Do LLMs Understand? AI Pioneer Yann LeCun Spars with DeepMind’s Adam Brown.

Yann LeCun (Chief AI Scientist) 00:00.160

work. So so you have to, you know, come up with those new architectures like JPA and stuff like that. And those kind of work, like we we have models that actually understand video.

Janna Levin (Professor of Physics and Astronomy) 00:08.680

And Adam, are people exploring other ways of building an architecture or imagining a computer mind? This the actual fundamental structure of a computer mind and how it would um how it would learn, how it would acquire information.

Janna Levin (Professor of Physics and Astronomy) 00:23.040

One of the criticisms, as I understand it, is it's a lot of the LLMs are trained for this one specific task of this discrete prediction. of these tokens, but something that is more unpredictable like how the audience is distributed in this room, what might happen with the

Janna Levin (Professor of Physics and Astronomy) 00:36.500

weather next, unpredictable more human experience based phenomenon.

Adam Brown (Research Scientist) 00:42.620

Um, certainly all kinds of explorations are being made in all kinds of directions including Yans and like a thousand flowers bloom. Um, but all of the resources, I mean, the bulk of the resources right now are going into large language models and large language models like

Adam Brown (Research Scientist) 01:00.860

applications in in including taking in text.

Adam Brown (Research Scientist) 01:03.660

To say to say that they are it's a specialized task predicting the next token. I think that's a not a helpful way to think about it. It is true that the thing that you train them on is given this corpus of text. I mean there are other things we do as well but the the bulk of the

Adam Brown (Research Scientist) 01:19.340

computer goes to given this corpus of text. Please predict the next word. Please predict the next word. Please predict the next word. Um

Adam Brown (Research Scientist) 01:26.500

but we have discovered something truly extraordinary by doing it. which is that given a large enough body of text to be able to reliably predict the next word or you know do it do it well enough to predict the next word you really need to understand the universe

Adam Brown (Research Scientist) 01:42.340

and we have seen the emergence of understanding of the universe as we've done that. So I I would like in it a little bit. I mean in physics we're very used to systems where you just take a very simple rule and you know by the repeated application of that very simple rule you get

Adam Brown (Research Scientist) 01:58.620

extremely impressive of behavior. Uh,

Adam Brown (Research Scientist) 02:01.540

we see the same with these LLMs. Uh, and another example of that would maybe be evolution. You know, in each stage and evolution, you just say uh biological evolution. You just say, you know, maximize the number of offspring, maximize the number of offspring, maximize the number

Adam Brown (Research Scientist) 02:14.540

of offspring. A very sort of unsophisticated learning objective.

Adam Brown (Research Scientist) 02:18.540

But out of this simple learning objective repeated many, many times, uh you eventually get all of the, you know, splendor of biology that we see around us and and indeed this room. So

Adam Brown (Research Scientist) 02:30.700

The evidence is that predicting the next token while a very simple task, because it's so simple we can do it at massive scale. Huge amounts of compute and once you do it a huge amount of compute you get an emergent complexity.

Janna Levin (Professor of Physics and Astronomy) 02:44.180

So I I guess the next question could be related to evolution. However, this intelligence emerges that you both imagine is certainly possible. You don't think there's anything special about this wetware

Janna Levin (Professor of Physics and Astronomy) 02:55.780

that there will be machines we just have to figure out how to launch them that will um have capacities that we align as a kind of intelligence or maybe consciousness. That's a

Janna Levin (Professor of Physics and Astronomy) 03:06.100

almost a different question. Will consciousness be a crutch? Machines don't need. I don't know, we can talk about that, but but is there a point in the evolution of these uh machines where they're going to say, "Oh, how quaint, mom and dad, you you made me in your image with

Janna Levin (Professor of Physics and Astronomy) 03:20.620

these human neural nets." But I know a way, a much better way having scanned 10,000 years of human output to make a machine intelligence and I'm going to evolve and leave us in the dust. I mean,

Janna Levin (Professor of Physics and Astronomy) 03:34.740

yeah, what are we why are we imagining that they would be limited at that capacity to the way we design them.

Adam Brown (Research Scientist) 03:41.220

Absolutely. This is this idea of recursive self-improvement where when they're bad, they're useless, but when they get good enough and strong enough, you can start using them to augment human intelligence and perhaps eventually just be fully autonomous and replace and make

Adam Brown (Research Scientist) 03:58.940

future of them.

Adam Brown (Research Scientist) 04:00.380

Once we do that, I mean, I think what we should do is just take this large language model paradigm that's currently working so well and just see how far we can push it. You know, it keeps Every time someone says it's a barrier, it pushes through the barrier over the last 5

Adam Brown (Research Scientist) 04:12.100

years, but eventually these things will get smart enough and then they can uh read Jan's papers, uh read all the other papers that have been made, try and figure out uh new ideas that none none of us have thought of.

Yann LeCun (Chief AI Scientist) 04:24.460

Yeah. So, I completely disagree with this. Um So, LMs are not controllable. It's not dangerous because they're not that smart as as as I explained previously.

Yann LeCun (Chief AI Scientist) 04:38.900

And they're certainly not autonomous in a way that we understand autonomy. We have to distinguish between autonomy and intelligence. You can be very intelligent without being autonomous and you can be autonomous without without being intelligent.

Yann LeCun (Chief AI Scientist) 04:53.220

Um, and you can be dangerous without being particularly intelligent. Um, and You can want to be dominant without being intelligent. In fact, that's kind of inversely correlated in the human species. Um