The Future Mark Zuckerberg Is Trying To Build
I'd love to start with these. 10 years of work right there. Someone on your team called these the real life Tony Stark glasses. Very hard to make each one of these... That makes me feel incredibly optimistic... In a world where AI gets smarter and smarter... This is probably going to be the next major platform after phones... I miss hugging my mom. Yeah haptics
is hard... How does generative AI change how social media feels?... We haven't found the end yet... The average American has fewer friends now than they did 15 years ago. Why do you think that's happening? I mean there's a lot going on to to unpack there... I'm about to interview Meta CEO Mark Zuckerberg. There are not that many people with more power over what our future might look like. Nearly half the total human
population now uses Meta products and I just tested some of their new tech that feels like science fiction. This is crazy! Mark Zuckerberg and the team at Meta are imagining a future that billions of other people might actually end up living in. So my goal for this conversation is to try to figure out what that future really looks like. To paint a picture of the future Mark Zuckerberg is trying to build so that you can decide for yourself what you think of it. Welcome to the first episode of our new series, Huge Conversations Hey, good to meet you! Thanks for doing this. Yeah looking forward to it. Awesome. I'd love to tell you what my goal is of this conversation. Go for it. We have a called
huge if true which is this very optimistic about science and technology and the potential futures that we can build and in every episode we're sort of exploring what does it look like if you play a certain technological future out and so my goal in this conversation is to try to help people see the future that you're imagining when you're building the products that you and the Meta team are building. What are you imagining this looks like in future? How are you imagining people use this? All of that. Cool. All right awesome. So I'd love to start with these. Let's do it. 10 years of work right there! I got to demo them a little bit earlier today. I heard someone on your team
call these the real life Tony Stark glasses? We're getting there. But I'd love to just hear in your voice what are these? Well these are the first full holographic augmented reality glasses I think that exist in the world. We've made I think it's a a few thousand or something right. Very hard to make each one of these but this is the culmination of 10 years of research and and development that we've done to basically miniaturize all the computing that you need to have glasses not a headset but glasses that can put full holograms into the world with a wide field of view. So you can imagine sort of in the future we'd be having a version of this conversation where you know maybe I or you are not even here it's like one of us is physically here and the other one is here as a as kind of a full body hologram and it's not just a video call you can actually interact you can do things I mean in the the demo we had the you know ping pong and games and things like that but I mean you could you can interact you can work together you can you know play poker play chests whatever like the holographic cards holographic board game. I just think it's going to be wild.
it's going to remake I think so many different fields that we think about today from how we work and productivity to a lot of things around science a lot of things around education entertainment fun gaming. But this is just the beginning you know this is the first version, it's a prototype version that we've made in order to develop the next version which is hopefully going to be the consumer one that we sell to a lot of people. Why build these? Well I think it's going to be the next major computing platform. So if you look at like the grand arc of computing
over time you've you've gone from like main frames to computers that basically like live on you know your desk or on a tower to phones that you have in your hand that you basically like you know can take with you everywhere that you want but it's it's pretty unnatural right it takes you away from the world around you and. I think that the trend in computing is it gets more ubiquitous it gets more natural and it just gets more social right so you want to be able to interact with people in the world around you and I think that this is probably going to be the next major platform after phones. I'll give these to you. These are the clear ones that show all the... The whole thing is a special edition and this is like a really special edition. There's not a single millimeter of of space. You know everything in here from the micro projectors that um basically shoot light into the wave guides right it's a special type of display system. I mean these aren't normal displays like you have in a phone or a TV or computer like the type of displays that people have been building for decades. It's a waveguide system. The projector
that's shooting light basically goes into these nano etchings across the wave guide that are what catches and creates the holograms. In order to synchronize that with your where you're looking there's eye tracking and little cameras, they illuminate your eyes and then of course there's all the basic stuff that you need all the computing, the batteries to power the whole thing, microphones, the speakers because it needs to be able to play audio and speak with you and the cameras and sensors to see things around you in the world so that way when it's placing holograms in the world it can do that in the right place and understand where you are so that probably is still not covering everything because there's a lot of things that need to go into syncing up the holographic images between the two displays because you don't just have a single display like you have in a phone or TV you have two and it moves around and you know physical things are hard and need to be synced up. There's also the radio that has to communicate with your other computing devices to do heavier computing um and the wrist based neural interface that you probably got to try out. We kind of miniaturized all of this and fit it into uh you know normal looking pair of glasses which is... you know when I told the team that we were going to do this 10 years ago you
know people weren't sure if we were going to be able to but I think you not only we're going to be able to do this but I think we're going to be able to get it cheaper and higher quality and even even smaller and more stylish over time. So I think this is going to be a pretty wild future. There are so many versions of trying to get a similar idea of digital objects in physical space. I'm thinking of for example of glasses that have heads up displays where it's headlocked and it's moving with my eyes, glasses that are really creating digital objects in physical space that don't move as I move, I'm thinking of these, I'm also thinking of the Snapchat Spectacles that they just announced, then on the other hand there are headsets like the Quest and also like the Apple Vision Pro that seem to fall into a different category. I'm curious how you would organize this landscape for people and how you think about people using these tools in their real lives in the near future? Yeah so when we were getting started on this about 10 years ago I thought that something like this was going to be the ultimate product for everyone. Right you get to you know
normal looking pair of glasses and we'll continue improving that that can have full holographic images. I think it's super powerful and it is sort of the science fiction future that I think we all hope to get to. On the journey we took a few other approaches as well um to help us develop towards that including building glasses that don't have displays to try to learn. Just take a stylish pair of glasses today and put as much technology into it as you can but really focus on the form factor and that's the Ray Ban Meta glasses and it's doing really well and initially we thought that that was sort of intro product for us to learn how to build this but one of the things that's clear now is you're going to be able to make that product a lot more affordable than this probably permanently. So I actually think that there are going to be a bunch of different of these paths that we've taken are going to be kind of permanent product lines that people will choose. I think you'll see display-less glasses like the Ray Ban Metas continue to get better and
better, great for AI, no display but you can talk to it, it can talk back. I think there's going to be something in between these that's basically a heads up display, so it's not a 70° field of view, maybe it's a 20° or 30 degree field of view, so that's not going to be what you want for putting kind of a full hologram of a person or interacting with the world around you but it's going to be great for you know when you're talking to AI, not just having voice but also being able to see what it's saying or being able to text someone with your wrist-based neural interface and then have their text show up rather than having it read to you, which is, we read faster than we can listen or getting directions right or just being able to search for information get all that. So there's a lot of value for heads up display that will be somewhat more expensive than the display-less but somewhat cheaper than this. Then I think you're going to get this. It's going to be probably the most premium and and expensive of glasses products but hopefully still something that you know like a computer is generally accessible to most people in the world but I think that there are going to be all of those and I I think people will like them. I also think that the headsets that people are using around mixed reality will continue to be a thing too because no matter how good we get at miniaturizing the tech for this you're just going to be able to fit more compute into a full headset. Fundamentally our mission is not you know build
something that is advanced and only a few people can use, we want to take it you the last mile and do all the innovation to get it to everyone. We you know just shipped or announced Quest 3S, the new mixed reality headset where we basically are delivering high quality mixed reality for $299. I was really proud last year when we delivered Quest 3, the first kind of really high quality high resolution color mixed reality device for $500, right it was like, it's like a fraction of the cost of of what the competitors are doing and I think it's actually higher quality in a lot of ways, and now we've just doubled down on that. So I think that they're all actually going to end up being important long-term product lines: display-less, heads up display, full holographic AR, full headsets. I think that they're all going to be important. Yeah. If you play out the future
of not just the hardware that we've been talking about so Meta Ray Bans, Quest, Orion, but also the Llama models, if everything goes according to you and the teams wildest dreams, I'd love for you to just begin to describe what that feels like. I mean I think that there are two primary values that we're trying to bring. On the AR and kind of mixed reality side, the main value we're trying to bring is this feeling of presence .Right so there's something that I think is just really deep about being physically present with another person that you don't get from any other technology today and I think that's the thing when people have a very visceral reaction to experiencing virtual or mixed reality what they're really reacting to is that they actually for the first time with technology feel a sense of presence like they're in a place with the person and that's super powerful. I
focused on designing social apps and experiences for 20 years that's sort of like the Holy Grail of that is being able to build a technology platform that delivers this like deep sense of of social presence. The other big track is around personalized AI and for that and that's sort of where Llama and Meta AI and all those things are going. There's all this development that's going into making the models smarter and smarter over time but I think where this is going to get really compelling is when it's personalized for you and in order for it to be personalized for you it has to have context and understand what's going on in your life both kind of at a global level and like what's physically happening around you right now and in order to do that I think that glasses are going to be the ideal form factor because they're positioned on your face in a way where they can let them see what you see and hear what you hear which are the two most important senses that we use for for kind of taking information and context about the world. I think that this is all going to be kind of really deep and profound stuff but it's basically those two things: It's this feeling of presence and this capability of really personalized intelligence that can help you. I'd love to talk about each of those two things. The first on presence, I owe a lot to being able to connect with people online. Right this job that I have is by definition that, also with my family. My parents don't live anywhere close to me. I video call them a lot and when I think
about the progress of technology like this in a timeline from the telegram to the telephone to video call to some feeling of presence with another person who's feels like they're right there in front of me, that makes me feel incredibly optimistic. I would love a future where like I can lose in Scrabble to my mom and feel like she's really there in front of me. Yeah and it feels like we're not that far away from something - I agree! - that persuades my brain that that's happening. Yeah
totally. And also I miss hugging my mom right like that never goes away. Yeah haptics is hard. Yeah and so my question is about that it's about this this feeling of like it's hard for me to imagine um a future where real physical presence is not different and special in some way where I don't miss literally hugging my mom and I'm curious how you think about the parts of human connection that are eye contact and physical touch and things that our ape brains value for connection with other people. Yeah well eye contact I think we're going to get to a lot before the the touch part. For haptics I do think we'll make progress on that but it's it's obviously there's a spectrum there too from kind of hands which is where if you you draw out the kind of like homunculus version of a person in terms of like what are what are our kind of sensory you know what what's like the majority of what we're sensing it's like yeah yeah so I think being able to do that for your hands is probably the most important place to start and you have a rough version of that with controllers today. I think that that'll get even more over time. We have this demo playing pingpong where you have a controller where as the digital ball hits the ping pong paddle you feel it hit the as if it's hitting the ping pong paddle wherever it is so you actually have a sense of like where it's it's hitting the the the paddle so I think that was that was just a wild demo so I think we'll get some of that the most extreme version of this is wanting force feedback right so I mean like for doing a lot of sports right it's it's like okay we can kind of do a good approximation of like boxing today or you get like good feedback on your hands but it would be hard to do a virtual reality version of Jiu-Jitsu where you're like grappling with someone and you need like real kind of force feedback on that so that's probably like the hardest thing right to go do but I think we'll get there. You know I think like most science fiction it's not this binary thing that you just like wake up one day and we're like oh we've realized all the dreams but but I I do think that these platforms are going to be the first time that I think that there's a realistic sense of presence in all the ways that that's special to people for most things that people want to do which are not the most physical ones and even some of the basic physical ones I think we'll get. But then there's a long tale of other stuff I mean smell is also
really important for people yeah right it's I think it's disproportionately important for memories and that's not really a thing that I think in the next few years we're going to have in any of these devices I mean that's a very difficult and challenging thing on its own. What is the piece of that that you feel most interested in, that you keep coming back to in your mind? This has the frustrating property to develop that the sense of presence is almost like when you're designing something that that's sort of trying to artificially deliver it you're delivering an illusion to a person and more than any one thing that provides a sense of presence it's actually more the case that any one thing done wrong breaks the sense of presence. You kind of know that you're interacting with technology but it's so convincing that um that you just kind of go along with it. You're like okay yeah no this person feels like it feel like they're there right. When I did that pingpong demo I like at the end of it I dropped the pingpong paddle on the virtual table and it shattered so that was not the best for for our internal development but like that's winning in our in our development right it's like when when you feel like something is is kind of so realistic that you you're just convinced that um that it's there now and there are a lot of things that can break that right so I think a a field of view that's too low right so something feels real but then you turn your head and it's not there um latency read physics that don't behave like realistic physics. It also is interesting in some ways what people can accept as physically real even though it's not right so like we've done a ton of work on avatars we we have this whole work stream on Kodak avatars to do these photo avatars and it's I think it's going to be incredbly compelling and people are going to love it but one of the things I found interesting is the ability to mix photorealistic and expressive kind of the cartoony avatars with photorealistic worlds and kind of more cartoony computer game type worlds so you can have the a Kodak kind of photorealistic avatar of a person in what is clearly like a video game or cartoon world and people are generally pretty fine with that it's like okay that that feels pretty good and similarly having a photorealistic world but good increasingly good kind of cartoon avatars as long as the avatars move in a way that feels authentic to the person you're interacting with it actually feels pretty good you know it's when you look at a 2d still frame of it some of the stuff can look a little bit silly and and we've certainly you know had had a our share of memes around that but um but when you're in there you know and you you've played around with lot of the stuff it feels realistic because it's basically mimicking the kind of authentic mannerisms of of a person that you're interacting with and even if it's not a Kodak photo realistic avatar if it's kind of a more cartoony expressive one so I I think that that's it's very interesting to see kind of which pieces you need to unlock and what where you just need to be like very technically excellent and consistent but it's um this isn't a space where it's like you deliver one thing and it's good this is like there's a wide breth of things that you need to nail and then have it all come together and that's why these are you know 10 year projects. It seems like an interesting way to learn about the human brain and what we actually care about with respect to what feels real. I was wondering about, there was this moment in an
interview that you did with Lex Friedman, you quoted research that says that the average American has fewer friends now than they did 15 years ago and I was so interested in that because it seems like if we want to get to a world where there's more human connection this is the trend that we're going to have to grapple with and just to give some data on this in the American Time Use Survey over the last 20 years the amount of time American adults spend socializing in person has dropped by nearly 30%. For ages 15 to 24 according to the Surgeon General it's nearly 70%. and I look at that data and I think to myself well maybe if we're all socializing digitally that doesn't matter so much maybe there's a future where that's actually fine but there's also data that suggest that we're struggling somewhat. The number of Americans who say that they don't have a single close friend - yeah it's really sad - that share has jumped from 3% to 12% in the last 30 years. It feels to
me like with all the tools that we've built for human connection, we're struggling to connect and I'm curious why do you think that's happening? I mean there's a lot going on to to unpack there. A lot has changed sort of economically and socially during that period and a lot of those trends go back before a lot of the modern technology. So I mean this is something that a lot of academics and folks have have studied but it is an interesting lens to look at this though because I think whenever you're talking about building digital types of connection one of the first questions that you get is is that going to replace the physical connection and my answer to that especially in the case of something like this is that no because people already don't have as much connection as they would like to have. It's not like this is replacing some sort of better physical connection that they would have otherwise had. It's that the average person would like to have 10 friends and they have two right or three and there's just more demand to socialize than what people are able to do given the current construct and giving people the ability to be present with people who are in other places physically just seems like it will unlock more. It's not going to make it
so, if I have glasses, it's not going to make it that I spend less time with my wife, it's going to make it so that I spend more time with you know my sister who lives across the country. And that's, I think that's good. I think people need that. As for the rest, I I think we could probably spend a multi-hour podcast just going into all of the different kind of socioeconomic political dynamics that are going on but none of the trends that I've seen does it seem like the primary thing that's going on is that because people are interacting online they're now not interacting with their with people physically. Now certainly I think you you I do interact with people online who I also like to interact with physically but and I think that that's kind of like a combination um like more combined richer relationship that you have overall but I think that there's a lot going on with the loss of of kind of social capital and connections that really predates a lot of the modern technology. The goal of what, I'm what I'm trying most to learn about is how we can structure the technologies that we use in the future to get toward this future I think you're imagining of more human connection in more ways. I'm curious, you
brought up the other big pillar of AI and in some of your conversations, I'm thinking of a conversation with Tim Ferris in particular, you talked about a lot of different use cases of AI and they seem to me to fall on somewhat of a spectrum. Like for example you mentioned automatic real-time translation, like basically the Star Trek Universal translator. We're pretty much there! Yeah and that's one example on one end of the spectrum where some people might argue that there is a chance that someone is less likely for example to learn a language because we can all speak to each other in real time in different languages. I think nobody would really argue
that therefore we shouldn't have that kind of universal translator. People still learn Latin and Greek. Right exactly and so I think that end of the spectrum is something like um technologies that really measurably unlock our humanity because they remove a struggle between people and then on the other end of the spectrum there are a lot of educational things for example where the struggle is kind of the point right? Like it's like building a muscle. I can think of so many times
in my life where like the reason why I was doing something was not the output it was the fact that I was trying so hard to do it. There's one example in the Tim Ferris interview where you talked about your kids struggling to articulate themselves emotionally and adults very much had the same problem and you talked about AI as a way to help them articulate those emotions. Yeah and I thought about all of the many times in my life where I have struggled to articulate my emotions and how I really could have used some help in those moments and I also found myself thinking about the times when that was really building a muscle where like the act of struggling to communicate with someone and understand what they wanted from me was was important to my development. And so my question is if you think about that as a spectrum between things that are really important to our humanity where and the struggle being removed is helpful versus things where the struggle is the point and it unlocks something about our humanity and is important to preserve like building a muscle, how do you draw the line between those things and how do we ensure that the muscles that we're building for this future are stronger and not weaker? Yeah it's interesting I mean I think we're always going to find new things to struggle with and I mean it's you can always get better at communicating with other people and kind of expressing yourself and understanding other people so having a tool that can help you do that better isn't going to mean that like oh now we perfectly understand every you know it's I mean I think the maybe one of the most functional aspects of this you're already seeing a lot of these AI models really help people with coding right like a generation ago um before I was getting started a lot of coding was like really low-level system software and you know then by the time that I got into it there was a little bit of that but um you you can make websites pretty easily make apps pretty easily and I think in 20 years or a lot sooner than that you're going to basically be in a world where kids will be able to just describe the things that they want and build incredibly complex pieces of software so it's um in that world are kids going to be not struggling I I don't think so I think that they're going to be just expressing their creativity and and it'll it'll be this kind of constant iterative feedback loop around like okay like yeah I you know took a few minutes to describe this thing and like yeah this whole like amazing virtual world was created that I can have see on my glasses or whatever but like these things are not exactly what I want them to be so now I need to like go back and edit them it just I don't know I think that there's always more. Another way to get this - it's one of the things that I think makes makes people so good. It just there's there's always more to do. We'll always find the struggle? Yeah. Another way to get at this is if you if you play this out to make the tools even better in like 10 years let's say your kids are in high school are there ways that you would want them using AI because you think it would accelerate them intellectually and ways that you would advocate for them not to use it or things that you would have concerns about? I mean I think that there's some things that you need to be able to do yourself. I think that's a lot of the basic fear that people have around this is
that while we're building these amazing tools we get away from this self-confidence and ability of being able to do like this basic stuff yourself so it's like all right you have a calculator but it's still good to be able to do kind of basic math in your head because there are a lot of things that come up throughout the day that you just want to have a general numeracy around right that often they're not expressed in numerical terms but just in terms of understanding trends or understanding arguments that people are making, you you kind of need to understand the shape of how numbers come together and so I think one of the big debates is like should we still teach our kids to program computers even though you're going to have these tools in the future that are just so much more powerful than anything that we have now to produce incredibly complicated pieces of software. I think the answer to that is probably yes because I think teaching someone how to code is teaching them a way to think rigorously and that even if they're not doing most of the code production I think it's important that you kind of have the ability to think in that way and I think it's going to just make you generally a better thinker and better person so yeah maybe that's like this generation's version of calculators it's like so you you want to you want to use the calculator but you'll also want to be able to generally do without it. Other ones like language I don't know I mean different people can come out I think this is one of the interesting questions about parenting these days is like is is just kind of like what what's important to teach your your kids and in an era where so much is going to change over the the time that they're even in school. Language I think you can make similar arguments. I think there's a lot of it's like it's probably going to be less functionally relevant in the future to learn multiple languages but it sort of helps you think in different ways, you know I found from the languages that I've studied that a lot of it you learn about the structure of your own language, you can you know you also learn about the culture right because so much of how things are expressed in different places is tied to the nuance and the history of kind of what how so I think like you that's all valuable and interesting stuff to get into but then I don't know at the same time we only have so many hours in the day so people need to prioritize what they're going to learn and it may be that okay in a world with perfect translation which by the way we basically just announced on the Ray Ban Metas that now you're going to be able to just like you go to countries yeah we're starting out with just a few languages but we'll roll it out to more and you know you'll be you could be traveling anywhere and you have your glasses and they just translate in real time in your ear. So it's wild, yeah so I think people are going to need to choose what
what what they want to focus on going forward. How do the developments that we've been talking about in AI intersect with social media and the platforms that most people use today? There's a future where there's images and generated text and maybe AI influencers. How does generative AI change how social media feels in the future? Yeah I mean I think that that's a really deep one. You know there's already been one big shift which is that social media started
out as people primarily interacting with their friends and now it is you know at least half of the content is basically people interacting with creators or content that's not created by people who they kind of personally know so we sort sort of already have that paradigm and I think AI is probably going to accelerate that. It will give all these people additional tools right so your friends will create kind of funnier memes and more interesting content um that'll come from a lot of different ways. I think some of it will be okay your friends have glasses and they capture a bunch of stuff and before they might have not been able bble to edit it to make it interesting or maybe it was just too much work or they didn't even realize that they captured something amazing but now the AI is like hey I like made this thing for you out of your content um it's like okay that's awesome like people will enjoy that. Creators obviously kind of much more specialized skills are going to be able to use even more advanced AI tools to make more compelling content but then I think that there will be a bunch of kind of green field type stuff where maybe in the future there will be content that is purely generated by AI by the system personalized for you maybe it's summarizing things that are out there that that are going to be interesting maybe it's um just producing something funny that makes you laugh this is going to be like a very kind of deep zone that there's a lot to to experiment with. I think there are going to be AI creators as well,
as creators building AI versions of themselves, I mean that's a thing that we just showed too at Connect is basically I mean if you're a Creator one of the big challenges is like all right there are only so many hours in the day and your community probably has a nearly unlimited demand to interact with you and you want to interact with them because you're trying to grow your community. I mean that's both socially and from a business perspective that's sort of you know growing the community is an important part of what every creator does so okay if we can make it so that each creator can basically make an like an AI artifact that their community can interact with people be clear it's not the actual creator themselves but it's almost like a piece of digital art that you're producing like an interactive sculpture or something that it's like it's like you train it to here's the context that I wanted to have here's the topics I wanted to communicate on here's stuff that I wanted to stay away from you're giving your community something to interact with when you can't be there to to kind of answer all the questions and I think that's going to be super compelling so there's like these interesting things but I think it's I AI it's kind of like the internet in a way where it's probably going to change almost every field and almost every feature of every application that we use um it seems sort of hyperbolic to say that but I do think that's true and it's just hard to sort of enumerate all the different things up front but I think that over the next 5 to 10 years we're just going to explore the impacts in each of these areas and it's going to be like an amazing amount of innovation and really exciting. I feel two things simultaneously when you say that. I feel both like I really want to be optimistic about the future of these
platforms and I obviously have gained so much from an enormous pace of change right like everything that we're doing now and what I actually feel is worried. I feel some specific concerns around the way that you know I might communicate with an audience and the way that they might respond to that or the way that human communication might change but also more generalized just sort of fear of the pace of change and and worry and I don't think I'm alone in that feeling. Yeah and you're supposed to be the optimist! I know! And I'm curious like how you talk to people who feel that way. What concerns do you feel are most legitimate and what do you feel most misunderstood? I think the pace of change is always a concerning thing right it's there is a lot of uncertainty about how how things will go in the future and we're all going to get really amazing new tools to do both our hobbies and our jobs and they'll make it so we can do better work and have better lives but at least on the professional side it's going to be our responsibility to keep up with that or else it's going to be difficult for us to compete with other people who are doing a good job of kind of keeping up with the new trends. So I get it. I mean I think you know especially in the you know line of of work of being a creator and it's a very sort of competitive space, I don't think that like creators necessarily think about it as competitive but it is right it's like it's you know and um and so I get it. I think that this is going to make it so that like the quality of work that people produce and how interesting it is and how much they can communicate and like really efficiently is is just going to kind of go through the roof but but when you're staring down a set of changes like you know that there's some big change coming and you don't know what it is that's always a time of anxiety so I get it. If I take my creator hat
off and I'm just a person who is youngish starting out my career-ish, starting out building a family, how would you advise someone like me to prepare well for the future that we're headed toward to be able to learn new skills now or just think about this future in an educated way? Yeah I mean I just think maintaining curiosity about things is is important. I do think we can overstate to what extent the next 10 years is going to be sort of different from the last 10 or 15. I mean a ton of stuff changed over the last 10 or 15 years too. It's not like this is the only time in history where there's some technology it's going to make it so there's new opportunities and things change the internet coming into maturity and everyone having smartphones has already rewired things dramatically and I mean maybe the next period will be a somewhat bigger change or maybe it won't I think it'll feel different to different people but I don't think this is like going from zero to one it's not like okay everything's just kind of been normal and now like now it's about to change it's like the technology of evolves over time and and like the opportunities that we have evolve and improve and I think that's like the people who do well I think are are people who are generally curious about it and and dig in and and try to use it to live better lives rather than the people who who basically you know try to fight it in in some way. One thing that I really want to ask you about is open source. Yeah. I think imagine that we're talking to an audience that has maybe heard that term but doesn't have any real idea of how that might impact them in the development of AI.
How would you explain the reasonable debate that people in your field are having about this right now? Well I think there are two pieces. I mean so what does open source mean? It means that people can build a lot of different things right so at a high level I look at the vision that a bunch of companies have right so Open AI, Google, they're building an AI right like one AI that I think in general they're like okay this is going to be it's like you're going to use they think you're going to use Gemini or ChatGPT for like all the different things that you want to interact with and at a high level that's just not how I think the world is going to go. I think we're going to have a lot of different AI systems just like we're going to have we have a lot of different apps. I think in the future every business just like they have a website and a phone number and an email address and a social media account is also going to have an AI that can interact with with their customers to help them sell things to help them do support. I think a lot of creators will have their own AIs right I think like a lot of people will interact with with a bunch of different things. There's a question of okay do you want a future that's fundamentally kind of very
concentrated and where you're interacting with kind of one system for everything or do you want one where a lot of different people are building a lot of different AIs and systems just kind of like you probably didn't want there to be you know just one app or just one website. It's like a richer world when there's a diversity of different things so that's one piece is is just giving people the ability to build it themselves and what open source does it makes it that everyone can take and modify the model and build stuff on top of it which is different from the kind of closed and centralized approach. The safety debate is a specific part of this which is in a world where AI gets smarter and smarter, what's the way that we have the highest chance of of having a a a kind of positive future and and not having a lot of the safety concerns? And I think some people think that if we keep the model closed and don't give it to a lot of developers that should make it safer because then you don't get bad developers doing bad things with the model.
Historically I think what we've seen with open source is actually the opposite which is that this is not the first open source project right I mean this is obviously this has been a thing in the industry for decades and I think what we've traditionally seen is that open source software is safer and more secure largely because you put it out there more people can scrutinize it because they can see all parts of the system and then there are inevitably issues with any software there are bugs there are security issues and initially with open source people thought hey if you're putting the software out there and there are holes in it isn't everyone just going to go exploit those holes and especially the bad guys but it turned out that it sort of in this counterintuitive way that by making by adding more scrutiny to the systems the holes became apparent quicker and then were fixed and then people roll out a new version just like we roll out a new version of our models right Llama 3, Llama 3.1, Llama 3.2 everyone upgrades, so I think the same thing is going to happen here I think it's sort of this counterintuitive thing where even though I I think there's some concern around all right are bad guys going to do bad things with these models. I actually think you just get a kind of smarter and safer model for everyone the more it's rolled out and the more kind of scrutiny is on it and then part of that is we get feedback and we make the model safer so that is we roll it out to to more people it's safer for more people to use. So I think that the history of open source in the software industry generally
would suggest that open source is going to lead to a more prosperous and safer future. Our show is called Huge If True and what I mean by that is kind of testing the most optimistic non-obvious thing and so my question to you is what is the biggest open genuine question on your mind right now? In which field? You're in so many! I am particularly curious about the combination of AI and hardware but I realize that we've covered a lot so I'm curious the direction you'd take this on a question that occupies you right now. Gosh I mean I think maybe one that's a little more AI specific is there a current set of methods that seem to be scaling very well right so with past AI architecture you could kind of feed an AI system a certain amount of data and and use a certain amount of compute but eventually it hit a plateau and one of the interesting things about these new transformer based architectures over the last you know 5 to 10 years is that we haven't found the end yet. So that leads to this dynamic where Llama 3 you know we could train on
you know 10 to 20,000 gpus, Llama 4 we could train on you know more more than 100,000 gpus, Llama 5 we can plan to scale even further and there's just an interesting question of how far that goes. It's totally possible that at some point we just like hit a limit and just like previous systems there's an asymptote and it doesn't keep on growing but it's also possible that that limit is not going to happen anytime soon and that we're going to be able to keep on just building more clusters and generating more you know synthetic data train the systems and that they're just going to keep on getting more and more useful for people for quite a while to come and it's a really big and high stakes question I think for for the company is because we're basically making these bets on how much infrastructure to build out for the future and this is like hundreds of billions of dollars of infrastructure so like I'm clearly betting that this is going to keep scaling for a while but it's one of the big questions I think in the field because it is possible that it doesn't. You know that obviously would lead to a very different world where it's I mean I'm sure people still figure it out eventually just need to make some new fundamental improvements to the architecture in some way but that might be a somewhat longer trajectory for okay maybe you know the the kind of fundamental AI advances slow down for a bit and we just take some time to build new products around this or it could be the case and that's what I'm betting on that the fundamental AI will just continue advancing for quite a while and that we're going to get both a new set of products that are just really compelling in all these ways and that the technology landscape and what's possible will just continue being dynamic over like a 20-year period and that's probably what I'd guess is going to happen but it I think it's one of the bigger questions in the industry and kind of for technology across the world today. Is there anything else that you want to say? I
don't know! Awesome. We're good. Amazing yeah thank you so much for doing this. Yeah no thank you...
2024-10-02 06:48
