Alexandr Wang, Meta’s new chief AI officer and head of its Superintelligence Labs.Alexandr Wang, Meta‘s new chief AI officer and head of its Superintelligence Labs, spoke to Himanshi Lohchab and Sruthijith KK about the tech giant’s plans to deploy AI at scale across its apps, guided by Mark Zuckerberg‘s vision of “personal superintelligence”. Wang, 29, is in India for the AI Impact Summit. Excerpts:
Meta‘s focus on open-source LLama models has taken a backseat… There’s more focus on AI productisation. Is Meta no longer in race to build frontier models?
There’s a bunch of pieces to this. So the first is we’re committed to open source and we intend to continue releasing open source models just as we have in the past. We’ll probably release a mixture of closed and open source models, but we’re still very committed to open source as a company.
The second is being committed to the frontier. And this is a really important point. One of my priorities when I joined, about seven months ago, was ensuring that our research programme was set up to be able to innovate at the frontier, to deliver the very best models in the world and to really get back to scientific fundamentals to be able to continue pushing the frontier. We’ve been working on our upcoming series of models.
Those will be good and we’ll release them in the coming months. But we are very focused on productisation too and ultimately while the models themselves are very powerful and exciting, what we really, really care about is that we can actually deploy this technology to everyone in the world. Mark (Zuckerberg) wrote this incredible essay on personal superintelligence, which is what could the world look like if we actually are able to deploy AI to every single person in the world, empower them to build the businesses they want or explore their passions or build that project they always wanted. And so we’re working a lot on ensuring that not only we’re building great AI in the lab, but that we’re also able to deliver that to our products in new and interesting ways.
We have seen much excitement around OpenClaw lately. Meta has just acquired Manus. Is the agentic era truly upon us? Is that the next wave on how consumers are going to use AI?
We are very excited about this. Manus released Manus agents on Telegram, and we really think this will be the year of the personal agent. We’re moving away from these AI systems that are passive and are sort of more search-based to ones where they’re working for you 24/7. They understand your goals. They’re able to take actions on your behalf. They’re able to constantly find ways to make your life better, to improve your life, to help you accomplish your goals and your tasks.
At Meta, we have three and a half billion people who use our platforms every single day. And we have an opportunity to really make personal agents globally ubiquitous in a way that we think that very few companies have the opportunity to. It’s an area that we’re going to continue investing in over the course of this year.
Will the era of agents mean that all commercial transactions over the internet, like ads, you’re trying to market things using eyeballs… Will all of that fundamentally change and will it all be agent-to-agent transactions?
It’s a good question. I think that it’s definitely true that I think the future of agents that we believe in are agents that you would trust enough to ultimately make purchases on your behalf and ultimately coordinate very complex logistics on your behalf. So in that sense, we really do believe in the world of agentic commerce and where agents are doing more and more commerce.
But is there a concern that it would undermine Meta’s core business of advertising and also take away the value of the social graph?
I’m not as concerned about this because I think that, if you look at the evolution of Meta’s products through the years, what you continue to see is that, consumer technology will continue evolving because new technology enables entirely new formats, entirely new technology, entirely new products for our users. I think that in a world where personal agents enable dramatically more commerce, there’s, if anything, a lot more opportunity to help businesses.
How is Meta planning its spending on infrastructure? Meta’s proprietary AI chip-the MTIA accelerator-has reportedly gone into production with TSMC this year. How soon do you see Meta becoming meaningfully self-sufficient in AI compute?
I think what we’re seeing is really twofold. First, there’s incredible returns to continue investing compute into our research programmes. And I think we’re seeing that overall as an industry, the speed at which the models are improving is still incredibly exciting.
And we can expect to continue seeing great progress as we continue investing. The other is on inference. Meta, unlike many of the other companies… We have a responsibility to deploy global platforms and we have billions and billions of people who use our products. And we need to ensure that we’re able to actually perform the inference to serve these billions and billions of users.
In terms of our own investments into Silicon and our chips, I think that we’re really focused on having a multi-Silicon strategy. We want to be utilising and leveraging many providers of Silicon and ensure that we have an infrastructure strategy that enables us to be able to scale with the demand as we see it and not be reliant on any one Silicon provider. And this includes investing into our own Silicon.
Are there conversations with Google to buy the TPUs (Tensor Processing Units) as well?
Not sure I can go into all of that, but we’re very thoughtful about having a multi-silicon strategy.
India is witnessing huge investments by global hyperscalers in infrastructure. There are supporting policy incentives from the government as well. Is Meta thinking about investing in data infrastructure here?
So, India is one of the most important countries for Meta, full stop. We have more than half a billion daily active users in the country. It is one of the fastest growing digital economies. We’re seeing a lot of that happen on WhatsApp, on Instagram, on Facebook. India is so, so important for Meta as a market and as a country. And one of the areas where we talk about is personal superintelligence and actually enabling every user in the world to have a personalised, very powerful agent.
India is one of the few countries where we could actually make that happen very quickly because of the reach and ubiquity of some of our platforms in the country. So we want to invest very, very deeply into India. And some of the things we’re doing, for example, are a lot of our investments into connectivity in the country and a lot of continued investments into supporting enterprises. And we have a joint venture with Reliance Industries on enterprise AI adoption. And so we’re always looking at ways to continue investing into India, but not much more to say right now.
AI is seen to be disrupting white-collar employment, and we are a country that has a very substantial service economy. Should we be concerned at a very large scale disruption of sector?
It is a complicated question. But one of the things that we see, even on our platforms consistently, is when we empower more and more small businesses, more creators with AI tools and AI products and AI platforms, the opportunity vastly outstrips any of the disruption. I think the version of the world that we believe in is, for sure… AI coding tools and a lot of the progress of AI on knowledge work is very real. But if anything, we think that will be utilised more as a tool for small businesses, entrepreneurs and new businesses to start.
One of the things that struck me as I mentioned, we had this roundtable yesterday with many of the sort of Indian founders and the Indian investors in the ecosystem. One of them had cited there’s more Indian consumer AI companies than there are American consumer AI companies. And so I think that if to the degree that India and America and frankly, every country is able to really embrace entrepreneuriality and supporting new and small businesses, that is a nearly endless sort of growing sector for the economy and one that can support a lot of the growth in the coming years.
Tell us about your own personal journey with AI. When did you discover these tools? What was your reaction as a teenager to it? How did you get the idea of starting a company and what has it been like? And how do you view the capabilities of today’s models?
I originally got exposed to AI when I was in high school. And this was back in a much simpler time when Google announced a model that could detect cats in YouTube videos. And that was a state-of-the-art algorithm at the time. And then when I was in college at MIT, I was studying AI. And the project that I tried to do in college was a camera inside my refrigerator that would tell me when my roommates were stealing my food. So I’ve been teeming with the technology and been quite excited about the technology since very early on.
And that’s really what inspired me to start my own company when I was 19, because I was inspired by the sort of potential of AI as a concept. But it was still nowhere close to obviously what it is today. And at that time, in 2016, when I started my company, the big focus was really on autonomous vehicles and autonomous driving. And that was sort of the holy grail use case. Fast forward to today, it’s only a decade later. And I think it has been staggering to watch the progress, to put it lightly. I personally had many kinds of ‘aha’ moments over the past few years as I’ve used it.
What is the one point on which you and Mark Zuckerberg are most in sync? And where are your views most divergent?
I think these days we’re pretty in sync on most things. I think we’re very in sync on the opportunity for personal superintelligence, the opportunity for personal agents wanting to really deploy this technology globally and the importance of being at the frontier and pushing on research. One of the things that I really respect about Mark is he’s always open to feedback. He’s a very strong debater, that’s for sure. But if you’re able to debate with him, he really does listen.
You’ve been a very successful builder in the domain of AI. Comments have also been made more recently about your relative youth and inexperience in research. How do you respond to these comments? Is there a builder versus researcher schism within AI right now?
I think AI as a field is just so incredibly dynamic and is changing so quickly. It’s an industry where you’ve seen a lot of disruption. For a long time, Google was by and far the leader in AI. And then you had OpenAI come along and now you’ve seen Anthropic come along. And so it’s an industry which really is quite dynamic.
And there’s always opportunities for fresh perspectives to be able to enable organisations to succeed and accomplish things that even the incumbents aren’t able to. My opportunity with Meta is to bring a fresh perspective, bring more of a builder’s perspective and ultimately enable us to approach the problem in a way that’s different, but also that enables us to do things that will differentiate us in the marketplace. And so I don’t know if I view it necessarily as a disadvantage, per se.
The AI arms race is insanely talent focused. What is your approach to attracting and holding on to talent?
Yeah, a lot of my personal focus is ensuring that they have the best possible environment because for most of these researchers, they’re so inspired by the technology and the science and what is going to be possible. And it’s really my responsibility and Meta’s responsibility to build the best possible environment for them to perform their science and to continue pushing the frontier.
One of the things is having a very talent-dense group. Smart people want to be surrounded by other very brilliant people who can help push them and they can learn from and create a very special environment that is frankly not possible in other circumstances. The other one is compute, enabling researchers to have lots of resources to be able to conduct their research and explore their scientific ideas and having an organisation that’s really focused on scientific basics. One of my focuses while coming in was really kind of resetting the organisation’s focus on the basics. We’re going to focus on scientific rigou0r. We are not going to have artificial deadlines for any of the work that we’re doing. We’re going to approach this stuff in a way that is proper and correct for the long term.
But is this ecosystem creating unrealistic expectations on compensation? Does that make the movement of talent altogether unstable?
Obviously, some of our hiring created headlines, but long before that, there were others in the ecosystem who are compensating researchers very well. I think it is a very dynamic environment. And researchers have created breakthroughs in science that are somewhat unprecedented in maybe even the history of humanity, like there are such deep advancements. But more than the compensation, we really focus on creating the best environment.
OpenAI is soon going to have ads. Researchers have sounded warnings that maybe OpenAI is making far larger mistakes than Facebook did when it started embracing advertising. What are your thoughts on it?
One of the things we think a lot about when it comes to deploying personal agents is how important trust is, because ultimately your personal agent that is truly personalised to you will be a technology that is more intimate than any technology has ever been historically. It’ll know a lot about you and be able to make a lot of decisions on your behalf. And so if we believe in that world, trust is incredibly, incredibly important.
And the only way that we’re going to be successful as an organisation in that world is if we’re able to build trust with consumers and trust with the billions of billions of people on our platform. So a lot of our decisions around AI and the technology in the future are actually quite rooted in how do we build that level of trust? And how do we build that in the long term, as kind of a North Star and first principle?
Superintelligence used to be a very elaborate term denoting a technology smarter than any human. But we’re seeing a tendency for people, including Mark, to speak of superintelligence in very general terms. What really is superintelligence?
We think about it as AI that is able to outperform humans on any measurable task. So it has a technical term. And I think a lot of the reason that we named Superintelligence Labs, was to speak to the ambition of our research agenda. And I think a lot of our research is really focused on the power of technology or a sort of ambition of the technology that is quite expansive. I agree with you that these terms have the tendency to become buzzwords and in ways that they get a life of their own to some degree. But ultimately we really care about the scientific definition of very powerful AI.
Leave a Reply