Mark Zuckerberg on Llama, AI, & Minus One

Okay, I'm going to kick us off. First off, good afternoon and welcome to South Park Commons. I think most of you know, but SPC is for talented builders and technologists who are trying to figure out what's next, which we define as the minus one to zero phase. We help our members build conviction to do their life's work. I am so excited to welcome Mark to SPC. He was my first boss ever. He needs no introduction as the founder and CEO of Meta, but today I really want to highlight him as a builder, a hacker, an original thinker, Zuck, which also happens to be his handle, and really discuss, you know, how he's maintained the minus one to zero founder mindset through the decades building Meta and keeping Meta at the forefront of innovation. They recently launched and open sourced their large language model, Lama 3.10. I don't have to tell anyone here. Yeah, woo! We celebrated in true SPC fashion with a hackathon over this weekend, so thank you for that as well. But with that, let's kick it off. Mark, you've had a glow up in the past couple of years. What has led to this personal transformation? Well, I injured my knee. So I was like, all right. It was kind of like I had my hair cut short because I was fighting and training before, and then I was like, I can't do this for nine months. By the way, this week, I'm officially healed, so I'm ready for some violence. But I kind of had this sort of grow thing when I was younger. I think Priscilla or someone was like, hey, you should grow that out. I was like, you know, I'm not doing anything else for the next nine months. It's looking great. It's looking great. Yeah, it's looking great. On that note, you've also been the main character in technology for the last 20 years. How would you delineate the chapters you've seen, and what chapter do you think we are in now? I don't know. I'm trying to be less of a main character. I think it's kind of working. It's working a little bit for me to just be a foil for other people. I don't know. I mean, we've gone through all these different phases, right? I mean, there was the kind of early Facebook, get it to work, survive phase. Then, like I said, the last 10 years, just kind of all the weight of the politics and volatility and responsibility around all that stuff has been just like a very different phase for the company. Obviously, that's not done, but I do think we sort of now kind of understand the different dimensions of that and have done a ton of work to do the things that we need to do. I feel better about that. I think that's allowed us to pivot more to just offense and doing more kind of proactive, awesome things. When I think about the next phase, there are a lot of ways to do awesome things, I think, in the world. The way that I think social media ended up being good is it gives everyone a voice. It helps you connect. Lots of people use it. I don't actually think it's super groundbreaking. It's basically you're giving people an ability to do something that's pretty fundamental and basic, and it just required really good execution. I think it's profound at large scale, but I think the scale is the thing that makes it sort of interesting. It's very much 1 to N, to use your analogy. I kind of think for the next phase of my career and what I want to do for the next 10 to 15 years, it's more working on just sort of breakthrough things that are more awesome and inspiring regardless of scale. Obviously, for the next computing platform stuff that we're doing with AR glasses and VR headsets and all that stuff, I do think we're going to get to a point where billions of people are going to have these things. It's like billions of people already wear glasses. It's kind of like everyone who just had a not smartphone easily upgraded to a smartphone, and then the rest of people were a little slower. I think here it's like everyone who has glasses is going to get smart glasses. Maybe everyone else will kind of come along over time or have different form factors, but kind of regardless, that journey might take like 10 or 15 years. I think that that's pretty normal for building something that's that fundamental in the world, but the experience when you just try on the AR glasses for the first time, and we have like our first working prototype of this now, and it's like it's awesome. Actually, we were working on this, and I wanted it to be our first consumer product, but I actually want the first consumer version of it to be even better. So we're just like, all right, let's just take the first one, and we'll kind of have it be a prototype, but when people try this on, they're just like, it's just giddy. It's inspiring. It's just a thing that you haven't experienced before, just being able to kind of like play with holograms and things in the world as if they're just like physically there in like a normal form factor pair of glasses. It's just this wild thing. So for me, I mean, that's a lot of that. I mean, I think that there's that. There's the AI version of that. I assume we'll spend some more time talking about that, but I don't know. It's more just like doing awesome things. I kind of have this saying that I like to tell people at Meta that there's a difference between doing good things and doing awesome things, and it's like a difference in inspiration, and good is good, too. I mean, you can be awesome and not good. They're kind of two different almost orthogonal directions, but I don't know. I think this is a phase to try to be awesome and try to build some awesome things. I mean, it's interesting. I think that one of the catch lines that we use for SPC is that you come here to throw away the good ideas in favor of the great ones. So I think I'm going to update that. So you throw away the good ideas in favor of the awesome ones. No, I mean, really, because I think that life is too short. I mean, the biggest scarcity that we have is just the amount of time that we have. So I think we can all have good ideas, but it's the awesome ones that we're going for. Speaking of 10-year arcs, I think one of the things that we tend to forget in the goodness of these LLM releases, and particularly with Lama, is that you started FAIR almost 10 years ago. In fact, one of the last projects I remember working on with you at Facebook personally was Coefficient back in 2010, trying to put a weight on every edge on our graph using machine learning techniques. Yeah, good one. Yeah, I think it worked out well. So it ended up being somewhat important. I'm grateful for your work. Thank you. But when you started investing in FAIR, can you take us through the early days of maybe what minus—I mean, back in 2014, we were still kind of in the AI winter. I think that—so what were some of the seeds that you were trying to plant, maybe back in the minus one to zero days of FAIR? Yeah, so after—I guess like around 2012. So that was when we went public and Facebook reached a billion people, and I was kind of trying to figure out what's next, right? Because obviously Facebook goes on to scale to three-plus-billion people, and it's still growing, which is sort of mind-boggling to—I think people kind of assume it's just tapped out, but it's not. Still doing awesome. But at that point, I was kind of like, okay, well, what's the next step? And part of it was, okay, I built one social app, so that's good, but there's kind of this vibe of, okay, you can do it once. Maybe there's some luck in that, but can we just go build four more billion people apps or take things that started to be pretty small and build them to be really big? So that was one part, and then it was like, okay, well, what are the next platforms? And this is something I thought about a bunch because social media and the apps that we're building kind of got started around the same time as smartphones, right? So we didn't play any role in defining smartphones, and we tried later, right? It's like, there's Android. We took this cut at this phone thing. The quick lesson from that is like, all right, this platform's pretty stable and solidified, and we're not going to really be able to influence that much. So even though it might take 15 or 20 years for the next major platform to come about and be at the same scale as phones are today, that's what I got to do, right? I'm not going to hit my head against the wall just trying to create another phone variant when that thing is similar. And also, there are a couple of issues with sort of building on other platforms, and one is just the lack of control over your own destiny, which I think is just really frustrating. I think when you're starting, you're navigating all this stuff, and the wind is blowing in all these directions, and there's a million things that can kill you, so it's not that big of a deal. But then when you get to be a bigger company, you're trying to I don't know, you want to make longer-term bets, which means that you want more kind of stability of the assumptions of your environment. So having more control over the infrastructure that you have, and knowing that you build something, someone's not just going to tell you you can't ship it, or they're not just going to change the ground underneath you or whatever, it's pretty important. Another thing is, I mean, this may just be kind of more specific for me, but when you're building social apps, it's kind of this weird thing that you primarily are delivering them through this tiny little screen that people carry around. It's like a really in some ways antisocial form factor, and it's part of the reason why I care so much about glasses. It's just like such a more natural thing. So 2012 was right around when we kind of started looking at what's next, and we started FAIR, and we bought Oculus, because at the time I was kind of, this is actually something I think back about, because I'm like, all right, could we have started working on VR and AR without Oculus? I'm not sure. I mean, I think like in theory, yes. I mean, we put a ton of work into it, but at the time, I mean, you remember when you were there? It was kind of like the ethos of the company at the time was not actually oriented around, let's just go have some groups sitting in a corner for five years building something separate. It was almost like we needed the separate seed of a thing to kind of go incubate it, to even get our culture to accept that it was going to be a thing, even though I'd kind of decided I'm like, no, we're doing this, right? But so yeah, so we kind of started the AI thing with FAIR. I did want to buy DeepMind, but they went to Google. Demis was good, by the way. He totally did a very good job of playing me off of Google to get a good price, which I respect and power to him. But yeah, so we kind of did it ourselves, and I guess because we didn't end up having this closed lab, we ended up getting all these academic folks, and that sort of laid the foundation for like the kind of philosophical approach being like overwhelmingly open source in the work. And we'd done that with a bunch of other stuff at MetaFacebook at the time, but the AI stuff, I think, especially was like really like it was kind of academic, open, Yann LeCun, right? Like a bunch of the kind of foundational people he brought in, but I don't know. At the time, there wasn't like a product, right? I mean, one of the things I struggled with for until really the latest wave of the AI assistants came out is, you know, we really looked at this mostly, you know, it's like we did fundamental research. It got its way into all of our ranking and recommendations and ads things and delivered really big results for the company, but it wasn't like a single product, which sort of lacks a sense of awesomeness in kind of the sense that we were talking about before, because in the way that like the AlphaGo stuff or the AlphaFold stuff that DeepMind did, I just think it's like really inspiring, right? It's like you look at that, and you're like, okay, this is like it's just a really kind of singular big contribution to the field. So it kind of hit my head against the wall on that on AI for a while, but I think more recently with the advent of being able to build different kinds of assistants, I just think that's like the product, right? So we're going to have meta AI. It's on track to being the most used assistant in the world. I feel good about that. We're going to make it so every creator and every small business and every person can create their own virtual assistants or virtual companions or the different things that they want. You'll have NPCs in the metaverse playing games, like all this kind of stuff. I think that that's going to be awesome, but it did take a while to figure that out. But I think it was back to 2012, 2014 was sort of around the time where we're like, all right, these are the two next things that we need to go do, AI and the metaverse. I mean, it's incredible to me just to kind of see how long you have to plant some of these seeds for. Yeah, no, it just takes so long. Honestly, it's way longer than I would have thought. So I think that's one of the lessons with entrepreneurship is it's almost like you have to be really excited about what you're doing because otherwise it's really too painful to actually do it. It's not rational. So one of the big open questions with AI has been kind of one of the, I would say, talking points that a lot of people say is that most of the value will accrue to either the bigger kind of like players, either through either the bigger labs or kind of like some of the bigger kind of, whether it be a Google or a Microsoft, and that there's very little for the startups to essentially compete for simply because they don't have scale. And you're obviously both a big company, but you're also kind of providing this incredible resources to all of us to kind of like push in some ways from the other side. So how do you see, like, what advice would you have to this room full of founders as they try to navigate, like this particular question of like scale versus not scale? Like how do you compete against the bigger players and so on? Yeah, I mean, I think that there's going to be almost every product category I think is going to get disrupted. There's going to be stuff that So some things are fundamentally very large investment-oriented things, and those are not well-suited to startups. So for example, building large foundation models that are going to train, requiring tens of billions of dollars to train, it's like, all right, there'll probably be a relatively smaller number of companies that do that. But I think the good news is that that hopefully will be somewhat of a commodity over time, right? And I'm not sure that that's where most of the value is going to be. Whereas what you can build on top of that, I just think like all these different categories of things are going to get built. I mean, I look at like, every part of what we do is going to get changed in some way, right? Feeds are going to go from, you know, it was already friend content, and now it's like largely creators. In the future, a lot of it is going to be AI generated. Ads are going from like advertisers targeting stuff to now it's like an advertiser is just going to give us a business objective that they want, and we're going to like produce the creative for them and find the people. It's just like, it's just kind of wild, right? I mean, all the metaverse stuff, it's like you go from having all these developers building out these worlds to it just being more generated, almost like a dream, like a lucid dream as you're kind of like walking through it. It's going to be wild. So there's going to be all this stuff. You can point to almost any product experience that's out there, and I think that they'll change. And some of that, there may be some larger company that has some distribution advantage. But like, I don't know, large companies are slow, and they lack conviction, and I think it's one of the advantages, I think, of being a kind of founder driven large company. But one of the kind of things that I've thought about from time to time is, like, why was it the case that we were able to build Facebook and that some other company didn't, right? It wasn't like it was a super novel idea, right? I mean, there was Friendster before, there was MySpace, there was all this stuff. Google, Microsoft, Yahoo, they all like had versions of it. Why didn't they do it? And I think it's like, I mean, we kind of watched them, in the early days, it's like fumbling around. And it's not that they had a lack of talent, right? I mean, we were like a ragtag group of children, right? And they had all these serious engineers and serious infrastructure. And so I kind of think the reason is because people doubt new ideas before they come to fruition, right? So the narrative with social networking is like, ah, just like this college kid thing. It's like, OK, fine, maybe not college kids, but it's probably a fad. Oh, OK, maybe it seems like it's going to be around for a while, but it's probably not going to make money. Oh, it's making money, but the switch to mobile is going to be pretty hard. And then it's like, OK, by the time we figured that out, it was too late for anyone to go. The companies had lost their advantage. And so what was the issue? I don't know. There's probably some team buried deep inside those companies that believed in it, and probably some VP person who was like, eh, that's probably not the biggest priority. And pour some sand in the gears or whatever analogy you want. And so I do think that even for the things that look like they belong to large companies as opportunities because they have a big distribution advantage, I would guess that big companies are going to fumble two-thirds of those, right? And then there are all these things where there is not an apparent big advantage because it plugs into an existing distribution channel. And those are just kind of free and kind of open to take. But I don't know. I think that this is going to be huge. And the llama thing that we're doing, I mean, I believe a lot in open source. I think it's good for the world more broadly. We're not doing this because we're altruistic, right? I mean, we're doing it because we want to build a platform that we know that we can rely on having llama as a thing. And the reality is this is an ecosystem, and it's not a singular piece of software that we could just build and deploy ourselves, right? It gets better when you have all the silicon providers optimizing all of their stacks for the thing that we're doing. And when you have all these other companies or startups or different folks who are building different distillation tools or inference tools to make it go faster, more efficient, and all that stuff. And all these people are building stuff on it. And we released the segment anything model. It's different from llama. But even in like the first week or two since it's out, you just have like all these people running all these videos through, and the model gets better, right? So I think this is going to be good for everyone. I think open source is kind of this question of like, I don't know, when you think about the future of computing platforms, I think there's this huge amount of recency bias where people assume that because like iPhone won and, you know, it's like that the closed model is going to win. That's just sort of like where the world is. And I don't know. I think that future is not written. It's like if you look back at PCs, like Windows was the leading platform. And there's kind of always an open approach. There's always a closed approach. Each has its own pros and cons. But I think a lot of the stuff just depends on like who goes and does it. And I think that's true both for the startup opportunities versus whatever kind of lazy big company there is. And I think that that's true for whether the open platform versus the closed one ends up being the kind of leading one in the next spaces, whether that's AI or Metaverse. I'm going to jump ahead to the next section and just one anecdote and kind of story I want to share that's actually been somewhat kind of, I would say, a defining moment of defining 30 seconds of my life, which is that back in 2005, pretty close to when I first started at Facebook, maybe like this is week two or week three, trying to figure out what to do. Nobody was really giving me much direction. And I remember Mark walked up to me. Not like a very clear management structure. Yeah, exactly. Chaos. Controlled chaos. I remember Mark walking up to me and he's like, Hey, dude, I think you should write a search engine for Facebook. Yeah, just you. Yeah. And it's like you and what team? It's really important because when people come onto Facebook, the first thing they want to do is search for people. So make sure you don't **** it up. And then my first reaction is like, Mark, I don't know how to write a search engine. I've never done this. Why don't we go hire somebody from Google or Yahoo to do this? And he looked at me. He's like, Dude, if I can build Facebook, you can build a damn search engine. And you did. And I did. And I say it was the like 30 seconds that I still think about, which is that I just think that we don't like all of us can build whatever we put our minds to. And this is kind of the culture that for me still defines Facebook, right? Which is that there is this hacker can do total ownership culture. But if I could actually tell every startup in the world to like take one thing, which is that all of you can do whatever you want, as long as you put your mind to it, it's probably easier than you think. There's a question in here. I think it's sometimes it takes longer than it takes longer. But like, but it's but you can write. And I think like a lot of the stuff that seems obvious, it's just not there's just not someone else who's actually like gonna go do it, right? So I think that that's one of the things that's the weirdest about about this is it's like, kind of comforting to think, hey, there are all these things that that kind of people should be on and someone else has got this, right? And it's like, No, no one else. I think we're the grownups now. I want to go back to Lama and open sourcing Lama for a second. I'm kind of curious, like, what kinds of real world examples do you want to see built on Lama? I don't know. I mean, I actually, I think that there's a lot of awesome stuff that will get built. But I actually don't have a specific vision for that. From my perspective, I just want everyone to be using it. Because I think that the more people who are using it, the more the flywheel is going to spin for making Lama better. I mean, this is maybe like a very selfish and parochial answer. But then that makes it so I can build the things I want to build better. But honestly, I think people should take comfort in that answer. Because I think one of the big questions that people have about our open source strategy is like, why are you doing this? You kind of train this model and then you just give it away? Is that sustainable? I mean, when Lama four, Lama five takes many billions of dollars to train, are you just going to give it away? And it's like, yeah, yeah. I don't view it as giving it away. I view it as like, you guys all making it better for me. So I actually don't care what you all do, as long as it's responsible. And I think the safety is important. There are all these questions around AI ethics and safety and security. And I think that that stuff ends up being really important. We take a huge amount of time. This is probably one of the biggest things that's part of training that makes open source difficult. Because if we were building it for just ourselves, there are certain assumptions that we can make, like that we'd have multiple layers of the stack in addition to just the model itself that could filter out bad stuff. So bad queries that come in, outputs that it generates that are kind of unintentionally bad. And we've tried to replicate some of that with the open source stuff. And we built this whole LamaGuard system. And it's meant to be deployed as an overall system. But fundamentally, we have to put way more energy into training the actual underlying model because it can be used as just a model with all those other layers. So maybe this is sort of a negative answer to your question. But I think part of what's good about having open platforms is people go build stuff that you didn't even imagine. So I don't know. I don't know what to ask people to build. You all figure that out. But I think the thing that I kind of worry about with it is I think some of the both the opportunities and the I do think that it's not a lot of the open source safety debate has been framed around like, okay, because of what I just said, open source must be fundamentally sort of less safe. I actually think once you take that, if you do a good job of it, it gets more safe for the reason that open source software has been more secure over time. It's just it's open. It's, you know, it's like, people can scrutinize it more. It's counterintuitive, right? At first with open source, people like, hey, well, like hackers can see all the holes, isn't it going to make it less safe? It's like, no, you just fix the holes faster than everyone patches it. And the same thing is going to happen with llama, I think, is that, yeah, it's like people point out issues. We're going to keep on doing dot releases. We'll keep on doing big releases. I think developers will keep on wanting to roll out the most recent models, and they'll apply their fine-tuning patches to that, to the most recent models, and it'll be awesome. So, yeah, I don't know. But I think there's like the whole stack, everything from kind of the silicon and the data centers, and we did the whole open compute thing before, sort of similar, where it was like, all right, we're designing our own data centers and our own servers, and let's just see if we can give this away, then maybe the whole industry will sort of standardize around our designs, and then the supply chains will get developed, and the prices will go down for everyone. And we have the advantage of we were kind of like, we're after Google, right? So Google built all that stuff first, so it wasn't like some proprietary thing for us. We're just like, all right, whatever. As long as it exists, and it gets cheap, that's awesome for us, and I think we're sort of in a similar zone on AI stuff. I mean, I'm curious, I mean, given the front row seat that you have, what's your, what is this, what is your best take on the scaling frontier for these models? I mean, they continue to follow the scaling laws. How much more runway do you think we have? I mean, yeah, Llama 4, I think, or, you know, whatever, CP5, all of that makes sense, but like, how much, what's your current take on, like, the runway? And what are the limiting factors? Yeah, it's hard to know. I think that there's, the compute thing, I mean, I, so without knowing exactly how far this scales, I'm clearly betting that it does scale, right? Because we're like, doing all this infrastructure, right? So it's, and so, so it's hard to guess the exact timing, and this is like, you know, five to seven year build outs of like, all this stuff, or to get to the massive scales that you want for like, Llama 10 down the road, but it's, you kind of have to work on that in advance, because you can train the model in six or nine months, but like, to get the energy that you need. So I think it will, I think it will work, but I think a lot of it is going to come down to data, and a lot of it is going to be in different domains, right? So, so that there isn't data on the web before, right? So right now, it's like, you kind of think about, this is like, companies go out, they look at all the content on the web, you kind of try to find all the patterns in that, that's your pre-trained model, then you, you kind of fine-tune it or post-train it in order to kind of build the different applications or things that you want. As you start getting towards more like, agentic type behavior, and it's just that data set doesn't exist yet, right? So now you have to like, go make that data set. And I think a lot of the new work going forward won't be just about kind of like, taking data that's out there, and then building compute to kind of crunch and train on it. I actually think more of the training going forward will look like inference today, to either be kind of trying to go experiment and in kind of some spaces, right? It's like, how do you create an agentic data set? It's like, okay, well, it's probably some amount of manual work, and some amount of just like, letting the system play, and like, and just kind of experiment and generating data from its own kind of game playing. And that's just one example, but it's like a type of thing that just doesn't exist today. So I think that there's a question of, how far does that go? I would guess pretty far. And yeah, I mean, I'm definitely an optimist on this, but I mean, tough to know. Ruchi, do you want to rewind us back to some of the early days of Facebook? I mean, just related to this, I, you know, Aditya has been talking about all his early experiences at Facebook. I remember, you know, we were working on News Feed, and I was part of so many war rooms. You were part of like, the pod of six people who built the first News Feed. And I was sitting next to you all the time. It's not always a good thing. But you know, there were so many war rooms and lockdowns in the early days of Facebook, particularly after we launched News Feed. I don't think I left the office for a whole week. Yeah, well, that was because there were like protesters in the streets. And the security was like, yeah, you probably shouldn't go out the front door. Because people are really, really angry about this. Right. So I'm kind of curious, I want to live through one of your more recent war room experiences, maybe like around Lama, or when the weights around Lama got leaked, or you know, like, what is a war room at Facebook look like today? So I think that there's two different pieces to what you're, so I think, I'm not really in those like operational war rooms as much now. So as much as it was News Feed and things like that in the early days, I think how we process these like big strategic things that come up. I don't know, I feel like one of the things that I'm trying to do is learn how to run the company a little more smoothly over time, I clearly have not succeeded at this yet. But it's like, like, if you look at like the volatility in the company, it's like, but, but I'm just like, alright, like, at some point, like in this life or the next, I will, I will, like, hopefully figure out how to do this a little bit more smoothly without compromising, obviously, trying to do really cool things, right. And I think that the obvious way to make things more stable is to stop pushing the, the put your foot down on the gas as much, but I am not able to do that as a person. So but I do think that part of it is like, I think we did used to run around with our hair on fire a lot. And I think some of that was, I don't know, you like makeup for lack of experience with more effort or something. But but I think some of this is like, you try to, like, as time goes on, be a little more stable about how you do stuff. And so it's like, all right, when when the first version of llama, it's like, we intended it to be an academic release, and it, it like got out. And I thought it was pretty cool. And we were, we're trying to figure out exactly like how to process this. And it wasn't like a pair on fire situation. I think it was bad. It was awesome. Like people like really wanted it. And we just need to figure out what were the right ways to handle this. A lot of the different crises that we have now, I mean, this is maybe an unfortunate artifact is it may be both being a bigger company, but also just where the world is a little bit more, I'd say more of the challenges we have are kind of like more social and political rather than technical, right? So the, you know, it's like the newsfeed that what were the big challenges? I mean, yeah, it was like this big change to the site. But also, it's like you launch newsfeed, and then immediately traffic on the site went up by like 50% overnight, because it was just like, okay, now people can find all this interesting stuff that's being surfaced to them. And it's like, now we have to like, how are we going to deal with that? Right? It's like, so you like corralled all the people and like, all right, let's, let's kind of go solve that. And that's great. People want certain controls, and we need to figure out like which controls to give them. But now I think a lot of it is okay, so people want this open source llama thing that we kind of accidentally half did, right? It's like it was intended to be open source, but sort of like an academic thing to start. And then it immediately, you know, so we had two debates internally. One is like, is this good? And should we keep doing it? And that got resolved pretty quickly. It was like, yeah, we think this is good. We should try to do this. Then the second one, which I think it's because sometimes the some of the political or social debates, they move so quickly, I think the open source debate, I think is going to be one of the most important technology policy debates of the next five to 10 years. And we have made a lot of really good progress already. And I think part of the reason why, why we're making progress is because so many people are embracing and using llama. And, and I mean, frankly, I think like, like startups and entrepreneurship are generally popular, and like way more popular than big tech companies, right? So, so I think like, like, it's like as a thing that we're doing, like, I don't know, is that gonna be popular by itself? I don't know. Maybe, maybe yes, maybe no. Like when, when you get like this groundswell of people who are using it, that's like, hey, yeah, I think like people are gonna, like people listen to that a little more. And they're like, hey, that's good. We want there to be new companies. But in the early days before we actually did llama too, which was the kind of like first commercial open source thing was this huge debate, which is, you know, it's like the, I'd say the, the kind of like concern risk side of it was really, I think, dominating the public discourse until there were actually a lot of like, startups that were using this stuff. And that was really what we had to sort through internally was how do we do this in a way that is really responsible and safe, given the somewhat unique nature of open source relative to closed. And, and kind of my view is, well, if we do a good job, and if we and if we kind of like over index on really making sure we do a good job on safety, that over time, you'll build trust, and we'll build this community. And like, they're not going to want to take away open source from, like, it's I don't know, if you think about the quarter, the big companies today, it's like they were all built on open source software, right? It's like, if we if we if we if this debate goes well, I think the next generation of major companies are going to be built on open source AI. And I just think that there's some of these things where it's like, yeah, there will be challenges along the way that we need to deal with. But I don't know, I think that this is going to be one of the most important debates and making sure we navigate it carefully. And that's why we've had some more restrictions on what we do. And, and sometimes it's like a little bit slower. But, but I don't know, it's less like, okay, we got to get in a war room and figure this out overnight. It was I mean, for from llama one leak to llama two, it was, you know, probably how long was it? It was it was several months, right? Like a bunch of months. Yeah, six months, eight months, something like that to kind of sort through all that stuff. And we were the team was cranking on the model. But in the background, we were like, I think figuring out all the responsibility stuff to make sure that we really nailed that and did a good job on that and laid the groundwork. Awesome. Okay, let's switch tacks a little bit to address some of the questions that the audience has submitted. What is the one piece of advice you would give technologists who want to be entrepreneurs? You know, working with a lot of entrepreneurs here, I find that they often hit local maximas. What kind of advice would you give them? Well, I think the whole minus one to zero idea that you have really fits my philosophy of how you want to go explore the space. I mean, so when I think about when I was when I was getting started, right, and there's a lot of people who decide I want to start a company and then they're like, get really committed and do it. But the problem is that once you get a bunch of people like doing a thing, it's really hard to change what you're doing. And it's like, yeah, you can do it through willpower, but it takes time. And early on, you're just exploring like a very dynamic and broad space. And you need to be able to, I don't know, just have like meetings and make decisions inside your own head and then just be able to like go do things differently, like come out and like change your mind. And it's like, all right, I'm going to like do a different thing now than I was doing a few hours ago. And so I don't know. So there's this the like when I was in college before I built the first version of Facebook, which I didn't think was going to be a big company, by the way. I mean, it's do you remember when like when Dustin and I first came out, we basically I started Facebook. He joined me because I was still taking all these classes and I was walked away in the computer science dungeon and like PHP do my problem sets and didn't have cell phone receptions when like the site went down. And like I was just I was just like, oh, I just like get out from doing a problem sets. The site's been down for a while. That sucks. So Dustin joined and he was awesome. And like basically just kind of like the first engineer and kind of ops person helping to keep it running. And like the company wouldn't have worked without him. Dustin's like an amazing person. And but yeah, Dustin. So after that first year, first spring, because I launched in February, we came out to Silicon Valley and we're like, hey, this will be a great place to spend a summer. This is like where all the companies come from. We'll learn something. It's like surely the thing that we're doing now is not going to be the company that we end up working. We'd already started Facebook. It had like almost a million people using it. And we were like, nah, this is like not going to be that big of a company. So I think, OK, so what's what's the lesson from this? So like when I was in college, I actually built like a bunch of different things. I just love building stuff. Some of them were I like wasn't sure what classes to take. So I built a service that like went scraped the course catalog and let everyone input their like what classes that they had taken and what they were planning on taking. And like it showed all these correlations of like people who took this one were kind of interested in this and here's how they ranked it. And like and and part of this was it was just interesting because it's like people like I kind of started that because I wanted to do this like crowdsourced question of like to answer what like what class I should take. But it's actually just people just like sat and clicked through all the classes that were taken because it's actually just like people really want to learn about other people. Right. So it's like what like what classes are you interested in taking? That's like interesting signal on on you and like an interesting way to get to know other people. So all that stuff, I think, pointed towards OK, I put a lot of those things together into the first version of Facebook. But there were probably like 10 different things like that that I built. Yeah, that was in high school with D'Angelo. And yeah, I mean, there were like all these. It's like I probably the peak was at some point I was working on some project and then I had finals coming up. And there's this class I was taking the Rome of Augustus. It was like a history class at Harvard. You needed to take like some lit arts or I don't know what they call it now. But so you needed to learn for the final. What were the what is the like you just basically show some piece of art and it's like what's the historical significance? So it's like, all right, that's like I have like I'm spent my whole study period building some other project. So let's like quickly in a few hours, just like threw together a thing like scraped all the pieces of art, sent out a link to the class. I was like, hey, I built a study tool. If you guys want to like look at this, it's like it just shows a random piece of art and you can say what you think is significant about it and then you can see what everyone else wrote. And the scores had never been higher on the exam. But honestly, it's like so I really believe in this approach of just like you want to build different stuff and not get committed too quickly. I think especially when you're a startup, right? Because it's I think when you're bigger and you can have it's actually a lot easier to predict where things are going to be in the world like five or 10 years from now than it is to know like what is your to get there. I think that's like the really hard thing to figure out. So I don't know. I mean, there's different styles on this. Some people have succeeded by just saying I want to go build a startup. I'm going to like go hit my head against the wall until I get this to work. But I've always much more believed in like just go build a lot of stuff that is like thematically interesting and like you'll learn different things from it and try to not get too committed too quickly because it's actually pretty hard to learn and fully kind of pivot through the space of things that you need to go do. So I think and from that perspective, my understanding is that like really aligns with what you're what you're all about. And I think it's just really good. Thank you. But I mean, a huge compliment to all of us here, all of us here. But it's like, I mean, and you know, just like you're good people, you know, it's like, keep going. You got a nice chain. He's like, you're a great surfer. Okay. But you know, in that vein, when people ask me like the different things I've worked on at Facebook, I can tell them that a few things that I launched that are successful, but I also worked on a bunch of things that kind of just didn't make it right. And I think even internally, we kind of had this attitude, which is that you should be pushing really hard towards kind of like what you think might be a great product, but also not hold on to it so tightly that you end up almost getting your ego attached to it. Because, you know, like the whole point of building and trying is that a bunch of stuff won't work. And that's okay. And that's something I still admire about Facebook today, which is that it feels as though that you're constantly pushing, but with the knowledge that not everything will work. And that's okay. That's a feature. Yeah, no, I mean, I think I've made more public mistakes than like anyone else in the industry. And there's there's some like Michael Jordan ad for this that I always like find very inspiring. Like, it really speaks to me. It's like, you know, it's like this ad about how he's like, like, I've failed over and over again. And then at the end, he's like, that's why I succeed. But but I think it's, I don't know, the Einstein biography that Walter Isaacson, I think sort of makes the same point. It's like, like Einstein got all these theories wrong, right? And but it's like, okay, you got some good ones, right? Right. And it's like, that's kind of what matters. Yeah, I think you talked about this. But maybe I'll ask another question, which is that, you know, I've worked for many startups, and I've helped many startups as well. But, you know, when I think of that early team at Facebook, it gives me the chills. Yes, we were inexperienced, but it totally gives me the chills. And I kind of I'm curious about what you get right with the hiring? What are like, you know, some important hires you made? And why were they the best people? Well, I think it was sort of this tale of two cities. I think that there were a lot of really excited and inspired people who were coming out of college because they use the product. And they were just like, this thing is awesome. Like, I want to work at at this company. And then the other side, which I think I just failed at for a long time, was building an executive team. And we went, I mean, we like flamed through so many executives. And it's like, because I had this like platonic model in my head of like, here's what a head of engineering should look like. And I think that they were they were probably all pretty reasonable people. But it was just like this big culture mismatch until finally, Peter Thiel took me out to lunch one day, and he's like, all right, like, this clearly is not working well, right? It's like, you like, you like, you like, aren't jiving with these executives. And like, you should basically just take people who you want to spend time with. Because like, you're, you pretty much like live at the office, or you're, you know, you're, you're like working on this with your whole life. And I think there's just sort of this kind of bandwidth thing with the early team, where it's like, you want to be like speaking the same language and have a lot of the similar kind of assumptions. And obviously want to challenge each other to write, I mean, the folks who were there had different views on things, but I think also had enough of kind of like an overlapping foundation of values. And I think that was sort of when it started to click a little more. We when to your point, it's like embracing the chaos of some of the stuff being. Yeah, it's like the people aren't quite as experienced, but we're like, a little more on the same page. And I think one of the things that's been awesome about the company and where we are now, is all these people have now grown up. And now, they're all like, really seasoned, amazing executives, who by any definition across the industry are extremely experienced. And we've all worked together for like 15 years. And it's like, like, so people are like, Okay, well, how do you build that? It's like you can't, right? It's I mean, it's like, you basically Yeah, you like, I mean, what I try to do is, I basically, I have a relatively large management team, right? Because we just have a lot of different products. And I, I basically, I like I want to be pretty hands on. And to me, part of that is, I don't want to like work through just like six or seven people. I have a group of like 25 people, right? Who like run as maybe like 15 product groups, you know, everything from the major apps to ads to AI, AI research to smart glasses to MR headsets, where all the different things, then you have all the business functions. And it's all there. And we just spend like hours together every week. And it's just like part of my style of running the company is like, have it be kind of this bonded group. And I don't know, I like one of the things I'm really proud of is of the product folks. I think on the business side, it's a little bit different. Like you want to hire a general counsel who has a lot of experience as a general counsel, right? It's like, that's like, I think I want to be really, really important thing. Yeah. But although Chris Cox, who is the chief product officer and is like, yeah, yeah. And I mean, Dustin, I had this huge fight about this because I Chris was like, he's clearly a very promising young man. And it was like, all right, like, Dustin was like, we need like, I really need him to be an engineering manager. I'm like, Dustin, you already have five engineering managers. It's like, I really need a head of HR. And like, like, I don't know, it's tough. It was tough. It was tough to hire. He was a good HR head. He was great. He was great. And he and he was very generous to do it for a few years before going back to product. But But yeah, no, it's um, but today, even I mean, one of the things that I'm most proud of is of all the kind of like top product group leads. None of them started at the company as a product group. They all started kind of like one or two clicks down. And we hire a bunch of people in as directors. I mean, some of the people who are at the company, I mean, a lot of them start off as individuals, and one of them start off as an admin. Actually, two now, who are on the management team, and just kind of like work their way up doing different things. You know, people started non technical roles, technical roles, technical roles, go to other stuff. So but I don't know, I think that that's, it's been really cool. Yeah. And it's just like a really good group. And it's, I mean, you can like power through a lot of stuff yourself. But I really think that this I mean, it's, by the time you get a little further along with the idea, it really becomes more of a team sport. And the amount of fun that you will have, and how painful the lows are really just depends on the people you have around you. And it is hard, right? And the stuff always takes longer than you expect it to longer, you like, you have all these issues. So I don't know, I think, to me, that's probably one of the reasons why, like, I've remained excited and happy doing the stuff that I do is just like, I love the people I work with. They're like my closest friends. I just following up on that, how have you managed the emotional ups and downs of starting a company running a company? And how has that evolved over the last two decades? Yeah, I don't know. It's tricky. I think some of the time when I think as the company has gotten bigger, the cycles are longer, right? So it used to be like, you're saying it's like, all right, we're launching newsfeed. It's like, we built the whole first version of newsfeed in a pretty short period of time. And then it's like, there's a spike. And then it's like, ah, like, some people are upset. And it's like, okay, a week later, it's like sort of resolved. And it's okay. But like, I think at the scale of what we're doing now, I think that's one of the things that's fun, but it's also quite frenetic about building a startup. And it's just that you do kind of always feel like you're, you go, you bounce between frequently. I don't miss that. It's tiring. It's tiring. I don't think I could do that again. But so now the cycles are longer. And I don't know, I think it's like, it's some amount of willpower, some amount of like the people around you keep you going. And some amount of maybe you don't fully realize how bad certain things are until you're at them. And I think that that's like, I mean, I think some of the stuff that we've dealt with, I don't know. I mean, it's just, I think some of it, you just like try to put one foot in front of the other, and you keep doing your work. And then you look back afterwards, and you're like, wow, that was really something. I feel like just hearing you say that, I mean, even for Mark, the activation energy to start again would be really, really hard. So I think the only thing you can do is put one step ahead of the other and just keep going and not give up. I don't think you guys have a choice. Don't give up. Certainly not with Ruchi here. So maybe switching gears to some fun stuff. So during the pandemic, I personally picked up a bunch of new sports, learned how to ski, call myself an intermediate professional at surfing, learned how to wake surf, I can foil a little bit. And it's been amazing, right? Like all these sports just kind of give me an outlet to just be out in nature and learn a bunch of new things. You obviously picked up a bunch of new things yourself, MMA, foiling, a bunch of other things. What do you love about kind of doing these sports? Has there been some amount of transfer learning between these sports and how you apply it to work? So for me, the reason why I love MMA is it is, like I was just talking about how the cycles are very long when you're running a bigger company, especially when you're doing hardware, right? Or it's like you're training like some foundation model. It's like, all right, we'll go work on this for nine months. So there is something that's nice in a bounded way about being like, all right, I need to focus on this or else I'm just literally going to get punched in the face. And the payoff is pretty big. It's like, I get to punch someone else in the face. It's like, they'll never let me do that in the office anymore. You know, all this ends with Alex Pereira kind of like in your- What's up? Oh man, Alex is awesome. But yeah, no, so I think similar journey to you. I mean, I used to, when I was younger, be pretty competitive athletically and then started the company and stopped doing most of that stuff. And basically, I mean, I'd stayed in shape, like I lifted some weights or something, but I never, I didn't have like a fitness goal for a while. And then at the beginning of the pandemic, when everything kind of went remote for a bit, it was like, all right, let's like get more into kind of running. And, you know, our family spent some time out in Hawaii and it was, got into surfing and I was like, all right, this is really good. Right. It's like, because it's like you're out in nature, it's beautiful. It's like, you're kind of physical, but you're, but it has a purpose. You're like, you don't want to get crushed by the wave, but it's fun when you're on it. It's like strategic. And then like, since like, all right, so that's done, like still spend some time out in Hawaii, but I don't live there. So it's like, what's the land version of this? And it's like, I need to learn a martial art. And then it's like, everything just escalates. Right. So it's like, I started doing this with a few friends and we're training and we're like, all right, we'll just train. And one of my friends is like, I'm going to go to a competition. And I was like, all right, that's cool. Um, I'm really curious to see how this turns out for you. I'm like, I'm not going to do a competition because that will be just absolutely humiliating. Right. And it's like, I go and just get choked out by someone or something. But then he went and he did pretty well. And I was like, no, it's like, so this is like, all right, I'm going to the next competition. So we went and like, like won a bunch of medals and stuff. And I was like, all right. So I negotiated with Priscilla that I can do one competitive MMA fight before. It's like, all right, look, I can do jujitsu for a long time, but, um, but I feel like there's some limit to the head drama that you want to take in this line of work. Um, so it's like, I was going to do one, one competitive MMA fight. And then I tore my knee and now I'm back. He's back. I mean, what's the date for that one? What's the date? Uh, I, I haven't decided yet. I need to get confident with my knee. Um, but I, uh, I, I don't know. I would guess in the spring. I think like some, some more time to ramp up. You heard it here first. I mean, I mean, I mean, not to say that. I mean, let me give you another, another anecdote about Mark's like competitiveness. Yeah. It's just, it's just a little competitive. I mean, maybe back in, I want to say this is 2009. It was this, you know, Mark had a theme for, you know, multiple years. This was a serious year. So Mark was showing up in a shirt and a tie. But then we decided that we were going to have a pushup competition to see who would be the first person in the office who could do 10,000 pushups self-reported. And Alex Himmel won and he now runs our AR glasses. He won, but Mark was a, he was a close second. I still remember being, not because of that, just because he's been around for a while and has been doing awesome work. And he was just like a dude, like a random engineer at the time. And then like, now he's like one of the top people at the company. Right. Okay. I've got the sign to wrap up. So I'm going to ask the one last concluding question. So the world often asks me what you're like. I try to avoid answering. It's like don't make eye contact and walk away. But, you know, sometimes I tell them, you know, he's like one of my closest friends that I would turn to in my darkest hours. And what I find fascinating when I think about you is the more famous you've gotten, the more time you've made for us. And I just want to ask you, what are some things that are important to you that most people don't know to value? I mean, I think, well, we talked about the kind of physical activity part. And we talked a little bit about the relationships part and trust. But I don't know. I mean, going back to the first topic that we talked about with building good things versus building awesome things, I'd say for this kind of chapter of my life, there's one more dimension to it besides doing awesome things, which is doing awesome things with people who I really enjoy. And sometimes it's like I want to go do a project because I like that person and want to go out of my way to find ways to work with them because they're a good person. I want to help them. Or I'm going to learn something from them. They're interesting. And there's a lot of different things. Right? I mean, it's a lot of it. I mean, obviously the bulk of my energy is still it goes towards meta. And one of the really fun things has been sort of broadening out my relationship with Priscilla. Right? It's like obviously we have a family now. But we also have been running CZI. And that's just been really cool to see her grow and basically lead this huge organization. And then there's some stuff that's just sort of random. Right? It's like I want to work on this project to raise the highest quality beef in the world. And it's like why? I don't know. But I think doing it with some people who I think are awesome. But, yeah, I think it's like those things that's just what matters to me at this point. And maybe it's partially because it's like the company is big and successful. And I guess like, yeah, we need to keep on doing well in order to kind of keep a general momentum and keep things going. But it's like it was never like getting to some valuation or something that motivated me. It was always what we were building. And I don't know. I guess for this phase, it's kind of being able to work on great things and being able to build these really deep relationships. That's part of just doing stuff for a long time. When you're younger, you don't have friendships for 20 years yet because you're younger. And I think we're at a point now in our life where it's like, oh, yeah, I have a bunch of those and that's great. But we're also young enough that we can build them. They're going to take 20 years. But I think like you can do that. And that's just what's important to me. And you guys have said a lot of nice stuff about me. You guys are some of my closest friends, too. I think what you're doing is great. And the work that you did at Facebook early on, I'm eternally grateful for because we wouldn't be where we are without it. You guys are just really good people. And I think the people who get to work with you and get to be coached by you are really just lucky and fortunate. And that's part of the reason why I wanted to come and do this is just because I have a lot of faith in what you're doing. And I think philosophically it's the right kind of push in the world compared to other different kind of incubators that have different philosophies or different things like this. And I don't know. That's what matters to me at this phase of my life. Thank you.