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

Janna Levin (Professor of Physics and Astronomy) 00:01.480

Is it the Moravec paradox that what computers are good at humans are bad at, what humans are good at computers are bad at?

Yann LeCun (Chief AI Scientist) 00:09.680

Yeah, we keep running into the Moravec Yeah. paradox Yeah.

Janna Levin (Professor of Physics and Astronomy) 00:12.000

Now, Adam, I I know that you are less pessimistic about the potential of the current neural net deep learning um paradigm, and you see the potential for a great escalation in success, and you don't see it saturating. Um, what's your thought about

Adam Brown (Research Scientist) 00:30.180

that? I I um I don't. That's right. Um, And so yeah. we have witnessed over the last five years the most extraordinary run up in capabilities in any system I've ever seen. This is what transfixed my attention. It's what transfixed many other people uh in AI and neighbouring

Adam Brown (Research Scientist) 00:56.500

fields to focus all of our attention on this matter.

Adam Brown (Research Scientist) 01:01.780

I don't see any slowdown in the capabilities. A year ago if you just look at all of the all of the metrics we use to judge how good these large language models are, they're getting stronger and stronger and stronger. Things that they you know the model from a year ago today

Adam Brown (Research Scientist) 01:16.140

would be you know table stakes would be considered extremely

Adam Brown (Research Scientist) 01:19.180

poor. Every few months these things push the capabilities and if if you track their capabilities, on all of these tasks, they're heading towards superhuman on on almost all of them. It's already better gives better legal advice than than a lawyer. It gives better um to a better

Adam Brown (Research Scientist) 01:39.220

poet than almost every poet you will come. In my

Adam Brown (Research Scientist) 01:42.060

little area In my little area of physics uh I I use it because like there's something I kind of should know, but I don't. I'll ask the language model, and it will not only tell me what the right answer is, it will patiently um I should say non-judgmentally, listen while I

Adam Brown (Research Scientist) 01:57.020

explain my misconception to it and it will carefully debunk my misconception.

Adam Brown (Research Scientist) 02:03.540

The extraordinary run up in capabilities that we've seen over the last five years and it continues up to the present is extremely tantalizing to to me and many other people in San Francisco. And and maybe maybe Jan is correct that we're just going to suddenly saturate and all of

Adam Brown (Research Scientist) 02:21.260

these uh straight lines that have been going up steadily for the last five years are suddenly going to stop going up,

Adam Brown (Research Scientist) 02:27.380

but I am mighty curious to see uh whether we can push it further, and I've actually seen no indication whatsoever that it's slowing down. Every indication I've seen is that these these are improving. And we don't have far to go, because once it's a better coder than almost all

Adam Brown (Research Scientist) 02:42.620

our best coders, it can start improving itself, and then we're really in for a wild ride.

Yann LeCun (Chief AI Scientist) 02:47.540

Well, we we've had better coders than the original coders of the 1950s, you know, for six decades also, that's called compilers. I mean, we we we keep getting confused about the fact that it's not because machines are good at a certain number of tasks that they have all the

Yann LeCun (Chief AI Scientist) 03:09.620

underlying intelligence that we assume a human having those capabilities will have, right?

Yann LeCun (Chief AI Scientist) 03:15.100

We're fooled into thinking those machines are intelligent because they can manipulate language. And we're used to the fact that people who can manipulate language very well are implicitly smart. But we're being fooled. Now they they're useful, there's no question. You know, we

Yann LeCun (Chief AI Scientist) 03:34.580

can use them to do what you said. I use them for similar things.

Yann LeCun (Chief AI Scientist) 03:39.220

Great, they're great tools like computers have been for the last five five decades. But let me make an interesting historical point.

Yann LeCun (Chief AI Scientist) 03:49.740

And this is maybe due to my age. This mean generation after generation of AI scientists since the 1950s claiming that the technique that it just discovered was going to be the ticket for human level intelligence.

Yann LeCun (Chief AI Scientist) 04:07.140

You You see declarations of Marvin Minsky, Newell and Simon, um you know, Frank Rosenblatt who invented the perceptron, the first learning machine in 1950, saying like, "Within 10 years, we'll have machines that are as

Yann LeCun (Chief AI Scientist) 04:21.060

smart as humans." They were all wrong. This generation with LLM is also wrong. I seen three of those generation in my lifetime, okay? Um,

Yann LeCun (Chief AI Scientist) 04:32.020

so, you know, it's it it's just another example of being fooled. And um in the 50s, Newell and Simon, pioneers of AI, came up with a program. They said, "Well, you know, really what what humans are doing is in reasoning is really a

Yann LeCun (Chief AI Scientist) 04:49.060

search, right? Every reasoning can be reduced to kind of a kind of search. So, you're forming a problem, you write a program that tell you whether a particular proposal for a solution is a solution to your problem, and then you just have to search for all possible combinations,

Yann LeCun (Chief AI Scientist) 05:04.820

you know, all possible hypotheses for one that actually matches uh satisfies the the constraint, and