Zuckerberg 赢下 AI 竞赛的秘密计划——Meta AI 掌门人揭示未来
Zuckerberg's Secret Plan To WIN The AI Race | Meta AI Chief Reveals Future
What's it like working with Mark? What are some things that you've had a completely different experience with? >> I think one of the things that's really struck me in working with him is how quickly he sees the future. Ladies and gentlemen, Mr. Alexander Wang. Back in 2022, he became the youngest self-made billionaire in the world. >> Chief AI officer at Meta. >> Started a company when he was still a teenager, went through Y Combinator, and then built Scale AI.
>> Alex Wang is the most expensive aqua hire yet, 14.3 billion dollars. >> He's a fascinating character within the valley. And he now leads Meta's superintelligence labs division. You work very closely with Mark. You're building Meta superintelligence. What is Meta superintelligence? >> You know, the discoveries that we make over the course of the next 5 years in terms of AI are going to be some of the most monumental discoveries of human civilization has ever made.
>> You know, I have the Meta Gen 2 Ray-Bans, right? And I just feel like maybe you guys are running a very old version of Lama there. It just doesn't feel like modern AI yet. Do you have an estimate for when we'd have modern AI on the glasses? Uh, very soon. Ladies and gentlemen, I'm with somebody who I think is very special, who I think, um, you know, is somebody who spearheads the AI race, but is slightly invisible to the public.
Like, you're much less public these days compared to everybody else in AI. So, thank you so much for doing this. My last interaction with you was, I think, at breakfast. We did breakfast at Meta, uh, at the campus. And, you know, one of the questions I had for you at that time was you had this fidget spinner. And it was a [clears throat] very, very different fidget spinner. I I I don't know. I've never seen anything like it.
What was it called? There's, uh, there's this great, I think it's a British company called MetMo. They make these, uh, very, uh, uh, high-end fidgets. Uh, it's a it's a good company. It's very cool. >> We're sending MetMo some sales. >> >> Uh, but but, you know, Alex, I want I want to start with this question, right? Like, what is it that you do now? I know you from the Scale AI days, right? Uh, but today I know that you work very closely with Mark.
You're building a Meta superintelligence. What is Meta superintelligence, uh, for everybody, you know, watching? I mean, first, Mark and and myself, we very strongly believe that this is a very special time in human history and you know, the discoveries that we make over the course of the next 5 years in terms of AI are going to be some of the most monumental discoveries of, you know, frankly, that human civilization has ever made.
And so, MSL, Meta Superintelligence Labs, was entirely dedicated towards how do we build and develop the most optimal organization to be able to both deliver the breakthroughs and the scientific advancements necessary to deliver superintelligence, and then also build the products that will enable this technology to be deployed to billions and billions of people worldwide. And one of the things that makes Meta such a special place for all of this is the fact that we have such incredible scale and reach through our products, you know, 3 and 1/2 billion people utilize our platforms every single day.
That puts us in an incredible position to actually bring this technology to the world in a way that's really different, we think, from many of the other AI labs out there. And then we also, you know, um we had the opportunity to 7 months ago, when I joined, really uh kind of design the org um and design the team from a blank slate on what is the optimal team look like for the future of superintelligence. So, you know, we really embraced how do we build the best possible scientific foundations for this organization, how do we have the highest talent density, how do we bring the very best people together and build the best possible environment for breakthrough research to occur.
Interesting. Um how do you balance commercializing some of these products along with research? Because from my understanding um and whatever is available publicly, it seems that you come more from a product's perspective. Like, you've done Scale AI, you've helped some of these product companies really grow. Uh how does it feel like going to research and how How you balance both? At the core, we need to be research-led because fundamentally, um we're in such a special time and and moment when it comes to frontier AI technologies and breakthroughs that um I really think that that uh you need to be very focused on the research um and the opportunity to actually push the frontier to have very real breakthroughs uh in super intelligence is just is just magical.
And so, we're frontier or we're uh research-driven to be able to push the frontier, but we actually view it as like a flywheel internally for all of Meta. So, um by building frontier models and building models that um are uh are pushing the boundaries of super intelligence, that enables to build incredible products. Um and it's kind of a base material that allows us to build some of the most interesting and innovative consumer products in the world.
Those products, as they gain scale, then give us the ability to grow our infrastructure footprint and to build um you know, large-scale infrastructure, some of the large-scale in the entire world. And then that will enable us to build even greater models and continue scaling our research efforts. So, um it actually is like one virtuous flywheel within Meta, and I think that's a lot of what excites me as well is that we're not um uh I think we view this as a very evolving uh and um and continuous discipline to continue advancing the models, products, and infrastructure all in tandem with one another.
And so, what's the team like? Like, how do you structure the research team and where where is handoff to to products? Like, how does it Like, if you can give me like, for example, uh an example of something you're working on, what's the team structure for that like, and then when is handoff to product? That'll be very useful. Yeah, so so actually, one of the things that has really struck me has been um if you look at a lot of the most successful AI products and a lot of the most successful AI developments that have happened, they They from a um uh a tandem effort between research and product.
So, um we're we're I think past the phase where it's just about research in, you know, a corner and then that hands off to product people and then they deploy that. I think if you look at a lot of the biggest breakthroughs like ChatGPT or Cloud Code or a lot of the things that we're working on, it comes from um you know, researchers who are thinking about the product and product people who are thinking about research and then working hand in hand with one another to sort of co-develop the best possible products.
Some of the things that we're really excited about are personal agents. You know, recently um Manas released uh agents that are working 24/7 on behalf of our users um and are constantly working to uh you know, accomplish your goals or make your life better. And we really see that as um this is the first of uh a a whole series of of um products that we're excited to release around personal agents. It's one of the areas where we think there is some of the greatest opportunity to actually give a more powerful version of AI to every single person in the world.
And I think we'll be one of the things that we'll we'll look back on a decade from now and view as one of the big breakthroughs in AI productization. Yeah, I I I really like Manas. I remember using it, you know, a few months ago. I I I keep going back to it again and again. It's a cool product. Um do you have a sense of what Meta's identity is in the AI battle? I feel like you know, Anthropic has one. It feels a very It feels like It feels very um you know, Machines of Loving Grace style.
OpenAI has one that's very consumer pop-friendly one. But like Meta's identity is still very much on device. It at least to me as a consumer, it feels like on device, it's there, but the device is like up front and center. But is there an identity you're you're you're building for for Meta's superintelligence play? Yeah, I think um I think what we really believe in are personal agents deployed globally. So, one of the things that makes us unique is that we are a global company and we have half of the world using our products every single day.
I mean, that is just an incredible amount of reach and it means that as we deploy powerful personal agents to everybody in the world, it creates totally new opportunities I think are very hard for any other lab to actually fully accomplish in terms of what does that mean for entire communities? What does that mean for countries? What does that mean for you know, the whole world as we all onboard onto this technology together?
The other thing that we're really excited about is the sort of continued um what does the hardware vision look like, right? And this is an area that we've been investing in for many, many years. Our wearables and sort of the the next form factors for consumer hardware and we think that with personal agents, um that vision has never been more real where I think you're going to want your personal agent to on a constellation of peripherals in the future and you're going to we're going to expand beyond the phone into a world where you're going to want your personal agent to be with you in a bunch of different ways that that that will always be on, be you know, see what you see, hear what you hear and will be able to just help you in a way that's much deeper than and then even devices today are able to help them.
Cool, like a like an always-on friend who's with you on all your devices. That's very cool. I think there are a lot of other companies trying to do that, but I think you have the step ahead because I've already worn the Meta hardware products and I'm already comfortable with it, so it's very easy to say, well, here's an update which which allows you to to have, you know, this got to your intelligence on. Which actually brings me to a an aside point which is, you know, I have the Meta Gen 2 Ray-Bans, right?
Um and I just feel like maybe you guys are running a very old version of Lama there. It just doesn't feel like modern AI yet. Yeah, yeah. Do you have an estimate for when we'd have modern AI on the glasses? Uh, very soon. Um, you know, I think I think uh we've been, you know, when I got when I got to Meta something 7 months ago, um, the entire focus was let's set up this organization uh, in the right way for the long term.
So that we're not we're not just setting us setting ourselves up to cut a corner here or to opt in for some short-term outcome, but but mortgage the long-term opportunity. We're going to set up it with the right set it up with the right scientific foundations. We're going to set up with the incredible talent density and the focus on the long-term science. We're going to remove artificial deadlines so that we're actually building the technology at the best pace.
And what we've seen from that is actually over the past 7 months we've built the foundations incredibly quickly and now we're at a moment where I think over the coming months you're going to see incredible velocity coming from us and that'll continue throughout the course of the year. We think that um, the over the course of the full year we will really be pushing the frontier in a very exciting way across many dimensions of the technology.
Um, so uh, so stay tuned. I know it's been a a long wait for many people, but uh, we're really excited about what's coming. Yeah, because I I just feel like that's very easy you know, sort of upgrade for me, right? Or everybody else which is just have those glasses be super because I I just feel like they're being artificially restrained by a very old AI on it and I just know what modern AI can do and you already have access to the camera, you already have access to audio.
Uh, you can do wonders, right? Uh, but to your point we've already sold millions of of units and it's already a ubiquitous technology and and so I I think the opportunity is just immense. Yeah, I think it's one software update away from superpowers. So I'm very very excited about that. But I think it's incredibly mature of you to come in and first build out the org and it's something that I've learned like this year, uh, compared to, you know, 5 or 10 years ago to build out the org for velocity a year later.
Um how do you learn all this? Like what's the difference between you as an 18-year-old entrepreneur, you know, starting Scale AI versus today building, you know, what you're building today? Like what's the difference between you as an as as an entrepreneur? Yeah, I mean, I I feel like I've learned just so many different things. Um I think I think one thing when you're when you're young, you're very impatient, right?
And uh and I think you know this as as as well as I do. Um you know, I started my company uh right out of college I dropped out of college and um you know, you you're so impatient to make things happen that and that's both a great strength and a great weakness. Like I think on the one hand, you do you can make things happen faster than than other people would expect, but you're also um uh not necessarily setting things up to be long-term sustainable and to be um to create long-term advantages.
And one of the things that um I've thought a lot about, you know, if you sort of study the history of uh of great businesses in Silicon Valley or even just broadly in the in the sort of like history of business, the ones with true staying power have built some sort of foundation that is actually very difficult to replicate. And it's something that is it's almost like a seed that is planted and grows over the course of of of, you know, in many cases decades.
And so I think a lot of um a lot of how we think about how we thought about building MSL and how we think about building teams and um you know, accomplishing big things going forward is how do you set something up such that uh it has durability and it has a um it has a differential point of view all the way down to the organization and that enables it to grow and expand and develop in a way that is um that will be continued to be differentiated along to the future.
So I I think this sort of mix of you know, I think I think you know, to put it pithily, I think you can't let your impatience drive you um too far and you need to um it's important to always be thinking about what foundations are you building and how what is the long-term story look like. Yeah. You know, if I had to summarize the last 10 years of my, you know, entrepreneurial career, I'd come up with the same insights.
Just slow down, intentionally slow down, build build for the long term, and then eventually and also put the right people together, right? Because, you know, and and they need like 3 or 4 years to bloom. So, I I totally get you. Hey, what was the Scale AI exit like? Like I mean, I don't even know if you can call it an exit, right? So, They're still going. Yeah, they're still going, right? Like I I would say, what is the what was the relationship?
Why did you go decide to work at, you know, Meta Superintelligence? Like how did that conversation Mark happen? Give us some some insider information. Yeah, yeah. Well, um I mean, it was incredibly non-standard um and I think the way it happened was even um uh was very surprising cuz it took, you know, Mark is obviously quite a bold and visionary leader and I think it really took um you know, uh that level of vision for for the whole thing to come together.
Um and I mean, it was an incredible I think milestone for Scale and everything that we've done um and Scale continues on and I think that that team is continuing to execute and deliver uh on really incredible um outcomes for enterprises and governments and um continues to to uh crush it. But, you know, the the opportunity that we saw really was um you know, there's it was an incredible milestone for Scale and it was a way to kind of give back to all the people who who have uh supported Scale's um you know, success to date including the investors and the employees and and everybody involved while also um setting Scale up for the future.
And and frankly, the opportunity that I saw with with meta was was just astronomical. I think that um you know, sometimes in the middle of these AI races everything just feels so um pressurized. So you can sort of lose sight of things and and um I think at that time you know, a lot of people were weren't giving meta the credit it deserved in terms of I mean it has all the ingredients for incredible success in AI. It has the distribution, it has the billions and billions of users, it has the scale, it has the business model, it has um the incredible talent, it has the infrastructure.
And so all the pieces were were really there and I think the opportunity to really um kind of lock everything in in a way that allows us to allows meta to to really um succeed and thrive in the long term was very exciting. Very cool. You know, I have I have this question that I ask almost everybody, you know, in the AI race, right? Which is Uh you know, I'm sure you have some ideas and thoughts about AI that your peers don't agree with, right?
Some way it's going to go in the future or some, you know, method of training or whatever it could be, right? Uh do you have an insight that you feel your peers might not agree with? I'm talking about your peers in AI. I think some some people in the industry agree with me on this, but um one of the things I think is is of paramount importance is developing the technology with extreme responsibility. And I think a lot of the concerns around safety, um you know, both the traditional concerns around AI safety as well as new concerns around how do you ensure that the technology is is used safely by the billions and billions of people are going to use it every day.
I think these are incredibly incredibly important. And I think we're seeing this as with any new technology as it gets deployed, um you know, they raise novel safety concerns and the responsibility really is on us to develop to develop the technology um in in in a responsible way. You know, the the other piece that's very important for this is the vision of the future that we see where where it is a personal agent, something that is sort of with you all the time that you really trust with your goals and your hopes and your fears and sort of everything in your life.
To build that technology effectively requires just a huge amount of trust from our users, from the public, from governments, from um from every stakeholder that you would possibly imagine. And so um we really take the sort of the need to build safely and and thoughtfully extremely highly. And I think this is shared by some in the the um AI community. I think some people are have, you know, um moved away from some of these commitments, but it's something that we're we're uh I at least take extremely seriously.
Would you have like a chief philosopher to to set the tone for your AI? I think Anthropic is somebody like this now. Um Gemini is very, like, you know, rate of refusals is very high. It's very bland. Uh would you have as Meta have a chief philosopher to to set the tone for how that AI behaves? Yeah, we we actually we collaborate with uh a number of both philosophers and psychologists to help us develop and build the behavior of the model in a way that we think will be most conducive for um being helpful and and empowering our users to accomplish their goals.
And and you know, it's funny, one of the things that we've spent a lot of time thinking about is how do you develop kind of like a um uh a mutual uh you know, in some ways like uh um uh like a mutual relationship between the between the humans and the agents where the humans obviously um want the agents to be successful and the agents want the humans to be successful Um and and figuring out how to engineer then design.
That is something that's I think very important for all of this to work out effectively. Very cool. I have one last question for you cuz I've I've taken a lot of your time. What's it like working with Mark? What's his working style like? What are some of the things that the public says that are true in terms of working style? What are some things that you've had a completely different experience with? What's it like working with Mark?
Well, I think first Mark is I mean Mark is one of the most you know notable individuals in technology for you know the past few decades and I think he um I think there's a lot out there about him that that is not fair from having worked with him very closely. I mean first he is just like in many ways a very regular guy. He's a total family man. He is very devoted to his children and his family and his wife and I think his his personal values I think are are quite commendable.
But what's more than that I think as a leader he is um incredibly bold and ambitious and and I think one of the things that's really struck me in working with him is how um how quickly he sees the future is maybe one of the one of the terms I would use. Like I think he is able to take a technological advancement or something that's happening all the way at the technology level and then really play that forward in terms of what does that mean for our users?
What does that mean for consumers? What does that mean for businesses? What does that mean for our entire ecosystem? And then work with everybody on the team to make that happen as quickly as possible through Meta. So I think he's been a very I've I feel very lucky to be able to work with him. One of the sort of like great entrepreneurs of our time and uh yeah, I think it's a real it's a real pleasure. He enables me and I think the whole team to dream bigger than than we would otherwise.
Very cool. Thank you so much, Alex. This was very enlightening. I learned so much about, you know, Meta super intelligence. I can't wait for, you know, superpowers to be on my glasses. I also can't wait for the new ones, right? I saw the displays at at Meta Connect, but you know, I I don't think they're available in India yet. So, at some point I want to get my hands on one and just, you know, use all of the AI you put in there.
I can't I can't wait to see all the cool stuff you do with the neural band as well. So, good luck and yeah, I'm going to continue buying and using your products. Yeah, it's going to be the the future will be here faster than we think. Yeah? This year? Yeah. This year? Awesome. Thank you so much and make sure you subscribe. Bye.