Yann LeCun: Human Intelligence is not General Intelligence

Summary

AI Inside Podcast Episode Summary

Episode Title: Yann LeCun: Human Intelligence is not General Intelligence
Release Date: April 9, 2025
Hosts: Jason Howell and Jeff Jarvis
Guest: Yann LeCun, Chief AI Scientist at Meta and Turing Award Winner

---

Introduction

In this enlightening episode of AI Inside, hosts Jason Howell and Jeff Jarvis engage in a deep conversation with Yann LeCun, a luminary in the field of artificial intelligence and the Chief AI Scientist at Meta. The discussion centers on the limitations of current AI models, particularly Large Language Models (LLMs), and the path forward towards achieving true artificial general intelligence (AGI).

---

Understanding the Limitations of Current AI Models

Yann LeCun begins by addressing the practical utility of LLMs, acknowledging their effectiveness in areas like coding assistance and general AI assistant roles. However, he emphasizes the gap between impressive demonstrations and the reliability required for everyday deployment:

"There's a big distance, it's much harder to make those systems reliable enough."
[04:00]

LeCun draws parallels with the development of self-driving cars, illustrating how initial breakthroughs often fall short of consistent, real-world performance. He critiques the recurring trend in AI where each new paradigm promises imminent human-level intelligence, only to reveal unforeseen limitations.

---

The Quest for a World Model in AI

Central to LeCun's vision is the concept of a "world model," an internal representation that allows machines to predict and reason about the physical world. He underscores the necessity of machines understanding the world to achieve reasoning and planning capabilities akin to humans and animals:

"We need machines that understand the physical world. We need machines that are capable of reasoning and planning."
[07:33]

LeCun explains that current AI systems lack this foundational understanding, making it challenging to tackle novel problems. He introduces the Joint Embedding Predictive Architecture (JEPA) as a promising approach to developing these world models, moving away from purely generative models towards more predictive and representational frameworks.

---

Challenges in Developing General Intelligence

When probed about the feasibility of AGI, LeCun provides a nuanced perspective. While he is optimistic about machines eventually matching human intelligence in various domains, he asserts that the path is fraught with complexities:

"It's almost certainly harder than we think and probably much harder than we think."
[19:26]

He critiques the notion of "general intelligence," arguing that human intelligence itself is not truly general but highly specialized for survival-related tasks. This specialization limits the applicability of the term "general intelligence" and highlights the challenges in replicating even this specialized form of intelligence in machines.

---

Future Directions: Beyond Language Understanding

LeCun emphasizes that true intelligence extends beyond language manipulation. He contrasts the capabilities of LLMs with the sophisticated, non-linguistic reasoning observed in animals:

"Most types of reasoning have nothing to do with language."
[25:30]

He highlights the importance of hierarchical planning and abstract representation in accomplishing complex tasks, such as planning a trip or manipulating objects. These processes require an internal world model that current AI systems lack, positioning them far from achieving the intuitive and adaptive problem-solving seen in living beings.

---

Meta’s Open Source Strategy and Its Impact

Jeff Jarvis brings the conversation to Meta's strategic decision to open-source their LLaMA models. LeCun explains that this move is designed to act as a "spoiler for exactly three companies" while enabling thousands of others to innovate:

"The best way we know how to do this in the context of academic research is you publish your research, you publish your code in open source as much as you can and you get people to contribute."
[37:42]

LeCun believes that open-source models democratize AI development, fostering a diverse ecosystem where ideas can emerge from anywhere. This approach not only accelerates innovation but also ensures that AI advancements are not monopolized by a few large corporations.

---

The Role of AI Assistants and Cultural Diversity

Discussing the future of AI assistants, LeCun envisions a scenario where intelligent virtual assistants accompany individuals, understanding and adapting to their cultural and linguistic contexts:

"You want a high diversity of AI assistant that first of all speaks your own language, whether it's an obscure dialect or local language."
[46:49]

He stresses the necessity for these assistants to reflect diverse value systems and biases, akin to the diversity seen in traditional media. This ensures that AI remains relevant and respectful of varied cultural nuances, preventing a homogenized digital experience.

---

Concluding Thoughts

As the conversation wraps up, LeCun reiterates his optimism about the future of AI, tempered by a realistic understanding of the challenges ahead. He underscores the importance of collaborative efforts and open-source contributions in overcoming the hurdles towards achieving advanced machine intelligence.

"We're not there yet. It's the big challenge of AI for the next few years."
[34:17]

Hosts Jason Howell and Jeff Jarvis express their gratitude for LeCun's insights, highlighting the value of his realistic perspective in navigating the rapidly evolving landscape of artificial intelligence.

---

Key Takeaways

• Current AI Models: While LLMs have practical applications, they fall short in reliability and general intelligence.
• World Models: Developing internal representations of the physical world is crucial for reasoning and planning in AI.
• AGI Challenges: Achieving human-level intelligence in machines is significantly more complex than anticipated.
• Open Source Strategy: Meta's open-source approach aims to democratize AI development, fostering widespread innovation.
• Cultural Diversity in AI: Future AI assistants must accommodate linguistic and cultural diversity to remain effective and respectful.

---

This episode provides a comprehensive exploration of the current state and future directions of artificial intelligence, offering listeners a grounded and insightful perspective from one of the field's foremost experts.

Transcript

[Yann Lecun]

As a contractor, I don't pay for materials I don't use, so why would I pay for stuff I don't need in my mobile plan? That's why the new My Biz Plan from Verizon Business is so perfect. Now I can choose exactly what I want and I only pay for what.

[Jason Howell]

I need right now with my biz plan. Get our best price as low as $25 a line. Visit verizon.com business to get started today. Price per month with 5 plus lines.

[Yann Lecun]

Includes auto pay and paper free billing.

[Jason Howell]

And special intro offer discounts, taxes, fees, economic adjustment charge and terms apply. Offers end June 10, 2025 I think you're on mute Workday starting to sound the same.

[Yann Lecun]

I think you're on mute.

[Jason Howell]

Find something that sounds better for your career on LinkedIn. With LinkedIn job collections, you can browse curated collections by relevant industries and benefits like Flexpto or hybrid workplaces so you can find the right job for you. Get [e-mail] jobs finding where you fit LinkedIn knows how Meta's chief AI scientist and Turing Award winner Yann Lecun joins us to talk about why current LLMs are less intelligent than a house cat, how developing world models that understand physical reality remains AI's biggest unsolved challenge, and why Meta's open source approach to LLAMA is enabling thousands of companies while disrupting just three. That's coming up right after this. This is AI Inside episode 63, recorded Wednesday, April 8, 2025. Human intelligence is Not General Intelligence with Yann Lecun. This episode of AI Inside is made possible by our wonderful patrons at patreon.com aiinsideshow if you like what you hear, head on over and support us directly. And thank you for making independent podcasting possible. Hello and welcome to AI Inside, the show where we take a look at AI that's layered throughout so much of the world of technology. I'm one of your hosts, Jason Howell, and this week's episode is definitely a little different and I want to get right to it. But just real quick before we do, huge thank you to those of you who support us directly on Patreon. That's patreon.com aiinsideshow Corky Garko, you know who you are, you're a huge supporter and we appreciate you. So thank you so much for that. All right, Jeff Jarvis, my co host and I had the chance to chat with Yann Lecun, very notable figure in the world of AI. This happened last Friday. That's when this interview was recorded and it was pretty incredible. So I'm not going to waste any time. Let's just jump right into it right now. Thrilled to welcome to AI inside Yann LeCun, Chief AI Scientist at Meta. Turing Award winner, known by many as the Godfather of AI. Welcome to the show, Jan. It's really nice to meet you.

[Yann Lecun]

Thanks for having me on.

[Jason Howell]

Yeah. Does it ever get old hearing someone introduce you as the Godfather of AI? You're kind of like, yeah, here we go again.

[Yann Lecun]

I shut my ears so I don't turn red.

[Jason Howell]

But you can accept it at this point because it's the truth. The kind of question that I have to kick things off is that we are so firmly implanted into the kind of current kind of realm of artificial intelligence, which really seems to be the LLM generation, and there's probably something on the horizon around that, but we're still firmly planet in there. And you've been pretty opinionated on the limits of LLM at a time when we're also seeing things like OpenAI securing a record breaking round of funding largely built on its success in LLM technology. And so I see, you know, diminishing returns on one side, on the other, companies betting everything on generative AI and LLM. And I'm. I'm curious to know what you think as far as why they might not be seeing what you're seeing about this technology. Or maybe they are, they're just approaching it differently. What are your thoughts there?

[Yann Lecun]

Oh, maybe they are. There's no question that LLMs are useful, I mean, particularly for coding assistant and stuff like that. And in the future, probably for more general AI assistant jobs, people are talking about agentic system. It's kind of still kind of not totally reliable yet. It's a bit like, for this kind of applications. The main issue, and it's been a recurring problem with AI and computer technology more generally, is the fact that you can see impressive demos. But when it comes time to actually deploy a system that's reliable enough that you put it in the hands of people and they use it on a daily basis, there's a big distance, it's much harder to make those systems reliable enough. Ten years ago, we were seeing demos of cars driving themselves in countryside streets for about 10 minutes before you had to intervene. And we made a lot of progress, but we're still not to the point of having cars that can drive themselves as reliably as humans, except if we cheat, which is fine, which is what Waymo and others are doing. But so there's been sort of a repeated history over the last 70 years in AI of people coming up with a new paradigm and then claiming, okay, that's it, this is going to take us to human level AI. Within 10 years, the most intelligent entity on the planet would be a machine. And every time it's turned out to be false. Because the new paradigm either hits a limitation that people didn't see or turned out to be really good at solving a subcategory of problem that didn't turn out to be the general intelligence problem. And so there's been generation after generation of AI researchers and industrialists and founders making those claims, and they're being wrong every time. So I don't want to poo poo. LLMs, they're very useful. There should be a lot of investment in them. There should be a lot of investment in infrastructure to run them, which is where most of the money is going actually. It's not to train them or anything, it's to run them in the end, serving billions of users potentially. But like every other computer technology, it can be useful even if it's not human level intelligence. Now, if we want to shoot for human level intelligence, I think we should, we need to invent new techniques. We're just nowhere, nowhere near kind of matching that.

[Jeff Jarvis]

I'm really grateful you're here, Jan, because I quote you constantly on this show and elsewhere because you are the voice of realism, I think, in AI. And I don't hear you spouting the hype that I hear elsewhere. And you've been very clear about where we are now. I think you've equated us to. Maybe we're getting to the point of a smart cat or A, A3. Not even. Exactly. And so you've, and I think you've also talked about the. We've hit kind of the limits of what LLMs can do. So there is a next paradigm, a next leap. And I think you've talked about, about the being understanding reality better. But can you talk about where you think research, where you're taking it, or where it should be going next? Where we should be putting resources next to get more out of AI?

[Yann Lecun]

So I wrote a long paper three years ago where I explained where I think AI research should go over the next 10 years. This was before the world learned about LLMs, and of course I knew about it because we were working on it before. But this vision hasn't changed. It's not been affected by the success of LLM, if you want. So here's the Thing. We need machines that understand the physical world. We need machines that are capable of reasoning and planning. We need machines that have persistent memory. And we need those machines to be controllable and safe, which means that they need to be driven by objectives. We give them. We give them a task, they accomplish it, or they give us the answer to the question we ask, and that's it. They can't escape whatever it is that we're asking them to do. So what I explained in that document is how we might potentially one way we could get to that, to that point, and it's centered on a central concept called world model. So we all have world models in our head, and animals do too. Right? And it's basically the mental model that we have in our head that allows us to predict what's going to happen in the world, either because the world is being the world or because of an action we might take. So if we can predict the consequences of our actions, right, then what we can do with that is if we set ourselves an objective, a goal, a task to accomplish, we can, using a world model, imagine whether a particular sequence of actions will actually fulfill that goal. And that allows us to plan. So planning and reasoning really is manipulating our mental model to figure out if a particular sequence of actions is going to accomplish a task that we set for ourselves. Okay, so that is what psychology is called. System two. Deliberate, sort of. I don't want to say conscious, because it's a loaded term, but deliberate process of thinking about how to accomplish a task, essentially, and that we don't know how to do, really. I mean, we're making some progress at the research level. A lot of the most interesting research in that domain is done in the context of robotics, because when you need to control a robot, you need to know in advance what the effect of sending a torque on a. On an arm is going to be. And so this process, in fact, in control theory and robotics, of imagining the consequences of a sequence of actions and then basically by optimization, searching for a sequence of action that satisfies the task. Even has a name, even has an acronym. It's called Model Predictive Control mpc. It's a very classical method in optimal control going back decades. The main issue with this is that in robotics and control theory, the way this works, the world model, is a bunch of equations that are written by someone, by an engineer. You want to control a robot arm or a rocket or something, you can just write down the dynamical equations of it. But what we need to do for AI systems We need this world model to be learned from experience or learned from observation. So this is the kind of process that seems to be taking place in the minds of animals and, and maybe humans, infants learning how the world works by observation. That's the part that seems really complicated to reproduce. Now this can be based on a very simple principle which people have been playing with for a long time without much success, called self supervised learning. And self supervised learning have been incredibly successful in the context of natural language understanding and LLMs and things like that. In fact, it's the basis of LLM, right? So you take a piece of text and you train a big neural net to predict the next word in the text. Okay? That's basically what it comes down to. These tricks are how to make this efficient and everything. But that's the basis of an LLM. You just train it to predict the next word in the text and then when you use it, you have it predict the next word, shift the predicted word into its viewing window and then predict the second word, then shift that in predict the third. That's auto ergonomic prediction. That's what LLMs are based on. And all the tricks is how much money you can afford to kind of hire people to fine tune it so we can answer questions correctly. Okay, which is what a lot of money is going into right now. Okay, so you, you could imagine using this principle of self supervised learning for learning representations of images, learning to predict what's going to happen in a video, right? So if you show a video to a computer and train some big neural net to predict what's going to happen next in a video, if the system is capable of learning this and doing a good job at that prediction, it will probably have understood a lot about the underlying nature of the physical world. Things that objects move according to particular laws, animate objects can move in things that are more unpredictable, but still satisfying some constraints, you're not going to have objects that are not supported fall because of gravity, et cetera. Now human babies take nine months to learn about gravity, a long process. Young animals, I think, learn this much quicker, but they don't have the same kind of grasp of where gravity really is in the end. Although cats and dogs are really good at this, obviously. So how do we reproduce this kind of training? So if we do the naive thing, which I've been working on for 20 years, doing similar thing as taking a piece of text, but just taking a video and then training a system to predict what happens next in the video, it doesn't really work. So if you train it to predict the next frame, it doesn't learn anything useful because it's too easy. Okay. If you train it to predict longer term, it really cannot predict what's going to happen in the video because there's a lot of plausible things that might happen. Okay? So in the case of text, that's a very simple problem because you only have a finite number of words in the dictionary, and so you can never predict exactly what word follows a sequence, but you can predict a probability distribution of all words in the dictionary, and that's good enough. You can represent uncertainty in the prediction. You can't do this with video. We do not know how to represent appropriate probability distribution over the set of all images or video frames, or video segments for that matter. It's actually a mathematically intractable problem. So it's not just a question of like, we don't have big enough computers. It's just like intrinsically intractable. So until maybe five, six years ago, I didn't have any solution to this. I don't think anybody had any solution to this. And one solution that we came up with is a kind of architecture that changes the way we would do this. Instead of predicting everything that happens in the video, we basically train a system to learn a representation of the video and we make the prediction in that representation space. And that representation eliminates a lot of details in the video that are just not predictable or impossible to figure out. That kind of architecture is called a JEPA Joint Embedding Predictive Architecture. What may be surprising about this is that it's not generative. So everybody is talking about generative AI. My hunch is that the next generation AI system will be based on non generative models, essentially.

[Jason Howell]

This episode of the AI Inside podcast is sponsored by BetterHelp. I've noticed a big shift in recent years towards taking mental health seriously, and I welcome that change because I recognize firsthand the benefits of taking care of my own mental health. Therapy can be a transformative experience, and it definitely has been for me. But no question it can be pricey. Traditional in person therapy can run anywhere between 100 to $250 per session, and that adds up. And it really should not stand in the way of getting the help that's needed when it counts. BetterHelp is online therapy that can save you on average up to 50% per session. With BetterHelp, you pay a flat fee for each weekly session, and that adds up to big cost savings over time. And not only that, BetterHelp is much easier to access than traditional therapy because it's an online experience that meets you where you are at with quality care from more than 30,000 therapists at a price that makes sense. You just click a button to join your therapist is there from wherever you happen to be. You can get support with anything from anxiety to relationships to everyday stress. And if you just aren't feeling it with your current therapist, you can easily switch to another at any time. It's mental health within reach and it's totally worth it. I know firsthand. I used BetterHelp a few years ago myself. It was incredibly convenient and more importantly, impactful to my life. I felt heard and supported and that's what I really needed. Your well being is worth it. Visit betterhelp.comaiinside today to get 10% off your first month. That's betterhelp h e l p.comaiinside and we thank betterhelp for their support of the AI Inside podcast. Trust isn't just earned, it's demanded. And whether you're a startup founder navigating your first audit or a seasoned security professional scaling your GRC program, proving your commitment to security has never been more critical or more complex. That's where Vanta comes in. Businesses use Vanta to establish trust by automating compliance needs across over 35 frameworks like SoC2 and ISO 27001, centralized security workflows, complete questionnaires up to five times faster, and proactively manage vendor risk. Vanta not only saves you time, it can also save you Money. A new IDC white paper found that Vanta customers achieve $535,000 per year in benefits, and the platform pays for itself in just three join over 9,000 global companies like Atlassian, CORA and Factory, who use Vanta to manage risk and prove security in real time. For a limited time, our audience gets $1,000 off [e-mail] AI inside that's v a n t a dot com AI inside for $1,000 off what occurs to me in kind of hearing, hearing you talk about the real limitations of where we're at. We're right on the precipice of AGI artificial general intelligence. And here's the reason why. It depends on who you ask, right? Some people are like, it's right around the corner. Other people are like, oh, it's already here. Take a look at this. Isn't that amazing? But when you're talking, others are, it'll never be here. But yes, well, and yes. And then others are it'll never be here. I think often on this show where, you know, we kind of talk about this topic a little bit in disbelief, and I think what you just said kind of punctuates that a little bit for me. When you put it that way, it really makes me more confident that AGI is not right around the corner, that AGI is really this distant theory.

[Yann Lecun]

Okay? So, first of all, there is absolutely no question in my mind that at some point in the future we'll have machines that are at least as smart as humans in other domains where humans are smart. Okay. That's not a question. People had big philosophical question about this. A lot of people still believe that the, the human nature is kind of impalpable and we're never going to be able to reduce this to computation. I'm not a skeptic on that dimension. There's no question in my mind at some point we'll have machines that are more intelligent than us. They already are in narrow domains. Right? So then there is the question of what does AGI really means exactly? Really mean? Does it mean general? What do you mean by general intelligence? Do you mean intelligence that is as general as human intelligence? If that's the case, then okay, you can use that phrase, but it's very misleading because human intelligence is not general at all. It's extremely specialized. We are shaped by evolution to only do the tasks that are worth accomplishing for survival. And we think of ourselves as having general intelligence, but we're just not at all general. It's just that all the problems that we're not able to apprehend, we can think of them. That makes us believe that we have general intelligence, but we absolutely do not have general intelligence. So I think this phrase is nonsense. First of all, it was very misleading. I prefer like the kind of phrase we use to designate the concept of human level intelligence within Meta is ami, advanced machine intelligence. Okay. This is kind of a much more open concept. We actually pronounce it ami, which means friend in French, but let's call it human level intelligence if you want. Right. So no question it will happen. It's not going to happen next year. It's not going to happen two years from now. It may happen or happen to some degree within the next 10 years. So it's not that far away. If all of the things that we are working on at the moment turn out to be successful, then maybe within 10 years we'll have a good handle on whether we can reach that goal. Okay. But it's almost certainly harder than we think and probably much harder than we think, because it's always harder than we think. Over the history of AI, it's always been harder than we think. It's the story I was telling you earlier. So I'm optimistic. I'm not one of those pessimists who say we'll never get there. I'm not one of those pessimists that says all the stuff we're doing right now is useless. It's not true. It's very useful. I'm not people who say we're going to need some quantum computing or some completely new principle, blah, blah, blah. No, I think it's going to be based on deep learning, basically. And that underlying principle, I think is going to stay with us for a long time. But within this domain, the type of things that we need to discover and implement is, we're not there yet. We're missing some busy concepts. The best way to convince yourself of this is to say, okay, we have systems that can answer any question that has a response somewhere on the Internet. We have systems that can pass the bar exam, which is basically retrieval, information retrieval to a large extent. We have systems that can shorten the text and help us understand it, that can criticize a piece of writing that we're doing, that can generate code. But generating code is actually to some extent relatively simple because the syntax is strong and a lot of it is. We have systems that can solve equations, that can solve problems as long as you've been trained to solve those problems. If they see a new problem from scratch, current systems just cannot find a solution. There was actually a paper just recently that shown that if you test all the best LLMs on the latest Math Olympiad, they basically get zero performance because there are new problems they've not been trained to solve. So we have those systems that can manipulate language and that fools us into thinking that they are smart. Because we're used to smart people being able to manipulate language in smart ways. Okay, but like where is my domestic robot? Where is my level 5 self driving car? Where is a robot that can do what a cat do? Even a simulated robot, they can do what a cat do, what cats can do, right? And the issue is not that we can't build a robot, we can actually build robots that have the physical abilities. It's just that we don't know how to make them smart enough. And it's much, much harder to deal with the real world and to deal with systems that produce actions than to deal with systems that understand language. And again, it's related to the problem I was mentioning before. Language is discrete, it has strong structure, the real world is a huge mess and it's unpredictable, it's not deterministic, it's high dimensional, it's continuous, it's got all the problems. So let's try to build something that can learn as fast as a cat.

[Jeff Jarvis]

I've got so many questions for you, but I'm going to stay on this for another minute. Should human level activity or thought even be the model? Is that limiting? There's a wonderful book from some years ago by Alex Rosenberg called How History Gets Things Wrong, arguing that the theory of mind, he debunks the theory of mind, that we don't have this reasoning, that we go through that. In fact we're kind of doing what an LLM does in the sense that we have a bunch of videotapes in our head and when we hit a circumstance we find the nearest videotape and play that and decide yes or no in that way. And so that does sound like the human mind a bit. But the model we tend to have for the human mind is one of reasoning and weighing things and so on. And also, as you say, we are not generally intelligent, but the machine conceivably could do things that we, it right now does things we cannot do. It could do more. So when you think about success and that goal, what does that model? Is it, is it a cat? Would be a big victory to get to the point of being a cat. But what's your, what's your larger goal? Is it human intelligence or is it something else?

[Yann Lecun]

Well, it's a type of intelligence that is similar to human and animal intelligence in the following way. Current AI systems have a very hard time solving new problems that they've never faced before. Right? So they don't have this mental model, this world model I was telling you about earlier, that allows them to kind of imagine what the consequence of their actions or whatever. They don't reason in that way, right? I mean an LLM certainly doesn't because the only way it can do anything is just produce words, produce tokens. So one way you trick an LLM into spending more time thinking about a question, a complex question and a simple question, is you ask it to go through the steps of reasoning and as a consequence it produces more tokens and then spends more computation answering that question. But it's a horrible trick. It's a hack. It's not the way humans reason. Thank you. Another example that LLMs do is for writing code or Answering questions. You get an LLM to generate lots and lots of sequences of tokens, all that have some decent level of probability or something like that. And then you have a second neural net that tries to evaluate each of those and then picks the one that is best. It's sort of like kind of producing lots and lots of answers to a question and then have a critique of kind of telling you which of those answers is the best. Now, there is a lot of AI systems that work this way, and it works in certain situations, like if you want a computer system to play chess. This is exactly how it works. It produces a tree of all the possible moves from you, and then from your opponent, and then from you, and then from your opponent. That tree grows exponentially. So you can't generate the entire tree. You have to kind of have some smart way of only generating a piece of the tree. And then you have what's called an evaluation function or value function that picks out the best branch in the tree that results in a position that is most likely to win. And all of those things are trained nowadays. They're neural nets, basically, that generate the good branch in the trees and select it. That's a limited form of reasoning. Why is it limited? And it's, by the way, a type of reasoning that humans are terrible at. The fact that a $30 gadget that you buy at a toy store can beat you, a chess demonstrates that humans totally suck at this kind of reasoning. We're just really bad at it. We just don't have the memory capacity, the computing speed and everything. So we're terrible at this. What we are really good at, though, is the kind of reasoning, and what cats and dogs and rats are really good is sort of planning actions in the real world and planning them in a hierarchical manner. So knowing that if we want to. Let me take an example in the human domain. But there are similar ones in sort of animal tasks, right? I mean, you see cats learning to open jars and jump on doors to open them and open the lock of a door and things like that. So they learn how to do this and they learn to plan that sequence of actions to arrive at a goal, which is getting to the other side perhaps to get food or something. You see squirrel doing this. They're pretty smart, actually, in the way they learn how to do this kind of stuff. Now this is a type of planning that we don't know how to reproduce with machines. A lot of it is completely internal. It has nothing to do with language, right? We think as humans, we think that thinking is related to language, but it's not. Animals can think, people who don't talk can think. And there are types of reasoning. Most types of reasoning have nothing to do with language. So if I tell you imagine a cube floating in the air in front of you, in front of us. Now rotate that cube 90 degrees along a vertical axis. Okay, so probably you made the assumption that the cube was horizontal, that the bottom was horizontal. You didn't imagine a cube that was kind of sideways. And then you rotate it 90 degrees and you know that it looks just like the cube you started with because it's a cube, it's got 90 degrees symmetry. There's no language involved in this reasoning. It's just images and sort of abstract representations of the situation. And how do we do this? We have those abstract representation of thought and then we can manipulate those representations through virtual actions that we imagine taking, like rotating that cube and then imagine the result. Right? And that is what allows us to actually accomplish tasks in the real world at an abstract level. It doesn't matter what the cube is made of, how heavy it is, whether it floats in front of us or not. I mean, all those details don't matter. And the representation is abstract enough to really not care about those details. If I plan to, I mean, New York, right? If I plan to be in Paris tomorrow, I could try to plan my trip to Paris in terms of elementary action I can take, which basically are millisecond by millisecond controls of my muscles. But I can't possibly do this because it's several hours of muscle control and it will depend on information that I don't have. Like I can go into the street and hail a taxi. I don't know how long it's going to take for a taxi to come by. I don't know if the light is going to be red or green. I cannot plan my entire trip. So I have to do hierarchical planning. I have to imagine that if I want to be in, in Paris tomorrow, I first have to go to the airport and catch a plane. Okay, now I have a sub goal. Going to the airport. How do I go to the airport? I'm in New York, so I can go down on the street, have a taxi. How do I go down in the street? Well, I have to walk through the elevator or the stairs, hit the button, go down, work out the building. And before that I have to. Now step will go into the elevator or to the stairs. How do I even stand up from my chair? Can you explain in words how you climb a stair or you stand up from your chair, you can this is low level understanding of the real world. And at some point in all those sub goals that I just described, you get to a situation where you can just accomplish the task without really planning and thinking, because you're used to standing up from your chair. But the complexity of this process of imagining what the consequences of your actions are going to be with your internal world model, and then planning a sequence of actions to accomplish this task. This is the big challenge of AI for the next few years. We're not there yet.

[Jason Howell]

Let's talk about something we don't talk about enough. What happens to all the data we share with AI platforms like ChatGPT or Claude? Every question we ask, every idea we brainstorm, it's all being collected and tied back to us as individuals. But then what does it get sold to advertisers, corporations, maybe even governments? We've also grown accustomed to social media companies selling our data over the last decade, and I'd like to think that maybe we've learned a thing or two so we don't make the same mistakes again. That's why I've been using Venice AI, who's sponsoring today's episode. Venice AI is private and permissionless, using leading open source models for text, code and image generation. And it's all running directly in your browser, so there's no downloads, no installs. In fact, your chats and history live entirely inside your browser. They don't even get stored on Venice's servers. Their pro plan is where things get really interesting, though. You can upload PDFs to get insights and summaries. You get a user controllable safe mode for deactivating restrictions on image generation. You can customize how the AI interacts by modifying its system prompt directly. And finally you get unlimited text queries along with high image limits that I couldn't even hit if I tried. We talk often on the podcast about the benefits of open source AI, and that's exactly what Venice AI is using. If you care about privacy like I do, or you just want an uncensored and truly open AI experience, Venice AI is worth checking out. Go to my Sponsor link Venice AI AIInside. Make sure to use the code AI Inside to enjoy private, uncensored AI. Use my code and you'll get 20% off a pro plan. That's Venice AI AIInside with code AIInside for 20% off the pro plan. And we thank Venice AI for sponsoring the AI Inside podcast.

[Yann Lecun]

This episode is Brought to you by Chevy Silverado. When it's time for you to ditch the blacktop and head off road, do it in a truck that says no to nothing. The Chevy Silverado Trail Boss. Get the rugged capability of its Z71 suspension and 2 inch factory lift, plus impressive torque and towing capacity thanks to an available Duramax 3 liter turbo diesel engine. Where other trucks call it quits, you'll just be getting started. Visit chevy.com to learn more.

[Jeff Jarvis]

So one question I've been wanting to ask. This has been a great lesson, professor, and I'm really grateful for that, but I also want to get to the kind of the current view of Meta strategy on this and the fact that Meta has decided to go. We call it open source or open or available or whatever, but LLAMA is a tremendous tool. As an educator myself, I'm grateful. I was emeritus at cuny, but now I'm at SUNY Stony Brook. And it's because of LLAMA that universities can run models and learn from them and build things. And it struck me, and I've said this often, that I think that the Meta strategy, your strategy here on Lama and company, is a spoiler for much of the industry part, but an enabler for tremendous open development, whether it's academic or entrepreneurial. And so I'd love to hear from the horse's mouth here, what's the strategy behind opening up Llama in the way that you've done?

[Yann Lecun]

Okay, it's a spoiler for exactly three companies. Yeah, exactly.

[Jeff Jarvis]

Well, exactly. Yes.

[Yann Lecun]

Okay. It's an enabler for thousands of companies.

[Jeff Jarvis]

Yes.

[Yann Lecun]

So obviously, you know, from the pure ethical point of view, it's obviously the right thing to do. Right? I mean, llama, llama 2. The release of llama 2 in qualified open source has basically completely jump started the AI ecosystem, not just in industry and startups, but also in academia. As you were saying, academia basically doesn't have the means to train their own foundation model at the same level as companies. And so they rely on this kind of open source platforms to be able to make contributions to AI research. And that's kind of one of the main reasons for Meta to actually release those foundation models in open source is to enable innovation, faster innovation. And the question is not whether this or that company is three months ahead of the other, which is really the case right now. The question is, do we have the capabilities in the AI systems that we have at the moment to enable the products we want to build? And the answer is no. The product that Meta wants to build Ultimately is an AI assistant or maybe a collection of AI assistant that is with us at all times, maybe lives in our smart glasses that we can talk to, maybe it displays information in the, the lens and everything. And for those things to be maximally useful, they would need to have human level intelligence. Now we know that moving towards human level intelligence is not going. So first of all, it's not going to be an event. There's not going to be like a day where we don't have AGI and a day after which we have AGI. It's just not going to happen this way.

[Jeff Jarvis]

I'll buy you the drinks if that happens.

[Yann Lecun]

Yeah, yeah, well, I should be buying too, because it's not happy, but it's not going to happen this way. Right. So the question really would be how do we make fastest possible progress towards human level intelligence? And since it's one of the biggest scientific and technological challenge that we faced, we need contributions from anywhere in the world. There's good ideas that can come up from anywhere in the world. And we've seen an example with Deep SEQ recently, Right. Which surprised everybody in Silicon Valley. Didn't surprise many of us in the open source world that much. Right. I mean, that's the point. It's sort of validation of the whole idea of open source. And so good ideas can come from anywhere. Nobody has a monopoly on good ideas except people who have like an incredibly inflated superiority complex.

[Jeff Jarvis]

Not that we're talking about anybody in particular, right?

[Yann Lecun]

No, no, we're not talking about anybody in particular. There's a high concentration of those people in certain areas of the country. So, you know, and of course they have a vested interest in sort of kind of disseminating this idea that they somehow they're better than everybody else. So I think it's still a major scientific challenge and we need everybody to contribute. So the best way we know how to do this in the context of academic research is you publish your research, you publish your code in open source as much as you can and you get people to contribute. And I think the history of AI over the last dozen years really shows that. I mean, the progress has been fast because people were sharing code and scientific information and a few players in the space started climbing up over the last three years because they need to generate revenue from the technology. Now at Meta, we don't generate revenue from the technology itself, we generate revenue from ads. And those ads rely on the quality of products that we built on top of the technology, and they rely on the network effect of the social networks or conduit to the people and the users. And so the fact that we distribute our technology doesn't hurt us commercially. In fact, it helps us.

[Jeff Jarvis]

Right.

[Jason Howell]

You mentioned the topic of wearables and glasses, and that of course, always sparks my attention. I had the opportunity to check out Google's Project Astra Glasses last December and it's stuck with me ever since and really kind of solidified my view of that being a really wonderful next step for contextualizing the world. The line that I've been able to draw in talking with you between where we are now and where we're going potentially is not only the context that that experience gives the wearer, but for you, for meta and for those creating these systems, smart glasses out in the real world, taking in information on how humans live and operate in our physical world could be a really good kind of source of knowledge to pull from. For what you were talking about earlier, am I on the right track or is that just one piece, one very small piece of the puzzle?

[Yann Lecun]

Well, it's a piece, an important piece. But yeah, I mean, the idea that you have an assistant with you at all times that sees what you see, hears what you hear, if you let it, obviously if you let it for sure. But to some extent is you're confident and can help you as perhaps even better than how a human assistant could help you. I mean, that's certainly an important vision. In fact, the vision is that you won't have a single assistant. You will have a whole staff of intelligent virtual assistants walking around with you. It's like all of us would be a boss. I mean, people feel threatened. Some people feel threatened by the fact that machines would be smarter than us, but we should feel empowered by it. I mean, they're going to be working for us. I don't know about you, but as a scientist or as a manager in industry, the best thing that can happen to you is you hire students or engineers or people working for you that are smarter than you. That's the ideal situation. And you shouldn't feel threatened by that, you should feel empowered by it. So I think that's the future we should envision. Smart collection of assistance that help you in your daily lives, maybe smarter than you. You give them a task, they accomplish it perhaps better than you. And that's great. Now that connects to another point I wanted to make related to the previous question, which is about open source, which is that in that future most of our interactions with the digital world will be mediated by AI systems. And that's why Google is a little frantic right now, because they know that nobody is going to go to a search engine anymore. You're just going to talk to your AI assistant. So they're trying to experiment with this within Google. That's going to be through glasses. So they realize they probably have to build those. I realized this several years ago. So we have a bit of a head start, but that's really what's going to happen. We're going to have those AI sitting with us at all times and we're not going to. They're going to mediate all of our information diet. Now, if you think about this, if you are a citizen anywhere in the world, you do not want your information diet to come from AI assistant built by a handful of companies on the west coast of the US or China. You want a high diversity of AI assistant that first of all speaks your own language, whether it's an obscure dialect or local language. Second of all, understand your culture, your value system, your biases, whatever they are. And so we need a high diversity of such assistants for the same reason we need a high diversity of the press. Right. And I realize I'm talking to a journalism professor here, but like, am I right?

[Jeff Jarvis]

Amen. In fact, I think that's what I celebrate is what the Internet and next AI can do is to tear down the structure of mass media and open up media once again. At a human level. AI lets us be more human. I hope.

[Yann Lecun]

I hope too. So the only way we can achieve this with current technology is if the people building those assistants with cultural diversity and everything have access to powerful open source foundation models, because they're not going to have the resources to train their own models. We need models that speak all the languages in the world, understand all the value system and have all the biases that you can imagine in terms of culture, political biases, whatever you want. And so there's going to be thousands of those that we're going to have to choose from, and they're going to be built by small shops everywhere around the world, and they're going to have to be built on top of foundation models trained by a large company like Meta or maybe an international consortium that trains those foundation models. The picture I see, the evolution of the market that I see is similar to what happened with the software infrastructure of the Internet in the late 90s or the early 2000s, where in the early days of the Internet you had Sun Microsystem, Microsoft, hp, IBM and a few others kind of pushing to provide the hardware and software Infrastructure of the Internet, their own version of Unix or whatever, or Windows nt and their own web server on their own, you know, racks and blah, blah, blah. All of this got completely wiped out by Linux and commodity hardware, right? And the reason it got wiped out is because, you know, running, you know, Linux is a platform software. It's more portable, more reliable, more secure, more cheaper, you know, everything. And so, you know, Google was one of the first to do this building infrastructure on commodity hardware, and open source operating system Meta, of course, did exactly the same thing. And everybody is doing it now, even Microsoft. So I think there's going to be a similar pressure from the market to make those AI foundation models open and free because it's an infrastructure like the infrastructure of the Internet.

[Jeff Jarvis]

How long have you been teaching?

[Yann Lecun]

22 years. 22 years.

[Jeff Jarvis]

So what differences do you see in students and their ambitions today in your field?

[Yann Lecun]

I don't know. It's hard for me to tell because in the last dozen years or so I've only taught graduate students. So I don't see any significant change in PhD students other than the fact that they come from all over the world. I mean, there is something absolutely terrifying happening in the US right now, right, where funding for research is being cut and then there's sort of threats of visas not being given to foreign students and things like that. It's completely going to destroy the technological leadership in the US if it's actually implemented the way it seems to be going. Most PhD students in STEM, science, technology, engineering, mathematics are foreign and it's even higher in most engineering disciplines. At the graduate level, they're mostly foreign students. Most founders or CEOs of tech companies are foreign born.

[Jeff Jarvis]

French universities are offering the opportunity for American researchers to go there. I've got one more question for you. Do you have a cat?

[Yann Lecun]

I don't, but our youngest son has a cat and we watch the cat occasionally.

[Jeff Jarvis]

I wonder if that was your model.

[Jason Howell]

Right. Well, Jan has been wonderful. I know we've kept you just a slight bit longer than we had agreed to for your schedule, so we really appreciate you carving out some time. Yeah, it's been really wonderful and it's wonderful to kind of hear some of this, as Jeff said from the horse's mouth earlier, because you come up in our conversations quite a lot and we really appreciate your perspective in the world of AI and all the work that you've done over the years. Thank you for being here with us. This has been an honor.

[Yann Lecun]

Thank you.

[Jeff Jarvis]

And for the sanity you bring to the conversation.

[Yann Lecun]

Yes, thank you so much. It's really been a pleasure talking with you.

[Jason Howell]

Huge thank you once again to Yann Lecun for joining us on AI Inside. Hope to have him back sometime down the line to check in on how things are going. And of course, a huge thank you to my co host, Jeff Jarvis. He's obviously not here right now. Jeffjarvis.com if you want to check out all of his wonderful books, everything you need to know about this show, this podcast can be found at our site. Just go there, AI Inside. You're going to find all the ways to subscribe and audio, video, all the details are there. And if you love this show, you know, give us a review. If you liked this episode especially, give us a review. Leave a comment if you're watching this on YouTube. We really want to hear from you as far as what you think about this interview. And finally, if you really love this show, you can support us on Patreon. That's patreon.com aiinsideshow you get ad free shows, a Discord, community access. You get an AI inside t shirt if you become an executive producer like this Week's executive producers, Dr. Dew, Jeffrey Maracini, WPVM 103.7 in Asheville, North Carolina, Dante St. James, Bono Derick, Jason Neffer and Jason Brady. Y'all are awesome. Thank you so much for your support each and every week. And thank you for being here. I'm Jason Howell. I hope to see you next Wednesday on another episode of the AI Inside podcast. Bye, everybody.