pregnant or something, you're trying to get a table at a restaurant	所以,如果你有过敏或你怀孕了什么的, 你试图得到一个桌子在餐厅
377	and you don't want to have to tell them your medical condition to make sure that it's going to be a good place for you,	而你不想告诉他们你的病情以确保这对你有好处
378	I mean, humans have common sense and the model so far,	我的意思是,人类有常识和模型到目前为止,
379	if you're building a coding agent that works in an environment where everyone has access to the same context,	如果你正在建造一个编码代理在每个人都能进入同一环境的环境中工作
380	discretion isn't particularly important.	谨慎并不特别重要。
381	So there are all these things that I think are just kind of more important for us.	所以,有所有这些东西我认为只是一种对我们更重要。
382	Aspects of coding where the code needs to work, but the consumers who are using it may not even ever see the code.	代码需要起作用的部分编码,但正在使用它的消费者甚至可能从未看到代码.
383	I mean, that probably matters more in a personal superintelligence context rather than serving software developers.	我的意思是,这也许在个人的超级智能背景下更重要, 而不是为软件开发者服务。
384	So I think these things are just a bit different.	所以我认为这些事情只是有点不同。
385	Ultimately, we care about making useful models.	归根结底,我们关心做有用的模型。
386	So we want models when you put them into people's hands, they have a great experience using.	所以我们需要模型,当你把它们放在人们的手中, 他们有很好的经验使用。
387	And I think that is the most important metric for any model that we or other people produce.	我认为这是我们或其他人生产的任何模型最重要的衡量标准。
388	I think that really, really focusing on the external benchmarks,	我认为,真的,真的专注于外部基准,
389	if you really dig into the supply chain a little bit,	如果你真的挖进供应链一点点,
390	it's some PhD student who made these benchmarks as part of the course of them doing their PhD.	是一些博士生把这些基准当作他们完成博士课程的一部分
391	And so you're kind of outsourcing all of your product thinking to that PhD student who happened to make a viral benchmark.	所以你将你所有的产品外包给那个碰巧做病毒基准的博士生。
392	So clearly that's wrong.	很明显这是错误的。
393	And so that's why we care a lot.	这就是为什么我们在乎很多。
394	A huge part of what a lot of people at Meta and MSL are focused on is crafting amazing evals.	Meta和MSL的很多人关注的其中一大部分是制作惊人的爱华。
395	And now that we have them, we can work on our own.	现在我们有了它们,我们可以自己去工作。
396	But if you're building like a ramping up of your lab, right?	但如果你正在建设就像你的实验室,对不对?
397	It's like those public ones are not a crazy place to start, but it's just not where you want to end up.	这就像那些公共的不是一个疯狂的开始的地方, 但这只是不是你想要的结局。
398	At that point, you're just looking for signs of life.	这时候,你只是在寻找生命的迹象.
399	But then when you're like really in this mode, as we are now, where it's like, you know,	但当你真的喜欢在这个模式,像我们现在, 它就像,你知道,
400	we want to be very intentional about how we're improving our model over time to build better and better, better products.	我们想非常有意识地研究我们如何随着时间的推移改进我们的模型, 以建造更好更好的产品。
401	Like a lot of the cases that Mark just mentioned, like these are actually detailed very rigorously	就像马克刚才提到的许多案子一样这些都是非常详细的
402	in like the spreadsheet of all the little capabilities and nuances of the product that we think really, really matter to the end user.	就像电子表格一样包含产品中所有微小的能力和细微的细微差别我们认为对终端用户来说都很重要
403	Maybe each of them matters like 2% to the user.	也许每个对用户来说都很重要
404	But in aggregate, that's like the obsession that will ultimately result in these models that feel magical.	但总而言之,这就像迷恋最终会导致这些模型感觉神奇。
405	Can you guys describe for me just what goes into training an RL model or sorry, a leading edge frontier model?	你们能给我描述一下在训练RL模型时有什么用还是抱歉领先的前沿模型?
406	I'm just curious in that I think a lot of people listening to this have a rough idea where there's, you know,	我只是好奇的是,我认为很多人听这个有一个粗略的想法在那里,你知道,
407	there's a pre-training and post-training step.	设有训练前和训练后步骤.
408	So pre-training is you kind of ingest this huge amount of data.	所以,预训是一种你摄入这个巨大的数据。
409	You know, you look at a lot of pictures and these are cats and these are dogs.	你看很多照片这些是猫和狗
410	And after a while, you start to, you know, develop a sense for what is catness and what is dogness and what is fur and, you know, everything like that.	过了一段时间,你开始,你知道, 发展一种感觉,什么是猫,什么是狗,什么是皮毛,你知道,所有类似的东西。
411	And so that's kind of the pre-training step where you just hoover up data and maybe you can talk about the data acquisition and just kind of what goes into that.	所以这是训练前的一步, 在那里,你只是收集数据, 也许你可以谈论数据获取, 以及进入数据的内容。
412	And then there is this RL stage, which is the new-ish advance over kind of the past few years of AI.	接下来是RL阶段,这是过去数年AI的新进步.
413	And again, it sounds like that is where a lot of the specific nuance of this model has come in,	再说一遍,这听起来就是这个模型中很多具体细微的细微变化之处
414	where you guys have better, have chosen better domains, better, you know, hills to climb for the model.	在那里你们有更好的, 选择了更好的领域, 更好的,你知道的,山丘爬向模型。
415	But just at a high level, describe how many people, how long it takes, what the steps are, how you think about those in.	但是在高水平上, 描述一下有多少人,需要多长时间,步骤是什么, 你如何看待这些。
416	How much time do you have?	你有多少时间?
417	对我知道 Yeah, I know.
418	At a high level.	在高处.
419	This happens to be like what we spend all of our time on.	这恰好就是我们所有的时间所花的
420	Exactly.	没错
421	And so summarize it in three minutes.	三分钟后总结一下
422	I can recommend some YouTube courses here.	我可以在这里推荐一些YouTube课程.
423	Yeah, I think at a very high level, you know, these are the stages.	是的,我认为在一个非常高的水平, 你知道,这些是阶段。
424	It's pre-training and post-training.	修习前修行后修行.
425	Ironically, training is the superset of all of it.	具有讽刺意味的是,训练是所有训练的超级集.
426	But how many people are on this whole effort, like when you develop these models?	但是,有多少人在整个的努力, 像当你开发这些模型?
427	So this I think you're getting into, like, what is interesting about MSL and the approach that we've taken, which is, I mean, from the very beginning, I think when I first reached out to you,	所以,我认为你正在进入,喜欢, 什么是有趣的关于MSL和我们采取的方法, 我的意思是,从一开始,我想当我第一次接触你,
428	I think we're just looking at this because we're like, just like, wow, how far we've come.	我认为我们只是看这个因为我们就像,就像,哇, 我们已经走了多远。
429	And I was just like, I pulled up the screenshot of like when I first reached out to Alex and my first thing was like, hey, do you have time to catch up and talk?	我就想,我拉了截图一样当我第一次接触亚历克斯我的第一件事是一样, 嘿,你有时间赶上和说话?
430	And then that's like, all right, we talked for 44 minutes or whatever.	我们聊了44分钟什么的
431	And then then after that call, two days later, Alex sends me this long thing, which is like, well, here's what I think I would do if I were in your if I were in your position.	两天后,亚历克斯给我发了这么长的电话, 就像是,嗯,我想如果我是你的话,我会怎么做。
432	And the very first thing is that you don't need a lot of people and talent density is the thing that matters.	首先,你不需要太多的人,人才密度才是最重要的。
433	So I'd say like from the beginning of the plan, this was like the core of your strategy, right, is basically like you actually don't need a massive number of people.	所以我想说,从计划一开始, 这就像你战略的核心, 对,基本上就像你实际上不需要大量的人。
434	But what you do need is for like each of the people to be as good as possible, which kind of informed this kind of crazy recruiting push and like very intense filtering that we had to take, you know, some of the best people from inside meta and a very int	但你需要的是,像每个人尽可能的好, 什么样的信息这种疯狂的招聘推力, 像非常激烈的过滤, 我们不得不采取,一些最好的人从内元, 一个非常紧张的过程从外元。
435	But I actually think like a lot of what has made MSL special and why a lot of people have wanted to join is because I think it's in some ways it's just like more fun to work on talent and smaller.	但我实际上认为,像很多使MSL 特别的原因, 为什么很多人想要加入, 因为我认为在某些方面它只是像更有趣工作在人才和较小。
436	I've heard it feels like the startup that's operating in some ways it has all the benefits of a startup with the stability and resources of a larger company.	我听说,它的感觉像启动以某种方式运行它拥有所有的好处启动在一个更大的公司的稳定和资源。
437	And this was like this environment that I think this was like point one in your kind of plan for what Ford Alex was like, this is what we need to go do.	这就像这种环境我认为这是你计划中的第一点对于福特·亚历克斯是什么样子, 这就是我们需要做的。
438	So I don't know the answer of like how big is a pre-training team.	所以我不知道一个训练前团队有多大的答案。
439	You know, it doesn't need to be that big.	没必要那么大
440	But I think if you look at some of the other labs, it's probably quite big in some other places.	但我想如果你看看其他的实验室, 在其他地方可能相当大。
441	So I think part of the reason why these models are doing well is because you have people who we're betting on and they have a lot of resources because we're taking our compute.	所以,我认为这些模型之所以能做得很好部分原因是因为你有我们所相信的人并且他们有很多资源因为我们的计算。
442	We're not like peanut buttering across a lot of competing efforts where basically just we're investing in like the key people, like a lot of a lot of compute per researcher.	我们不喜欢花生酱跨越很多相互竞争的努力, 基本上我们像关键的人一样投资, 像很多计算每个研究人员。
443	And then they can they can do their work.	然后他们可以做他们的工作。
444	Was wonder the one big training at first like that?	一开始是这么训练的吗?
445	That was the next pre-training.	这是接下来的预训。
446	It was climbing up this ladder.	它爬上这个梯子。
447	So, I mean, we have this tiering system and then kind of spark.	所以,我的意思是,我们有一个分层系统, 然后是火花。
448	I mean, internally, like avocado was the codename for the new spark line.	就像鳄梨是新火花线的代号
449	And then watermelon was the codename for this one.	然后西瓜是这个的代号。
450	And we're like out of that we're out of fruits.	我们就像从那我们没有水果。
451	Yeah.	对
452	So we're going to have to switch to a to a new convention.	因此,我们不得不转向新的会议。
453	But I think the only fruit that is technically larger is the durian.	但我认为唯一技术上更大的水果是榴弹。
454	And you don't want to go to a durian.	而你不想去一个榴弹。
455	No, no, you want to stay far away.	不,不,你想远离。
456	A durian is bigger than a watermelon.	榴弹比西瓜大.
457	I think it's definitely heavier.	我觉得它绝对是更重。
458	That's horrifying.	这太可怕了
459	Definitely heavier.	绝对重。
460	Spikey.	师曰.
461	Yeah.	对
462	But no.	但是没有。
463	Yeah.	对
464	I think as as Mark mentioned, like, you know, each of the I mean, each of the stages stages I'll describe, like it's tens of people working on it.	我想正如马克提到的, 比如,你知道,每一个我的意思, 每个阶段,我会描述的, 仿佛是数十人的工作。
465	A meta versus, you know, hundreds.	一个元对,你知道的,数百。
466	So meta is a huge company.	所以meta是一个巨大的公司。
467	This is like a really tiny effort.	这就像一个很小的努力。
468	It's a meta.	谓之元.
469	Yeah.	对
470	The research team is less than 200 people.	研究团队不足200人.
471	Yeah, it's very it's very small.	是啊,这是非常小。
472	It's intentionally very, very tight.	这是故意的非常,非常紧。
473	I think that the closer to 100.	我认为,接近100。
474	And then there's a bunch of info.	后有众生消息.
475	Yeah.	对
476	And then it's it's a lot of info work.	后是多处信息工作.
477	So I think at a very high level to go back to the to the to the what is it question.	所以,我想在非常高的水平上回到问题所在。
478	Yeah.	对
479	Well, how does it happen?	怎么会这样?
480	How does model happen?	模型是怎么发生的?
481	That's basically what I'm asking.	基本上这就是我要问的。
482	Where do models come from?	模特儿从哪里来?
483	You you take a huge amount of data for pre-training.	你收集了大量的训练前数据
484	And here you're you're here like roughly.	在这里你在这里就像大致。
485	It's like all the, you know, public Internet and like all the books you can get your hands on and just like as much data as possible.	这就像所有,你知道的,公共的互联网和所有的书籍你可以得到你的手和就像尽可能多的数据。
486	But different types.	但种类不同.
487	Plus the proprietary metadata.	加上专有元数据.
488	But also different modalities.	但也不同的方式。
489	So if you want it to be good at perception, then you want to make sure it has a lot of images and audio and video and this kind of stuff.	所以,如果你希望它能善于感知, 那么,你想要确保它有很多的图像, 音频和视频, 之类的东西。
490	You know, if you want it to be good at certain domains, you don't just want to RL that in at the end.	你知道,如果你想它在一些领域是好的, 你不只是想要RL 在结尾。
491	You want to have it have exposure to that from the beginning.	你想让它从一开始就接触到这个
492	Yeah.	对
493	So and in this stage, you're kind of you're much less you care a lot less exactly what the data looks like or how it's formatted or all of everything about it.	所以,在这个阶段,你更不关心确切的数据是什么样子或者它是如何格式化的或者所有的一切。
494	But it's but it's just lots and lots of data.	但它只是很多很多的数据。
495	When you talk about this being the best personal agent model, was there data you had to get for the pre-train or it's like huge amounts of restaurant reservation called book?	当你谈论这是最好的个人代理模型, 是否有数据你必须得到的预训或者它像大量餐厅预订书?
496	I don't know what it is, but like kind of want to dump in large volumes of the modalities that people are going to care about.	我不知道是什么但就像想把人们会关心的大量模式丢掉一样
497	I mean, there were a bunch of things.	我的意思是,有一堆东西。
498	I mean, I don't actually think we talked about this when we're talking about where Muse Wonder excels.	我的意思是,我并不认为我们谈论这个当我们谈论哪里Muse Wonder的优异。
499	But some of the areas that it's like, all right, what do people care about?	但是,有些地方,它像, 好吧,人们关心什么?
500	Right.	对
501	It's I mean, there are things like shopping that people care about a lot.	我的意思是,有的东西像购物人们关心很多。
502	Personal advice that people care about a lot.	人们关心的个人建议。
503	People have personal legal questions, personal health questions, personal tax questions.	人们有个人法律问题,个人健康问题,个人税务问题.
504	Muse Wonder, I mean, is like quite good.	Muse Wonder, 我的意思是,是相当不错的。
505	I don't know if there's any quantification that we want to give.	我不知道我们是否想要给出任何量化.
506	But it's let's say it's it's it's it's very good on those.	不过,我们还是要说,这些东西都很好。
507	It's good.	不错
508	Yeah.	对
509	And so those types of things, I mean, you can are are all them to some degree.	所以,这些类型的东西,我的意思是,你可以在所有程度上都是。
510	But you like just want to put in a large amount of data.	但你喜欢只是想放大量的数据。
511	Like you can imagine health is a very there's a lot of data out there about health.	就像你能想象的健康是一个非常有很多关于健康的数据。
512	And in pre-training, you just wanted to like see all these different things.	在训练前,你只是想看看这些不同的东西。
513	And so you kind of just put these different corporate.	所以你就把这些不同的公司。
514	Yeah.	对
515	So so so as as Mark was mentioning, like the we one of the highlights of Muse Wonder actually is its ability on domain,	所以正如马克提到的那样就像我们Muse Wonder的亮点之一其实就是它在域上的能力
516	various domain expertises.	各种领域专门知识。
517	And I think our philosophy here is that, you know,	我认为我们的哲学是,你知道,
518	the best personal agent should feel like a team of experts in your corner.	最好的私人特工应该感觉像一个团队的专家在你的角落。
519	Like it should feel like it is your legal team and your finance team and your medical team and and and and and.	感觉好像是你的法律团队财务团队医疗团队还有还有...
520	And so on a bunch of these domains, we Muse Wonder is actually state of the art on, I think,	因此,在很多这样的领域, 我们Muse Wonder实际上是最新技术,我认为,
521	a lot of the finance evals as well as a lot of the legal evals.	有很多金融电子报还有很多法律电子报
522	It is the state of the art model.	这是艺术模型的状态。
523	And that these these domain areas, that starts in pre-training.	而这些领域,从训练前开始。
524	We look for a lot of domain data to put into pre-training.	我们寻找很多域数据投入到培训前。
525	Yes.	对
526	But a lot of the a lot of the art of pre-training, I mean, pre-training is there's this one rune tweet where it's like pre-training is full of the researchers who like chew glass and stare out into the distance into the into the abyss or whatnot.	但很多预训的艺术,我的意思是,预训是, 有一种Rune的微博, 它就像预训充满了研究者, 他们喜欢咀嚼玻璃, 并凝视到遥远的深渊。
527	But but I think there's a truth to that because it is like it is both the highest stakes of the research areas because you know,	但我认为这是事实因为它就像它既是研究领域最高的利害关系因为你知道
528	我们投入了很大比例的计算来训练Muse Wonder 到预训Muse Wonder。
529	And it was very high stakes.	赌注很高
530	I think that's probably both of us at various points in that process.	我认为,在这个过程中,我们可能都处于不同阶段。
531	You went through troughs of sadness, troughs of uncertainty.	你经历了悲伤和不确定
532	You know, all the seven stages of grief happen while you're while you're pre-training the model.	所有七个阶段的悲伤都发生在你训练模特的时候
533	And it is the most sort of like fundamentally scientific in an interesting way because, you know, you just don't have very many samples.	它是最像本质上科学的方式一个有趣的方式,因为,你知道, 你只是没有太多的样本。
534	So like your ability to infer exactly what will happen scientifically really depends on your ability to sort of like, you know, big brain reason about everything that's happening this process.	所以,就像你能够准确推断科学上会发生什么一样, 真正取决于你的能力, 比如说,关于正在发生的每一件事,大脑的大理性。
535	OK, so in the pre-training stage, everyone who's training frontier models basically is trying to get as much data as possible for the pre-train with maybe a bent towards if you want Muse Wonder to excel as legal and finance and tax questions and things li	好吧,所以在训练前阶段, 每一个训练前沿模型的人基本上都在努力获得尽可能多的数据对于训练前,也许有一个倾向,如果你希望Muse Wonder 杰出的法律和金融,税务问题和类似的事情,
536	then you want to have as much of that specific corpus as possible in the pre-train.	那你就想在预训中尽可能多地拥有这些具体的工具
537	You still want as much intentionality about where ultimately you want your model to be.	你仍然希望你最终想要你的模特儿在哪里
538	But again, it just, you know, benefits from a lot more data.	但是,它只是,你知道, 从更多的数据中获益。
539	Yeah.	对
540	And this is sort of the massive data, massive compute stage.	这是大量的数据,大量的计算阶段。
541	Do you hit some saturation in the data you put in the pre-train or ultimately is it just more data is better?	你在预列车中的数据中是否达到一定的饱和程度,或者归根结底只是更多的数据更好?
542	Well, there's a relationship between if you want the model to be bigger, you need more data.	如果你想让模型变得更大, 你需要更多的数据。
543	Otherwise it overfits.	否则就太过份了
544	Yes, but my point is like, is there data where it's like, oh, we can't distract the model with that.	是的,但是我的观点是,有数据它像,哦,我们不能分心模型用它。
545	We can't clutter up its brain attic with knowing that stuff or is it just more data is better?	我们不能把大脑的阁楼弄得一团糟还是更多的数据更好?
546	There is some.	有的还有的
547	There are some of these effects, I think.	有一些这些效果,我想。
548	This is part of the science.	这是科学的一部分。
549	We're going to secret sauce.	我们要去秘密酱汁。
550	No, it's not about it.	不,这不是它。
551	Just like, I mean, like, I think that there are all these questions about like, OK, well, we put a lot of these tokens in.	就像,我意思是,我想, 有很多这样的问题比如,好吧,好吧,我们放了很多这些符号。
552	Is there any zero-sum-ness and a bunch of places there isn't or it's actually positive transfer?	有没有零和和和一堆地方没有, 或者它实际上是正转移?
553	But then, like, depending on how it's set up, like, you can have some negative transfer.	但是,像,取决于它是如何设置的, 像,你可以有一些负转移。
554	And so, I don't know, there are all these interesting questions.	所以,我不知道, 有很多有趣的问题。
555	But in general, one of the bigger challenges of scaling bigger models is it's not just about how many flops and how much compute.	但一般而言, 缩小更大的模型的更大挑战之一不仅仅是有多少软体和多少计算。
556	You want to be able to scale data with that.	你想用它来缩放数据。
557	I mean, this is like part of what when the original kind of scaling laws for pre-training were written.	我的意思是,这就像当原始类型的缩放法对于训练前。
558	I mean, this is sort of what came out of that is that.	我的意思是,这是某种程度上,从这个结果是。
559	So if the Internet isn't growing that quickly and you want to scale bigger models that can be more intelligent,	所以如果互联网没那么快发展起来并且你想放大更大的模型可以更聪明,
560	I think one of the interesting problems is, all right, where's the where's the data coming from?	我认为一个有趣的问题是, 好吧, 数据从哪里来?
561	Yes.	对
562	So that probably gets to the end of what we're going to talk about on that.	所以,这也许到了我们将要谈论的结尾。
563	But we can go to the other.	但我们可以去另一个
564	Excellent CEO energy over here.	这里有很好的CEO能量
565	So post-training.	故后训.
566	So then I think post-training is kind of is a very exciting field now, I think,	所以我认为后训练是一种非常令人兴奋的领域现在,我认为,
567	and is an area where I think actually I think almost on a continual basis,	我认为我几乎可以持续地思考
568	the entire field is learning about the optimal recipes to be able to post-train these models.	整个领域都在学习最佳的食谱才能完成这些模型的训练
569	And again, for people like post-training is where you give the model task that you expect to do greatest on how it does.	再说一遍,对于像后训练这样的人来说, 就是你给模式任务的地方, 你期望它做得最好。
570	And so if it's chess playing model, it plays chess games, learns that, you know,	所以,如果它是象棋游戏模式, 它玩象棋游戏, 学习,你知道,
571	actually two stages.	实际上是两个阶段。
572	这是SFT。
573	So there's supervised learning.	所以有监督学习。
574	So that's like a lot of when you take this kind of raw thing and you give it both the question and the answer you would want it to have.	所以,这就像很多当你拿这种生的东西, 你给它的问题和答案你希望它有。
575	All right.	好吧,我们走。
576	If you're having a chat conversation, how you want it to respond in this case.	如果你正在聊天,你希望它如何回应本案.
577	So when you're training things like personality, the SFT is actually like a very important part for molding it.	所以当你在训练像个性一样的东西时, SFT实际上就像一个非常重要的角色来塑造它.
578	But then I think most of the advances in the recent years have been in RL.	但是,我认为近年来的大部分进展都是在RL.
579	Yeah.	对
580	Yes.	对
581	So I think I think one mental model that can be helpful is like in SFT is where you kind of initialize a lot of these supervised fine-tuning supervised fine-tuning.	所以我认为一个精神模型是有用的, 就像在SFT中,你在那里初始化了很多这些有监督的微调,有监督的微调。
582	Yeah.	对
583	Where you initialize a lot of the behaviors that you want the model to be able to exhibit later on.	在那里,你初始化了很多行为, 你想让模型能够展示以后。
584	And then, you know, maybe at the end of SFT, it has those behaviors kind of baked in, but it'll only do it like 10 percent of the time.	然后,你知道,也许在SFT的结尾, 它有那种行为烘焙, 但是它只会做像10%的时间。
585	And then in reinforcement learning, you really sort of give the give the model reinforcement in various situations to bring the elicitation of those behaviors dramatically up.	然后在强化学习中,你确实在各种情况下给予模型强化, 使这些行为的引领大增。
586	And I think that the yes.	我认为是的。
587	So I think there's a bunch of interesting things in that that are happening in RL and and post-training.	所以,我认为有很多有趣的事情正在发生在RL和后训练。
588	I think one is that this just the scaling of RL and post-training has been has been one of the clear axes, just spending more compute and gathering more data.	我认为其中之一, 仅仅是RL的缩放和训练后一直是明确的轴心之一, 只是花费更多的计算和收集更多的数据。
589	Just doing much more of it before you release a model.	在你释放模特儿之前做更多
590	Yes.	对
591	Doing much more of it, especially as the pre-trained scale up, doing much more.	做更多的事情,特别是随着预先训练的扩大,做更多的事情。
592	It's like anything in scaling laws, like it has to go with the rest of the factors or it doesn't work.	它就像任何在缩放定律, 像它必须去其余的因素或者它不起作用。
593	Yeah.	对
594	In general, I mean, if I were to say like one broad statement about AIs, just scale everything.	一般来说,我的意思是,如果我说得像一个关于AI的泛泛的言论,只是把一切放大.
595	But in particular, you have to scale everything.	但特别的,你必须放大一切。
596	Like scaling one thing in isolation does not work.	就像孤立地缩小一个东西是行不通的。
597	You technically can.	技术上可以
598	You can scale one dimension.	你可以放大一个维度。
599	You just will get much more optimal performance if you scale many things all together.	你只要把很多事情结合起来就能得到更好的表现
600	Well, I mean, so one way to think about like Muse Wonder, right?	嗯,我的意思是,所以一个方法想像Muse Wonder,对不对?
601	It's like, all right, so we did the pre-trained.	这就像,所有的权利, 所以我们做了预训。
602	We've done some RL.	我们做了一些RL。
603	Right.	对
604	当你看着它像Muse Spark,对不对?
605	So version one to one point one, one point two, one point three.	所以版本1到1点,1点2,1点3.
606	We didn't change anything about pre-training, right?	我们没有改变任何关于预训,对不对?
607	We just continued doing more RL on top of it.	我们只是继续做更多的RL在上面。
608	So so you when you think about the early versions of a model of a new class, they're actually relatively less like the RL to pre-training ratio.	所以当你想到一个新班级的模型的早期版本时, 它们实际上比起RL到预训练的比例要少。
609	The ratio is going to be lower for RL than the later versions.	相对于后来的版本,RL的比例将会更低.
610	That is kind of what the point releases in models are.	这就是模型中的点释放。
611	Well, I think every company has its own convention.	我想每个公司都有自己的会议
612	For us, that's what it is meant.	对我们来说,这就是它的意思。
613	But for a lot of companies, like the big number is the pre-train and then the.	但对于很多公司来说,像大数字是预训,然后是.
614	I think it kind of lost that semantic value.	我认为它有点失去了语义价值。
615	Yeah, I think now you actually.	是啊,我觉得现在你实际上。
616	OK, but it is true at Meta, at least.	好吧,但这是真实的在梅塔,至少。
617	We haven't even done it.	我们甚至没有做到这一点。
618	We kind of decided.	我们决定了
619	It's true.	这是真的。
620	It is true.	这是真的。
621	确实,Muse Spark和Muse Wonder是预先训练的。
622	And we decided to kick the can down the road on this question for Muse Wonder by not giving it a number yet.	我们决定把这问题踢下去给Muse Wonder 通过不给它一个号码。
623	But when we have our first dot release, we will have to decide.	但是,当我们有我们的第一个点释放, 我们必须决定。
624	I think the RL stage, you know, part of what has been, you know, part of what's very interesting is first is just producing the environments and producing the tasks.	我认为RL阶段,你知道的, 一部分是,你知道的, 一部分是非常有趣的, 首先,只是生产环境和生产任务。
625	And I think that is one of the areas where there's a lot of intentionality in training these models.	我认为,在培训这些模型时, 有很多有意的行为。
626	It's also one of the areas where, you know, you notice very funny things like if you, you know, in a lot of cases, less is actually more when it comes to data in RL.	这也是其中的一个领域,你知道,你注意到非常有趣的事情, 比如,如果你,你知道,在很多情况下, 更少的其实是更多的,当涉及到RL的数据。
627	And so I think I think the sort of like the the feel and the science of it like is pretty flipped, generally speaking, from from pre-training.	所以我觉得这种感觉和科学通常都是从训练前开始的。
628	The other is that I think it can give very magnified kind of instruction to the model, if that makes sense.	另一个是,我认为它可以给模型一种非常放大的指令,如果这有道理的话。
629	So I think that like a lot of safety cases or alignment cases that we care a lot about are like, you know, and, you know, where you would see when you do large high compute RL runs on frontier models are where you apply a lot of RL or optimization pressur	所以,我认为,像很多安全案例或者我们非常关心的对齐案例一样, 就像,你知道,你知道,当你在前沿模型上做大高计算RL运行时, 你会看到, 当你应用很多RL或者优化压力的时候, 在模型成功的唯一方法就是做一些错误的对齐或者做一些不安全的事情。
630	And so that's another thing that we had to be very careful about and thoughtful about in training these wonders, ensuring that we curated the RL environments very thoughtfully to ensure that all of them were eliciting and driving towards more aligned beha	这是我们在训练这些奇观时必须非常小心和深思熟虑的另外一件事, 确保我们非常周密地研究RL环境, 以确保它们都能从模型中引领到更一致的行为。
631	So the point that you're saying about quality, I think is an important one, because I think a lot of the previous questions back and forth were about the quantity and have the flavor of like, oh, you need to scale the quantity.	所以,你所说的质量问题,我认为是一个重要的问题, 因为我认为以前很多问题都是关于数量的问题, 并且有类似,哦,你需要扩大数量。
632	But I mean, I do think like a huge part of this is just is is the point that you just made of like the even a smaller amount of high quality.	但我想,我确实认为这其中的很大一部分只是你刚刚做的点就像更小的高质量。
633	Yeah, I think I think in RL like we, you know, as we've seen pretty big jumps in our new spark line, I think these have been like accompanied with, you know, real meaningful improvements to the underlying recipe.	是的,我想在RL中,就像我们,你知道, 当我们看到在新的火花线上, 跳得相当大, 我认为这些就像,你知道,
634	And each of these improvements in the recipe have been like, you know, I think the improvement between new spark one to me spark one point three feels night and day.	食谱中的每一个改进都像,你知道,我认为, 新的火花的改善对我来说,火花的点三日夜的感觉。
635	I think probably to most people, if you ask them, like, are these like, you know, do these feel like they came from different pretrains?	我想对大多数人来说,如果你问他们, 喜欢,这些,你知道, 这些感觉像他们来自不同的前幕?
636	I think they probably would say yes, because I think you can, you know, I think it's happening industry wide for what it's worth.	我认为他们可能答应了, 因为我认为你可以,你知道, 我认为它正在发生整个行业的价值。
637	Just like we are learning to squeeze so much capability and so much and elicit so much agenda capability from, you know, these existing base models.	就像我们学着挤压如此多的能力并吸引如此多的议程能力从,你知道,这些现有的基础模型。
638	Now, that's not to say like we believe scaling is very important.	这并不是说我们相信缩放非常重要
639	That's why we have to use Wonderland in the first place.	故先须用仙境.
640	But but, you know, this sort of like these two things are happening at the same time, which is also, by the way, for for us, this is why we're seeing so much progress in general.	但是,你知道,这种事情是同时发生的, 顺便说一句,对于我们来说,这就是为什么我们总的看到这么多进步。
641	Like, I think it's the fact that we're scaling both pre-training and scaling our recipes in RL kind of like concurrently in a nice club.	喜欢,我想这是事实,我们正在扩大预训和放大我们的食谱在RL 类似同时在一个好俱乐部。
642	Yeah.	对
643	And then there's these other scaling axes, too.	还有其他的缩放轴
644	I mean, like making sure that you're the agent harness is good and that you can co-optimize these things is probably just another aspect of our dimension for scaling.	我的意思是,像确保你是代理的拉带是好的并且你可以合作优化这些东西可能只是另一个方面我们的维度放大。
645	But if I'm understanding you guys correctly, are you saying that alignment is not about coming up with some alignment specific RL steps?	但是,如果我理解你们正确, 你是说对齐并不是要提出一些对齐特定的RL步骤?
646	And instead about making sure the entire RL process is kind of steering the model.	而不是确保整个RL过程是一种引导模型。
647	You have to be thoughtful about it in the entire process.	整个过程你都要仔细考虑
648	Yeah.	对
649	So I think this is a it's a it's a holistic effort, I think.	我认为这是一个整体性的努力。
650	Yeah.	对
651	But also, I think different labs have very different theories.	但是,我认为不同的实验室有非常不同的理论。
652	I mean, alignment is a very philosophical concept of like, who do you want it to be aligned to?	我的意思是,对齐是一个非常哲学的概念像,你想要它与谁对齐?
653	And a lot of our theory is like, OK, we care more about it being aligned to what we're trying to give people agents.	我们的理论大多是这样的, 好吧,我们更关心的是它与我们试图给人们的代理。
654	Right.	对
655	And it needs to if you ask it to do something, you need to have a high confidence that it's going to do what you ask it to do.	并且它需要如果你要求它做一些事情,你需要有一个高度的信心,它会做你要求它做的事.
656	Yes.	对
657	You know, like you wanted to do that with high reliability.	你知道,就像你想这样做有很高的可靠性。
658	You don't want a moralizing lecture for why you're wrong for asking it to do the thing.	你不希望一个道德化的演讲,因为你为什么错误地要求它做这件事.
659	Obviously, if you're trying to do something that's catastrophic, it needs to gracefully without pissing you off.	很明显,如果你想做一些灾难性的事, 它需要优雅而不会惹恼你。
660	Right.	对
661	Like not do it.	喜欢不这样做。
662	But in general, we view alignment as more of a question of an instruction and values alignment with the person who it's intending to serve.	但总的来说,我们认为校正更是一个问题一个指示和价值校正与它打算服务的人。
663	And I think that's that's kind of a different mode of thinking about it.	我想这是一种不同的思考方式。
664	It's very different from, I think, how some of the others do.	这和,我想, 与一些其他人做的非常不同。
665	But one of the more interesting things, I think, that is maybe a question related to your original question is also just how long the process of training one of these models takes.	但其中一件更有趣的事情,我想, 也许是一个与你最初的问题有关的问题, 就是训练其中一个模型的过程需要多长时间。
666	像Muse Wonder,我的意思是,总的来说, 就像我们计划Muse Wonder 从上次,你知道,August。
667	So, yeah, I really the offside when we were.	所以,是的,我真的很离谱当我们是。
668	So this is like when they first presented this.	这就像他们第一次提出这个。
669	So I think all of the stages that we just talked about are like, you know, you really work on them for quite a long time.	所以,我认为所有我们刚刚谈论的阶段都像,你知道, 你真的工作了相当长的时间。
670	并不是说当我们完成Muse Spark后我们没有开始研究Muse Wonder。
671	It's that we've actually just been planning and working on this for quite a long time.	这是因为,我们实际上一直在计划和工作这一点相当长的时间。
672	And we really view this spark, if anything, is like I think we mentioned this in the first blog post where we wrote about it.	我们真的看到这个火花,如果有的话, 就像我想我们在第一个博客文章中提到了这一点。
673	那是通往Muse Wonder的阶梯
674	And that really was how we thought even at the time you thought about it.	即使在你考虑的时候我们也这么想。
675	And in terms of how long it takes, I mean, we're also even just talking about the research part.	至于需要多长时间, 我的意思是,我们也在谈论研究的部分。
676	I mean, there's also the compute part where we've been investing in building up, first of all, a huge amount of compute overall.	我的意思是,还有计算部分我们一直在投资建设, 首先, 大量的计算总体。
677	But a lot of part of our philosophy on this is that you can kind of measure the compute strength of a lab by how big its largest single computer is.	但我们关于这个的哲学有很多部分是你可以用一个实验室最大的计算机来测量它的计算强度。
678	Supercomputer for being able to do these these large training runs.	超级计算机能够完成这些大型训练
679	So, I mean, we built this Prometheus cluster.	所以,我的意思是,我们建立了这个普罗米修斯集群。
680	We call it in Ohio that's scaling to larger than a gigawatt.	在俄亥俄州,我们称之为规模大于千兆瓦。
681	And now we're building this Hyperion cluster in Louisiana that's going to scale to five gigawatts.	现在我们正在路易斯安那州建造一个超离子星团, 规模将达到5千兆瓦。
682	And those actually predate MSL because like building a whole like data center site from scratch, at least the plans for those, they're coming online now.	而那些实际上比MSL早, 因为像从零开始建立一个像数据中心站点一样的整体, 至少这些站点的计划, 他们现在上线了。
683	But that was you guys thought it might be the feed, right, that you just have used for a lot of compute generally.	但你们觉得可能是饲料, 对,你们只是用了很多计算法。
684	No, no, no.	不,不,没有。
685	This is this is kind of predating even going back to the to kind of the llama program.	这是种预兆甚至回到某种拉玛程序。
686	I mean, it was probably in 2023 or either 2023, 2024 was when we basically started making the plans to build those clusters.	我的意思是,大概是在2023年,或者2023年,2024年我们基本上开始计划建造这些集群。
687	They're now kind of fully coming online.	他们现在已经完全上线了。
688	Well, I guess the Louisiana one will come online over over a couple of years.	我猜路易斯安那州那家过几年就会上线了
689	But but if you want to go really end to end on undoing this stuff, then I mean, obviously the the kind of full stock and I mean, met as a full stack tech company.	但是,如果你真的想结束关闭这些东西, 然后,我的意思是,很明显的, 那种完整的股票,我的意思是, 遇见一个完整的堆技术公司。
690	I mean, we build we build our own data centers.	我的意思是,我们建造了我们自己的数据中心。
691	We build our own chips.	我们自己造芯片
692	We build our own low level infrastructure.	我们建设自己的低水平基础设施。
693	And now this is kind of the next component, which I mean, arguably is probably the most important component in all of this for the next wave of technology is is kind of building really good models and then delivering them to people.	现在,这是下一个组成部分, 我的意思是,可以说, 可能是所有这一切中最重要的组成部分, 因为下一波技术是一种建立真正好的模型, 然后把它们交给人们。
694	But I mean, yeah, if you want to go like fully end to end, I mean, this is a yes, it's a yeah.	但是,我的意思是,是的, 如果你想去像完全结束到结束, 我的意思是,这是肯定的,这是对的。
695	Yeah, I do want process.	是的,我确实想要过程。
696	I do think this is like a I mean, maybe it's well appreciated by folks in the tech industry, but like it really is, you know, it feels like models are coming out constantly.	我觉得这就像一个我的意思,也许它受到科技行业的人的好评, 但就像它真的,你知道, 它感觉模型不断涌现出来。
697	So maybe you'd have the mental model that like, oh, a company like launches a model and then they figure out how they're going to make the next model and or something like that.	所以,也许你会有一个精神模型, 比如,哦,一个公司像发射一个模型, 然后他们想出他们要怎么做下一个模型,或者类似的东西。
698	But it's actually these like very long planning processes and these sort of like very long work streams that kind of all come together at the very end and then results in these beautiful models.	但实际上,这些就像很长的规划过程, 和这些很长的工作流, 这些都集中在最后,然后形成这些美丽的模型。
699	Yeah.	对
700	How do you build good products when the feedback cycle is so long, like ultimately you start planning the model in August and you don't know until more than a year later if it's good.	如何在反馈周期如此长的时候构建好产品,比如最终在8月开始计划模型,直到一年多后才知道是否良好.
701	And you know, Silicon Valley famously likes to run on metrics.	你知道,硅谷著名的喜欢运行的度量衡。
702	But like ultimately, do you just need someone who is the estate and so it's nice when you have metrics.	但归根结底,你需要一个人谁是庄园, 所以当你有衡量标准,这是很好的。
703	But I think the like state of being an entrepreneur is like, how well do you make decisions in an information poor environment?	但我认为当企业家的状态就像, 在信息不良的环境下,你做决定有多好?
704	Right.	对
705	So but there's like that on steroids, right?	但是类固醇上也有类似的东西,对吧?
706	Mega steroids.	巨型类固醇
707	Yeah, I think that this is and I think Mark gets a huge amount of credit for this.	是的,我想这是而且我觉得马克得到巨大的信用这一点。
708	Like the because of how long everything takes, like there there were many points along over the past year since I've been here, but also, you know, as you mentioned, like long before that, where you have to make very large investments or make very big dec	比如说,因为一切都需要多长时间, 比如说,自从我到这儿以来,过去一年里, 有很多问题,但是,你知道,正如你提到的, 很久以前,你必须做出非常大的投资, 或做出非常大的决定, 你知道,你打算如何投资或开发技术, 或者,你知道,你将要做出什么决定远没有证据, 你知道,你是否甚至事情正在起作用。
709	And so I think this is this is like, you know, in some sense, like entrepreneurship on hard mode, it's like the opposite of lean startup.	所以我觉得这就像,从某种意义上说, 像在硬模式下的创业, 和精益启动相反。
710	Like lean startup is like, yeah, you like, see if you get something working and then you double down and then you see if that works and you double down.	就像精益求精的启动就像,是的,你喜欢, 看看如果你得到的东西工作, 然后,你加倍下来, 然后,你看如果工作,你加倍下来。
711	Like you cannot do that in AI.	就像你在人工智能里不能这么做
712	You have to like weave.	你必须喜欢编织。
713	I mean, we've been investing into these data centers since, you know, for years and years and years.	我的意思是,我们一直在投资这些数据中心从,你知道, 多年来和几年。
714	We've had to and you'll be radically behind if you wish to see things are working.	我们不得不这样做,如果你希望看到事情正常的话, 你会被彻底的落后。
715	We have these conversations where it's like the end of last year.	我们有这些对话,就像去年的年底。
716	We're like, all right, we are quite excited about the progress.	我们喜欢,所有的权利, 我们非常兴奋的进展。
717	We haven't seen the models yet.	我们还没有看到模型。
718	Like, it's like, I'm going to like, you know, we're going to write this check for like, I don't know, like a hundred billion dollars or two hundred billion dollars.	就像,我要,你知道, 我们要写这张支票像,我不知道,像千亿或两千亿。
719	I'm like, Alex, you sure this is going to work?	我很喜欢,亚历克斯, 你确定这会工作?
720	It was it was so extreme that, you know, we had we had like the whole exec team.	这是如此极端,你知道, 我们有我们像整个执行团队。
721	Yeah, well, yeah, no.	是啊,是啊,没有。
722	Well, it's like the the joke every time we talked about would just be like, yeah, and we hope it's all going to work and be good.	就像我们每次谈论这个笑话都会觉得,是的,我们希望一切都会好起来
723	But like there were literally like the end of last year, other than me giving a vibe report or like research was going and, you know, I could sort of show some plots and whatnot.	但是,就像去年的年底, 除了我做一个感官报告,或者说研究正在进行, 我可以展示一些情节和什么的。
724	But it was it was like I was like, I mean, I was feeling good about it, but it was like certainly not clear at all that it was going to work.	但它就像我一样,我的意思是,我感觉很好, 但是,它当然不是很清楚它会成功。
725	You guys are saying that the essence of training a frontier model is you have to push your chips all in before you've been dealt the cards.	你们说训练边疆模特的本质就是在牌被处理之前把芯片都推进去
726	You know, you just have to decide that you're going to develop a frontier model and you don't get a meter.	你知道,你只需要决定你要开发一个前沿模型而你没有得到一个表。
727	At this scale.	依此规模.
728	I mean, if we like back a few years ago, the clusters didn't have to be this big, but now they do.	我的意思是,如果我们喜欢回到几年前, 集群不必这么大, 但现在他们做到了。
729	So I think this is part of the the excitement of like booting up a frontier lab now, right?	所以,我认为这是兴奋的一部分像启动前沿实验室现在,对不对?
730	Is is basically like the ante is is extremely high.	因为它基本上就像 ante是极其高。
731	And I think this is and like I kind of like viscerally remember this.	我觉得这就像我有点喜欢直觉地记得这个。
732	We were in Hawaii at the end of last year is kind of like the whole exec team got together.	去年年底我们去了夏威夷就像整个执行团队在一起一样
733	And and yes, Susan pulled me aside many times.	是的,苏珊多次把我拉到一边
734	It was like, you know, I just want to tell you if this doesn't work, like, you know, this is going to be really, really bad.	这就像,你知道, 我只是想告诉你如果这不工作, 像,你知道,这将会是真的,真的很糟糕。
735	It's going to be very, very bad in financial terms.	从财政角度来说,这将会非常非常糟糕。
736	I really need to impress on you that this model has to work like it has to be good.	我真的需要给你留下深刻的印象这个模型必须像它必须好。
737	If it's good, it's all good.	如是诸善.
738	We're great.	我们很好。
739	But but if it's bad, it's really, really, really bad.	但是,如果它是坏的, 它真的,真的,真的。
740	And Susan, who you know, that was useful motivation.	苏珊,你知道, 这是有用的动机。
741	It was well, it was like it was one of these very funny moments where I was like, you know, obviously the amounts of money are like so like obscenely large.	感觉好极了,这就像一个非常有趣的时刻, 我当时喜欢,你知道, 很明显,钱的金额就像是如此下流的庞大。
742	What about the amounts of money in training or whatever you can share?	训练中的钱或者任何你能分享的东西呢?
743	But just like, well, you can do the basic math.	但就像,嗯,你可以做基本的数学。
744	I mean, a gigawatt cluster.	我的意思是,一个千兆瓦集群。
745	I mean, it's like, I don't know.	我的意思是,这就像,我不知道。
746	I think what you'd see out there would be, you know, to operate a gigawatt a year, call it on the order of 15 billion dollars.	我想你会看到,你知道, 每年操作一千兆瓦, 把它叫做150亿美元。
747	OK, so it's roughly a 15 billion dollar effort to train to operate to operate a gigawatt cluster for a year.	大约是150亿美元的努力, 用来训练操作一个千兆瓦集群一年。
748	If you're talking about building the CapEx, it's probably closer to, you know, you pay for it up front.	如果你在谈论建造CapEx, 它可能更接近,你知道,你支付它前面。
749	So it's probably, you know, 50 or 60 billion dollars.	所以大概是500或600亿美元。
750	So but then that's just kind of one cluster for one part of training.	所以,但这只是一个集群的一部分培训。
751	Then you have all the other parts.	你还有其他的
752	Research isn't just doing the run.	研究并不只是在运行。
753	Right.	对
754	It's like that's like probably an equal amount of it is because for each change that you make,	可能是因为你所做的每一件改变
755	you're testing all the ladders all the way up to the big the big scale.	你正在测试所有的阶梯一路到大尺度。
756	And then they're serving the stuff.	然后,他们正在服务的东西。
757	So part of the stuff that in some ways of billions of dollars to train a frontier model.	因此,一部分的东西,在某些方面数十亿美元来训练一个前沿模型。
758	Yeah, definitely the whole I've seen the financial world tens of billions of dollars is a good estimate for.	是啊,我所看到的金融界上百亿美元是一个很好的估计。
759	Yeah.	对
760	And then but then but then you want to do something with it.	然后,但后来,但之后你想做一些事情。
761	Right.	对
762	So I actually think like some of the more complicated questions that we had were like, all right.	所以,我实际上认为,像一些更复杂的问题, 我们曾经喜欢,好吧。
763	So we're training this model and we know we know we want to train the model because this is like key.	所以我们在训练这个模型我们知道我们想训练这个模型因为这个模型就像关键
764	But like how much product demand do we think we're going to have?	但是,像多少产品需求我们认为我们将有吗?
765	So now it's like, all right, well, the product doesn't exist.	所以现在它就像, 好了,嗯,产品不存在。
766	Like and as you're trying to build data centers to serve muse for a product that doesn't exist.	就像和当你试图建立数据中心服务缪斯对于一个不存在的产品。
767	Yeah.	对
768	And we have to make in years in advance.	我们必须提前几年才能完成
769	Right.	对
770	So this is probably where like when we were talking about it last year was like, huh.	所以这大概就是去年我们讨论的时候的样子了
771	There was a moment where I think a bunch of people were like, yeah,	有一瞬间,我想一群人喜欢,是的,
772	maybe maybe we should just like be a little more conservative.	也许我们该更保守一点
773	And it's like, no, we're doing this.	而它就像,不, 我们正在这样做。
774	It's like it's like we're going to make it work.	如来如是成就.
775	This is going to work.	这会成功。
776	And but it was it was like, I mean, even I mean, I, too, am an entrepreneur and founder and have, you know,	但它就像,我的意思, 即使我是,我也是一个企业家和创始人,并且,你知道,
777	at various points in life taken risk.	在生命的不同阶段冒险。
778	I remember the visceral feeling at the time was like, this is a fuck ton of risk.	我记得当时的内脏感觉就像,这是他妈的一吨风险。
779	Like, I I don't know if I know it's only risky.	喜欢,我不知道,如果我知道它只是风险。
780	Well, I do think this is like one of one of the like things that I think Mark is very impressive about you,	嗯,我确实认为这是类似的事情之一我认为马克是非常令人印象深刻的你,
781	which is I think you have really internalized this concept of, you know,	我认为你已经把这个概念,你知道,
782	it's about how big it is when you win, not like, you know, failure is OK.	关键是当你赢了它有多大, 而不是,你知道,失败是好的。
783	It's just like how big how big is the win when you when you do win.	就像你赢的时候赢多少
784	And this was I think Mark has fully internalized power laws in a way that many people I think quite deeply.	我认为马克已经把权力法完全内部化了, 和许多人一样,我深思熟虑。
785	Well, I think the other pieces, I don't actually think it's quite as risky as this.	我认为其他的片段, 其实我并不认为它很冒险。
786	I mean, the numbers are big.	我的意思是,数字很大。
787	But I think META has a backstop in terms of how it can use confused.	但我认为META在如何使用混淆方面有后盾.
788	Yes, I think there's two things.	是的,我想有两件事
789	One is if AI works overall, plenty of other people want all our compute.	一个是如果人工智能能全面工作很多人想要我们的计算
790	Right.	对
791	So they're willing to pay like, yes.	所以,他们愿意支付像,是的。
792	I mean, if if like the cost is whatever, 15 billion a gigawatt, we could get a lot more for that.	我的意思是,如果像成本是什么, 150亿千兆瓦, 我们可以得到更多的为此。
793	So so I think that you in order for it not to work out, it's not just that we have to fail.	所以我觉得你为了不成功不只是我们得失败
794	AI overall has to fail.	总的来说,大赦国际必须失败。
795	Yes.	对
796	And then even in that happens, we still have all these other uses to productively use compute.	即使在这样的情况下,我们仍有其他的用途用于生产。
797	So in the grand scheme of things, we were definitely building out this compute for inference ahead of knowing that the product was going to be ready.	因此,在大计划里,我们肯定在计算这个推论之前, 已经知道产品已经准备好了。
798	But your AI pulled enough that you said, and like you use online and powered will be useful to someone.	但你的人工智能拉得够多了就像你在网上用电一样对某人有用
799	But also like we believed in what we were building.	但就像我们相信我们建造的东西一样。
800	I mean, it's like I think and you can look at the kind of the trajectory where it's like, all right.	我的意思是,这就像我想, 你可以看那种轨迹,它像,好吧。
801	First, there's chat.	首先,有聊天。
802	GPT is the text based assistant.	GPT是基于文本的助手.
803	Then, you know, Anthropic got coding agents to work and no one had built the kind of personal consumer version of that.	然后,你知道,Anthropic得到了编码代理工作,没有人制造这种个人消费版本。
804	And it just seemed like this is the thing that META should build in the world.	这似乎是META在世界上应该建造的东西。
805	And I think that that was just like pretty clear to you and me and not like from the beginning.	我认为这对你我来说是相当清楚的,而不是从一开始。
806	Right.	对
807	That that's like that's kind of what we're here to do.	这就是我们在这里要做的
808	And so I don't know.	故不识.
809	I mean, there's always like the real question for me on the Muse launch even was you never know when you launch something, how it's going to be received.	我的意思是,总是有像真正的问题对我来说在Muse的发射甚至你永远不知道当你发射的东西, 如何会得到它。
810	Right.	对
811	So there's always the case where you get a homerun, which in this case it was.	所以总有你得到一个全垒打的情况, 在这个例子中它是。
812	And it's just like extremely well received right off the bat.	它就像极好得到的球棒。
813	But the normal case would be you kind of it's fine.	但正常的情况是,你还是可以的
814	And you have to like iterate a few times to get product market fit before you really grow.	在你真正成长之前,你必须喜欢过几次脚踏实地才能让产品市场适应。
815	In some ways like I actually think the product we maybe could have even grown it faster, but we didn't have all the growth stuff really ready to go when we shipped it because we are like our median expectation was like, all right, we're gonna have to like	在某些方面,我觉得我们甚至可以更快地发展它, 但是当我们发货时,我们并没有真正准备好所有生长的东西, 因为我们就像我们的中位期望一样, 好吧,我们得喜欢一下这个。
816	And like, that'll be that'll be fine.	就像,那会是罚款。
817	Don't know.	不识.
818	I think like if it takes a few iterations, it's like not the end of the world.	我认为,如果它需要一些迭代, 它就像不是世界末日。
819	That's normal.	这是正常的。
820	Yes	对
821	You can have launches where it's like you get like it's kind of catastrophic and then it takes a while to recover	你可以有发射的地方它看起来像你得到像它是一种灾难性的然后它需要一段时间恢复
822	But that's not good.	但那样不好
823	But but it's like but I mean even if it took a few iterations, it would have been fine	但是它就像,但我的意思是,即使它需要一些迭代, 它会一直很好
824	So, I don't know.	所以说,我不知道。
825	I guess it just seemed to me like this was gonna be yeah, the numbers were big	我想这对我来说就像会是, 数字很大
826	But I it seemed like it was probably	但我看起来可能
827	Was it bonding for the executive team to be?	执行团队的感情是紧密相连的吗?
828	Holding hands taking this risk together.	牵着双手一起冒险
829	You're talking about Susan taking UXI.	你说的是苏珊服用UXI。
830	Is that a bonding moment?	那是一个亲密的时刻吗?
831	It's like I mean, it's stressful.	这就像我的意思, 它的压力。
832	It's I think it was	我觉得是
833	Yeah, being in a stressful situation together is generally bonding.	是啊,在一个紧张的情况在一起是一般联系。
834	Well, one thing that I think is quite notable about meta	嗯,有一件事,我认为是相当显著关于元
835	is	这是
836	the degree to which like first of all, most of the exec team has been there quite a long time like Nat and I were	我和纳特在一起很久了
837	really newbies and	真的新来的和
838	and	和
839	You know most people have been through	你知道大多数人都经历过
840	crazy	疯了
841	situations many times in the past as well and I think I think meta as a company is like an extremely resilient company that has	过去也发生过很多次,我想作为一家公司, 元像一个极有弹性的公司
842	been through like, you know	一直这样,你知道
843	many situations that also at the time would have seemed incredibly daunting and so everyone's kind of like	很多当时也似乎非常令人生畏的情况所以每个人都喜欢
844	Knows the program so to speak in terms of you know, this obviously on	这么说来,这很显然
845	On paper and especially when like Susan Lee our CFO is sort of like pulling me aside in the meetings	在纸上,特别是当像Susan Lee 我们的首席财务官有点像把我拉到会议一边
846	Does you know does seem like a particularly	你知道这看起来很特别吗?
847	Scary situation, but I think I think most people are like, yeah.	情况很可怕但我认为大多数人都喜欢
848	Yeah, no, but we're meta	是啊,没有,但我们是元。
849	We're gonna get through it	我们会熬过去的
850	Like I think it was like a like a real camaraderie around that and and sort of a like quite a nice	我觉得这就像一个真正的友情周围和类似一个相当不错的
851	Quite a nice energy around that	周围的能量不错
852	Which which which really I did appreciate because I think it would be very easy for	哪个我真的很感激因为我觉得这很容易
853	You know, it's the other executives at the company who are	你知道,是公司的其他高管谁是
854	Operating and running her big bills.	操作和运行她的大帐单。
855	Yeah, one of the greatest businesses of all time	是啊,有史以来最伟大的企业之一
856	and and it'd be very easy for them to be	和这将是非常容易他们成为
857	quite a quite annoyed by this whole endeavor, but but I think I think everyone is also very excited about their what the	对这一切相当恼火但我想大家都很兴奋
858	What the opportunities for the company with AI are and I think like, you know taking a step back.	和AI公司有什么机会我想,你知道退一步。
859	I think one of the	我觉得其中之一
860	one of Mark's jobs is like the founder and CEO of the company is to ensure that the company is always	马克的其中一项工作就像公司的创始人兼首席执行官是要确保公司永远
861	pushing towards the most ambitious and	向着最宏伟的
862	Expansive version of what it can be and I think that like one of the great things about	精致的版本它可以是什么我认为就像一个伟大的事物
863	this entire MSL and AI push is that like I think this is like one of the rawest expressions of that is like	整个MSL和AI的推力就像是我认为这是最原始的表达
864	You know, we we knew that the you know	你知道,我们知道,你知道
865	AI certainly is a very very important technology and it's you know, one of the bigger technologies right now	AI肯定是一个非常重要的技术,你知道的,一个更大的技术现在
866	but as a	但作为一个
867	you know if it if we were able to build it effectively and properly integrate it into everything that we do as well as like	如果我们能有效建造它并把它适当融入我们所做的一切
868	You know as a full-stack technology company like build it into you know	你知道作为一个完整的技术公司就像把它建进你知道的
869	these incredible products that we know will exist in the future this could be like immensely expansive for Metta and	这些不可思议的产品,我们知道将来会存在这可能像巨大的扩张对于梅塔和
870	I think this is true.	我认为这是真的。
871	Like I think Muse is an incredibly exciting product	我觉得Muse是个令人兴奋的产品
872	It has like, you know, you can go in a thousand different directions	它就像,你知道, 你可以走一千个不同的方向
873	I think is one of like probably the most exciting products to just like, you know	我觉得是其中之一可能是最刺激的产品只是喜欢,你知道
874	When like late at night and you just think about all the ways it can go like I think not	当深夜一样,你只是想想所有的方式它可以走像我想不
875	Mark and I just like, you know shoot across messages late at night of just like oh we could do this and we do this	马克和我只是喜欢, 你知道在深夜拍摄消息,就像哦我们可以做到这一点,我们这样做
876	And you know, we can make this move and we can like build this product and I think it's just like	你知道,我们可以做这个动作我们可以喜欢建造这个产品我想它就像
877	you know, very	你知道吗,非常
878	It's very exciting time in technology.	科技时代非常激动人心.
879	Yeah, it's very blue ocean in that way.	是啊,这是非常蓝色的海洋的方式。
880	Yeah.	对
881	Yeah, I'm struck by as we talked about	是啊,我被我们谈论的
882	Muse and	缪斯和
883	Just how it clearly feels like the next platform an agent operating on your behalf	就像下一个平台代理你的经纪人一样
884	You've talked for a long time about not wanting to be	你已经说了很久不想成为
885	Reliant on other platforms and how kind of the experience of building on	在其他平台上重新亮相,以及如何利用
886	Android and iOS kind of burned you on being dependent on other and often when you talk about say reality labs is the importance of owning	Android and iOS 那种烧你依赖他人, 并且经常当你谈论现实实验室是拥有的重要性
887	the platform of the future and kind of many years working on reality labs with the intent of own the	未来和许多年的平台在现实实验室工作目的是拥有
888	platform of the future and now	未来和现在
889	This new	这个新的
890	Modality comes along that feels pretty clearly like how you do a lot of productive work and is the platform of the future	模式的出现感觉很明确就像你如何做很多生产性工作, 是未来的平台
891	Yeah, and now men is really well set up.	是啊,现在男人们都准备好了。
892	Yeah.	对
893	No, I mean, it's less about	不,我的意思是,它没有
894	wanting to own a platform for its own sake it's about being able to build the best things and	想要拥有一个平台为了它自己,它就是能够建造最好的东西和
895	It's some level, you know, if you look at like other social apps	你知道,如果你像其他社交应用程序一样看,这有点关卡
896	For example, I think part of the reason why meta is so much better is because we don't just build the app	比如说,我认为元之所以这么好, 部分原因是因为我们不光是构建应用程序
897	A lot of people think that we're like this app company.	很多人认为我们就像这个应用公司.
898	We're not an app company, right?	我们不是一个应用公司,对不对?
899	It's like we're like a deep technology company and we do this because we are world-class of machine learning and like all the	这就像我们像一个深科技公司我们这样做,因为我们是世界一流的机器学习像所有的
900	Recommendation systems are operating at a scale that basically few others are in the world	推荐系统的运作规模世界上几乎没有其他的
901	I think it's basically like just us and by dance on that	我觉得它基本上就像只有我们和通过跳舞
902	and	和
903	And then like okay, we just kind of go all the way to the metal and to me	然后像好的,我们只是那种一路到金属和我
904	Being able to also deliver go the last mile through the device.	也能够送货通过装置走最后一英里。
905	It does come from	它确实来自
906	Awhile of having these experiences where we couldn't build the experiences that we wanted on phones	有了这些经历之后我们无法在电话上积累想要的经验
907	but I just think for the future the	但我只是认为未来
908	The models are going to be by far the most important part of delivering the technology experience	这些模型将是提供技术经验的最重要部分
909	And I mean, this is something that I think	我的意思是,这是一件我认为
910	We kind of bonded on agreed on like very early is this notion that like in every generation of technology companies	我们很早就同意了这个概念就像每代科技公司
911	There's like no, they're usually not one company that kind of is the is like wins everything	好像没有,他们通常不是一个公司那种是像赢得一切
912	I don't think that that's how this is going to go either	我也不这么认为
913	But with the internet there were like okay five or six a very big companies and then a whole ecosystem of other companies that	但随着互联网,有像好五六个非常大的公司, 然后整个生态系统其他公司
914	Are great, but the companies that did the best	很棒,但公司做得最好的
915	Kind of had the core technology to be able to	有一种核心技术可以
916	Serve people the best there and I think in this next era	给人最好的服务,我想在下一个时代
917	They're going to probably be a small handful of companies that have the ability to build frontier super intelligence	他们很可能是少数有能力建立前沿超级情报的公司
918	And I think that those companies are going to be the ones that effectively drive the agenda and can build most of the interesting the most	我认为,这些公司将有效推动议程,并且能够建立大多数有趣的公司。
919	interesting	有意思的,你觉得呢?
920	Experiences for consumers for enterprises hardware these different kinds of things	企业硬件对消费者的经验这些不同类型的东西
921	So when I just look at it, it's okay.	所以,当我只是看它, 它没事。
922	Like is it even really a choice, right?	就像它真的是一个选择,对不对?
923	It's like it like yeah, okay meta could will probably exist into the future regardless of whether we do this	这就像是的,好吧元可能存在在未来,无论我们这样做
924	but if we want to be kind of an interesting company that is like	但如果我们想成为一个有趣的公司就像
925	Inventing new things that matter to people in the future	发明未来对人很重要的新事物
926	I don't see how you can do that if you don't have the ability to build end-to-end your own	我看不出你怎么能做到这一点如果你没有能力建立端到端自己的
927	Superintelligence and build them to the products that people use and I think this is something that you just like believe from very early on	超级智能并把它们建在人们使用的产品上我认为这是你从一开始就喜欢相信的东西
928	Is basically it's a guy they're gonna be like a handful of these	基本上是一个家伙他们会像一个小的这些
929	Like it's like there's a kind of personal and consumer oriented version of this and like this lane is open because everyone else seems to just	像是有个人和消费者的版本这条车道是开放的因为其他人似乎都只是
930	Be obsessed so much with just automating knowledge work	这么着迷于把知识工作自动化
931	I mean isn't even that charismatic version of the future right because it's like all right	我的意思是,甚至不是魅力的版本未来的权利,因为它像所有的权利
932	like a big part of our belief too is that the kind of highest use of superintelligence is invention and creation not automation and	就像我们信仰的一大部分一样超级智能的最高用途是发明和创造而不是自动化
933	So, I don't know it seemed like there's like a big place in there, but you gotta you kind of need to own the stack	所以,我不知道这似乎有像一个大的地方在那里, 但你得种需要拥有的堆
934	how is	怎么样
935	the muse mediated future going to shake out with	缪斯调解的未来将动摇与
936	Businesses need to okay businesses variously need to build support for	企业需要为企业争取各种支持
937	For	对于
938	Muse where it might try to computer use its way in but that might be janky in certain ways	在它可能试图使用计算机的地方, 它可能在某些方面是粗糙的
939	So you kind of want everyone to have first-class support for agents talking to things	所以你希望每个人都有头等舱的支持代理商说话的东西
940	Yeah, then the bigger question is they might actively try to	是啊,那么更大的问题是他们可能会积极尝试
941	Agents from interacting with them and this might be because they have a misguided sense where like it's actually in their interest to support agents	和他们互动的特工可能是因为他们有错误的感知而这种感知实际上符合他们的利益
942	But they don't want to it might actually not be in their interest to support agents	但是他们不想支持特工可能并不符合他们的利益
943	You can imagine where maybe they have a very thin moat and it all comes from you know a business that only	你可以想象他们哪里有一条很薄的护城河所有这一切都来自你认识的生意
944	Succeeds by consumers spending time on the site or only succeeds by overcharging consumers then like they should	消费者成功在网站中度过时间或只成功过重的消费者
945	Rationally not want to allow agents because the agents are like wow	理性地说,不允许代理商因为代理商就像哇
946	Those are gonna get so	这些会变得如此
947	Talk to me just through the second-order effects of like how the economy works	跟我谈谈经济运行的二阶效应
948	Well, I think that it starts off with I mean we trained into the models very good browser use	嗯,我想它从一开始,我的意思是我们训练的模型非常好的浏览器使用
949	Right so that way we want to make it so that no one no one has to do anything to integrate with muse	这样我们就能让任何人都不必做任何事来融入缪斯
950	And then but then of course for we're kind of just going down the list of the most important partners and	然后,但是,当然,我们有点只是去名单中最重要的伙伴和
951	For people who want to work with us	对想和我们合作的人来说
952	We'll just get direct API access	我们直接进入API
953	Right and then it's like okay, then it ends up being way more efficient and faster for people to use muse	右,然后它就像好的, 然后它最终是更有效率和更快人们使用缪斯
954	It's more efficient because now you don't have an agent crawling your website	效率更高,因为现在你没有代理爬到你的网站
955	It's just going so anyone who wants to do that we eventually want to get there	它只是去,所以任何人谁想做,我们最终想得到那里
956	And then I think the fullness of time is that you know just like today every business has	然后我觉得时间的充裕就是你知道,就像今天一样
957	A website and social media accounts and email addresses and all these things every business is gonna have agents, too	网站、社交媒体账号、电子邮件地址和所有这些东西每个企业都会有代理商
958	so and we're also building that on the other side of it and	所以,我们也正在建造在它的另一边,
959	And so of course muse agents will talk to other types of agents	当然Muse探员会跟其他类型的探员谈
960	But they'll work really well with our agents right because I mean we'll we'll be able to integrate that	但是他们和我们的特工们合作得很好因为我想说,我们会把这件事融为一体的
961	Just I mean so muse using websites just a moment in time thing	我的意思是,如此缪斯使用网站只是片刻的时间的事情
962	They just to make it easy for everyone you guys want to have it as a backstop	他们只是为了方便大家你们想当后援
963	So it's reliable for people using it yes and easy for businesses to get bootstrapped	所以对使用它的人来说是可靠的,对企业来说是容易的
964	But we have we have way more businesses that are calling us asking to integrate deeply	但是我们有更多的企业要求我们深入融合
965	And very few really	很少有人真的
966	Who have tried to block it and my guess is that the ones that try to block it it will end up?	谁试图阻止它,而我猜那些试图阻止它的人最终会结束它?
967	Not being the right strategic thing for them, but we'll see this will this will come play out over time	对他们来说不是正确的战略手段但我们会看到这会随着时间推移而出现
968	It saves them a lot of time it saves them a lot of money like I think these are	省了他们很多时间省了他们很多钱就像我想的
969	These are like clearly good things for the consumer so in the long arc you'd clearly expect that	这些对消费者来说都是好东西所以在漫长的弧线里你显然会期待
970	Demand will shift from current avenues more and more towards	需求将越来越多地从现有渠道转向
971	Agentic demand and you know there always will be	代理需求,你知道总会有
972	Demand that lives elsewhere just in the same way that like there's still brick-and-mortar demand in commerce	要求住在别处就像在商业上还有砖块和摩尔塔一样
973	But there's a lot of e-commerce demand and kind of in the same way like I think	但电子商务需求很大和我想的一样
974	agentic commerce demand will grow as a portion of the pie the others will still exist will grow as a portion of the pie and	代理商业需求会增长部分馅饼其余的会增长
975	I think there's sort of these questions for	我想有些问题要问
976	businesses in the long run like do they want to participate in that slice or that portion of the economy over time and	从长远来看,他们是否希望长期参与经济的一部分或部分?
977	I think it's an incredibly exciting area.	我认为这是一个令人难以置信的刺激领域。
978	I think it's like very very exciting to imagine	我觉得这就像非常令人兴奋的想象
979	What will happen in this agentic commerce future?	在这个代理商的未来会发生什么?
980	I actually think Patrick wrote a great	我觉得帕特里克写得很棒
981	Yeah	对
982	Tweetical is that what we call it these days.	令人惊叹的是,现在我们称之为它。
983	I don't know Patrick has his own thing going on with tweets	我不知道Patrick自己在推特上也有问题
984	It's like they're buried amidst.	好像埋在他们中间一样
985	I noticed the brother Karmatsov was a was a quite a yeah	我注意到Karmatsov哥哥是个很厉害的人
986	That was long quite a good literary analysis, but uh yeah	说来话长文学分析不错不过是的
987	Well, you know I think I think like I think clearly there's this is a	嗯,你知道,我想我觉得像我想显然有一个
988	discontinuous event for how	如何终止事件
989	You know commerce will happen broadly speaking or just generally speaking have like the entire how all the business pipes will work	你知道商业将发生广义上,或只是一般地说, 拥有像整个如何所有的商业管道将工作
990	Maybe even in a broader sense, but I think that you know you just go back to this is clearly a lot better for the consumers	也许从更广义的意义上来说,但我想你知道你刚刚回到这个显然对消费者更好
991	It's a lot better for the customers, so there's a natural gravity towards that reality one of the exciting things	这对顾客来说要好得多所以对现实来说自然有引力令人兴奋的事情之一
992	I'm curious what you guys see too	我也好奇你们看到什么
993	But I think I think news is one of the first products where you're actually people are fully delegating	但我认为新闻是最早的产物之一在那里,你实际上是人们完全授权
994	Purchases to the agents, and this is like a this is kind of like a you know a step change, right?	给特工们买东西这就像一个... ...你知道一个步骤的改变,对不对?
995	Like I think you know we've been talking a lot about agentic shopping and like the possibility of it	像我想你知道我们一直在谈论代理购物,喜欢的可能性
996	But now it's like really happening and a lot of people are trusting the agents to shop for them	但是现在就像真的发生了一样很多人都相信代理商会替他们买东西
997	And I think the agents are finding them you know in my case like great teas or you know	我觉得特工们正在找到他们你对我而言就像喝茶一样
998	Who's ever case like great products for them to purchase?	谁会像大产品一样买?
999	Which is quite exciting and and so I think that there's	这是相当令人兴奋的, 所以我认为有
1000	There's clearly	很明显
1001	An exciting thing happening	令人兴奋的事情发生了
1002	I think I think part of the value prop well one one part of the value prop for the users just reduce friction like hey	我想我认为部分价值道具好一部分价值道具给用户只是减少摩擦像嘿
1003	I don't feel like click on all these websites and do all this research to try to find the exact products like that step is	我不喜欢点击所有这些网站并做所有这些研究试图找到精确的产品像这个步骤是
1004	Taking care of for me and also like criterias can just be automatically applied like hey	照顾我和像标准一样的只能像嘿嘿一样自动应用
1005	I need something by Wednesday, so you know the agent will only look for things that could get to me by Wednesday	周三之前我需要一些东西,所以你知道经纪人只会在周三之前找到可以找到的东西
1006	There's I mean keyword search it was insane that we ever relied on those because that's a very reasonable constraints to want to shop by	我的意思是关键词搜索,这是疯狂的我们曾经依赖这些因为这是一个非常合理的限制想要购买通过
1007	Yeah, exactly and and and and so I think the like old way of like searching for stuff just feels so obviously primitive now	是的,准确的,而且,所以我认为像旧的方式喜欢寻找的东西只是感觉如此明显原始现在
1008	now that you can do this then there's this sort of like	现在,你可以做到这一点, 那么有这种喜欢
1009	Back part of the funnel so to speak, but just like the ease of friction reduction of just yeah filling up	说来话长但就像摩擦减少的轻便程度
1010	Just like getting it ordered	就像得到它订购
1011	Is quite nice, and I think the long term thing that we're quite excited about is is greater personalization like ultimately	很不错,我觉得我们很兴奋的是个人化程度更高
1012	We think I think one of the exciting things about news is that it is has sort of like this holistic	我觉得新闻的令人兴奋之处之一就是它有点像这个整体
1013	Context into you as a person and all of the aspects of your life	一个人的背景和生活的各个方面
1014	So it's sort of like can holistically understand things about your family or things about your co-workers or things about your friends and then like	所以,它有点像可以整体理解的东西关于你的家庭或者事情你的同事或者事情关于你的朋友,然后喜欢
1015	You know sort of like help you co-optimize across all of those to get you the exact right gift for your you know	你知道那种类型的帮助你合作优化在所有那些给你准确正确的礼物,你知道
1016	Best friend in the context of your friend group based on what it knows about your fire	根据你对火情的了解你朋友小组里最好的朋友
1017	Like I think there's just like a lot of exciting opportunities	就像我觉得有很多刺激的机会
1018	that that are somewhat new and novel and when it comes to like these personalization these recommendations because historically a lot of these recommendation systems are	这些建议有些新颖,如果说喜欢这些个性化建议,因为历史上许多建议系统都是
1019	Quite siloed and quite isolated	相当孤僻和孤立
1020	and so they're sort of like they don't see like a holistic view of	因此,他们有点像他们不认为一个整体的观点
1021	You as a person and muses has an opportunity to offer much greater levels of personalization	你作为一个人和缪斯有机会提供更高层次的个性化
1022	Yeah, I think people are already seeing that where if they ask	是啊,我想人们已经看到在那里如果他们问
1023	Like with muse if you ask it to buy something for you quite frequently	喜欢与缪斯如果你要求它买东西给你
1024	It'll add in some bit of context that it pulled in from a different, you know conversation.	它会在某种语境中加上它从不同的语境中抽出来,你知道的。
1025	You had a good example this	你有一个很好的例子
1026	Yeah, well, so yeah, so my my three-year-old daughter	是啊,那么,所以是啊, 所以我的三岁的女儿
1027	Loves baking she loves cooking and so my god, I'll start baking stuff	喜欢烤她喜欢做饭所以我要开始烤东西
1028	It's like I'm normally into you know smoking meats	就像我平时喜欢抽烟一样
1029	But like but I'll do some baking in order to make to make a really happy and so we've gotten into this	但我会做一些烘焙为了让一个真正的快乐, 所以我们得到了这个
1030	really cute routine where	可爱的例行公事
1031	You know, we I've been playing with muse voice mode, which isn't fully rolled out yet, but we're we're playing with it and	你知道,我们一直在玩缪斯语音模式, 它还没有完全推出, 但我们是在玩它,
1032	It does this great thing.	它做这个伟大的事情。
1033	It segments the the voices who it knows which one is me and it knows which one is not me	它分辨出谁知道谁是我,谁知道谁不是我
1034	based on that and we just like, you know	在此基础上,我们只是喜欢,你知道
1035	throughout the week we're talking about what we'll want to bake that weekend and	整个星期,我们谈论什么我们要烤那个周末,
1036	she'll like	她会喜欢
1037	Give it ideas or like she'll kind of come up with crazy things that are not real recipes	给它一些想法,或者说她会想出一些疯狂的东西而不是真正的食谱
1038	but like the bounce it off with with with muse and and then	但就像用缪斯弹出然后
1039	Muse will just make sure that we every you know Saturday when it's time to bake	穆斯会确保周六你都认识是时候烤了
1040	I get the ingredients from Instacart and they they kind of show up and they're on time	我从Instacart那里拿到原料他们来了他们准时来了
1041	Which I mean the biggest thing is like if I were just starting the activity on Saturday	我的意思是,最大的事情是如果我刚刚开始活动在星期六
1042	You'd have to get everything and it would take some time, but now everything's ready whenever we're gonna go sit down and do it	你必须得到一切,它需要一些时间, 但现在一切都准备好了,只要我们要去坐下来做。
1043	This weekend it knew that watermelon was launching.	这个周末它知道西瓜会发射
1044	So it threw a watermelon in the bag	所以它把西瓜扔进袋子里
1045	It was very very sweet.	非常贴心
1046	It was very heartwarming	感觉很温暖
1047	It deduced that that I that the recipe it did not have enough	我推断出它没有足够的食谱
1048	There was not enough I guess	看来还不够
1049	Expense in it to get the delivery so it needed something else and it was like you're gonna and it talked it up with	花很多钱去送货所以它需要别的东西就像你要去和它谈过一样
1050	Like do you want this watermelon?	你想吃西瓜吗?
1051	I was like, yes.	我很喜欢,是的。
1052	I'm like, yes, that's very cute	我喜欢,是的,这是非常可爱的
1053	Which I think you do you is that the watermelon you?	我想你做的就是西瓜吗?
1054	Okay, but the watermelon the live stream was that was just like you just got it for me	好吧,但西瓜活溪是就像你刚得到它给我
1055	Yeah, just got it for me	是啊,刚给我的
1056	But but I think there's this that was a very memeable moment of that very memeable situation	但我认为这个时刻是一个非常美好的时刻
1057	So we have we have muse to thank for it	所以我们有缪斯来感谢它
1058	So but here's another interesting thing about the whole thing is I didn't really use Instacart that much before	但还有一件有趣的事情是我之前没有真正用过Instacart
1059	You know, it's like every once in a while maybe twice a year	你知道,这就像每隔一段时间也许一年两次
1060	我想用Priscilla的账户什么的
1061	Yes	对
1062	I went and got an Instacart account to give muse and now I use it like once or twice a week because like I'm talking	我得到了一个Instacart账户给缪斯现在我使用它像一周一两次因为我在说话
1063	With Aurelia with muse and we're coming up with these recipes and it's delivering them and now it's like a core part of my workflow	带着Muse的Aurelia,我们想出这些食谱它正在送来,现在就像我工作流程的核心部分
1064	So I actually think that the opportunity here isn't just okay	所以我觉得这里的机会不是那么好
1065	It's gonna be 20% more efficient on the conversion rate	转换率会提高20%
1066	Like I am a new Instacart's like subscriber and I just wouldn't have done this before	就像我是一个新的Instacart's 像用户,我只是不会这样做之前
1067	So I do think that there's this question and then there's this question, which I think is these sort of interesting	所以,我确实认为有这样一个问题, 然后有这样一个问题, 我认为是这种有趣的
1068	Like this is maybe the more interesting economic question is is that Instacart account really for me or is it for the muse?	也许这更有趣的经济问题是 Instacart的帐户真的是给我的还是给缪斯的?
1069	Because I think like over time	因为我觉得像随着时间的推移
1070	I think that the question is is the total addressable market for all of commerce just increasing dramatically	我认为问题在于整个商业市场都能解决
1071	Yes, because it's not just the people anymore like now agents are effectively going to have accounts	是的,因为现在这样的人不仅仅是特工们会有账户
1072	Yes, and I think that's where this starts to like become a really very exciting opportunity where anyone who sits this out	是的,我想这就是开始喜欢成为一个非常令人兴奋的机会在那里,任何坐在这里的人
1073	I think is gonna end up feeling sad over time	我觉得以后会很伤心的
1074	And why did you have to sign up for an Instacart account like we're solving this abstract projects for developer stuff	为什么你要注册Instacart账户就像我们正在解决这个抽象的项目开发商的东西
1075	Or your coding agents destroy integration, by the way is great.	或者你的编码员破坏整合,顺便说一句,是伟大的。
1076	You guys you guys did a great job of this	你们干得不错
1077	I think we're very	我觉得我们很
1078	We've been very excited about agent and commerce for a long time obviously because again	我们一直很兴奋经纪人和商贸显然因为再次
1079	it's felt to us like clearly how people are going to buy things in the in the future and	我们感觉很明确人们将来会怎样买东西
1080	that was both apparent because no one needs all that friction in their life from the actual buying stage and	这都很明显,因为没有人需要生活中的所有摩擦从实际购买阶段,
1081	How early product research with AI had product market face like even from kind of the earliest AI's	早期的AI产品研究如何有产品市场面貌甚至从某种最早的AI。
1082	It really feels like computer use was like the last big unlock that we needed to make it happen because now you can ask your	感觉就像电脑是最后一个大解锁因为现在你可以问你
1083	To buy anything more or less and then to just figure it out with computers, right?	买东西或多或少,然后用电脑解决,对吧?
1084	And so it's like the ultimate backwards compatibility.	所以它就像最终的反向兼容。
1085	Yeah, it's business a set of things that not really obsessed over	是的,这是生意一套事情,没有真正迷恋
1086	Yeah, I think I think there was a lot of like I think a lot of magic of Muse is the fact that like a lot	是的,我想我有很多感觉我觉得很多魔法缪斯的事实是,像很多
1087	Of the stuff just works and so like clearly if everything was just like a little bit less reliable	如果一切都那么不可靠的话
1088	It would feel like a total pain in the ass and nobody would use it and people just throw away the app	感觉像一个完全痛苦的屁股没有人会使用它人们只是扔掉应用程序
1089	And so I think the fact that so much of this has gotten to the point of reliability where you know	所以,我认为事实, 这么多的这一切已经到了可靠性的点
1090	It just works is like a huge part of the magic.	它只是工作就像魔力的一大部分。
1091	Yeah	对
1092	That was obsessed with oh, well, I mean that's not kind of	这很迷恋哦,好吧,我的意思是,这不是那种
1093	Everything.	一切众生.
1094	Yeah	对
1095	No, it's great.	不,这是伟大的。
1096	It's great.	赞曰.
1097	But this is I mean, but yeah, I mean this getting this experience to be good	但这是我的意思,但是的,我的意思是,这使这个经验是好的
1098	I think it's something that that he in a core group of people on that team really focused on it's kind of funny	我觉得这是他在一个核心群体在团队中真正专注于这是一种有趣的
1099	I think I had a dinner with Patrick like I want to say it was	我想我和帕特里克一起吃过晚饭了就像我想说的那样
1100	2023 where I was he was like, what do you think we should be doing?	2023年我和他一样你觉得我们应该做什么?
1101	I was like, oh you're you're like you're already way behind the eight ball on this agent to commerce stuff	我当时想,哦,你就像你已经落后八球在这个代理商的东西
1102	Like you got to figure this stuff out	碞钩眔稱快猭
1103	Like he just gonna be buying things and like you're not prepared for it	好像他会买东西,你还没准备好
1104	And you were thinking this pattern or part of seeing this I was saying this to Patrick	你当时在想这个模式或者是看到这个我对帕特里克说
1105	Well, cuz Patrick was very politely asked.	因为Patrick很礼貌地问了
1106	What do you think we should be doing?	你觉得我们应该做什么?
1107	And I sort of had this like very like urgent spiel.	我有点像急事一样
1108	Yeah, and I think it's very	是啊,我觉得这是非常
1109	It's funny how that was like you for your provocation.	很有趣,因为你挑衅我
1110	It was it was directionally correct	这是方向正确
1111	but but like when we were three years early in 2023, I think and so	但就像2023年我们早3岁时一样,我想是这样
1112	It's very interesting to look at the agenda commerce	考虑一下议事日程商业很有意义
1113	stats at stripe because	数字在条纹上,因为
1114	We started developing all this stuff	我们开始开发这些东西
1115	I mean we had a whole bunch of announcements at sessions back in May and then it only as of September basically started going vertical	在五月份的会话上我们有一大堆的通告直到九月份才开始垂直
1116	Met is building increasingly capable models including muse wonder and their models are available on open router alongside hundreds of others	Met正在建造能力越来越强的模型,包括Muse奇迹,其模型与数百个其他模型一起在开放的路由器上可供使用。
1117	Stripe has agreed to acquire open router who's the leading model gateway	Frede已经同意购买开放路由器谁是领先的模型网关
1118	Open router already routes more than 10 trillion tokens a day and handles much of the complexity of working with different models through a single API	打开路由器已经每天运行超过10万亿令牌,并且通过单一的API处理与不同模型合作的许多复杂问题
1119	Making it easier to experiment find competitive prices and automatically route around provider outages	使试验更容易找到有竞争力的价格,并自动绕过供货商停电
1120	Try it for yourself at open router dot AI	在开放路由器点AI上自己试试
1121	Hey, could you order a watermelon to stripes office in South San Francisco?	你能叫西瓜去旧金山南部的办公室吗?
1122	I'm doing a podcast here.	我在这里播客。
1123	This would be funny	这很有趣
1124	Funny podcast idea.	有趣的播客创意。
1125	I'll check that watermelon delivery now	我现在就去检查西瓜
1126	Great.	不错
1127	We'll see how we get on	看我们怎么上路
1128	Well, okay, this actually raising an interesting question	好吧,这其实是一个有趣的问题
1129	People have been trying to make the how	人们一直试图使
1130	9,000 it's working.	九千有为.
1131	I really like this	我真的很喜欢这样
1132	It's tough to prop up.	坚固拥护.
1133	Yeah	对
1134	And	还有
1135	People even trying to make the speak to a computer AI vision happen for a really long time	人们甚至试图让说话计算机AI的视觉发生很长一段时间
1136	where Apple launched theory in	苹果公司在
1137	2014 2015 something like that and	2014年2015年类似的事情
1138	Obviously, there's been voice mode and various AI products for a few years now	很明显,有语音模式和各种AI产品现在几年了
1139	Why is it so hard?	为什么这么难?
1140	like it's still not the case if you were driving your car and you wanted to	就像你开你的车你想开的一样
1141	Check on the baking or move a meeting or something like that.	检查一下烘焙或移动一个会议或类似的东西。
1142	Yes, you probably wouldn't do it with the voice	是的,你可能不会做用声音
1143	Maybe almost now	也许现在差不多了
1144	Now is actually a good example as we're talking over each other and I don't know when you're using different so is it just technical limitations	现在是一个很好的例子,因为我们在谈论彼此,我不知道当你使用不同的,所以这只是技术限制
1145	That's why we haven't done it.	故未办.
1146	Well, I well I think that there's a few things and one is you actually want it to go do	好吧,我很好,我认为有几件事,有一个你真的希望它去做
1147	Stuff so there's a question of when you speak with it, even if it understands.	因此,有一个问题你说话的时间, 即使它理解。
1148	Yeah, how's that gonna work?	是啊,这怎么行?
1149	Then I think that there's a lot of stuff about the way that people have conversations	我觉得有很多关于人们谈话的方式
1150	It isn't typically I will give a prompt and now I will wait and you will speak and then so it just	我一般不急着说,现在我等着你说话,然后说
1151	goes back and forth so	不停地走来走去
1152	How many of your sessions when you're talking to an agent in voice mode?	在语音模式下与代理交谈时,您有多少次会话?.
1153	Do you basically know it starts saying something?	你基本知道它开始说些什么吗?
1154	But then like I know it's like I kind of feel like if I pause for three seconds	但后来我仿佛知道如果我暂停三秒钟
1155	It's gonna think I'm done and want to speak and then if it starts speaking	它会认为我已经说完了, 想说话,然后如果它开始说话
1156	But then I cut back in is it gonna think that it's like a whole new question	但后来我又退了一步,它会觉得这就像一个全新的问题
1157	When there's a lot of the stuff about the conversational dynamics that you want to get right, but then there's this other issue, which is that	当有很多关于对话动态的东西你想得到正确的, 但还有另一个问题,这就是
1158	For voice overall	整个语音
1159	It's quite high bandwidth for you to communicate with voice, but in terms of consuming information	与声音交流的带宽相当高但从消费信息来看
1160	Having someone speak to you is very inefficient compared to having it be visual	和你说话的人和直观的人相比效率太低了
1161	Yeah, so I think like that's why you know, partially when we do these	是啊,所以我想这就是为什么你知道,部分当我们做这些
1162	These interfaces I think it's it's kind of working on its task	这些界面,我想是它的工作
1163	I'm fired off now.	我现在被解雇了。
1164	So I think the ideal interface is probably not just strictly voice voice	所以我认为理想的界面可能不只是语音
1165	Although I think voice	虽然我觉得声音
1166	Output from you is pretty useful.	您的输出相当有用。
1167	Yeah, I think I think they're very clearly already is an explosion of voice input	是的,我想他们已经很明确了是声音输入的爆炸
1168	Like I think you even see this encoding.	我觉得你甚至看到这个编码。
1169	We're like, you know, we're speaking to them	我们就像,你知道, 我们正在和他们说话
1170	Yeah, most of the researchers just kind of like whisper into their into their keyboard because it is like as Mark mentioned	是啊,大多数研究者只是有点像悄悄地进入他们的键盘因为就像马克提到
1171	It's like the the highest bandwidth way to get your like thoughts out or usually the lowest friction way to get your thoughts out	这就像最高带宽的方式让你喜欢的想法出来或者通常最低的摩擦方式让你的想法出来
1172	Yeah, I think the most extreme version of this by the way is like I think how	是的,我认为最极端的版本顺便说一句,就像我想
1173	Like college students now use AI is maybe the most interesting where they just like I think a very common modality of how to use	就像大学生现在使用人工智能一样也许最有趣的是他们就像我想一个非常常见的方式如何使用
1174	AI is that they just will	AI是他们只是会
1175	Completely word vomit for like 10 15 minutes	满词呕吐10个15分钟
1176	Yeah, like being full full sort of like stream of consciousness shift out of like all the things and then have AI clean it up	是啊,就像完全一样意识的流转移出来像所有的事情,然后有AI清理它
1177	And then for yeah and have AI effectively like do all of this sort of like	然后因为是的,并且让AI有效地做所有这些类似
1178	Organization of it and like pulling things together and and structuring everything.	组织它,就像把事情拉在一起,安排一切。
1179	It's like it's quite interesting	好像很有趣一样
1180	It's a very interesting like yes, I think modality of tech usage, but um, yeah, I think voice input has been big	这是一种非常有趣的,就像是的, 我认为方式技术使用, 但是,嗯,是的,我觉得声音输入很大
1181	I do think agents is like that's a big step change because I think the mental model	我觉得特工们就像一个很大的改变因为我认为精神模型
1182	Is just very different like if it's just like an	只是非常不同,如果它只是像一个
1183	Informational thing that you're talking to that like you will clearly get into situations	你和这种人说话的信息肯定会陷入困境
1184	Which is feels useless whereas like now that I can do things for you	这感觉没用,而现在我可以为你做一些事情
1185	It's like much closer to the experience of talking to a human total or to a person	这就像更接近与人类或人交谈的经验
1186	And so I think there's all these like interesting uncanny valleys and like interacting with	所以,我认为所有这些像有趣的不可思议的山谷像互动
1187	AI's that until they've been able to take actions like I felt kind of strange and now that we have agents like I think one of	AI的,直到他们已经能够采取行动像我觉得有点奇怪现在,我们有代理像我想
1188	the cool things about	很酷的东西
1189	Muse is like this confluence of a number of things like I think the fact that Muse is embodied and it like is this little guy	缪斯就像一些东西的交汇点我认为缪斯就是这个小人物
1190	That like you feel like is working on your behalf is also only possible and a coherent like	就像你觉得自己在代表你工作一样也只有这样才能做到
1191	representation and embodiment of the product because it has persistent memory and it has	产品的代表性和表现,因为它具有持久的记忆力,而且具有
1192	The ability to do things for you and the ability to take actions and has like a little bit of a personality like is the	为你做事情的能力以及采取行动的能力就像个性有点像
1193	Confluence of all these things coming together then make this like embodiment concept	把这些东西拼凑在一起然后把事情变成化身
1194	Even work in the first place and it turns out okay.	一开始甚至工作结果也不错
1195	Actually, this is the real glasses, right?	其实,这是真正的眼镜,对不对?
1196	That's like what keeps a lot of people want it even with just like no display or a small display	这就像让很多人想要的东西即使没有展示或是小展示
1197	The camera and audio is enough of a useful product.	相机和音频足以成为有用的产品.
1198	Yeah.	对
1199	Yeah, and I and then	是啊,我和后来
1200	We started shipping the display glasses that aren't the kind of wide field of view full augmented reality	我们开始运送显示眼镜不是那种广的视野
1201	Which I mean, I felt we awesome too, but though even the small display	我的意思是,我也觉得我们真棒, 但是即使小展示
1202	It's pretty powerful.	这是相当强大的。
1203	So I think you'll start getting some interesting interactions there that are kind of fully	所以,我认为你会得到一些有趣的互动在那里是那种完全
1204	Multimodal right so it's voice in and out and it'll be able to see what you see and it'll be able to kind of	多模式右,所以它的声音进出,它可以看到什么, 它可以有点
1205	Paint the pixels and one of the I mean the other exciting thing that were where we've done here is news gadgets	画像素和其中之一我的意思是另一个令人兴奋的事情我们曾经在这里做的是新闻节目
1206	which is really I think unleashing the ecosystem to	我认为这真的释放生态系统
1207	Try cool device formats and form factors that could be really exciting like since releasing that we've gotten actually even a number of phone	试一下很酷的装置格式和形式因素这可能是非常令人兴奋的,因为发布后我们得到了很多电话
1208	Manufacturers, you know talk to us about excited, you know	制造商,你知道跟我们谈论兴奋,你知道
1209	We're like we're just gonna integrate news gadgets into our phone and see how well it works and like we're really excited about that	我们就像我们只是要整合新闻设备在我们的手机, 看看它如何工作,
1210	like I think I think part of what is cool about the current moment of technology is that like	我觉得目前科技的酷点是
1211	It's very clear agents are really exciting.	很明显的特工们真的很激动人心
1212	It's really clear that like this should unlock lots of things	很明显,这样可以解开很多东西
1213	Will you guys come back to the household devices?	你们能回到家用设备吗?
1214	Like this is obviously a personal device	这显然是个人的装置
1215	but we actually had during Coovers the	但我们在科弗斯期间
1216	Portress the portal portal portal.	移植门户门户门户.
1217	There we go.	开始了
1218	Yeah, I mean it like is it finally the	是啊,我的意思是,它就像终于
1219	One of the gadgets that we just shipped is the little home link	我们刚运来的装备之一就是小的家用链接
1220	It's like the size of a little like nickel or something, but with a with a USB, right?	这就像一个有点像镍什么的大小, 但有一个与USB,对不对?
1221	Yeah	对
1222	So you can basically connect that and then your muse can connect to everything in your home	所以你基本上可以连接,然后你的缪斯可以连接到你家里的一切
1223	This is actually one of the one of the most fun things too is just having it	这其实也是最有趣的事情之一就是拥有它
1224	Yes, connect to your Sonos and connect to all your lights	是的,连接到你的索诺斯连接到你所有的灯光
1225	I have it basically like when it's time for me to start winding down in the evening	我拥有它基本上就像当我开始在晚上的风下
1226	I just have it like dim all the lights across the house and	我只是有它像昏暗所有灯光过的房子和
1227	It's it's nice.	谓善哉.
1228	How are we gonna fund the AI build out?	我们要怎么资助人工智能?
1229	Everyone's worried about you know, Google went free cash flow negative	大家都很担心,谷歌是免费现金流负数
1230	You know, you're talking about one gigabyte	你知道,你说的是一个千兆字节
1231	Yeah, then five gigawatts and everything like this that is not free cash flow negative.	是啊,然后五千兆瓦和像这样的东西,不是免费现金流负数。
1232	Yes, but there's always more capex to	是的,但总是有更多的披风到
1233	Defund and so	退款之类的
1234	No, I mean look I think I	不,我的意思是看,我认为我
1235	Think that there are gonna be a lot of ways to create value from this	认为有很多方法可以创造价值
1236	so what we're trying to do with muse specifically is	所以,我们正试图与缪斯具体是
1237	Make it so that it can be	让它可以
1238	Widely used because I think that they're gonna be some kind of	被广泛使用,因为我认为他们会是某种
1239	Valuable network effects at scale as people start being able to interact with each other's and that's gonna be great	随着人们开始互相交流网络效应在规模上是有价值的这将是伟大的
1240	So I think there are all these reasons why you want it to get to large scale	所以,我认为有所有这些原因你希望它得到大规模的
1241	Also, you know, we're talking about partners partners are gonna want to integrate more the more people are using it	而且,你知道,我们谈论的合作伙伴会希望整合更多人使用它
1242	So it will get more valuable if we can get it to scale	所以,如果我们能把它放大,它就会更有价值
1243	The best way to do that is to basically have a model where it is affordable and free for people to use	最好的办法是基本上有一个模型能够负担得起和免费人们使用
1244	But there are gonna be a lot of ways that I think we can we can scale revenue	但有很多方法我认为我们可以扩大收入
1245	Subscriptions is gonna be a big one	订阅会很大
1246	You know basically	基本上,你知道的,你应该知道
1247	Just because the base version of it is free and we want the free version to be excellent	因为它的基版是免费的我们希望自由的版本是优秀的
1248	Doesn't mean that there are things that just naturally aren't going to be economical to include in that	这并不意味着有些事情自然是不经济的
1249	so, you know if you want the the kind of	所以,你知道,如果你想要的那种
1250	We talked about this earlier if you want to start using muse wonder in news	我们之前谈过了如果你想在新闻中用缪斯的奇观
1251	Tomorrow then you know the way that you'll do that is you'll you'll pay for a subscription	明天你就会知道自己会怎么做, 你会支付订阅费
1252	So I think that that will be a big business	所以,我认为这将是一个大生意
1253	There are a bunch of other things around commerce that we're quite excited about too	商业上还有好多东西我们也很兴奋
1254	Where we basically think we believe that muse is actually gonna make people money and save people money by by using it and that	我们基本上认为缪斯会通过使用它来赚钱拯救人们
1255	Over time there are lots of different ways to effectively take a small cut of what you are saving or making for people and that	随着时间的推移,有许多不同的方法有效采取小削减你节省或制造的人,
1256	Will end up being a very large	会变成一个非常大的人
1257	Revenue stream and then there's all the other things right so	收入流,然后有所有其他事情是正确的
1258	You think news will generally just cost negative money for people because it helps	你觉得新闻一般只会给人们带来负面代价因为它有助于
1259	Already all these things that I think are interesting where people are like go look through my emails find a bunch of subscriptions	这些东西我都觉得很有趣在那里人们喜欢去查看我的电子邮件找到一堆订阅
1260	I don't actually use cancel them whatever	其实我不用取消什么
1261	Yeah, so yeah, I mean that	是啊,所以是啊,我是说
1262	I don't know if that's a thing that we would monetize but but like but you get you get you kind of get the point	我不知道这是不是我们想要货币化的事儿但是... ...但是...
1263	And where it helps you make money	让你赚钱的地方
1264	I mean one of the ideas that we kind of talk about him	我是说我们谈论他的想法之一
1265	This is this is something that I think you've been kind of obsessed with is making it like a business in a box for small businesses	我觉得这是你迷恋的东西把它变成小生意
1266	So someone can just kind of it can help them administer all these parts of what they're doing and like you can imagine that being	所以有人可以只是一种它可以帮助他们管理所有这些部分他们在做什么,像你可以想象
1267	Very valuable and being very profitable both for the business owner and for meta	非常有价值,对企业主和元都非常有利可图
1268	But I think one of the things that's interesting is like okay if we're building the models	但我认为其中一件有趣的事情是像没关系如果我们建造模型
1269	To build the personal agents and the business agent	建造私人代理和商务代理
1270	There is this large surplus of value because now the model it's very general right so it doesn't just do those things	因为现在的模型很普通所以它不只是做那些事
1271	So what we'll sub the API that'll be profitable a big part of why?	那么,我们如何潜入API, 这会盈利很大的一部分为什么?
1272	We want to start the meta enterprise platform and have CJ decide join us to build this is you know	我们想启动元企业平台让CJ决定加入我们建造这个
1273	We have this view which is like all right.	我们有这样的观点,这就像所有的权利。
1274	Yeah, sure	当然
1275	So traditionally meta hasn't been an enterprise company	所以传统上,元不是企业公司
1276	But we believe that a huge part of the challenge is building the leading models in the first place	但我们相信,挑战的一大部分是首先要建立主要的模型
1277	they're very general and once we do that if we make a	他们非常普通,一旦我们这样做,如果我们做一个
1278	significant investment in in building an enterprise group that can	对建设能够
1279	Provide that in the way that enterprises need to consume it then I think a lot of enterprises are gonna want to use muse wonder	提供企业需要消费的方式我认为很多企业会想使用缪斯的奇迹
1280	it's gonna be an excellent model and affordable and like it's so	这将是一个优秀的模型和负担得起的,就像它是如此
1281	So I think that that is gonna end up being a very big business too and all of these things put together	所以,我认为,这将是最终会是一个非常大业务太和所有这些事情汇集在一起
1282	Coupled with the fact that we were already starting off as a very profitable business that can that has good credit and can kind of underwrite	再加上我们已经开始一个非常有利可图的生意能够有良好的信用并可以担保
1283	These large investments.	这些大投资。
1284	I think puts us in a pretty good position as a lab to be able to build out the compute that we need	我认为,我们处于一个相当好的地位作为一个实验室能够建立我们所需要的计算
1285	to build leading products build up the network around this and position the company to	建设主要产品围绕此建立网络定位公司
1286	Really lead in the area for decades to come okay?	在未来几十年里,这里真的领先
1287	So you think ultimately there'll just be a lot of revenue from these products, and so they'll be self-funding and fairly short order	所以,你认为最终会有很多收入从这些产品, 所以他们会是自筹资金和相当短的订单
1288	Yeah, I mean that's a much shorter answer than mine	是啊,我的意思是,这是一个更短的答案比我
1289	The question was are we going to be able to make money from this	问题是我们能从中赚钱吗?
1290	Yeah, the answer better be yes	是啊,答案最好是肯定的
1291	But the history is very front-loaded expense right well the history of AI is an industry	但是,历史是非常前置的费用正确的是,AI的历史是一个行业
1292	I think to some degree is like you know people have always sort of been hand-wringing about you know	我想在某种程度上就像你知道人们总是有点手摇手摇你知道
1293	How are these businesses gonna make money?	这些生意怎么赚钱?
1294	Yeah, and I think you know every year	是啊,我想你每年都知道
1295	People are sort of generally surprised the outside of that super intelligence is is a useful	人们总觉得很惊讶超级智能的外在是有用的
1296	Yeah, this is kind of metas whole playbook	是啊,这是一种metas整个游戏本
1297	As we we build something we get it to a billion people then we scale the revenue while	当我们建造一些东西,我们得到它 10亿人然后我们放大收入,同时
1298	Growing our revenue primarily in the meantime by continuing to scale the things that are already kind of at scale and growing	主要通过继续扩大已经规模和增长的事物来增加收入
1299	Yeah, I was also one of the dynamics like I think	是的,我也是动态之一,就像我想
1300	You know I think Darius talked about this as well	你知道,我觉得大流士谈到这一点
1301	But like just in general because it requires so much advanced planning to build	但就像一般的,因为它需要这么多的高级规划建设
1302	These models like if you look at the R&D expense in any given time that is like over in overlapping R&D expense of	这些模型就像如果你在任何特定时间内研究与开发费用与重叠的研究与开发费用一样
1303	Multiple model generations at once yes, and you know these they require a long time to work on and then ultimately	多个模范一代一时是的,你知道他们需要很长的时间工作,然后最终
1304	You know you see like immense payoff when when you ultimately deliver them	你知道,你看到巨大的回报当你最终交付他们
1305	And this is kind of what we saw with the muse spark family of models like yeah	这就是我们看到的与缪斯火花家庭像模特儿一样
1306	you know again, we started working on these a long time ago and	再说一遍我们很久前就开始研究了
1307	Seeing like incredible consumer adoption through the news product the other thing is that we use the models to explore a core business	从新闻产品上看消费者的收养是不可思议的另一回事就是我们利用模型来探索核心业务
1308	Right so when you have this core business	好吧,当你有核心业务
1309	That's hundreds of billions of dollars and still growing extremely quickly and now like if you can grow it a few percent faster by using	几千亿美元,而且增长仍然非常迅速现在,如果你能通过使用更快地增长几率
1310	These models to accelerate these things then that ends up funding a huge amount of this too is your scaling it out	这些模式可以加速这些事情, 最终也为大量的资金提供资金, 也就是你的推广。
1311	The meta core business is a is a like eighth wonder of the world	元核心生意是世界第八奇迹
1312	It's an extremely good really is quite quite incredible and is growing it	这是极好的真的相当不可思议并且正在成长
1313	Sort of it.	有点
1314	You know astonishing rates	你知道惊人的比率
1315	Yeah, yeah, you guys really cooked with that	是啊,是啊,你们真的煮用这个
1316	Yeah, I think you're also getting at something Alex	是啊,我想你也在得到的东西亚历克斯
1317	Which is like there's an interesting accounting artifact where if you go build a data center	这就像有一个有趣的会计文物如果你去建造一个数据中心
1318	You have capex and you get to depreciate that over a useful life of many years and people debate	你有卡佩克斯,你得到贬值,在多年的使用寿命人们辩论
1319	You know what that useful life is but the point is you do get to depreciate us	你知道什么是有用的生命,但关键是你确实得到贬低我们
1320	A lot of the AI costs are also just in researcher salaries and those salaries are being spent on you know model training	很多人工智能费用也只是研究者的工资而这些工资都花在了你知道的模范训练上
1321	Thus also has a useful life of many years, but you do not get to capitalize generally	因此,使用寿命也很多年,但一般不能利用
1322	That's that's also a financial advantage for us building out the data centers as opposed to a bunch of the other companies that are paying	这对我们建立数据中心来说也是一种经济优势而不是其他那些付钱的公司
1323	Clouds, but as they're paying the annual rate, but then at some point the	云,但当他们支付年率,但之后的某个时候
1324	The GPUs will be fully depreciated	GPU将全部折旧
1325	but we're probably still going to be using them for a while because they I mean the value of like	但我们可能要用一段时间因为它们意味着喜欢的价值
1326	even like an h100 at this point is higher than it was last year right just because if compute continues being constrained, so	甚至像现在的h100 比去年还高仅仅因为如果计算继续受到限制,所以
1327	Yeah, I think it's yeah, I think any advice for us with open router	是的,我想是的,我想任何建议我们开放的路由器
1328	You know we have agreed to acquire them.	你知道,我们已经同意获得它们。
1329	It has not closed yet.	它尚未关闭。
1330	Yeah, I think I think	是的,我觉得
1331	You know part of the exciting	你知道一部分令人兴奋
1332	Part of what is what we're starting to see in terms of the models	我们开始看到的模型的一部分
1333	Especially the frontier models that are being developed like news wonder versus let's say a fable or an Astra	特别是边疆模式, 正在开发像新闻奇异和让我们说一个寓言或 Astra
1334	Is that we actually are starting to focus quite?	我们真的开始专心了吗?
1335	In quite different directions from the other models	在相当不同的方向与其他模型
1336	And I think there's a lot of intentionality that goes in the model development process that we've discussed as part of this	我认为在模型开发过程中, 有很多的故意行为, 我们讨论过作为这个的一部分
1337	You know in this conversation that is genuinely different from how other models are focusing	你知道在这场谈话中与其他模特儿的专注完全不同
1338	And so you know when these models kind of get collapsed to the same set of benchmarks that	所以你知道当这些模型崩溃到了同样的基准
1339	elides all of the this is a good point the interesting sort of like	eliides 所有这一切都是一个好点有趣的类型喜欢
1340	Texture that differs between the various models, and so I think one thing that is an exciting opportunity for open router generally speaking is like you know	不同型号的纹理不同, 所以我认为一件事是一个激动人心的机会开放路由器总的来说,就像你知道的
1341	helping to	帮助
1342	Helping developers understand what these sort of subtle differences are so that they can make the most informed decisions about what models	帮助开发者理解这些微妙的区别是什么以便他们能做出最明智的决定
1343	Ultimately they select and then also you can imagine the flip side of that which is you know	最终他们选择,然后你也可以想象翻转的一面,你知道
1344	You have all these developers using these models	这些开发者都使用这些模型
1345	And so those developers are great people to ask about what are the interesting textual differences between these models that are like more	因此,这些开发者是伟大的人问,这些模型之间有趣的文字差异是什么
1346	You know more than just like tweet length like this models good this models bad like you know more than just that level of	你知道更多不只是像微博长度像这些模型好这个模型坏像你知道更多
1347	Insight and I think this is like a you're getting to the wine sommelier	洞察力和我想这就像一个你得到的酒sommelier
1348	You know version or form of understanding the models, and they're sort of like relative strengths and weaknesses	你知道版本或理解模型的形式, 它们有点像相对强弱
1349	I think would be quite exciting okay	我觉得会很刺激的
1350	Based on your experience the open router view that there will be a diversity of models in the future	根据你的经验开放式路由器的观点未来会有多种模式
1351	And they're optimized for different tasks, and I agree with that exactly that's just your guys experience based on yeah	他们被优化不同的任务, 我同意这恰恰是你的人的经验基于是的
1352	Like we we are focusing on different things.	就像我们把注意力集中在不同的事情上一样。
1353	We don't care a lot more about personal professional agents	我们更不关心私人职业特工
1354	Yes, my observation than the other major labs	是的,我的观察比其他主要的实验室
1355	Which is good like there should be differentiation in the economy.	经济应该有所区别
1356	This is like the point of capitalism to some extent	这在某种程度上就像资本主义的观点
1357	So	这么说
1358	No, I just really agree with that point	不,我真的同意这一点
1359	I mean	我是说
1360	I think that part of what I love about you guys working on open router and being able to scale it is	我觉得我喜欢你们在开路由器上工作并且能够放大的部分
1361	It just the more infrastructure that can get developed that supports a diversity of players in the ecosystem	它只是更多的基础设施可以得到开发支持多样性的生态系统参与者
1362	I think is good overall.	我认为总的来说是好的。
1363	I also for the same reason and	我也出于同样的原因
1364	Quite a fan of some of the recent and video moves to support the open source ecosystem	热衷于支持开源生态系统的近期和视频动作
1365	No, but this just kind of goes with our whole theory and philosophy about like what is a good future look like we think	没有,但这只是符合我们的整个理论和哲学关于什么是一个美好的未来看起来像我们认为
1366	That a good future is one where you achieve a like like you basically get prosperity by empowering people	一个美好的未来就是你通过增强人们的能力实现你这样的繁荣
1367	So it's kind of the personal superintelligence thing that we're doing, but there's the whole business version	所以,这是一种个人超级智能的事情,我们正在做, 但有整个商业版本
1368	But for a lot of the important safety topics like our view is that the way to get there is having a	但是,对于很多重要的安全话题,比如我们的看法是到达那里的方式是有一个
1369	Diversity and having checks and balances by kind of having a lot of different things out there and accessible to people	多样性和有制衡,因为那里有很多不同的事物,人们可以使用
1370	Not like just having a couple of good models that are only in a couple companies hands.	不只是几个好模特只在几家公司手上
1371	I think that that would be	我认为,这将是
1372	from my perspective, I think that is	从我的观点来看我认为
1373	probably a riskier	可能更冒险
1374	Configuration for the future.	未来配置。
1375	Yes, then	对,那么
1376	Then lots of people having access to different things	然后很多人可以接触不同的东西
1377	Do you guys have a theory as to why that is where I feel like maybe people predicted?	你们有理论吗? 为什么这就是我的感觉也许人们预测到的地方?
1378	More of a winner-takes-all dynamic in AI or one person clearly like a story that people told back in 2023	更像是AI里一个赢家的动画或者是一个人明显像2023年人们说的故事
1379	Is that one AI lab will start to develop an advantage and that will allow them to get recursive self-improvement going	一个人工智能实验室将开始开发一个优势让他们可以进行自修
1380	And then they'll just pull further and further away from everyone else.	然后他们就会越来越远离其他人
1381	Whereas it has	虽然它已经
1382	Remained extremely competitive where there's kind of multiple people at the frontier you guys with wonder today are you know coming along with your	仍然极具竞争力在边境上有很多人你们今天很奇异
1383	Frontier class model open source models have done an impressive job of keeping up all told	边疆类模型开源模型做了一个令人印象深刻的工作保持所有被告知
1384	And so we had this story a few years ago that someone was just going to run away with the prize	我们几年前有个故事说有人要带着奖品逃跑
1385	Why hasn't that happened?	为什么没发生?
1386	I think there's a lot of	我觉得有很多
1387	underlying reasons and mechanics that that drive this like one certainly is just	根本原因和力学驱动这样的一个当然是
1388	execution like I think	就像我想的那样
1389	You know there for any given frontier lab there are periods of strong execution and weak execution and I think you know	你知道,对于任何特定的前沿实验室, 有一段时间强处决和弱处决我想你知道
1390	That is true across all the labs.	所有实验室都是这样
1391	And so that's sort of like, you know starts to average itself out over time	所以,这有点像, 你知道开始自我平均一段时间
1392	I think another diplomatic	我认为另一个外交
1393	So, yeah, is it that like some of the players in the lead screwed up I	所以,是的,它是像一些玩家在领先搞砸了我。
1394	Think well, I think if all players when they were in the lead executed perfectly they would probably be able to stay in the lead	仔细想想,我认为,如果所有玩家当他们处于领先状态时被完美地执行他们可能能够保持领先
1395	So I think that's one factor	所以,我认为这是一个因素
1396	I think	我觉得
1397	The the power and strength of distillation is certainly another factor that's really benefiting the open-source ecosystem quite a bit	蒸馏的能量和强度无疑是另一个真正有利于开源生态系统的因素
1398	I think this is just like a there's an anti	我觉得这就像一个有反
1399	Hegemonic force in the phenomenon of distilling a model	蒸馏模型现象中的重力
1400	Yeah, and and this is just a quality of the technology like I don't think nobody it's not like anyone planned for distillation to work	是啊,这只是一个质量技术,就像我不认为任何人它不是像任何人计划蒸馏工作
1401	Super duper well, but it just turns out that you can pack, you know, you can get a lot of capabilities	超级酒保的好,但事实证明你可以打包,你知道, 你可以得到很多能力
1402	Naturally through some form of distillation and and and this is you know	自然通过某种蒸馏和,这是你知道的
1403	broadly speaking distillation is quite a good force for the entire ecosystem because it means that you can pack like	从广义上讲,蒸馏对于整个生态系统来说是相当好的力量,因为它意味着你可以像包装一样包装。
1404	Very strong intelligence capabilities into smaller and smaller models	将非常强大的情报能力转化为较小的模型
1405	Which enables us to serve them at much wider scale and that's certainly like one of the very exciting stories	这让我们能更大规模地为它们服务这当然像一个非常令人兴奋的故事
1406	It's just like really small models today are like, you know much more powerful than the than the biggest models a year ago or two	这就像真正的小模型今天是喜欢, 你知道比最大的模型比一年或两年前
1407	years ago	数年前
1408	and then I think the last is that there's just sort of this um, I think there's like I actually think a lot of	然后,我想最后一点是, 只是有种这个嗯,我想有像我真的想了很多
1409	people in the ecosystem	生态系统中的人
1410	Want this future where there it is not just this like, you know	想要这个未来,那里不只有这样,你知道的
1411	Definitely one final company in all of capitalism	绝对是所有资本主义的最后一家公司
1412	Kind of kind of and outcome like I think this is one of the reasons not just individuals but companies putting dollars against things	我觉得这不光是个人的原因也是公司用美元对付事物的原因之一
1413	Yeah.	对
1414	Yeah, and I and I think like, you know	是啊,我和我想,喜欢,你知道
1415	This is actually one of the reasons why a lot of people were excited to join meta	这其实是很多人兴奋起来加入meta的原因之一
1416	and meta super intelligence labs in the first place was that they were just sort of like, you know, I	和一个超级智能实验室一开始是他们只是那种,你知道,我
1417	Want to work at a place that is focused on giving this technology to billions of people that is like not the stated goal	想要在一个集中力量给数十亿人提供这种技术的地方工作,而这种技术不是既定的目标
1418	necessarily of all the other frontier labs and so I think there's like a lot of sort of just like natural forces where	一定是其他的前沿实验室所以我觉得有很多自然力量
1419	many people in the industry are sort of like grappling with okay, what I what is the actual world state that I want to live in in the	这个行业的很多人都喜欢努力与好, 我是什么真实的世界状态,我想生活在
1420	future and	未来和
1421	there's lots of sort of like	有很多类似
1422	decisions that people make at all levels in terms of where they work or who they do business with or what business deals they do	各级人民在工作地点、与谁做生意或从事何种商业交易方面作出的决定
1423	That's sort of like drive towards a more equalized outcome.	这有点像推动一个更平等的结果。
1424	You're getting to the	你即将到达
1425	Some of the MSL startup which I also want to ask about which is it just seems fundamentally very hard still to	有些MSL的启动,我也想问一下究竟哪一个根本上还很难做到
1426	Again, we talked about what goes into developing a frontier model, but also	我们再次谈到如何发展前沿模式,但也谈到
1427	From a standing start like when was the standing start for MSL?	从一个站立的开始,像什么时候站在MSL的开始?
1428	Well, we launched	我们开机了
1429	Llama for on April 4th, I think it was	4月4日的拉玛,我想是
1430	2025 and	2025年和2025年
1431	then	接下来
1432	I think was it April 20th	我想是四月二十号
1433	or 21st was I think when I was like hey when I when I	或者21岁的时候,我想当我像嘿,当我
1434	When I started calling people for advice	当我开始呼吁人们的建议
1435	Okay, what to do so I remember I was I was in a staff meeting and I got like the whatsapp notification like	好吧,怎么办,所以我记得我是在员工会议,我得到了喜欢什么 app通知喜欢
1436	After I'm for call I was like, huh?	我打完电话后,我就像,是吧?
1437	Interesting, okay.	有意思,好吗?
1438	So MSL is a little over two years old.	所以MSL是2年多一点。
1439	What do you know one year one year?	你知道一年一年吗?
1440	You did 25?	你做了25个?
1441	MSL is a little over one year old.	MSL有点超过一年。
1442	And yeah, what are your reflections a little over a year in on?	对,你一年多的反射是什么?
1443	What was load-bearing as the eyes would say in how it came together	眼睛会说些什么?
1444	Well, I mean, I think that there's this whole thing early on which is just like first we had to pull together this core group	嗯,我的意思是,我认为有整个事情的早期, 这就像我们第一次召集这个核心组
1445	right, so	对,所以
1446	So not and Alex and I we were like talking a bit up front	所以不要和亚历克斯和我我们就像在前面说话
1447	and	和
1448	You know at first we were	你知道我们一开始
1449	Brainstorming yeah, like I was talking to Nat about could we is there a way that we can like structure this?	集思广益是的就像我和纳特说的我们可以有办法来安排吗?
1450	So it's like a win for Alex's shareholders and like we can like get them over here and then so then we were talking about	所以,这就像一个胜利亚历克斯的股东, 像我们可以让他们在这里,然后我们谈论
1451	like what would the plan be and	像什么是计划和
1452	And then basically I was just I was in this mode where I was just like asking a lot of different people what we should	基本上,我只是在这种模式,我只是想问很多不同的人,我们应该
1453	Do and like every time	每次都这样
1454	That that we would talk	说我们要谈谈
1455	I'd come away with like three more things of like, oh, yeah, this is what we should do	我会带着三样东西离开, 哦,是的,这是我们应该做的
1456	it's like very like high signal very like a lot of strategic clarity and	这就像非常像高信号非常像很多战略清晰度和
1457	So we were brainstorming and working on this for a bit and then that and Daniel joined and we pulled together a few of the	于是我们开始集思广益并研究了一下然后丹尼尔加入了我们
1458	Initial researchers but like in the beginning it was sort of like the supernova thing where we just kind of like jammed a bunch of	最初的研究者,但就像一开始, 它有点像超新星的事情我们只是有点像卡住了一堆
1459	People together and I don't know.	人们在一起我不知道。
1460	It's like I like I don't think either of us slept that much for the first couple of months	我觉得我们头几个月都睡不着觉
1461	I like don't even I don't like I kind of this mental block almost where I like I kind of almost um	我喜欢不甚至不喜欢我那种精神障碍几乎我喜欢的地方那种我几乎嗯
1462	cleared my	清净我
1463	Calendar for like a month or two and was just meeting	一两个月的日历,只是开会而已
1464	Researchers and talking to them and I remember even before we announced the deal where you were gonna come over	研究人员和他们交谈我记得在我们宣布协议之前
1465	We were basically like you were recruiting people people like what is going on like that's like, what is this?	我们基本上就像你在招募人一样像现在这样,这是什么?
1466	But	不过
1467	But then there was the whole question which was a really hands-on data labeling	但接下来是整个问题一个真正的操作数据标签
1468	But but that was that was this whole question to where it's like, all right	但是,这就是整个问题到它是什么样子,所有的权利
1469	Well, like, okay	嗯,像,好吗
1470	So if we're not happy with the trajectory of the current program like who were the right set of people	所以,如果我们不满意当前的方案轨迹像谁是正确的人群
1471	to lead the program going forward and you know Alex I had spent a bunch of time talking over the years and maybe like whatever once	领导节目向前发展你知道亚历克斯... ...我花了很多时间... ...
1472	every few months or something we'd talk and I was always just very impressed by how he thought about things strategically and	每隔几个月什么的,我们会谈论, 我总是非常深刻的印象他如何思考战略,
1473	And I kind of had this view of like alright.	我也有这种感觉
1474	Yeah, currently he's doing this this very operational data business	是的,目前他正在做这个非常操作性的数据业务
1475	But it always felt to me like	但我总觉得
1476	he was	他曾经是
1477	The sharpest entrepreneur of his generation of the ones who I had had a chance to to meet and	他那一代最尖锐的企业家我曾经有机会遇到过
1478	and	和
1479	Like it always just struck me.	就像它总是打我。
1480	It's like alright, like this is a guy who yeah	碞钩琌
1481	He's like running this like kind of very grindy operational business, but like lovely business still by the way, what's up?	他就像在经营这那种非常磨碎的业务, 但像可爱的生意顺便说一句,怎么样?
1482	We're big investors	我们是大投资者
1483	There was more commentary on the on the difficulty of running the business	关于经营企业的困难,有较多的评论。
1484	The world a lot of businesses look like a grindy operation	世界上有很多企业看起来像一个磨盘操作
1485	It's not it's just like my view	跟我的观点不一样
1486	No, no, no	没有,没有,没有
1487	but	不过
1488	No, so it's uh	没有,所以是...
1489	so I just kind of had this room like all right, like like even though this is sort of I	所以,我只是有种这个房间像所有的权利, 喜欢,即使这有点我
1490	Think maybe when we got started a lot of people were like, okay, like how is this gonna work?	也许当我们开始很多人喜欢,好吧, 像这会如何工作?
1491	it's like the kind of like the	这就像那种
1492	Social media guy is working with the like data guy	社交媒体的人在和数据人合作
1493	like is that like are we gonna build a frontier lab here or what but but I mean but my but I mean my	就像我们要在这里建一个前沿实验室一样但是我的意思是...
1494	like really strong conviction because I talked to like all the leaders of all the labs like all the heads of research like	就像很坚定的信念因为我喜欢所有实验室的领导
1495	All this stuff and like I was like no Alex is the guy like he's the fucking guy and like and we're gonna and basically	所有这一切,就像我像没有亚历克斯的家伙像他他妈的家伙,喜欢,我们要去基本上
1496	like we're gonna basically be able to	就像我们基本上可以
1497	to to go pull together a team and recruit the the people that we need to go do this and	去召集一个团队招募我们需要去的人
1498	We need this kind of very	我们需要这种非常
1499	This kind of like strategic clarity that he has around the the different ways that we want to approach this the kind of extreme focus on	这就像战略清晰度,他有不同的方法,我们想要处理
1500	talent density on	有关人才密度
1501	On kind of like the clarity of communicating the different vision which we kind of fashioned together over this period	在那种清晰的沟通不同的视觉我们所塑造的这个时期
1502	and a bunch of things that I think were	和一堆我认为是
1503	kind of extreme but have	那种极端,但有
1504	Basically worked like having this	基本上是这样工作的
1505	This like very flat culture was something that you were kind of really obsessed with right?	这就像非常平坦的文化是那种你真的很迷恋对不对?
1506	It's I mean you have like was it almost 200 direct reports.	我是说你好像有近200个直接报告
1507	That's like we had it's not	这就像我们有它不是
1508	It's not quite that extreme in practice.	实践上没那么极端
1509	I think there's like my team leads and all and all the stuff my news helps me manage them	我觉得有像我的团队领导和所有的东西我的消息帮助我管理他们
1510	Yeah	对
1511	but like but there were these series of things that I think quite early we	但喜欢,但有这些系列的事情我认为很早
1512	you kind of articulated and I was like	你有点清晰,我就像
1513	Okay, this is like kind of extreme but	好吧,这就像一种极端但
1514	It seems like directionally right and like we should go do this and you know	似乎方向正确,而且我们应该去做这个你知道
1515	I went and like talked to other people or more like the people you'd conventionally	我过去喜欢和其他人说话或者更像那些你习惯的人
1516	Yeah, think of and it's like it was just very much like I don't know what you'd expect like a big company to operate	是啊,想想看,这就像它只是非常像我不知道你期望什么像一个大公司经营
1517	I was like, no, this is we got it	我当时想,不,这是我们得到了它。
1518	We want to do something that's as you describe this on those four things that I think even other	我们想做一些事情正如你描述这这四件事我认为,甚至其他
1519	opinionated effective founder CEOs would not have done	有意见的创始人CEO不会这么做的
1520	One you just called the recent prior efforts by a bunch of	一个你刚才所说的最近的努力由一群
1521	People at the company a failure and not good enough.	公司里的人失败了不够好
1522	I mean you're more diplomatic than that, but the active standing	我的意思是,你比这更外交, 但积极的地位
1523	trajectory	轨迹
1524	But the active standing up MSL, you know people can read between the lines	但主动站起来的MSL, 你知道人们可以在行间阅读
1525	So sometimes yeah, yeah, sometimes you have to reboot a team.	所以,有时是的,是的, 有时你必须重启一个团队。
1526	That's one	就是这个
1527	Thing two is you were willing to pay	第二点是你愿意付钱
1528	Ungodly amounts of money for effectively talent like you're paying a lot of money to bring Alex and some other people but you know	高贵的人才,就像你花了很多钱带亚历克斯和其他一些人来,但你知道的
1529	Yeah, a lot of money for for Alex in particular	是啊,很多钱亚历克斯特别
1530	over and	结束
1531	King three and 49% of this very good grindy business	国王3和49%的非常好的磨坊生意
1532	Let's keep the story	我们继续讲故事吧
1533	So that thing three is I'll notice Alex had never	所以第三件事是我会注意到Alex从来没有
1534	Trained a frontier model before and so you were taking some amount of a bet on	之前训练过边疆模特儿所以你要下注
1535	Alex's potential to do something where he had like done the stuff in AI	亚历克斯的潜质做一些他喜欢的事情在AI做的东西
1536	But like no offense you just had not done the specific thing that you were being hired to do and then fourth	但无意冒犯,你只是没有做具体的事情,你被雇用做,然后第四
1537	You were saying you cleared your calendar and just only did this for two months	你说你清理了日历只做了两个月
1538	I think a lot of CEOs would have had the business review and would have had the board thing and they would have had the you	我认为很多CEO会有业务审查, 和有董事会的事情他们有你
1539	Know customer stuff and things like that.	了解顾客之类的东西
1540	Those four seem to me pretty	这四个人看起来挺漂亮的
1541	Interesting where I think a lot of CEOs would not have done those	我觉得很多CEO不会这么做
1542	Interesting, I definitely think I mean one thing you may not appreciate this by yourself	有意思,我肯定认为我的意思是你可能不欣赏这件事
1543	but I think one thing that is quite special about mark is	但我认为马克最特别的是
1544	the degree to which like once he concludes that	当他得出这样的结论时
1545	Something is the right thing or something makes sense.	某种东西是正确的事情或有道理的东西。
1546	He just like	他只是喜欢
1547	fully shifts everything to be consistent with that belief set and	完全改变一切符合这个信念和
1548	It is quite like to your point	和你说的差不多
1549	I think like the inertia of most even six very successful founders who like are opinionated in an edgy or whatnot would just be to like	我认为,就像大多数六位非常成功的创始人的惰性, 他们喜欢在一种情绪或什么不会只是喜欢
1550	Oh, I still have like my whole calendar and like so I'll just like talk about doing this thing and hope that somebody does	哦,我还有我的整个日历喜欢,所以我只想谈谈做这件事,希望有人做
1551	It's kind of kind of situation.	这是一种情况。
1552	Whereas I think	我觉得
1553	No, I think when mark concludes that something is the thing he has to do or makes a lot of sense	不,我想当马克认为事情是他必须做的事或有很多道理
1554	then he really does like	然后他真的喜欢
1555	He's like maybe very high learning rate in that sense	从这个意义上讲,他可能学习率很高
1556	like he's very willing to like pivot entire things to be consistent with it and	就像他非常愿意把整个事情都切入其中
1557	And I think that like, you know	我觉得,就像,你知道
1558	one very underappreciated thing is the degree to which so much of the company had to pivot to support MSL as well like I	公司这么多人都不得不支持MSL的程度就像我一样
1559	think that	认为
1560	while it was a	当它是一个
1561	Kind of like this more startupy effort within a larger company, you know that the needs of	喜欢在更大的公司里做这种更新颖的工作你知道的
1562	MSL were still quite large like we needed a huge amount of help on infrastructure	MSL仍然很大,就像我们需要大量基础设施帮助
1563	We needed these large-scale data centers to be built.	我们需要建造这些大型数据中心。
1564	We needed like lots of support from many other parts of the company	我们需要公司许多其他部门的支持
1565	And especially as muse has been successful.	尤其是因为缪斯已经成功。
1566	We've also needed lots of yeah collaborate deep collaboration with the rest of the company	我们还需要很多和公司其他人合作
1567	And this is one thing that I think	我想这是一件事
1568	like kind of only is possible because mark is the founder CEO like and he can ensure that	因为马克是创始人CEO 他可以确保
1569	We have kind of like a global optimization towards this	我们有种全球优化
1570	This sort of like AI maximalist version of meta versus like the natural inertia of an organization towards like, you know	这就像AI的最大版本的元和像一个组织的自然惰性
1571	Just local optimization ad infinitum, but it's also a lot of how we work together is	只是局部优化广告无限, 但这也是很多我们如何合作
1572	It's important to me to be in the details of this work	这件事对我很重要一定要详谈
1573	So that we can get everything around the company to go do what we need to do, right?	这样我们就可以让公司周围的一切都去做我们需要做的,对不对?
1574	And so it's like I mean, yeah, I'll have opinions about this stuff	所以,这就像我的意思,是的, 我会有关于这些东西的意见
1575	But I'm not like I'm not the lead researcher, right?	但我不是首席研究员,对吧?
1576	It's like it's you know, it's um	这就像你知道,这是嗯
1577	but I think that's kind of a lot of I think how we work together well is that	但我觉得我们合作得很好
1578	yeah, it's I mean we're kind of like quite in the details of all of this stuff, but	是啊,我的意思是,我们有点像在细节所有这些东西,但
1579	You know running the lab is a big thing running the rest of the company in order to support	你知道经营实验室是大事情管理公司的其他部分为了支持
1580	This is like the effort is a big thing	这就像努力是一件大事
1581	You almost like you can't really get this done	你几乎是无法真正完成这一切的
1582	I think without both of those pieces and and one thing that I think was	我想没有这些片段还有一件我认为是
1583	quite important about when we set the whole thing up is you know, we	相当重要的是,当我们设置整个事情是你知道,我们
1584	Meta the company invested a huge amount to stand up MSL and then we we very clearly told everyone like hey	Meta公司投资了大笔钱来支撑MSL 然后我们非常明确地告诉大家喜欢嘿嘿
1585	There's no deadlines.	没有最后期限。
1586	We are going to take things at the right pace	我们要以正确的速度处理事情
1587	Which I think on the one hand could be like, you know	我认为一方面可能是像,你知道
1588	Absolutely terrifying is like you've just have like sort of this extremely expensive unit and you're telling everyone like hey	绝对是可怕的就像你刚刚有那种非常昂贵的单位你告诉大家喜欢嘿嘿。
1589	And you're worried inside a larger company that like okay seven years from now.	你担心的是,在一个更大的公司里, 7年后还不错。
1590	Maybe something will ship, right?	也许有东西会运送,对吧?
1591	Yeah, totally	是啊,当然
1592	But I think that like first I think one of the cool things about you know	但我认为,首先,我想一个很酷的事情关于你知道的
1593	Focusing on talent density and hiring all these very brilliant people have done so much already	专注于人才密度和雇用所有这些非常聪明的人已经做了这么多
1594	This very impressive is you know, everybody had like a lot of	这令人印象深刻的是,你知道, 每个人都像很多
1595	motivation to ship and and everybody wanted to sort of like leave their mark so to speak on the field and so	开船的动机每个人都想留下印记
1596	Everybody was like heavily highly self-motivated to to move as quickly as possible	每个人就像非常有自我动机尽快行动
1597	and one of the things that's that's been funny and I've had to	还有一件很有趣的事情我不得不
1598	Continue reminding the team of is like, you know, it is a team of of a lot of	继续提醒球队就像,你知道,这是一个团队很多
1599	insecure overachievers	不安全的过度成就者
1600	So like at any one point if I ask them like hey, how good do you think?	如果我问他们,你觉得怎么样?
1601	Where do you think will be in three months?	你觉得三个月后会在哪里?
1602	I always get the most like everyone always it's awful	我总是像大家一样总是很糟
1603	We're like these like very pessimistic answers and then I record it and I like three months later	我们就像这些非常悲观的答案然后我录下来,我喜欢三个月后
1604	I'm like, hey, you know, we just shipped this model	我很喜欢,嘿,你知道, 我们刚运了这个模型
1605	Let me just show you what all of you said	让我给你们看看你们都说了什么
1606	Where all of you said we would be in three three months ago and	你们都说我们三个月前到的地方
1607	And this has that caused them to get any more optimistic or no, they're still they're very self-critical	这使得他们更加乐观或没有, 他们仍然非常自我批评
1608	I think it's it's not it's not that they're pessimistic about I	我觉得不是他们对我感到悲观
1609	Think that the average person seems quite optimistic about MSL and very self-critical about their own project	想想普通人似乎对MSL相当乐观对自己的项目很自我批评
1610	Yeah, which I think is actually quite a healthy sure a healthy dynamic	是啊,我认为这其实是一个相当健康的确定一个健康的动态
1611	But but the trend continues like I think I think and and this is I mean in many ways Muse Wander is like this incredible	但趋势还在继续就像我想的在许多方面穆斯万德就像这个不可思议的
1612	Milestone for everyone because it is a the team was very very critical.	每个人的里程碑因为它是一个团队是非常关键的。
1613	Yeah, that was working on it along the way	是啊,这是工作沿途
1614	When did you guys know that Muse Wander was good?	你们什么时候知道Muse Wander很棒的?
1615	That's a good question.	问曰.
1616	I think that we	我认为我们
1617	So there's all these intermediate you're kicking in the oven and I know it seems kind of gooey still and yeah going back to it	所以,有所有这些中间体你踢在烤箱我知道它似乎一种古伊仍然,是啊回去
1618	I mean there are all these intermediate signals that you can get along the way that things are good or bad	我的意思是,有所有这些中间信号你可以走的方式事情好或坏
1619	What are the intermediate signals?	中间信号是什么?
1620	So in in pre-training you have a lot of perplexity based evaluation	所以,在预训中,你有很多迷惑性的评估
1621	So you have a it's sort of a different regime of evals	所以,你有一个它有点不同的制度Evals
1622	That can help you in pre-training and then you have sort of the more traditional benchmarks that can help you towards RL	这可以帮助你接受预训,然后你有更传统的基准,可以帮助你走向RL
1623	But but no, I mean, it's it	但是,不,我的意思是,这是它
1624	I mean this model kind of was you mentioned has been cooking for quite a while and you know	我是说你提到的模特儿已经煮了好一阵子了
1625	a lot of I think one of the funniest things about AI research is that you know, there's a lot of	我认为AI研究最有趣的一件事是你知道,有很多
1626	there's a lot of times where people just sort of like are kind of squinting at a bunch of dots and being like	有很多时候,人们只是那种在一束点和像
1627	Yeah, looks like it's good or it looks a little bit off.	是啊,看起来很好或者看起来有点脱落。
1628	Like one of the funniest things is the the	就像最有趣的东西是
1629	people have protractors	人们有前导器
1630	Like like analog protractors that they put against you know	就像他们用来对付的模拟推进器
1631	a set of lines on their monitor to like measure an angle and then	显示器上的一组线条喜欢测量角度然后
1632	So I did not come for this job.	所以我不是来做这个工作的
1633	Yeah, so I think there's I think there's like a huge amount of like	是的,所以我想,我觉得有像一个巨大的数量
1634	Like it's very there's more vibes	感觉好像很紧张
1635	Driving like we have this conversation where how did you know is good?	开车像我们有这个对话你怎么知道是好的?
1636	Well, we had this debate after his own pre-training	我们经过他自己的前期训练才开始辩论
1637	Where it was okay, how long will it take in RL?	在RL里需要多长时间?
1638	Before it is better than use spark and I remember the the folks on the pre-training team	之前比用火花好我记得训练前队的人
1639	We're like just try to set expectations	и碞钩琌稱稱稱快猭
1640	They're like look like sometimes it like this takes a while and then it was like immediately and so you're like, okay	他们看起来像有时像这样这样需要一段时间, 然后它就像立即, 所以你喜欢,好吗?
1641	I think I think we're like on on a good path right now that to me was was sort of yeah	我觉得我们现在在一条好道路上对我来说是有点对的
1642	I think when when people got yeah, we've been excited.	我想当人们得到是, 我们一直兴奋。
1643	I think we've been pretty excited about for a few months now	我觉得我们几个月来一直很兴奋
1644	Yeah, yeah, but there's a lot to do there	是的,是的,但有很多事情要做在那里
1645	But going back to the beginning I think one of the things that was very	但回到开始,我觉得其中一件非常
1646	interesting and in challenging and standing the thing up was like you really just wanted this like very high talent density effort and	有趣的是,在挑战性和站立起来事情就像你真正想要这一点,
1647	We were just kind of pulling some of the best people from meta in and recruiting a lot of the best people	我们只是拉一些最好的人从元和招募了很多最好的人
1648	And I mean you were asking it's like alright	我的意思是,你问它像没事
1649	Well, like okay for people who like if you're asking for it's like it well Alex hadn't necessarily trained	好吧,像可以的人谁喜欢如果你要求它就像它好亚历克斯并不一定训练
1650	Models like this but to me one of the big things was like all of the best people were very excited to work with him	像这样的模特儿... ...但对我来说,其中一件大事... ...就像所有最好的人...
1651	and I think like it's I mean you I'm sure see this and all of	我觉得这就像我是说,你我肯定看到这个和所有
1652	in a lot of the things that you do which is like, you know, it's like you	在很多事情中,你做的是喜欢,你知道, 它像你
1653	Most people are not kind of one-dimensional, right?	大多数人不是一维的对吧?
1654	And yeah.	对
1655	Yeah, you like you maybe you do a specific thing, but	是啊,你喜欢你也许你做了一个具体的事情,但
1656	but like your ability to grow and expand is kind of largely based on	但就像你的成长和扩张能力基本上是基于
1657	How effectively can you like just do people who are excellent in different domains want to?	在不同领域表现优异的人想要怎样有效?
1658	Want to come work with you on different things and the great people of followership	想和你们一起研究不同的事物和伟大的追随者
1659	Yeah, but and so I think that there was there was a little bit of that	是啊,但是,所以我认为有一点点
1660	But also just it's even for the people that don't know you yet or like or aren't you know?	但即使是那些还不了解你的人也不知道你是否了解你
1661	Don't know what you're about.	不知所终.
1662	Like are you	就像你一样
1663	Like an impressive person and a compelling person that people want to come work with	像一个令人印象深刻的人和令人信服的人人们想来工作
1664	Do people feel comfortable that you're gonna create?	人们对你创作感到舒服吗?
1665	This environment and I think you were like very thoughtful about this early on we were kind of	这个环境,我想你好像很周到这么早,我们有点
1666	Some of the stuff I think you kind of had fully formed as beliefs of how you thought we should do this	有些东西,我想你已经完全形成了信念,你认为我们应该怎么做
1667	But I also remember when we were having the initial recruiting conversations.	但我还记得我们最初的招募谈话
1668	We were sort of in real time	我们算是实时的
1669	Shaping it's like okay.	塑造如是.
1670	Well this construct that we're talking about with this person isn't quite gonna	那么,这个构造,我们谈论这个人是不会
1671	Work but like they seem good and like they have this thought and it was just like like really kind of	工作,但好像他们看起来很好, 喜欢他们有这种想法, 这就像真正的类型
1672	Coalescing this vision about what the shape of it should be in real time	凝聚这个想法关于它应该是什么形状的实时
1673	I mean, I mean, I'm sure you remember this or is like there were there were some of the the initial concepts of like	我的意思是,我的意思是,我敢肯定,你记得这一点或者像有一些最初的概念像
1674	I mean, I think probably the most extreme thing was I think you you had the concept of the talent density and	我的意思是,我认为可能最极端的事情是我想你有的概念人才密度和
1675	All and that kind of piece, but I think we weren't sure upfront that it was gonna have to be a full reboot, right?	所有和那种作品, 但我认为我们不确定它会是一个全面重启,对不对?
1676	that that basically	这基本上
1677	Started when we were out talking to a lot of the kind of best people and we realized that hey	从我们出去跟很多好人聊天时开始我们意识到
1678	We like in order to make this as attractive as possible.	我们喜欢尽可能的吸引人
1679	It should feel like okay	应该感觉很好
1680	We're starting this thing again from scratch and yeah	我们从头再来一次
1681	We'll pull in a lot of the great people from from the llama effort and fair and all of that, too	我们将吸引很多伟大的人从拉玛的努力和公平和这一切,太
1682	But that was a very intentional thing that we basically figured out and that you figured out as we were out talking to the	但这是非常有意的,我们基本上想通了而你也想通了,当我们出去谈话
1683	Initial set of people like what was gonna make this compelling?	最初的一组人喜欢是什么会使这个令人信服的?
1684	And it ended up being really smart and like making a huge difference in the culture of the lab and how it got and how it	它最终变得非常聪明就像在实验室的文化以及它是如何得到和如何得到的
1685	Got started, but I don't know	已经开始了,但我不知道
1686	It's like being able to be that dynamic about figuring out like what is the shape of this need to be that?	这就像能够充满活力地去想, 像这个形状需要是什么?
1687	It's not like that.	不是那样的
1688	I think that's a very uncommon trait.	我认为这是一个非常不寻常的特征。
1689	It's a very entrepreneurial trait, right?	这是一个非常创业的特点,对不对?
1690	So when I was thinking about like art like this field is gonna keep on evolving	所以当我想着艺术像这个领域会继续发展
1691	All right, so what is the thing that's gonna be important for someone who's gonna run a lab?	好吧,那对于一个要经营实验室的人来说是什么很重要?
1692	Well, it's not like are they an expert in whatever the latest modality is of like or kind of of	好吧,这不像他们是一个专家无论最新的模式是什么类似或类型
1693	Training technique of for a lab.	实验室的训练技巧
1694	It's um, is the person a clear?	这个人清楚了吗?
1695	Thinker and ability and able to kind of dynamically update and and move in like as as kind of an entrepreneur	思考者和能力, 并且能够有活力地更新和移动像像企业家一样
1696	Would I think this is just like an inherently very entrepreneurial space?	我会觉得这就像一个内在的非常创业空间吗?
1697	has this experience	有这样的经历
1698	changed your view on the value of	更改了您对
1699	going external at where both	向外走,在其中两者
1700	Your reboots a bunch of the people you brought in were	你带来的一群人
1701	external like Alex and not and Daniel and folks, but then also you're describing	外表像亚历克斯,而不是丹尼尔和人们, 但你也在描述
1702	Deciding whether you need a reboot based on going to the outside	决定您是否需要根据前往外部重新启动
1703	Intelligence of you know	你知道吗?
1704	the aggregate views of the researchers in the industry and getting a sense for the just do you feel like you will carry forward	研究者们在业界的总体观点和对正义的感知感觉你会继续前进
1705	That approach	这种办法
1706	Forward that approach	推进这一方针
1707	For future unrelated projects because obviously the risk of a large company is that everyone.	对于未来无关的项目,因为很明显,一个大公司的风险是每个人。
1708	Yeah.	对
1709	Oh, we got a great approach	哦,我们有一个伟大的方法
1710	Yeah, let me show you our hundred-page document on it	是的,我给你看我们的百页文件
1711	Not I mean obviously you can't a lot of what we had to do here	我不是说很明显你不能做很多我们不得不做的事
1712	Was rotate a lot of the rest of the company and like for things like the infrastructure team that now	公司里很多地方都在轮换就像现在的基础设施团队
1713	You know, it's like building these huge clusters to support this.	你知道,这就像建立这些巨大的集群支持这一点。
1714	I mean, that's obviously a huge thing	我的意思是,这显然是一件大事
1715	so huge parts of the company kind of had to have to pivot to do this and	公司里有一大堆东西不得不做这个
1716	Obviously can't do too many of those at the same time, right?	显然不能同时做太多,对吧?
1717	Only one thing can be the top priority.	只有一件事是头等大事
1718	Yes	对
1719	but I	不过
1720	Think it unbalanced it has made me somewhat more optimistic.	认为它不平衡让我有些乐观。
1721	I mean we brought in	我的意思是,我们带来了
1722	This guy Kunal Shah who it's I think he's like one of the top entrepreneurs in India	这个叫Kunal Shah的人我觉得他像印度顶尖企业家之一
1723	I mean he was building this great company cred and now he's running whatsapp	我的意思是他正在建立这个伟大的公司信用,现在他运行什么app
1724	Which I always thought should do more in commerce and we were sort of under underserved there and India is the biggest country that we're in	我一直认为在商业上应该多做一些我们在那里的服务不足而印度是我们最大的国家
1725	and	和
1726	But I mean that's it's a bit of a different situation	但是,我的意思是,这是一个有点不同的情况
1727	Because we're not like pivoting the whole company towards whatsapp.	因为我们不喜欢把整个公司都引向什么
1728	We just wanted to bring more entrepreneurial energy there	我们只是想在那里带来更多的创业能量
1729	This was like a level 10 out of 10 difficulty thing	这就像10个难题中的10个
1730	I don't talk about in the past tense as if it's done because it's still a level difficulty thing, but	过去的紧张状态里,我可没说因为这还是个关卡问题,但是...
1731	But I think	但我觉得
1732	the number of pieces that kind of	此类棋子的数量
1733	Had to work	不得不工作吗?
1734	It was yeah, it was it was quite all-encompassing and I think that that it's both difficult for the people doing it	是的,这是相当包罗万象我认为这对人们来说都是困难的
1735	But I do think that that also puts some level of strain on	但我确实认为这也给
1736	The rest of the company on the financial model of the company	公司其他部分的金融模式
1737	I think like you so I wouldn't want to conflate those two pieces, right?	我觉得像你这样这样我就不会把这两块拼起来了对吧?
1738	It's like I think there's yeah, it's good to have new blood and it's good to have	好像我觉得有,有新血是好的,有
1739	people who understand the context and you know	了解环境的人,你知道
1740	You were talking about when people have been around meta for a while and understand the ups and downs and it's like	你刚才在说人们在元器件周围呆了一段时间并且理解起伏
1741	If you've worked at meta for ten years, you've seen some shit, right?	如果你在meta工作了十年, 你见过一些狗屎,对不对?
1742	it's like it's like we've gone through a lot of the hard times and	就像我们经历了许多困难时期
1743	and like I think that that's kind of important context too, but	我觉得那也是很重要的背景
1744	but no, I think a lot of this is just	但是,不,我认为很多这是只是
1745	trying to	试图
1746	You you're whenever you're building a team.	你总是在建立一支队伍。
1747	You're you inherently	你本来就是
1748	You know	你知道吗
1749	It's like Alex had a theory for what the team shape needed to be and like I guess I had a theory for what like the	亚历克斯似乎有理论关于团队的形状需要是什么, 我想我有理论关于什么喜欢
1750	Shape of the leaders needed to be for it and	领导人的造型需要为它和
1751	And you know, it's like a year plus in it like seems	你知道,这就像一年加上在它看起来像
1752	Okay, like it's like first of all, it was just good to try something different right and then different than what everyone else is doing	好吧,就像它像第一, 它只是好尝试一些不同的正确,然后不同别人在做什么
1753	yes, and	对,而且
1754	Then yeah, I mean it's	然后是的我是说
1755	You know, I think there's some other dimensions here that are that are worth mentioning too	你知道,我认为有一些其他的维度值得一提。
1756	like one of the like when deciding we had to do the full reboot like, you know,	比如说当我们决定要全面重启的时候比如说
1757	We a lot of it was, you know, one of the main inputs was external but also one of the inputs	我们有很多,你知道,其中之一主要投入是外部但也有一个投入
1758	I don't think I've ever actually can't remember I told you this story	我从没有忘记过这个故事
1759	but one of the first meetings I had when I joined I was in this it was in this, you know meeting with some of the	但当我加入的时候,我第一次见面... ...就是在这,你知道,我和一些...
1760	leads of the existing effort and	B. 现有努力的牵头者和
1761	I was like, okay.	我当时想,好吧。
1762	I think the first thing we need to do is train a really good reasoning model by	我认为我们首先需要做的是训练一个非常好的推理模型
1763	You know by some time frame.	你知道一些时间框架。
1764	I don't know when but this is like clearly like the most important milestone and	我不知道什么时候但是这显然是最重要的里程碑
1765	One of the the key leads in the group	组里的关键线索之一
1766	Who no longer works the meta but was like, oh, we don't we don't need a reasoning model	我们不需要推理模型
1767	Does any are any of our do our consumers need a reasoning model?	我们的消费者需要推理模型吗?
1768	like I feel like we don't need to do a reasoning model we just do a normal model and	我觉得我们不需要做一个推理模型我们只是做一个正常模型
1769	I was I had like this heart-sinking moment.	我有一个心沉的瞬间。
1770	Oh, no, I've joined the wrong.	哦,不,我加入错了。
1771	I was like I was like, oh	我就像我一样,哦
1772	We I've got I gotta take like ten steps back	и璶10翴
1773	And I think I think this is you know, this is a kind of funny anecdote	我觉得这是个有趣的传闻
1774	I remember calling that after this meeting it was like that's like	我还记得我打过电话之后的这个会议就像是
1775	This is what they just said in the meeting.	这是他们刚才在会议上说的话。
1776	But but I think it was it was it was	但我认为是它
1777	specifically an indicator of something that I think like Mark when he talked to people I certainly also started to realize which is like	具体地说,是马克在和别人交谈时所认为的一个指标我当然也开始意识到
1778	we needed leadership in the company being much more AI pilled and much more sort of like forward leaning and front leaning on	我们需要领导公司更多的AI 和更多的类型像向前倾斜和前倾
1779	Like where the how quickly the technology is moving where it's gonna go to what the possibilities are and like, you know	比如说,科技发展得有多快, 它会发展到什么程度,像什么样,你知道的
1780	effectively like dramatically more AI pilled and	有效地像惊人的 AI丸和
1781	I think as	我觉得
1782	Mark started	开始标记
1783	Pivoting and shifting the whole company towards this direction and it was very clear that this is like, you know	引导和引导整个公司朝这个方向发展很明显,这就像,你知道
1784	I think Mark got AI pilled and then you know, everyone saw that he was AI pilled and ever realized like oh	我想马克被安顿了然后大家都知道他被安顿了然后意识到
1785	This is like where the company's going then I think you have like a lot of the existing leaders like	这就像公司要去的地方,然后我想你有像很多现有的领导喜欢
1786	very quickly pick it up and change and like Adam Messeri, for example, who runs Instagram has one of the most like	像亚当·梅塞里一样管理Instagram的人最喜欢
1787	Aesthetic but also like effective kind of like personal vibe code and personal mini app setup of anyone that I've seen	美学,但也像有效的类型像个人的氛围代码和个人的迷你应用设置任何人,我见过
1788	Like I think it's like he's like really gone quite far in the direction	我觉得他好像朝这个方向走了很远
1789	And I think a lot of the leaders at meta who have been a matter a long time are	我认为很多在米塔的领导已经是一个长期的问题
1790	Like all very adaptive like I think you know kind of to the prior point	像所有非常适应性像我想你知道那种前一点
1791	They've been in a highly turbulent like period of technology now AI is like maybe	他们一直处在一个高度动荡的时期就像科技的时期现在AI就像也许
1792	You know five or ten times more turbulent and faster moving than those pathways	你知道五到十倍动荡和移动速度比那些路径快
1793	But I think they're like they have adapted to new trends many times.	但我认为他们好像适应了许多次新的趋势。
1794	And so I think there's like	所以我觉得...
1795	I've been very impressed actually by seeing how	我非常佩服地看到
1796	Quickly a lot of meta leadership	快速的很多元领导
1797	You know if I compare the kind of conversations I would have with them when I first joined versus the kind of companies	如果我比较一下我第一次加入公司时和他们的对话
1798	I have now they're not super a build	现在他们不是超级建筑
1799	Yeah, and it's it's very impressive honestly for that to happen at like the large scale of you know meta	是啊,这是非常令人印象深刻的,诚实地说, 发生在像大尺度你知道元
1800	Which is one of the largest companies in the world	这是世界上最大的公司之一
1801	Yeah, we got them there quickly.	是啊,我们得到了他们很快。
1802	It was um, yeah, but but I mean like you're saying it's like if you know you start	嗯,是的,但我的意思是就像你说的,如果你知道你开始
1803	Start coding then it's helpful	开始编码,这样有用
1804	But I think a big part of what a big part of the culture that we were trying to create is also	但我认为我们试图创造的文化的一大部分也是
1805	You know we were talking about how you have all these direct reports the point of it isn't that	你知道我们在谈论你是如何有所有这些直接的报告这一点不是
1806	You have the direct reports.	你有直接的报告。
1807	It's that there aren't it's the absence of non-technical managers, right?	是因为没有非技术管理人员,对吧?
1808	So so I don't think you would say that you're like, you know that you're providing a ton of managerial support to 200 people at once	所以,我不认为你会说,你喜欢, 你知道,你正在提供一吨管理支助一次200人
1809	That's not a feature of the thing.	这不是事情的特征。
1810	I'm well known	我当然知道
1811	How you didn't check I think it's just well known among researchers that you know	你怎么没查过研究人员都知道
1812	it's there's this common patterns like you go from being very technical to you go to a role where you're	有一种常见的规律就像你从很技术到角色
1813	You're not like in the details anymore	你不再喜欢细节了
1814	And then you very quickly become not technical or don't kind of understand it much faster than you probably realize and it's quite	然后,你很快变得不技术或不太理解它比你可能意识到的快得多
1815	Fast-moving domain especially so I think that like this whole thing of just like you have to stay technical people have to use the	特别是快速移动的域,所以我认为像这样的整件事就像你必须保持技术人必须使用
1816	Agents people should be writing code.	特工人员应该写代码
1817	Yeah	对
1818	I think that that's that has been a really important part of this whole thing.	我认为这就是整个事情中一个非常重要的部分。
1819	Have you had any very funny disagreements?	你有任何非常有趣的分歧吗?
1820	We have a lot of disagreements.	我们有许多分歧。
1821	Yeah, I mean we're going to hang up until very recently	是啊,我的意思是我们要挂断直到最近
1822	Some of the things I appreciate about like so it's like one of his I mean what this is like I think it goes back to your	有些事情我很欣赏,所以它就像他的其中之一。我的意思是,这是什么样的,我认为它回到你的
1823	Philosophy you're talking about this for recruiting back at scale is like hire people who really care, right?	你在谈论这个哲学,因为招募以规模回来是像雇用真正关心的人,对不对?
1824	I think there was some expletives in it	我觉得里面有些夸张
1825	but	不过
1826	But I don't really care as it turns out	但我真的不在乎结果
1827	No, that's like one of the things that I thought was you know, I think that that's like really important, right?	不,这就像我想的事情之一你知道, 我认为这就像真的很重要,对不对?
1828	it's um, so I think part of what you you kind of get when you like have the supernova of like all these different	所以,我想你得到的一部分当你喜欢的超新星像所有这些不同的
1829	Very strong-willed people together is like you're gonna have like fun fights and not fun fights	非常有意志的人在一起就像你们要玩游戏而不是游戏
1830	but like but it's like to some degree I'd rather have people push for things that they believe in and like	但喜欢的,但是它喜欢在某种程度上,我宁愿有人推动的东西他们相信和喜欢
1831	Argue over things then you know, just have people think that they're like blindly executing something, you know	争吵的事情,然后你知道, 只是让人们认为他们像盲目执行的东西, 你知道
1832	It's like people need to feel ownership.	就像人们需要感觉主人翁感
1833	So yeah, how does mark a different boss than you expected?	那么,对,如何标记一个不同的老板比你预期的?
1834	Well, I don't have a huge comparison set.	好吧,我没有一个巨大的比较集。
1835	I can leave now	我现在可以走了
1836	No, I think I think that like the I think it has been very	不,我想我认为这是我认为它一直非常
1837	One of the very cool things about	一个很酷的事情
1838	Like working with mark over the past year has just been watching	就像去年和马克合作一样
1839	like the true degree of his boldness and like and this trait that we were talking about before of like	就像他的胆量和那种真实程度还有我们之前谈论过的这种特质
1840	yeah, once he internalizes certain beliefs like the	是的,一旦他内部化某些信仰像
1841	Speed and magnitude of like willing to like change, you know prior actions and and and	速度和规模喜欢喜欢改变, 你知道以前的行动和和
1842	Pivot towards the like worldview that is consistent with those beliefs.	向符合这些信仰的类似世界观的方向发展。
1843	Like I think these are	就像我想这些
1844	pretty unusual traits generally speaking like even	相当不寻常的特质一般来说就像偶
1845	Among the space of like founders are can actually ingest new information and fully updated worldview and then act in accordance with the newly installed	在像创始人一样的空间中,可以实际摄入新的信息和完全更新的世界观,然后按照新安装的
1846	worldview very patient and then and then willing to take	世界观非常耐心,然后愿意接受
1847	I've never gotten such an exclamation point before	我从没有这么感叹过
1848	What lovely priors you have	你的前科多可爱啊
1849	Yeah	对
1850	But the posteriors	但后遗症
1851	You're not allowed to talk about posteriors anymore after the whole incident	事发后不准再谈后事
1852	Yeah, no, but and then but then I think I think also just the degree of of boldness and I think that like the	是啊,不,但后来,但然后我想我也认为只是程度的胆量我认为喜欢
1853	This is it was like actually quite puzzling to me at first the degree to which you were like willing to make these massive bets	这就像实际上相当令人费解起初,你喜欢的程度愿意做出这些大赌注
1854	Something actually not and I would often like talk about was like wow, we're really going big here	其实有些事没有,我经常想谈就像哇,我们真的要去大这里。
1855	Like and both of us would be like would you go this big?	就像我们两个都一样你会去这么大?
1856	Would you go this big and so I think that this I think it's like I mean that is that is like quite a notable trait	你会去这么大所以我觉得这我觉得这就像我是说, 这是一个相当显著的特征
1857	And and something I really admire and I think I think is I think is very I actually like if I	和一些我非常钦佩的东西我觉得我想是我非常喜欢如果我
1858	look back at the history of meta and I	回首我和元的历史
1859	Imagine what the like scenario would have been in all these like pivotal moments.	想象一下在所有这些关键时刻, 会发生什么样的情况。
1860	I can see how like	我看得出来
1861	This is kind of the movie so to speak or like this is kind of how a lot of those situations go where it's like	这部电影就是这么说的或像这样的那种很多这样的情节
1862	there's I	还有我
1863	feel like probably	感觉可能是
1864	the way I	我这样
1865	Imagine is like they're clear these moments where like something very very big is happening	想象一下,就像他们清楚了这些瞬间一样... ... ......
1866	And it may not even really be showing in the metrics or may not be impacting user	它甚至可能没有真正显示在度量衡中也可能没有影响用户
1867	Counts or impacting revenue or whatever it might be like it's it's not really impacting things	计数或影响收入或它可能像它不是真正影响事物
1868	but clearly if you believe that thing is going to happen then it has like massive implications for the company and the product and	但是很明显,如果你相信事情会发生, 那么它像巨大的影响公司和产品,
1869	and I think	我觉得
1870	You know my you know post-hoc retelling of metas history is that like I think metas a company where every time that happens	你知道,我你知道后热重述 metas 历史就像我想metas一个公司,
1871	Meta takes it extremely extremely seriously and that has enabled the company to innovate and continue evolving	Meta极其认真对待,这让公司得以创新并不断发展。
1872	With a very fast-moving and fast-changing tech landscape	技术环境变化很快
1873	And it's very Bayesian.	而且非常巴耶西亚语.
1874	Yeah, very good soap Asian	是啊,非常好的亚洲肥皂
1875	Life is short.	寿命短短.
1876	Just do the right thing	做对的事
1877	Yeah, and I think I think AI's I mean, I think AI is maybe that like, you know times 100 or whatnot	是的,我想我想AI的我的意思是, 我认为AI可能是喜欢, 你知道乘以100或什么的
1878	but like I think	但我觉得
1879	the	联合国
1880	You know a lot of CEOs would have told you last, you know last summer	你知道很多首席执行官会告诉你最后一个, 你知道去年夏天
1881	Oh, yeah	对
1882	I think AI is the most important thing for my business like I think maybe every CEO would say that	我觉得AI是最重要的事情我的业务,我觉得也许每个CEO会说
1883	but then you look at like what they're actually doing and how consistent are their actions with that belief and I think this is like a	但是,你看着像他们实际上在做什么, 他们的行动如何一致与这种信念我认为这是一个
1884	like quite quite special thing about Meta and Mark	喜欢相当特别的事情关于梅塔和马克
1885	Last question because it were passed all of our good times so flattered.	最后一个问题,因为它已经过去我们所有的美好时光如此受宠若惊。
1886	Yeah	对
1887	What do you make of all of Alex's deranged posting on Twitter?	你对艾丽克丝在推特上发的疯消息有什么看法?
1888	I mean, I think most of it is great, but	我的意思是,我认为大部分这是伟大的,但
1889	But I I guess my main view but it doesn't matter how how much you fail	但我猜我的主要观点,但不管你失败多少
1890	It matters how win how big you win when you win.	重要的是你赢多少赢多少
1891	I think with	我觉得
1892	So I think with with communication	所以我想与沟通
1893	I think it's most important to just be authentic and I think when you're trying to be authentic	我认为最重要的是,只是真实的, 我想当你试图真实的
1894	Some of the things don't come out the way you want and I think that that is a that it is better to make the trade	有些事情不是你想的那么简单我认为这样比较好
1895	In that direction than it is to be overly scripted and not be authentic	在那个方向上,比它过分的脚本和不真实
1896	But I mean at most of it I just feel it's hilarious.	但我的意思是,大多数情况下,我只是觉得这很有趣。
1897	So	这么说
1898	Yeah, what do you think about my deranged tweets?	是啊,你觉得怎么样我的疯狂的微博?
1899	I'm not a I'm not a heavy X user	我不是个重X用户
1900	But but you you you are back on I'm back to full posting	但你现在又回到我身边了
1901	I mean, it's like a bunch of the researchers are there and we want to you were there	我的意思是,这就像一群研究人员在那里,我们希望你在那里
1902	but everything I post there I also post on threads and	但我贴的每件东西,我都贴在线上
1903	It's it's all good.	谓一切善根.
1904	But every time I load X I have like four Alex posts and it just like cracks me out	但每次我装X的时候我都有四个Alex的帖子就像是把我弄出来一样
1905	It's like some of these like it's like where do you go?	就像这些东西,你要去哪里?
1906	We're just talking about you're now a clearinghouse	我们只是说你现在是一个信息交换所
1907	Where people send you good stuff like where do you come up with this stuff?	人们会把好东西寄给你比如你从哪里想出来的?
1908	It's like, yeah, this is yeah, how does it work?	就像,是啊,这是是啊, 它是如何工作的?
1909	most like we were we were talking to my partner yesterday and he was you know, I think one observation was just	就像我们昨天跟我的搭档谈话他就是你知道的我觉得有一点只是
1910	There was actually probably still to this moment.	事实上,现在还可能存在。
1911	I might my thought pattern oftentimes is oh, that's funny	也许我的思维模式经常是哦,这很有趣
1912	I wonder what Jolly would think about that and then and then you know, you get good memes out of that that being said	我想知道乔莉会怎么想然后你知道,你得到好的迷因从刚才说的
1913	I have experienced the the	我经历过
1914	sort of like brutal	有点像残忍
1915	metabolism of the internet and how quickly it like metabolizes memes and so the sort of like	互联网的新陈代谢和它如何快像新陈代谢迷因,所以那种喜欢
1916	You know, you really have to stay	你知道,你真的必须留下
1917	Stay on the on the cutting edge of memeing to default behind.	保持最前沿的记忆以违约后。
1918	Yeah, which is maybe a good summer of this conversation	是啊,这可能是一个好的夏天这个谈话
1919	It's easy to fall behind	很容易落后的
1920	Stay in the details.	留在细节。
1921	Yeah, stay technical and stay at the frontier.	是啊,保持技术并停留在边境。
1922	Yeah.	对
1923	All righty.	好吧,我们走。
1924	We'll leave it there.	我们把它放在那里。
1925	All right.	好吧,我们走。
1926	Thanks guys.	谢谢各位。
1927	Thank you	谢谢
1928	Okay, so what is everyone drinking?	好吧,什么是每个人都喝?
1929	Well that what is this?	这是什么?
1930	That's a keg of Guinness that I brought you.	那是我给你带来的一桶吉尼斯
1931	This is Guinness from Dublin.	这是都柏林来的吉尼斯
1932	Oh, wow.	哦,哇呜。
1933	Yeah.	对
1934	Yeah	对
1935	I was just there this week.	我这周才到
1936	Oh awesome.	赞极了
1937	Yeah.	对
1938	Well, thank you.	谢谢
1939	We got the local stuff.	我们得到了当地的东西。
1940	Yeah, there you go	对,给你
1941	Yeah, no, I figured this would be a good occasion to	是啊,不,我想这将是一个很好的机会
1942	Get the Guinness.	得到吉尼斯号。
1943	Yeah	对

该视频共有字幕 1956 条。解锁更多字幕为会员功能，请移动到价格

下载安装