Meta CTO Andrew Bosworth: Our Path To Frontier AI, Renting Models, Consumer AI's Struggles

Well, thanks for having me.

Well, the only other ingredient I would add is great data. And you have that and we do have that as well. So, there are two stories here. The first one is, we go back to Llama 1, Llama 2, Llama 3. We were at the forefront and advancing things, and you of course know this. The Facebook fundamental AI research group goes back a decade more. That's where I first got queued into what was going on with AI when the AI messaging bot popped up in my feed, and then I met Yan and started to meet the FAIR people, and I realized this technology is progressing really fast, so Meta was on it very early. The real gap, which has been pretty public, was when we were pulling Llama 3 together, we had pulled in all the research and every stop we had, and unwittingly killed the pipeline. Researchers build a base, then people pioneer incremental versions and pathfind entirely new strategies. Unbeknownst to us at the time, and it speaks to the fact that we weren't focused enough on it, Llama 3 was a great model and well-received, but to get to that model, they had pulled forward all the future bets. That meant when it came time for Llama 4, we didn't have any of the pathfinding the other labs still had. So we fell behind on reasoning, mixture of experts, and a bunch of critical technologies. This was a pretty public disappointment a year ago and led Mark to shift from seeing AI as one of our bets to a bet that's foundational to the entire company. He flipped into founder mode, becoming so focused on getting us the compute and talent we needed. The researchers we signed landed about a year ago. Alexander Wang just hit his one-year metaversary, and I've loved working with him. We are seeing the fruit of that with Muse Spark.

It's not our frontier model, but it's the latest model we've released, a very well-received model. Depending on the benchmark, it does really well on things we care most about that are unique to our products. You're absolutely right about where we are in terms of public perception. We've built the team, I really believe in it, and we have all the compute and data we need. I'm very confident we'll be where we need to be. I'll add a second piece that is strategically very important: models are available. You can rent a model from Anthropic, OpenAI, Google. They're great. The real value we're going to create is the product. The vision we have for personal super intelligence is one we're uniquely suited to deliver. It's not just that we have data; we have a better chance of understanding you and what you're trying to do. Having the model is one piece, and you want that strategically to avoid dependency, but the model itself isn't the value. We'll get to a world where consumers don't care about the model version; they just want functionality. Today the discussion is about models, which suggests we're underindexed on the user side. We need to demonstrate the value to consumers.

Yeah, it's not just that. This is the whole industry. The era of the monolithic model died around Llama 3 launch. The idea of one model that rules everything is gone. We're now in a world where harnesses shop underneath to lots of different models depending on the task. For example, Gemini farms tasks out to different models for image generation. We've moved past one model ruling everything. What you want is a very expensive, intelligent model that you can distill down and use only when necessary, and otherwise use cheaper, faster models. Human tasks don't have infinite intelligence demands. I believe in scaling laws, so raw intelligence will scale up, but there will be stratification. It's not about one model to rule them all, but a collection of models that solve problems with the right balance of performance, price, and value.

There are two parts. We use lots of different models today. You want to provide consumers the best model for them, considering price, performance, and latency. Having your own model gives you the ability to control your destiny and stronger negotiating terms to get consumers the best available answer.

It's too early to tell. I don't know what the experience will be yet. I don't have access to it. For us, we're talking about personal super intelligence. We want to bring a tremendous specific capability to bear, not just general intelligence. We're seeing this as an entirely new way people interact with computers. It goes back to our work in Reality Labs. We've modeled ourselves after pioneers like Xerox Park, SRI, and Bell Labs. We're thinking about how to get information from our brains into the machine and back. AI is potentially the best tool for that, especially if it can observe things around us. Those are unique capabilities we're trying to bring to bear. It's not just the model, but the model's ability to work with novel inputs and create a closed-loop system.

No, we do rent models. We use them in development. We have a lot of development on our own models, but also on models from Google, Anthropic, and OpenAI.

There's no reason for us not to. When we're doing development internally, we use our own models and also models from others. The ability to be model agnostic and economically sensible hinges on having a competitive model, right?

It's also worth noting that whether it's a developer or a consumer, I don't want them to worry about the model over time. Today they have to, but over time they just have a goal to accomplish.

There are a bunch of pieces you have to connect in product, distribution, and consumer experience. The collection of all four is our superpower relative to competitors, most of whom only have one. I'm going to get into product deeper. But first, last time we spoke, you told me you wouldn't merge with AI, but the way you're talking about this sounds a lot like that. Have you changed your mind? No, I don't see this as merging with AI. I still want a very clear separation.

I know. We'll keep it going. It's a continuation of a trend where the bit rate between us and machines goes up. We've already been doing this with autocorrect, which is a little AI that improves the bit rate. QR codes are another example. If AI can understand things in human terms, that's a profound improvement. Combined with AI's ability to synthesize information, we've tremendously improved the bit rate. Doug Engelbart's idea was that human problems were getting harder faster than human capability, and he wanted human-computer symbiosis. He led the first video call, joint document editing, the mouse—all from wanting to increase the bit rate. AI is exactly that kind of thing.

Everyone's doing exciting work. The business Anthropic and OpenAI are pursuing is enterprise, building harnesses. That's where the money is, and they need capital. Their major focus is on work use cases. That's not our major focus; ours is 100% on how this helps consumers in their lives. I don't know that AIs become indistinguishable. There's a real question: you framed it yourself—these are like a personal assistant with access to information about you. It's a trusted assistant.

I don't know why we thought this one would be immune to the hype cycle. The hype cycle is an evergreen concept. It's not that the technology is fake; it's that people willing to go through hoops to make it work are a small percentage of the population.

My favorite example is agentic work. I was early on with Pi and Myclaw, building and playing with agentic frameworks. They're powerful but not user-friendly. They're hard to build and maintain, and they drift over time. I built one for my wife and I on WhatsApp. She never uses it; I use it all the time. She just asks me to do things. I'm the agent. It's hard to integrate into a workflow. We haven't made these things easy to use yet. We've done a great job with search and research use cases, and generative AI for content. But we haven't made it something people want to integrate into daily life. It's not easy enough, doesn't create enough value, and is too fussy. That's the product problem to tackle.

Yeah. I mean, I try not to judge the way people choose. No, I agree. Yeah. With technology, my instinct is that for the overwhelming majority of people, the major benefit of AI is going to be increased time for human contact with people that they care about, people they love. And I talked about this a lot in the context of augmented reality, for example. Even just the camera glasses that we have, when I'm with the kids, I'm able to both record something and share it with my wife, which is meaningful to us, and also be fully present, and I don't have a phone between me and them. And that's an important piece for me. I've talked about if you were able to be more effective with your work, that's more time that you're not spending commuting, that's more time that you're not spending away from your families, from the ones that you love. My personal sense is that for the overwhelming majority of people, the value of authentic human connection only goes up over time. It doesn't go down over time. And I think we're seeing that a little bit in how people's reactions to AI early on have been. I think people are worried that it's a replacement of technology. I don't find it that way myself, having been an avid user of it. And actually, mostly I'm spending more time not having to be at my computer thanks to it, not the opposite. So I think that's how the overwhelming majority of people will interact with it and how it will affect their relationship to media and to their loved ones, which I think puts a premium on authentic connection and authentic human moments. But I'm sure the entire distribution will exist.

Yeah, that's right. So we'll talk about that right after this.

I can't comment. I haven't seen them. I haven't seen them myself yet.

Well, this will be the first video of me wearing our new glasses. We'll see what happens. We'll let the market decide.

Yeah, phones are great. I mean, I love phones. I have two of them. I think they're wonderful devices. From the very beginning, the question we asked ourselves was this exact question. And we said, 'Okay, phones are great. What is something that you wish you could get access to that's on your phone without having to take your phone out of your pocket?' And we came up with camera and audio. It's just very simple. It's like, 'Cool, if I could just do that.' The AI has been this tremendous tailwind where actually it unlocks a much larger swath of potential capability over time than what the phone can do just through Bluetooth connections. So yeah, it's much more promising now than it looked two or three years ago. Two or three years ago, this looked like, hey, at some point you have to put a display on this and it has to become a standalone system and it has to have all these accessories attached to it. Now, it actually looks like there's enough room in the market for a big range of wearable devices. Glasses certainly, probably not just glasses, probably a lot of other things. People don't want to wear glasses, they want to wear different things. And some of those devices are just going to be input and output to your phone. That's cool. If it's just making your life more efficient in terms of how it's doing input and output, that's awesome. Some of them will be more complete. So for our band display glasses, for example, we just launched a vibe-coded platform for it, so anybody who wants to can go literally just build whatever app you want for the glasses. Right now, you kind of build the app and you put them on the glasses. But in the future, there's no reason that couldn't just be you wearing the glasses in real time, telling the glasses what app you want right now and having it on the fly build that app for you.

You know what I'm saying? And so I think we are headed towards a very cool zone where it's a little less app-garden specific. You're still going to have these content homes. Content continues to be an evergreen and important thing, as it has been on TV, as it has been on social media, as it has been everywhere. So there's still going to be places where media that you want to reach lives, and those look kind of like apps or channels or whatever, for lack of a better term. But there's a long tail of things like, why does my toaster need an app? Let me ask you this in seriousness. Like my toaster has an app.

I don't want that, right? I just want to tell my AI agent, get me the toast that I want. It's the same toast I have every day. Just get it for me. I don't want to have to go do whatever thing. What does your toaster app do? Does it let you...?