Ep 86: Yann LeCun on Leaving Meta, Breaking The LLM Paradigm, & Why Hinton is Wrong

Yann LeCun, Turing Award winner and former Chief AI Scientist at Meta, joins Jacob Effron. The conversation centers on Yann's contrarian thesis that LLMs are a dead-end on the path to human-level intelligence, despite being useful products — because they can't predict the consequences of their actions, can't plan, and fundamentally can't model the messy, high-dimensional real world. He unpacks his alternative architecture, JEPA (Joint Embedding Predictive Architecture), which learns abstract representations rather than generating pixel-level predictions, and explains why this approach is essential for robotics, industrial applications, and any system that needs to operate beyond the substrate of language. Yann also reveals the real story behind his departure from Meta (he had zero technical influence on Llama, contrary to public narrative), the genesis of his Tapestry project for sovereign open-source AI, why he believes LLMs are intrinsically unsafe, where he diverges from his fellow Turing laureates Hinton and Bengio, and why he predicts the industry will recognize the paradigm shift by early 2027. Throughout, he offers candid reflections on the tension between research and product at major labs, and why he intentionally headquartered AMI Labs in Paris with zero Silicon Valley VC money.

• 0:00Speaker 00

  You're one of the godfathers of AI. What's your kind of view of the path of progress here?

• 0:03Speaker 01

  Five years complete world domination. The best way to get breakthrough research is you hire the best people. You get the out of the way. Pardon my French.

• 0:11Speaker 00

  You shared the Turing Award with two others. When did your views start diverging?

• 0:14Speaker 01

  In 2023.

• 0:15Speaker 00

  How do you know it was time to leave Meta? It sounds like you were thinking through some of these things over a period of time.

• 0:19Speaker 01

  There is a big misconception about my role, my relation to Alex, and how AI was run at Meta.

• 0:25Speaker 00

  What's like one thing you've changed your mind on in the last year?

• 0:27Speaker 01

  I mean, the whole idea of.

• 0:28Speaker 00

  Yann LeCun is one of the godfathers of AI. He's an absolute legend in the field, someone I've admired for a long time. And so it was such a treat to get him on unsupervised learning. He's been a noted skeptic of LLMs in many ways. And so we dug into what LLMs can do, what they can't do, some of the limitations he sees and why he ultimately decided to pursue a different architecture. And we also talked about his time at Meta. You know, the things he's proud of in setting up FAIR, how the last few years proceeded and what ultimately led him to spin out and start his own company, Ame. I think it's just fascinating to get Yann's thoughts on everything happening in the AI ecosystem today. This tension between basic research and then pushing LLMs forward and how that's happening in a bunch of organizations today,

• 1:12Speaker 00

  as well as his thoughts on just where the whole space is headed. He's just an absolute giant in the field. And when I started this podcast, I hope we get guests like him. So it is just such a treat. I think folks will really enjoy hearing the conversation we had. Without further ado, here's Yann.

• 1:29Speaker 00

  Yann, this is such a pleasure. You're one of the godfathers of AI. I feel like when I started doing this podcast years ago, I was really hoping we might one day get someone like you on.

• 1:37Speaker 01

  You know, I don't like that term because I live in New Jersey. When you're a godfather in New Jersey, it doesn't mean the same thing.

• 1:43Speaker 00

  Very fair, very fair. You know, obviously, you know, your bet on neural nets when everyone doubted them is legendary. And I feel like today you're making a similar bet in many ways against LLMs and the kind of predominant generative architectures that so many believe in. You've recently started a new company behind this theme. And so, you know, our goal today in the conversation is to leave our listeners with a lot more information about Ame, what you're doing there, some of your work at Tapestry, you know, why you think the rest of the field is pointed in the wrong direction. Around some of these generative models and then also just get your reflections on the way the fields unfolded, your time at Meta and all that. So, you know, modest goals for a for for for a single podcast episode. I figured it'd be great to start with the meat because the company feels like the clearest statement of your technical thesis going forward.

• 2:25Speaker 00

  And so you recently launched the company focused on world models and scaling the Jepba architecture, which you obviously pioneered over at Meta. And so I'm wondering if you could talk a little bit about the origins of that architecture and the extent to which you drew inspiration from the human brain and the way that works.

• 2:38Speaker 01

  So first of all, I want to say there's nothing wrong with L L M's in the sense of L L M's, you know, are the basis for a lot of very useful AI products that all of us use, including me.

• 2:52Speaker 01

  They're great, okay, for what they do. They're just not a path towards human level or human like intelligence or even animal like intelligence. So that's my claim. Okay. I'm not saying L L M's are useless, right? I'm just saying they're not a path towards.

• 3:09Speaker 00

  I mean, you helped build some of the first major open source ones.

• 3:12Speaker 01

  Right, absolutely. So what is AMI? So AMI really stands for advanced machine intelligence. And the the the kind of subtitle, the motto, if you want, is AI for the real world. So basically, a lot of, you know, AI techniques that people know about today are good for language manipulation. Either human language or computer code or mathematics or legalese, which barely qualifies as human language.

• 3:43Speaker 00

  Unfortunately, a lot of human language used for it.

• 3:46Speaker 01

  Sadly, you know, language is very special in a way, and it's particularly well suited for the type of, you know, architectures that have been so successful recently. The, you know, large language models, GPT style architectures. But what about the real world? What about like understanding the physical world? Turns out, reality is way more complicated than language, because it's highly measured, it's continuous, it's noisy, it's messy. And training a system to understand. And the real world is much, much harder. So that's really what we're after. That's what I've been after for most of my career and really kind of,

• 4:27Speaker 01

  you know, working on in an accelerated fashion over the last five, six years or so and making significant progress over the last two years. And so it made sense to really do a startup around it and sort of go to into high gear, you know, and pushing that. And it became clear, you know, by the end of last year that Meta was really not the right place for that. So which is why I left and started Amin Hubs.

• 4:52Speaker 00

  I think it's an interesting like, you know, trend that we're seeing across the board, right? Where it feels like um there you're there's there's many folks spinning out of, you know, either some of the large companies or research labs, you know, that have a particular direction of research they're excited about. And you you'd have set an interesting vantage point of this from your time at Fair, this uh almost tension that exists between, you know, go pursue as many different research directions as possible in these companies versus, hey, something's really working. This is the thing that we're going to sell for the next six, 12 months. Like, go focus on that. You know, I'm curious your your thoughts on that and what you've kind of seen in the industry at large.

• 5:23Speaker 01

  Well, it's a strange uh trade off. There's really two modus operandi, right? There's a lot of exploratory research, a lot of direct research directions, right? And sometimes something kind of seems to work and you you need to push it further. And it's not research anymore. I mean, the people working on it are researchers or they're called researchers at least in the press, but uh but really it's becoming more engineering and pushing for for products, right? So

• 5:51Speaker 01

  that happened a number of times at Meta because of things that were started at Fair. Such a thing happened in, you know, early twenty twenty three, essentially, uh when, you know, Lama, which was developed at Fair, Lama one, was very promising. And uh Meta created a whole organization, Gen AI, to turn it into something real and a series of products uh and produce, you know, Lama two, Lama three, Lama four, which was a bit of disappointment. And because, you know, Mark Zuckerberg was disappointed by it, he kind of rebooted the entire organization, reorganized it and hired new people, et cetera.

• 6:31Speaker 01

  But what also happened over the last year is that basically the company, Meta, realized that they'd fallen behind a little bit. And so that kind of refocused the strategy on trying to catch up with the industry. And the sad side effect of it is that a lot of the exploratory research was basically not given high priority anymore. I mean, it didn't concern the stuff I was working on, all the JEPa and world models. Because, you know, Mark himself and Do Buzzwords, the CTO, and a bunch of other people in the company were really interested in that project and really believed in the long-term impact. But the rest of the company was just, you know, totally entirely focused on L L Ms and

• 7:16Speaker 01

  made it clear to me that Meta was really not the right place to push for that project anymore. And then we started to have good results. And so it was clear that, you know, we had to kind of make that transition between research and actually kind of developing the technologies, scaling it up and building products out of it. And we realized also that most of the applications were probably for things that Meta was not particularly interested in. A lot of applications of the kind of stuff that we've been working on is in the industry, like manufacturing industry and stuff like that.

• 7:51Speaker 00

  Obviously, you're you're kind of pursuing world models and in that broader world. And I think there's other people that have come at the world model pace from a more like generative approach. And so I think you've got folks, you know, you've got the Google folks in Genie and the video models, you've got folks, you know, building V L As on the robotics side, you've got Fei Fei and kind of like the 3D spatial models. As you think about kind of the the body of of evidence that got you excited about the JEPa models and how you kind of compare them to what the generative folks have done, you know, where do you think we are today in terms of like comparing these architectures and approaches?

• 8:22Speaker 01

  Okay, so world model is quickly becoming a buzzword.

• 8:25Speaker 00

  Yeah.

• 8:26Speaker 01

  Right now, right? Certainly in research, but also in industry to some extent. And and then there are two factions, if you want. I'm not going to talk about V L A because V L A. Is clearly now being seen as not going anywhere. Like, it's really not working. Uh, so VLA is, you know, vision, language, action models, right? So basically, use the LLM technology to train a system to produce actions for, like, controlling a robot or something like this, right? So you have vision in, language in, action out, maybe language out too. Uh, and that's pretty much now seen as a failure. Uh, not being reliable enough, requiring too much training data, you know, things like that. Okay, then there is world models. Okay, so what is a world model?

• 9:11Speaker 01

  Uh, a world model at a very general level is something that allows an agentic system to anticipate the consequences of its own actions. Okay? Predict the consequences of its own actions. From my point of view, I cannot imagine how you can even think of building an agentic system without that system having the ability to predict the consequences of its actions. I mean, that's pretty essential, right? When we act in the world, we have this ability. And when we, uh, take an action without thinking about the consequences, we're taking a big risk and very often, you know, other people think we're an idiot.

• 9:53Speaker 01

  Uh, we have plenty of examples on the international political scene at the moment of people who have, you know, no ability to predict the consequences of their actions. So that's a world model. That's all it is, right? Ability to predict the consequences of your own actions. If you, if you have this ability, then you can plan a sequence of actions to accomplish a task, to, you know, satisfy a goal. And you do this by planning, reasoning, uh, by a process of search and optimization. You don't do this by predicting one action after the other autoregressively, like a VLA we do. Uh, you do this by searching for a sequence of actions that will accomplish the task you set, you set for yourself.

• 10:37Speaker 01

  So the blueprint for this is completely different from what, you know, LLMs can do at the moment. Uh, LLMs do not have the ability to predict the consequences of their actions, and they do not have any planning abilities. Because inference is by predicting the next token, right? It's not by search. Okay, so right there, you have the two characteristics that I think are essential for intelligent behavior. Ability to predict consequences of your actions and second, uh, ability to plan by optimization, by search, um, find a good sequence of actions that will produce the correct outcome. And then there is a third characteristic, which is, uh, how you, how you predict the consequences of your actions.

• 11:20Speaker 01

  Okay, so, you know, if, uh, if I have a water bottle in front of me, I realize some people would just listen to this and not have the picture. So I have an open uncapped water bottle in front of me. If I push at the bottom, it's going to slide on the table. If I push near the top, it's probably going to flip. We can't predict exactly how the, the bottle will, will fall in which direction. Uh, we can't exactly predict how it's going to slide, you know, how the water will spill, you know, whether the table is tilted in one way and the water will, uh, you know, kind of flow in one direction or another. There's no way we can predict this at a pixel level.

• 12:01Speaker 01

  And so our mental model of the world predicts, but at an abstract level of representation.

• 12:07Speaker 00

  So as you were working on this architecture, was a lot of it inspired by the human brain? I mean, obviously like the, you know, the way you're articulating things is exactly how, how we do things.

• 12:13Speaker 01

  Right, or at least by, you know, cognitive science, right? Whether you can sort of translate this into a neural architecture and things like this, that's, there's a big gap there. Um, okay, so that, that, you know, certainly cognitive science was a bit of a motivation or, or, you know, what psychologists call system two, which is this idea of the way you behave in sort of deliberate reflective behavior is that you do imagine, predict the consequences of your actions and you plan, uh, accordingly. Contrary to system one, where you just act, you know, reactively and instinctively. So yeah, there is an inspiration, but also there is a lot of empirical evidence that you don't want to generate pixels.

• 12:56Speaker 01

  Okay, I've been, I've been really interested in that problem of learning models of the world by prediction for a very long time and then had an epiphany about five years ago, realizing that all of the architectures that have been successful to learn representations of images and videos are non-generative architectures and all the generative ones basically have been failures, right? So.

• 13:25Speaker 01

  VAE, right, variational autoencoders, or autoencoders more generally. It's kind of a natural way to think about like learning abstract representations of inputs, right? So you put an image at the input of a neural net and then you train it to just reproduce the input on its output. Now, with a big neural net. Now, if you just do it this way, your neural net will not do anything interesting, we just learn the identity function. Yeah. Completely uninteresting. It doesn't work. If you train a VAE to learn representations of images, you get something, but it's really not that great. Same with sparse autoencoders. Then you have another set of techniques, and it's kind of derivative of something called denoising autoencoder. Masked autoencoder is a version of this.

• 14:10Speaker 01

  BERT is a version of this for NLP. So you take the image, you corrupt it in some way, and then you train this big neural net to recover the original image. There's a huge project at FAIR on this called MAE, Masked Autoencoder. It was very disappointing. A lot of computation and not really great, satisfying result. Simultaneously, some of the same people working on MAE and some other people in Paris and in New York were working on other techniques using non-generative architecture, joint-embedding architecture. So take an image, corrupt it in some way, and then run the two images to encoders, and then try to predict the representation of the original image from the representation of the corrupted one.

• 14:54Speaker 01

  That's JEPA. Yeah. Okay? So JEPA means joint-embedding predictive architecture, right? So you have one encoder that makes an observation, another encoder that makes a different observation. You try to predict the representation of the first one from the second one with a predictor. And those techniques turned out to work much better for representing images and video. So things like DINO, DINO V1, V2, V3, project that is still going on at FAIR in Paris, projects like IJEPA, and then VJEPA, and then before that, there were like SimCM and MOCO and a whole bunch of different techniques, mostly for Meta, there was a bunch of others from other groups.

• 15:37Speaker 01

  But that turned out to be a much better way of learning representations of images than predicting pixels. Yeah. And so it just clicked in my mind, but you know, not just mine.

• 15:51Speaker 01

  That this was the way to go and predicting pixels was kind of a losing proposition.

• 15:55Speaker 00

  You know, it feels like there's all these robotics demos that are released, you know, from some of the model companies that are feeling increasingly impressive and maybe, you know, seem to resemble things like planning and reasoning when, you know, they maybe haven't seen a room or a specific, you know, a version of a task before and are still able to execute that task. You know, what would you say to our listeners, I guess, that observe that stuff and feel like, ah, it feels like we're trending toward some real progress with some of the generative approaches?

• 16:20Speaker 01

  Well, there is real progress and some of those demos are really impressive.

• 16:25Speaker 01

  But they are trained with enormous amounts of data collected either from teleoperation or from just, you know, human action with things you hold in your hand.

• 16:37Speaker 00

  Yeah, like grippers.

• 16:39Speaker 01

  Grippers that, you know, and you collected data for that or just, you know, tracking hands and fingers of a person and then translating this into kind of commands for a robot. And so those things are trained with imitation learning mostly, right? And a little bit with, you know, reinforcement learning to fine tune in mostly in simulation. So the issue with this is that you need a lot of data to train those systems to imitation. And it becomes expensive and it's a little brittle in the sense that, you know, you need to collect lots of data for every task you want the robot to solve. Whereas if the system

• 17:22Speaker 01

  had a world model that allows it to predict, you know, the outcome of an action, it would just plan an action to solve a new task without actually having to be trained to accomplish this task. So the degree of generalization you would get with a world model based system is much, much larger, you know, kind of wider spectrum of tasks with less training data that would be required than a system trained with imitation learning and, you know, fine tuning.

• 17:54Speaker 00

  No doubt those approaches require more data. And I guess this question of generalization really is the big question, right? You know, and I think some folks have shown some results around, you know, getting better at task A helps with task B, but that obviously feels like the still the big unanswered question, you know, around those architectures.

• 18:10Speaker 01

  I mean, you get this. Or synergy between tasks, so the more tasks that you train the system to solve, the more task is going to be able to acquire with a small amount of data, regardless of what technique you use. But the hope with well models is that the system can solve new task zero shot, which humans are completely capable of doing, right? And many animals as well. So that's really the hope, like, you know, solving a lot more problems with either a small amount of training data or no training data at all, and just a little bit of maybe, you know, RL style fine tuning.

• 18:49Speaker 00

  Yeah.

• 18:49Speaker 01

  Like, you know, how is it that a seventeen year old can learn to drive in like a dozen hours or maybe twenty hours? We have millions of hours of training data of, you know, people driving cars. We still don't have level five driving cars, right? So imitation learning obviously doesn't work even for just the task of autonomous driving.

• 19:10Speaker 00

  Yeah, I guess it will be a race between the ability to develop some of those capabilities, which may take time and lots of data versus this kind of architecture. I feel like there's this dream of using video models to just generate like tons of synthetic data for simulation. And even if it's not perfect, these video models, from a physics perspective, it's like helpful enough to, you know, improve robotics in the underlying physical world. What have you made of some of those approaches? Obviously, I think Nvidia has been focused there. Google seems to be going down that road.

• 19:35Speaker 01

  I'm sort of asking you again the question, you know, why can seventeen year old learn to drive in twenty hours? You don't need millions of hours of demonstration. And you don't need synthetic data. You don't need any of that. So, you know, I want a system that can learn as fast as that. If we crack that, then we don't need, you know, generated data, right? I mean, we might need to train the system in simulation, but not with the same amount of, you know, of time or trials as current systems require. It's really a question of data efficiency.

• 20:08Speaker 00

  And I was interviewed Jerry Tork on the podcast. He was at OpenAI and spun out to start his own lab. And you could sense a similar tension where I think he actually might even agree that, you know, if you continued scaling RL the way we're scaling, you get more, you know, you continue getting very impressive results. But I think he felt, God, there's just got to be some like way more efficient way to do this.

• 20:25Speaker 01

  Exactly.

• 20:25Speaker 00

  It's interesting. It's an interesting tension because you could imagine if you're OpenAI and you know something is going to continue. Like you could continue scaling it and it will keep getting better. There's not a ton of incentive necessarily from a business perspective to do something more data efficient.

• 20:37Speaker 01

  Right. There's no incentive for the other companies to do anything different either because they're all chasing the same. Like they can't afford to kind of fall behind the others, right? So they all work on the same thing.

• 20:50Speaker 00

  Yeah.

• 20:50Speaker 01

  And it's a bit of this sort of, you know, kind of herd behavior.

• 20:58Speaker 01

  And, and, you know, in, in mostly in Silicon Valley where everybody is digging the same trench.

• 21:03Speaker 00

  Yeah.

• 21:04Speaker 01

  And, you know, so I purposely set up the headquarters of AMI Labs in Paris.

• 21:16Speaker 01

  The American office being in New York, not Silicon Valley.

• 21:19Speaker 00

  It's really interesting because I think it, it, it points to a tension that, you know, exists in the broader ecosystem today where, you know, you could imagine the other side being, sure, maybe there are more data efficient methods out there, but like, almost who cares because we can keep scaling what we have to, to better and better results. And then obviously I think from both, you know, net new, new things you can accomplish from these models as well as just the joy of being a researcher and finding these new things. I get why there's such an attraction to, to, to these other architectures as well.

• 21:46Speaker 01

  And it's a bet.

• 21:47Speaker 00

  Yeah.

• 21:47Speaker 01

  But, you know, pretty confident because, you know, we, we have results already.

• 21:51Speaker 00

  Yeah. And as you think about like the, the kind of, um, the initial spaces you're most excited about for the AMI technology, like what gets, you know, where do you think, you know, the, the technology goes and what are you most excited about?

• 22:03Speaker 01

  Well, I mean, you know, AI for the real world.

• 22:07Speaker 01

  Like, you know, can, where is your domestic robot? Where is your level five self-driving car?

• 22:12Speaker 00

  Yeah.

• 22:12Speaker 01

  Where is, uh, and that's.

• 22:14Speaker 00

  When am I going to get a domestic robot? I'm excited about this.

• 22:17Speaker 01

  Well, so this is several years down the line. Okay. Despite the fact that there is like a huge number of companies building robots, none of those companies. Actually, I have an idea how to make them smart enough to be useful, right?

• 22:29Speaker 00

  Or trusted around with a baby in the house or something.

• 22:31Speaker 01

  Certainly not that, but even for like, you know, relatively narrow manufacturing task, right? You know, I mean, none of them really knows how to do this reliably, other than, you know, for by imitation learning for a small number of tasks. So how do we make those things useful? So that's kind of a relatively long-term objective. Shorter term, there is a huge amount of applications in industry where you need to have a system, an intelligent system, that has the ability of, you know, predicting what's going to happen if I change this control variable on this complex system, be it a jet

• 23:13Speaker 01

  engine, a chemical plant, a power plant, some manufacturing line, a patient, a human cell, right? Those are systems that are sufficiently complex that you can't model their behavior with a small number of equations, right? So the traditional way of modeling does not work. And what you need to do is train a neural net, a deep learning system, to, you know, model the dynamics of that system from data. And what you get at the end is a phenomenological model of that process, of that system. And if it's action-conditioned, then you get basically a world model of that system

• 23:58Speaker 01

  that allows you to control it optimally for whatever purpose you have. And I think the number of applications of this in industry is mind-boggling.

• 24:08Speaker 00

  Where do you think we'll be with, you know, JEPMA models over the next couple of years? Are there like, you know, milestones you'd point to or like, what's your kind of view of the path of progress here?

• 24:16Speaker 01

  Okay, a couple of years is a little short, like five years, complete world domination, essentially.

• 24:22Speaker 00

  So somewhere between on the path to world domination in five years.

• 24:25Speaker 01

  I mean, this is kind of a joke, obviously, but this is a quote from Linus Torvalds, right? You know, when people ask him, what's your goal with Linux? He said, total world domination. He actually managed to do that.

• 24:36Speaker 00

  Yeah, very fair.

• 24:37Speaker 01

  To first approximation, every computer in the world runs Linux, right? So, so that's kind of a joke. But, but in the end, I think this is the blueprint for intelligent systems of the future. There'll still be a small place for LLMs, you know, for like a language interface, basically.

• 24:55Speaker 01

  But what we're designing are our systems that are capable of thinking. They may not be capable of talking or listening initially, but they will do the thinking. And then you can add the talking and listening on top of that.

• 25:11Speaker 00

  I'm sure you and the team are eagerly working to kind of, you know, get the early proof points of this. And obviously, you've already had some in the work you've done. How do you think about like the interim steps of what you'll be able to show on that path to five-year world domination?

• 25:23Speaker 01

  Well, so I think, you know, within a year or so, we'll have, I think, a general methodology to train hierarchical world models on, you know, a very wide variety of modalities. We know we can do a good job on video

• 25:43Speaker 01

  with some techniques that we're not completely happy with because they have some shortcomings. But we have sort of small-scale demonstration of a methodology that we think is really what we want. So we need to scale that one up and get it to the same level of performance as the other techniques that are not as satisfying, if you want, on things like video, but also on other types of data sets that we would get from industry partners. Okay, so we'll have demonstrations that we can train world models, perhaps action condition world models, that allow us to plan for a number of different use cases. Some of them will be robotics. Some of them will be industrial process control of various types.

• 26:28Speaker 01

  Maybe some of them in health, healthcare as well, because we are partners in that domain. And that should be within a year to 18 months.

• 26:40Speaker 01

  And then we'll push this methodology and those models into those use cases with partners, some of which are investors already, you know, in our company, and gain experience on how to kind of essentially build a somewhat universal world model, if you want.

• 26:57Speaker 00

  I mean, and you've obviously had this, you know, this experience before of kind of making this really contrarian bet on neural nets and being certainly proven abundantly right in the history books. I guess as you think about this bet, which I think, you know, if you talk to the majority of people, maybe at the cutting edge of various parts of AI, maybe would say is contrarian today. In what timeframe do you think it will become apparent like, you know, that this was right?

• 27:20Speaker 01

  I think it'll happen faster than expected, perhaps, because, I mean, you can see that one model is already becoming a buzzword, right? At least at a research level.

• 27:34Speaker 01

  And it's starting to kind of permeate into the industry.

• 27:37Speaker 00

  Yeah.

• 27:37Speaker 01

  And a lot of people are realizing, like, VLA's suck. And, you know, LLMs don't work for real-world data. Industry has realized this already, certainly on the user side. And I think because of the importance of the robotics industry, you know, a lot of people are kind of trying to figure out, like, how do we get there? How do you make those robots useful? So I think the realization that you need a change of paradigm is happening as we speak. It will become completely obvious to people by early 2027, I think.

• 28:13Speaker 00

  Yeah.

• 28:13Speaker 01

  Now, that doesn't mean we'll have a solution by then. We hope we will, but, you know, we'll see.

• 28:18Speaker 00

  I guess, you know, switching gears to the LM side, you mentioned some of this work you're doing with Tapestry, which I think would be really interesting for our listeners. And so maybe just speak to that a little bit.

• 28:25Speaker 01

  Okay, so this is kind of a little bit orthogonal to AMI Labs.

• 28:30Speaker 00

  As if that wasn't enough to keep you busy.

• 28:32Speaker 01

  Well, it's a kind of an idea I've been forming over the last three years or so. It's the fact that people increasingly use AI assistance for various things, right? I mean, you see a decrease in the use of traditional search engines. And you just ask a question to your favorite AI assistant.

• 28:55Speaker 01

  And, you know, if the plan that Meta and others are developing of, you know, having smart devices like smart glasses and stuff like that, you know, is realized, basically you would just be talking to your AI assistant, you know, by voice with, you know, through your smart glasses or maybe some other smart device. And so all of your information diet will be mediated by AI assistance. And if you are someone, you know, somewhere in the world, let's say outside the US or China, and you have an AI assistant, and that AI assistant was built in California or, you know, Beijing, or Shanghai, or Shenzhen.

• 29:37Speaker 01

  It's not good for you. Like, you may speak a language that those systems really haven't been trained to handle particularly well. You may have a culture that is not particularly well-understood by people in Silicon Valley in China, not well-represented by the training data that is publicly available on the internet. You may have a value system that is absolutely not represented by, you know, people building those models. And certainly, you almost certainly have political opinions that are absolutely not represented by the handful of AI assistants you you might be able to get from the, you know, West Coast tech companies or from Chinese companies. So what is the solution to this? Like, how do you serve,

• 30:23Speaker 01

  you know, a farmer in India? Or even a philosopher in France? Or Germany? And what you need is a platform which basically is an open, free foundation model, LLM style, that is fine-tunable by anyone to cater to the interest of people speaking a particular language, having a particular culture, having particular value systems, political biases, creeds, whatever it is.

• 31:03Speaker 01

  And so what you need is a wide diversity of AI assistants. There's a lot of countries around the world

• 31:11Speaker 01

  that are neither the US nor China, who absolutely want some level of sovereignty for AI, not just for the industry, but also for the citizen. They don't want the citizen to get brainwashed by... Chinese model or a Californian model, actually. And so they want sovereignty. How do you get that? So the way you get a platform like, an open platform like this to get to the frontier is you just train it on more and higher quality data than the proprietary systems. If you talk to people in India, in France, in Vietnam, in Morocco, in Switzerland,

• 31:54Speaker 01

  in Korea, Japan, Kazakhstan, everyone wants basically sovereignty. And you tell them, like, you guys have been training your model, you know, locally. You don't have to share your data. So that's the crucial aspect of tapestry. You would have international contributors to tapestry, contributing to training a global model that would basically constitute a repository of all the world knowledge and culture, if you want. But the contributors would contribute data and computing resources, but they would preserve the control on their data. They would not have to share their data with the other contributors.

• 32:40Speaker 01

  What they would contribute is parameter vectors, right? So it would be a kind of federated learning style thing where you have a bunch of data centers.

• 32:51Speaker 01

  You know, they get the parameter vector from the global consensus of a model. Think of it as an average of all the parameter vectors of all the contributors, right? So all the contributors periodically tell everyone else through maybe a central server, here is my parameter vector. What is yours? Okay? And so you exchange parameter vectors like this. And a local worker, basically, whenever it updates its parameter vector, it tries to also make it as close as possible to the global consensus vector. So as the training of this thing kind of progresses,

• 33:32Speaker 01

  all those parameter vectors converge towards like a consensus model, essentially, which is kind of a repository of all human knowledge. Now you have an open model that is as good as if it had been trained on all the data in the world.

• 33:50Speaker 01

  And now you can fine tune it for your own purpose, your own political, cultural and linguistic biases.

• 33:57Speaker 00

  Yeah.

• 33:57Speaker 01

  Whatever you want, or centers of interest. And I think there is a natural force for this to happen. Because, you know, most countries that are not the US nor China want sovereignty. But also because AI is fast becoming a platform. And there is a natural tendency for platforms to become open. That's what happened with Linux, right? And that's what happened with the software infrastructure of the internet or the wireless network. It's all open source.

• 34:29Speaker 01

  It was proprietary initially, but that was all wiped out.

• 34:33Speaker 00

  It's a really clever way to get around, you know, it would seem that this trend of, you know, decreasing open source.

• 34:39Speaker 01

  And yeah.

• 34:39Speaker 00

  And obviously, I think there's been many fears that as like the closed source models get better, they'll be held back and they'll be used to train the next generation. And, you know, they'll, they'll kind of be this almost like escape scenario for, for closed source models where they get, you know, so much better than, than their open source counterparts.

• 34:53Speaker 01

  So remember what, you know, who the big players of the internet infrastructure were in 1996? Sun Microsystems, HP, Dell and a few others. So Sun Microsystems was selling you Solaris with their, you know, proprietary hardware. HP with HPUX. They were claiming, you know, Unix is so much more reliable than Windows. You're not going to run a web server on Windows. Dell was doing this, you know, with Windows NT, but like, who is running Windows NT now as a web server? All of this was totally wiped out by Linux, like the entire internet runs on Linux. Even Azure, right? Even Microsoft runs Linux.

• 35:35Speaker 01

  So basically, OpenAI, Anthropic, et cetera, today are the Sun Microsystems and HPUX of yesterday.

• 35:49Speaker 00

  Yeah, I mean, I guess implicit in that is, is obviously, you know, I think your, you know, your view of like the limitations of what like, you know, these models can only get so good. And so it'll be possible over time for, for the open source folks to, to catch up.

• 36:03Speaker 01

  They've already run out of data, right? I mean, the, the. The openly available, publicly available data, text data is already all used. There's not more of it, right? So what those companies are doing is licensing commercial copyrighted data or training on synthetic data.

• 36:25Speaker 00

  I guess I'm curious because obviously there's been some impressive results in the last few years that they have been able to drive, you know, post these large scale pre-trainings. You know, I M O Gold, you know, the meter task horizon benchmark keeps going up.

• 36:38Speaker 01

  That's, okay, that's very interesting. Now, think about those two domains, right? Mathematics and code. Those are two domains where the language itself is the substrate of reasoning. It's not the only substrate of reasoning, but a lot of, when you do mathematics, right? The formal way on a piece of paper, not the intuitive stuff, but the, you manipulate language, right? And L L M's are really good at this. So, you know, proving theorems and stuff like that, that's, that's what L L M's are really good at. They're not so good at the sort of, you know, coming up with like good concepts and definitions and things like that. It's more like, here is a problem, solve it. They're problem solvers. Mathematics is not just problem solving, right?

• 37:20Speaker 01

  Most of it is actually a creative act that those things don't do. And same for code. So L L M's are good programmers. They're not software architects. They're not computer scientists, right? But they can program for us. So they're not in a, in a state where they can just, you know, replace humans entirely. It changes the world of humans. So humans now, you know, kind of go one level up in the abstraction hierarchy. And our role is to decide where to build, but like building it, you know, you can, you can get help from L L M's. But okay, that's the, the important point is that L L M's are particularly successful

• 38:04Speaker 01

  at domains where the language itself is the substrate of reasoning, not for anything else.

• 38:10Speaker 00

  Yeah. What would an L L M like need to do to convince you otherwise?

• 38:14Speaker 01

  So like a zero-shot agenting system, right? You have an agenting system, give it a new problem. It's not been trained to solve that, that particular problem, doesn't have a script for it. Is it going to be able to accomplish this task that it's never been trained to solve? And unless this system has the ability of predicting the consequences of its actions, and then use that for planning, it's not going to be able to do it. And you're not going to do this with an LLM. You're going to do this perhaps with a significantly augmented LLM that is capable of, you know, search and planning, blah, blah, blah. And currently, you know, LLMs that do math and code actually do this.

• 38:56Speaker 00

  Yeah.

• 38:56Speaker 01

  Right? Because they search for, you know, sequences of tokens that actually accomplish a particular task and, you know, they can run the code or verify that the proof is correct or whatever. So you have like a way of checking whether something that's produced is correct. But that's not a very efficient way of doing planning. And it only works in domains where this type of search can be performed in token space. What I'm talking about with JEPa is you don't do this in token space. You do this in, you know, abstract thoughts space.

• 39:29Speaker 00

  And I'm sure some people listening might think, well, you know, hey, even if it's inefficient and it works and it works at, you know, at things that are done in token space, that's still a large part of the economy.

• 39:39Speaker 01

  I mean, if it works, it's fine. I mean, there's, again, there's nothing wrong with, you know, using it for what they're good at. It's just not a path towards you whenever you're missing, you know, like a huge domain.

• 39:52Speaker 00

  You seem like, you know, hey, it's going to tap out before it can become a software architect, whereas I'm sure.

• 39:56Speaker 01

  It's not going to tap out. It's just going to have like a limited, you know, ability to be deployed for like, it's going to become like increasingly difficult to kind of deploy it for an increasingly large number, you know, of use cases, because you're going to have to collect tons of training data for each of those use cases. And these are, basically, you're not going to be able to make those systems completely reliable, you know, without hallucinations or dangerous stuff or et cetera, unless those systems have the ability to predict the consequences of their actions, which means they're going to have to have explicit word models.

• 40:30Speaker 00

  Yeah, so I guess the bet against, you know, the 100% accuracy and then also the generalization across different tasks.

• 40:36Speaker 01

  Right.

• 40:37Speaker 00

  I guess, you know, one thing that's so interesting about the way the field has developed is obviously you shared the Turing Award with two others, and I feel like they seem much more convinced of like maybe the power or potential threats or safety risks of LLMs over time. I'm wondering, like, when did your views start diverging?

• 40:53Speaker 01

  Uh, in twenty twenty three.

• 40:56Speaker 00

  And what like drove that in your mind?

• 40:57Speaker 01

  I didn't change my mind. They changed their mind, okay? And at just about the same time, it was basically G P T four. I mean, Jeff basically had was not connected to any of that. He was never really interested in L L M's and discovered uh G P T four, you know, twenty twenty three when it came out. And basically he had an epiphany and said, oh, oh, my God, those systems, you know, are really close to human level intelligence and they have possibly they have subjective experience.

• 41:28Speaker 01

  And he did a quick calculation saying like, okay, the human cortex has about sixteen billion neurons. If you want to um do something like backprop, okay, the brain doesn't do backprop directly. But if it does something like backprop, like some sort of, you know, gradient estimation for some sort of objective function, you probably need like a network of a few neurons to kind of reproduce the functionality of a virtual neuron in a in a neural net. And so I like, let's assume, you know, maybe you need you need a circuit of ten actual neurons to reproduce what a backprop neuron does. Then all of a sudden, your your cortex is only one point six billion neurons. Oh, my God, G P T four is really close to this.

• 42:14Speaker 01

  Okay, so maybe it's a smart, you know, it's gonna get as smart as humans. I do not believe in this claim at all. This is kind of, you know, Jeff's uh uh way of saying, okay, basically, I can retire, I can declare victory. You know, I searched for the learning algorithm of the cortex of my career.

• 42:37Speaker 01

  Maybe I didn't discover what it really was, but backprop seems to be like a good substitute for it. It works really well. And so maybe that's all we need. So I can retire.

• 42:49Speaker 01

  And go around the world and give talks about, you know, the potential

• 42:56Speaker 01

  promises and dangers of AI. That's basically what, you know, I think what is intellectual kind of trajectory has been.

• 43:08Speaker 01

  He's much less vocal about the potential dangers now than he was a year or two ago. He kind of realized it's probably a way to design truly intelligent systems. So first of all, he probably, you know, he realized that current L L M's are not that smart, first of all. And second, that there's probably a need for a few breakthroughs, like conceptual breakthroughs, before we get to human-like intelligence. And third, that the blueprint of those systems will be quite different from L L M's. And we probably have a way of, you know, making them controllable and things like that.

• 43:46Speaker 00

  Yeah.

• 43:47Speaker 01

  I've been saying this for years, but, okay, he sort of discovered this recently.

• 43:51Speaker 00

  Yeah.

• 43:52Speaker 01

  Same kind of, there's a similar thing with Yashra. I think what they are both worried about is the ability of society and the political system to make sure that the benefits of AI will be maximized and AI would not, you know, just profit, you know, make a few rich people even richer and, you know, accentuate inequalities and, you know, cause major catastrophes because of bad usage. Okay, this is not like the doom scenario of AI taking over the world. It's more bad uses.

• 44:31Speaker 00

  Which seems possible with the L L M's of today.

• 44:32Speaker 01

  Which is a danger, but, you know, I don't think it's as apocalyptic as, you know, what some people have claimed it is, certainly not as apocalyptic as what even Anthropic has claimed and has tried to kind of lobby governments into, you know, scaring governments into kind of regulating AI because of that. I don't, I don't, I don't subscribe to this at all.

• 44:55Speaker 00

  They seem to genuinely believe it.

• 44:57Speaker 01

  I think they genuinely believe it, but also I think there is, you know, some kind of commercial, good commercial reasons for them to believe that and to kind of, you know, brainwash some people and government into thinking their systems are dangerous.

• 45:13Speaker 00

  And it sounds like, you know, with these other architectures, do you think they're, because obviously it doesn't, you know, as maybe bearish as you are on LLMs being the end state of everything, you know, you have some pretty ambitious timelines too for these new architectures. And so it doesn't seem like you think we're particularly far away from some very compelling capabilities. How do you think about, I guess, the safety around, you know, if it ends, if these breakthroughs end up coming from newer architectures and whether that should make us rest easier or not?

• 45:38Speaker 01

  I'm going to say something that's, again, might be controversial. And certainly, some of my colleagues at Meta didn't like me saying this, but I think LLMs are intricately unsafe. I don't think they can be made reliable and safe. Okay. They cannot be made reliable because you can't stop them from hallucinating. And if they are agentic, you cannot guarantee they're not going to like take an action that, you know, they didn't predict the outcome of and that.

• 46:05Speaker 00

  I mean, does it surprise you they can do these like 15 hour coding tests, given the concerns around reliability?

• 46:09Speaker 01

  Well, the coding is something where you can actually verify that, you know, the code that you generate, you know, satisfy your specification. But, but not everything is coding. And, and there are examples of, you know, coding agents like wiping up your your hard drive, right? So like, or doing stupid things, right? That makes you lose a lot of money or data or whatever. So I think, I think, you know, LLMs in their current forms are intricately unsafe because they cannot predict the consequences of their actions. And because the way the task that they accomplish is determined is, is subject

• 46:54Speaker 01

  to their training. You know, you give them a prompt, and then they will accomplish a task that corresponds to that prompt only to the, to the extent that their training has conditioned them to actually do the right task corresponding to this prompt. But it is no like, you know, hardwired constraint that will force them to accomplish this task and then, you know, predict that the task would be accomplished properly.

• 47:20Speaker 00

  Yeah. I mean, I think famously in the early days, right? They would, you'd ask them a question and they'd keep asking, they'd keep asking the question, right?

• 47:26Speaker 01

  Well, I mean, also, they don't have common sense, right? So, I mean, there's the the joke that was circulating like a month ago of, you know, I need to wash my car and, you know, the the car wash is a hundred yards from my house. Should I walk?

• 47:43Speaker 01

  I tried it again, like, maybe two weeks ago. Uh, they all say, yes, you should walk, except Gemini.

• 47:49Speaker 00

  So they're training on your video of of having done, uh, having given that speech before?

• 47:53Speaker 01

  It was not my video because I didn't come up with this.

• 47:55Speaker 00

  Or whoever came up with it.

• 47:57Speaker 01

  Yeah, right, whoever came up with it. But there are a few instances, right, where where I said, like, you know, an L N M can do this, and then six months later, it was people are doing it. And it's simply because, you know, as soon as people watch the podcast of me saying L N M's can do this, they, of course, type it into chat G P T. So now it becomes part of the training set.

• 48:14Speaker 00

  Right.

• 48:15Speaker 01

  And now, of course, you know, the next version has that, uh, you know, that that thing in the fine tuning set. And of course, it can answer the question, but it's not because it's it becomes smart all of a sudden. It's just because it was explicitly trained with that question. So L N M's are interestingly unsafe. Uh, I don't think there is any way to fix that in the current um paradigm. Um, and what I've been proposing is, uh, the architecture I've been talking about is objective driven A I. So basically, you give an objective to uh an A I system, which is accomplish this task. Now, how does the system knows it will accomplish this task? It has a word model and it predicts, uh, you know, the outcome of a sequence of actions

• 49:00Speaker 01

  it imagines taking. Uh, and if this uh outcome satisfies a cost function that, you know, describes to what extent the task has been accomplished or not accomplished, then that system, if if the way that system works is by optimize by optimization, finding a sequence of actions that accomplishes this task, minimizes this cost according to its word model, it can do nothing else.

• 49:29Speaker 00

  Yeah.

• 49:30Speaker 01

  Okay? And. Of course, there's many things that can go wrong there. In particular, the cost function might be inaccurate. It could be that the cost function you think is actually measuring to what extent the task had been accomplished, but perhaps it's not accurate. Okay. The one model might be inaccurate. So the prediction that the system makes is actually not the right one. So its prediction of what was going to happen as a consequence of its action wasn't right. Okay, so the system can still make mistakes, but it can predict the consequences of its actions to some extent, which is, I think, indispensable for any agentic system. Now, what you can add to that system is not just a cost function that guarantees a task

• 50:12Speaker 01

  has been accomplished, but you can also add a bunch of other objective functions, other cost functions, or even constraints that are safety constraints. They say, okay, you know, don't hurt anybody on the way, right? And you cannot specify this at an abstract level, but you can have, you know, low-level objective functions that, put together, will guarantee that the system will not be dangerous. And the system cannot violate those things. By construction, it will have to satisfy those conditions. Not the case for an LLM. The LLM can always escape. There's a gap between your training error and test error. There's always going to be a prompt where the system is going to do really stupid things.

• 50:52Speaker 00

  To talk through one specific space around LLMs, like, you know, I think you're obviously really excited about AME and healthcare, and I think, and people have been using LLMs in healthcare for all sorts of things. And so I'm curious how you think about, like, the set of things where LLMs are just not going to work in healthcare and you need, like, a model that understands the world better.

• 51:09Speaker 01

  So, I mean, designing a course of treatment for a chronic disease, for example, or even a non-chronic disease, for a particular patient, which may not completely fit into, you know, templates that you've observed before, but if you have a good mental model of the dynamics of the physiology of the patient, then you might design a course of treatment that will actually bring the patient to a good state.

• 51:34Speaker 00

  Yeah.

• 51:35Speaker 01

  When I'm seeing, and when I'm seeing a patient, it can be a cell. Okay, how do you tell a stem cell to turn into a pancreas beta cell that produces insulin? Okay, you have a patient with type 1 diabetes, and, you know, they have, you know, their immune system basically, you know, kind of... eats up their own beta cells, right? It's autoimmune. How do you keep making beta cells? You know, can you send a message? Do you have a model of a human cell that will allow you to figure out what sequence of message you need to send to a stem cell so that it turns into a beta cell?

• 52:16Speaker 00

  The less LLM pilled camp and the LLM pilled camp talk past each other. But it's like, I think it's actually very possible that both what LLMs can do, which is maybe scaling what a top doctor, the treatment you get at like the top doctor or the top place, scaling that around the world, like unbelievable potential impact of that, right? If you're able to do that. And then, you know, I think what you're talking about, which is certainly still on the come for, for a lot of these things is, okay, well, even better than the top doctor. Like, how do you, how do you go do that?

• 52:42Speaker 01

  But it's more than just a top doctor, right? Because, I mean, what a LLM can do well is, you know, it can sort of regurgitate knowledge that you can read in books, mostly. But if medicine was only kind of about accumulating declarative language, that declarative knowledge that exists in books, you can be a doctor by just reading books. And you can't be a doctor by reading books. You have to do, you know, residency and, you know, actually kind of listen to the heart and like press on the belly and things like that to, you know, diagnose appendicitis or whatever it is.

• 53:18Speaker 00

  Yeah. It's interesting. I would be very curious to see whether LLMs themselves can provide like, you know, top quality healthcare globally. We'll have to check back in on that one. It seems, it seems like they're pretty, pretty close. You know, I definitely also want to hit on your, your time at Meta because you spent over a decade building like one of the most respected research labs in the world. You know, obviously you recently left. As you reflect back on, on the time there, what do you think you got like most right and most wrong in your time running FAIR?

• 53:41Speaker 01

  So the thing we got right is, you know, building a, a, a top research lab that really sort of innovated, produced a lot of the sort of basic methods and science and tools like PyTorch that are useful to the entire industry, right?

• 53:59Speaker 01

  I mean, the entire industry is built on A hundred percent. basically, except for a few people at Google.

• 54:05Speaker 01

  And I think a culture of, you know, openness and, and, and kind of, you know, scientific process, which I think is necessary for breakthrough innovation. Yeah. Because, you know, there, there is a lot of, there's a whole chain of innovation, right? You have Blue Sky Research, new concepts, a lot of that takes place in universities. Some of that takes place in advanced research labs in industry, which can be counted on the fingers of one hand. You know, Google is a good one, you know, FAIR was a good one. Hopefully it will still be, I'm not sure. And, you know, a few others.

• 54:47Speaker 01

  Then you have, okay, this is a good idea, let's push it forward and see, see if it can be made useful, but still at a research level. In a, in a sense of, we're not going to fool ourselves. We're not going to try to just, you know, find a solution that just works for this problem. We, we're going to see if this technique that we imagine or we picked up from other people in the community can actually be pushed and, and be, be made practical, not as a product, but like we can show that it beats some record on, you know, some task or benchmark. And then the next stage is for the, the company that hosts the research lab to say, okay, now we're going to push the button,

• 55:32Speaker 01

  devote a, you know, big engineering effort to that, to that vision, and then push it forward. That is where a lot of projects fail. That's, that's where a lot of companies kind of fail to pick up. Meta was actually pretty good at this, okay? But far from perfect. It was not like, you know, textbook example of how you do it wrong, like, you know, Xerox PARC, like totally missing out on, Yeah. you know, GUI interfaces and, you know, mouse and windowing systems, right? Meta was, you know, kind of missed a few steps, essentially. And it, and it's partly, partly just organizational. It's partly because you need a, an organization that is pretty close to research,

• 56:17Speaker 01

  but not completely a product organization, to take the relay of, you know, pushing a technology a little further.

• 56:25Speaker 01

  Not making product with a three months deadline, but like, you know, pushing things.

• 56:30Speaker 01

  And we had that at one point.

• 56:32Speaker 00

  Yeah.

• 56:33Speaker 01

  At, at, at Facebook and Meta. And then we lost it. And FAIR was basically isolated within the company, had lots of ideas that nobody picked up on. And then in 2023, the Gen AI organization was created by basically taking about sixty or seventy scientists and engineers from FAIR, right, initially, and then it built up.

• 56:57Speaker 01

  But then it was under so much short-term pressure that basically that organization, Gen AI, didn't have time to talk to FAIR. And so instead of being at the forefront and innovating in LLM,

• 57:12Speaker 01

  Gen AI basically had to focus on short-term things and became very conservative. And so there was a gap, basically impetus mismatch between research.

• 57:21Speaker 00

  Yeah.

• 57:21Speaker 01

  And.

• 57:22Speaker 00

  Is that kind of what happened with Lama four?

• 57:24Speaker 01

  Yeah. Well, even with, you know, Lama three, starting with Lama three. So Lama one was a small project within FAIR. 2022, early 2023, Gen AI was created. The Lama people were basically moved to Gen AI. They started working on Lama two. And then a bunch of them realized, like, I could do a startup. So that was the genesis of Mistral.

• 57:49Speaker 00

  Yeah.

• 57:50Speaker 01

  Okay. Two of the authors of Lama one basically created Mistral with another guy from Google. And, and, you know, a few people kind of left and sort of did other things. This is not a kind of a happy time at, at Meta for various reasons. And so there were, you know, a bunch of people kind of left. And then the, the, the Gen AI organization, which kind of took over Lama.

• 58:20Speaker 01

  Two to some extent and La Maître, three and four was under so much short term pressure that it became very conservative. And, you know, it's a combination of disparity of the groups, but like pressure from the leadership. And, I mean, there's many ways things can go wrong and you can't blame anyone in particular. But, but yeah, that's kind of what happened.

• 58:43Speaker 00

  I mean, it feels like a lot of these organizations obviously are under short term pressure right now because there's just an incredible race going on. And so I'm curious, like, obviously, this, this, you know, FAIR setup you had and kind of there's a similar one, you know, at Google for many years and certainly many researchers running around OpenAI, Anthropic, trying many different things. Do you think like that is still possible going forward or like is the only, you know, is one of the only paths to leave and do your own company or, or, you know, are there still places within the industry that you think have this like original ethos of FAIR even amidst the race that is race dynamics that are happening?

• 59:15Speaker 01

  I think there are a few places within Google Research and DeepMind that where people actually do research. But increasingly, the industry has become more kind of closed, right? I mean, Google certainly climbed up and, you know, Meta and FAIR even is kind of going a bit in the same direction. There are restrictions on publication now, like more restrictions. And so it's sort of less appealing for people who really want to kind of do breakthrough research and, you know, they don't get as much resources. If they do something that is relevant in medium term, they are told not to talk about it. And, and so it's, it's not, you know, it's not a good atmosphere, I think, for, for breakthrough. It's not conducive, you know.

• 59:58Speaker 01

  And basically, the best way to get breakthrough research of the type that, you know, you were getting in the early days of FAIR and, you know, at Bell Labs in the good days and Xerox PARC is you hire the best people and those are people who have a good nose to know what to work on, what projects to kind of attack. You give them the means to succeed and you get the fuck out of the way. All right. Pardon my French.

• 1:00:27Speaker 00

  Yeah, I mean, I'm curious like what you what impact that then ends up having on the broader research communities. Obviously, one of the legacies of FAIR is you trained, you know, so many researchers, right? And like, they're all throughout the ecosystem. And it feels like now the maybe equivalents of those people that came in younger in their careers at FAIR, you know, they're joining these these labs with maybe shorter term priorities and focus. And I guess I'm wondering, like, you know, in this current ecosystem, where it feels like a lot of younger people getting into the field are thrust much more into these like, short term dynamics, does that change anything about the way the ecosystem evolves?

• 1:00:57Speaker 01

  Well, I mean, the people who tend to want to work with me are generally people who,

• 1:01:04Speaker 01

  you know, sufficiently crazy to do it, first of all.

• 1:01:07Speaker 00

  Very, very fair.

• 1:01:07Speaker 01

  And or, you know, kind of subscribe to the whole idea that in academia and during your PhD, you should work on the next generation of AI systems, you shouldn't work on the current generation.

• 1:01:20Speaker 00

  Yeah.

• 1:01:21Speaker 01

  Like, if you work on LLM in academia now, it's incredibly boring. At least, to me, it's boring. It's basically kind of studying how how and why LLMs work and explaining why they work or what the limitations are. It's like descriptive science. It's really not, you know, kind of creative, very creative. Like, I don't find that particularly interesting. It's useful.

• 1:01:41Speaker 00

  Yeah.

• 1:01:43Speaker 01

  And, you know, if you really want to kind of show how to do new things with LLM, like, you're not going to have the GPUs you need for that.

• 1:01:50Speaker 00

  Totally.

• 1:01:51Speaker 01

  So like, forget that. Like, don't work on LLM if you're doing a PhD. Like, there's no point. You cannot contribute.

• 1:01:57Speaker 00

  How do you know it was time to leave Meta? It sounds like it was, you know, you were thinking through some of these things over a period of time. You know, was there a moment that it crystallized or?

• 1:02:04Speaker 01

  Well, it was a combination of things, right? So first of all, you have to understand, a lot of people have like, completely wrong idea about what my role at Facebook and Meta was. So I joined in late 2013, really kind of started early 2014. The first four and a half years, I was director of FAIR. So I built the FAIR organization, set up the culture, hired the key people, and sort of managed it. And after four and a half years, I stepped down from that role for a number of reasons. And I became chief AI scientist. Okay, so the reason is, you know, I was

• 1:02:49Speaker 01

  Basically getting close to uh turning sixty. First of all, fifty eight and uh I just don't want to do management. Okay, I mean, I was ready to do it for a while to get the the organization started, but I'm just not good at it. It's not the thing I'm I'm more like a, you know, scientific or technical visionary and engineer and scientist. So uh other people are much better at management than I am. So I basically stepped down, uh two other people, uh Joel Pinot and uh Antoine Bard, basically Yeah. took over uh the directorship of of FAIR and I became chief AI scientist.

• 1:03:33Speaker 01

  So um I was reporting to the CTO and uh and, you know, had walls of

• 1:03:44Speaker 01

  basically restarting a research project that I thought was necessary because the ambition of FAIR was always to build intelligent systems, right? And I thought, you know, I put my own research in in parentheses while I was running FAIR. I just didn't didn't have the time. And I thought it was important to basically kind of design the architecture of of like human level, you know, human-like AI systems. Uh and, you know, I had come up with the concept that this was going to be based on self-supervised learning and on, you know, prediction from sensory signals like video, things like that.

• 1:04:27Speaker 01

  I mean, this is, these are old ideas and uh and word models. I actually gave a keynote at NERIP in twenty sixteen where I I said like this is the way AI research should go, like word models predict, you know, consequences of your actions and plan. And I said like, you know, RL is not the thing that will take us there because it's too inefficient. Supervised learning has shown its limits and so the future is self-supervised learning and word models. So how do we do self-supervised learning on wild models? And I started a few projects on this with like a few avenues that didn't pan out. Some projects on video prediction and stuff like that. And then came up with this concept that you could train self-supervised learning from video.

• 1:05:10Speaker 01

  But you have to train the system to make prediction in representation space. So that's the idea of JEPAR. And if you have JEPAR, you can turn it into a wild model by making it action condition. And then you can use it for planning. So I had this idea around 2020 and in 2022, I wrote a long vision paper. So I said, I'm just gonna write a paper with my entire vision, okay? Spill all my secrets, like I don't care. But maybe that will rally a bunch of people to that vision. And boy, did it work. Because not only did I rally, you know, a bunch of students who kind of came working with me at NYU or in Paris because they wanted to work on this, but also a whole team at FAIR who said like, this sounds great, that's what we want to work on.

• 1:05:53Speaker 01

  And then Joel Pinault said, well, maybe this should be like a major mission of FAIR. We called it Advanced Machine Intelligence. That was the internal name of the project.

• 1:06:05Speaker 00

  Interesting. And I'll actually leave with it.

• 1:06:08Speaker 01

  And now it's the name of the company. And, you know, Mark Zuckerberg, you know, kind of, kind of read that paper and knew what it was about and subscribed to the project and Andrew Botwurst, the CTO also. And Mark Schrepper, the previous CTO. Chris Cox, who was my, my direct manager, chief product officer, also loved the idea. So like, you know, there's a lot of support in the leadership about this project that we internally called AMI.

• 1:06:37Speaker 01

  And, you know, and, and it started really kind of working for, for video.

• 1:06:47Speaker 01

  But then, you know, company kind of refocused all of its effort on LLM. Despite support from Mark and Andrew.

• 1:06:57Speaker 01

  Buzz, we call him Buzz.

• 1:07:00Speaker 01

  You know, the all the layers below, like, didn't see the point, I think. And so politically, it sort of became a little difficult.

• 1:07:10Speaker 01

  The applications, as I said, of Japan War Model are, there are applications in like, you know, wearable agents and stuff like that, but and robotics, but. But Meta chose to get rid of its entire robotics AI group.

• 1:07:26Speaker 01

  That was led by Jitana Malik, who is now Amazon. And so, you know, clearly, it wasn't the right environment anymore. Most of the applications were in industry that Meta had no interest in. FAIR was increasingly getting pressure to kind of basically help MSL with LLMs.

• 1:07:48Speaker 01

  So, yeah, you know, it may clear, it may clear. And, and that, you know, throat ramming, worked really well with investors too. Because when I had to raise money for AMI, everybody knew my story. And anybody knew, you know, many investors, you know, staff at various VCs that read my paper and. Or I'd listen to my talks and had bought my story that were realizing, you know, LLMs had limitations. And, you know, were kind of, yeah, interested by the idea of like building the next generation AI systems.

• 1:08:24Speaker 00

  I guess was was like the scale acquisition, like part of this catalyst of of like the pure LLM focus internally?

• 1:08:29Speaker 01

  Yeah, definitely. I mean, there's probably some, you know, other reasons to it, I think, you know, maybe.

• 1:08:35Speaker 01

  I don't have any sort of inside information to comment on this, but it's possible that Mark sees in Alex kind of a potential successor to himself, like a younger version of himself.

• 1:08:45Speaker 00

  Yeah, I feel like, like, a lot of the popular narrative or, you know, in the media has been like, oh, like, you know, when Alex comes in, it then gets harder to run like a research organization. You know, I don't know if that to the extent you felt that or.

• 1:08:56Speaker 01

  Well, okay, so here's a big misconception about my role, my relation to Alex and how AI was run at Meta. I had zero technical contribution to Lama, like none whatsoever. My one contribution to Lama. Was to argue for open sourcing Llama 2, because it was a big internal debate whether we should open source. Like, the legal department was against it, the policy department was kind of against it, uh, the comms department was for it, all the engineering side was for it, like Buzz was for it. Uh, so there were like enormous internal discussions at a very high level, you know,

• 1:09:37Speaker 01

  forty people from Mark Zuckerberg down every week for two hours, uh, for months. So, so really it was, you know, kind of a, a big debate internally and I really, really, you know, pushed, um, argued for, for the fact that, uh, you know, and Buzz also was, was very vocal about it, that, um, the, uh, you know, uh, safety risks were basically overblown, uh, the opportunities to create an industry were extremely strong, um, and that we were gonna jumpstart the AI industry by open sourcing Llama 2 and in fact, that's exactly what happened. So, but I had zero contribution to, to Llama, positive or negative.

• 1:10:20Speaker 01

  Like, I, I didn't do anything to stop it or slow it down or anything. There was a lot of people working on LLMs within FAIR and it was fine, uh, yeah, I never said anything against it, okay, um, other than saying, this is not a path to human intelligence, but it's fine, uh, it's useful. Uh, you know, same thing for speech recognition or translation, right? Uh, so, uh, and particularly since, uh, twenty eighteen, when I stepped down from being director of FAIR, uh, I didn't have any direct influence on what people were working on, other than, you know, basically publishing my, my vision and then rallying people, uh, around, uh,

• 1:11:03Speaker 01

  around my project, but, you know, they, they were working with me because they wanted, not because I was their boss. I wasn't telling them to work with me. Um, and so, um...

• 1:11:16Speaker 01

  So I had no positive or negative influence on LLM, okay, within within Meta.

• 1:11:24Speaker 01

  And I had some influence on the strategy, but it was more like the long term and like how you maintain a research lab and things like this. And in the last year, and you know, I mean, starting maybe early 24, and certainly in 25, the the way FAIR was kind of the direction in which it was moved and managed basically did not correspond to what I thought was necessary to preserve.

• 1:11:51Speaker 01

  You know, innovation, research and breakthrough and preserve the good people, like a lot of good people have left already.

• 1:11:57Speaker 00

  Yeah, and I guess a lot of, you know, it probably was harder to get people to work on the stuff you were working on internally. And I'm sure there's pressure for you yourself to work on a lot of the LLM stuff.

• 1:12:05Speaker 01

  Yeah, yeah, no, but a lot of other people also have left.

• 1:12:08Speaker 00

  Yeah, no, it's fascinating. And one thing I'm struck by throughout our whole conversation is I feel like you're you've like had a remarkably consistent point of view, like, you know, on the in the space like for a long time. And you can go back to your, you know, a bunch of the earlier talks you referenced. You know, obviously it is a fast moving space and a ton of interesting things have happened in the last year. What's like one thing you've changed your mind on in the last year?

• 1:12:30Speaker 01

  I mean, the whole idea of what we used to call unsupervised learning that we now call self supervised learning. You know, until about 2003, the whole idea of unsupervised pre-training, where you get a good representation for the input data. And then you either fine tune the model with a little bit of supervised label data. And it sort of give us, you know, some evidence that this whole technique could work. I tried to apply this to video because ultimately what I wanted to do is train a system to understand how the world works by just watching the world go by, right? I mean, that's the basic idea. And sort of started to argue for this in the sort of, you know, early 2010s.

• 1:13:15Speaker 01

  Did some some work on. Simple video prediction, we didn't have GPUs, okay.

• 1:13:22Speaker 01

  And then sort of doing this more seriously about after the creation of FAIR, by doing pixel level video prediction, realizing that wasn't working. But then arguing for self-supervised learning, okay? This whole idea of like training a system generically not to solve a task, but to basically just predict. And then using the representation that is learned this way as input to a downstream task that you can train supervised or reinforcement or whatever. So that was a bit of the topic of my second half of my keynote at NIPS in 2016. It was still called NIPS at the time.

• 1:13:58Speaker 00

  Yeah, of course.

• 1:13:58Speaker 01

  In 2016. And then I kept kind of, you know, kind of pushing for this idea and tried to kind of discover some methods to get that to work. And what surprised me is that that became incredibly successful, but not for video, for language. LLMs basically are a blindingly successful example of

• 1:14:21Speaker 00

  That they are.

• 1:14:22Speaker 01

  Self-supervised learning.

• 1:14:23Speaker 00

  No, that they are. Well, I feel like that's almost like the perfect note to end on, but I want to make sure to leave the last word to you. I feel like there's, I mean, all our listeners are very familiar with you, but I want to at least give you the mic to point them to anything that you think they should, they should check out with some of the new stuff you're doing, or I don't know, any of your work you want to point to. The mic is yours.

• 1:14:41Speaker 01

  Okay, let me tell you one thing. An LLM works because when you have a sequence of discrete symbols, making predictions is easy. Because there's only a finite number of possible symbols in your language.

• 1:14:57Speaker 01

  A hundred thousand possible tokens or something like that, right? And you can have your neural net produce a probability distribution over all possible tokens. And then you can sample from that distribution, shift the token into the input and then produce the next token. And you can do autoregressive prediction, okay? So that's a special case. If you have the real world, you can't use a generated model. So now you have to train a system that learns a representation and makes prediction in the representation space. There's a big issue with this, which I didn't think until about five years ago that was. Easily solvable, even though I invented one technique to solve it, you know, decades before that. And it's the problem that if you take two inputs, let's say the initial segment of

• 1:15:42Speaker 01

  a video and the continuation of that video, or you take one image and a corrupted version of it, you run them both through an encoder and you train a predictor to predict the representation of one from the representation of the other. It's a very simple solution where the system basically predicts a constant representation and now the prediction problem becomes trivial. That's called a collapse, representation collapse. So the big question of self-supervised learning for JEPA, for the Joint Embedding Architecture, is how do you prevent collapse? The solution that I came up with many years ago, 1993, is contrastive learning. So basically you have examples of things that should be predictable from one another and an example of things that should not be predictable from one another.

• 1:16:27Speaker 01

  It turns out this method works, but it doesn't scale with dimension. It doesn't scale very well. There's another technique that was actually invented by Jeffington and Sue Becker in the late 90s, late 80s, I'm sorry, where you have those two networks and you try to maximize mutual information between them. Yükan Şmirüver is mad at me because he also came up with a version of this in 1992 and he says that's JEPA. It's not JEPA. It's just another way of preventing collapse of a joint-Embedding architecture. Okay. Which is fine, but it's not, you know, it's a particular way of doing it, which I don't think is particularly good.

• 1:17:11Speaker 01

  So, okay, so now you have this JEPA architecture. You have to come up with a good way of preventing collapse. And there is a couple of ways. So, as already said, contrastive methods, I think, is not a good, a good approach. There's another set of methods that are kind of called distillation methods.

• 1:17:33Speaker 01

  And they do prevent collapse, we don't know why. So a good example of that is Dino or Dino.

• 1:17:39Speaker 00

  Yep.

• 1:17:40Speaker 01

  Um, that's a joint invading method using the distillation method. Basically, one of the encoders trains the other one is like used as a teacher for the other encoder. Uh, and the encoder that is being trained, you do backprop through it. The one that is not being trained, you don't do backprop, but you share the weight with the other one with some exponential moving average. It's a collection of recipe. There was a paper from from DeepMind about it, could be way well. Bootstrap here on latent which uses this trick. That trick is derived from some intuition from reinforcement learning and somehow prevents collapse. But we don't know why. Okay. There's a few theoretical papers on it that explain why it possibly might work in some simple cases, but it's not satisfactory.

• 1:18:25Speaker 01

  The function, the cost function, you think you're minimizing, you're not actually minimizing, and so you can't monitor. It actually goes up when you train. So we don't like this method, but it works. And some of the models we've trained, large scale video representation learning system, VJPA, VJPA two, VJPA two point one. They train using this method, uh, IJPA also. But we're moving away from this. And now we have, uh, a few papers that came out recently on a explicit regularizer to prevent this collapse, which basically tries to maximize the information content coming out of the encoder. So it's in the same family as the Becker and Hinton from eighty nine and the Schmiedl-Burton nineteen ninety two and a bunch of others since then.

• 1:19:10Speaker 01

  And to some extent, also contrastive techniques, although it's not, although it's not sample contrastive.

• 1:19:17Speaker 01

  And then the question is, how do you measure information content? How do you maximize the information content coming out of a neural net? And the problem is, if you want to maximize the quantity, you either need to be able to measure it or you need to have a lower bound on it.

• 1:19:31Speaker 00

  Yep.

• 1:19:32Speaker 01

  Uh, information content, we only have upper bounds. We cannot measure it, we can only come up with upper bounds. And so we take an upper bound and we cross our fingers. Okay, and it kind of works. So the latest one is called Sigreg. That means sketch is isotropic Gaussian regularization. We had a previous one called. VC reg or VC reg variance, invariance, covariance, regularization. And the C reg stuff is really cool. So this is some work by Randall Balistrier, who was a postdoc with me, is a assistant professor at Brown now. And it basically consists in forcing the distribution of variables coming out of the

• 1:20:17Speaker 01

  anchor to be joint Gaussian, essentially, so maximize information, if you want. It's just a very different way of doing it than, you know, what Jürgen Schmidhuber and Sue Becker and Jeff Inte were doing. And so this is super promising, in my opinion, and we have, you know, variations of it, you know, one that can produce sparse representations, another one that can produce isotropic representations, but not necessarily Gaussians. And we have a paper with Randall, a student at Mila, Luca Maes, that where we train a world model with this. It's still a small scale, but we think it's super promising.

• 1:21:02Speaker 01

  So if you want to read one paper, read that paper, it's LE World Model, L E World Model.

• 1:21:10Speaker 00

  Awesome, we'll definitely link to it, too.

• 1:21:11Speaker 01

  Yeah, I'm not responsible for the name. Randall picked up the name.

• 1:21:15Speaker 00

  Amazing. Well, Jan, seriously, thank you so much. It is such a privilege to get to spend the last bit of time with you and really appreciate coming on the podcast.

• 1:21:23Speaker 01

  Thanks for having me, that was fun.

• 1:21:24Speaker 00

  I'm Jacob Effron, and this has been Unsupervised Learning, a podcast where I get to talk to the smartest people in AI and ask them tons of questions about what's happening with models and what it means for businesses in the world. As I hope is clear, I have a ton of fun doing this. It's a nights and weekends project in addition to my day job as an investor at Redpoint. But our ability to get these incredible guests on really comes from folks like you subscribing to the podcast, sharing it with friends. It's really what ultimately makes this whole thing work. And so please consider doing that. And thank you so much for your support and listening. We'll see you next episode.