Yann LeCun (Chief AI Scientist) 00:01.100
They don't learn to the same level but they but they do learn faster. And And so this type of learning that we need to reproduce. Um, we'll do this with back prop with neural net with deep learning. It's just that we're missing a concept, an architecture. Um, so I've been I've
Yann LeCun (Chief AI Scientist) 00:16.460
been making proposals for the type of architectures that could possibly uh learn this kind of stuff.
Yann LeCun (Chief AI Scientist) 00:22.720
You know, why is it that LLM's handle language so easily? It's because um as Adam described, you you train an LLM to predict the next word or the next token, doesn't matter. There's only a finite number of words in the dictionary. So, you can never actually predict exactly which
Yann LeCun (Chief AI Scientist) 00:42.640
word comes after a sequence, but
Yann LeCun (Chief AI Scientist) 00:45.000
you can train a system to produce essentially what amounts to a score for every possible words in your dictionary or a probability distribution over every possible words. So essentially what a lemma does is that it produces a long list of numbers between 0 and 1 that's some to 1
Yann LeCun (Chief AI Scientist) 00:59.520
which for each word in the dictionary says this is the likelihood that this word appears right now. You can
Yann LeCun (Chief AI Scientist) 01:05.280
represent the uncertainty in the prediction this way. Now, try to translate it um the same principle instead of training a system to predict the next word feed it with a video and ask it to predict what happened next in the video. And this doesn't work. I've been trying to do
Yann LeCun (Chief AI Scientist) 01:22.480
this for 20 years. And it it really doesn't work.
Yann LeCun (Chief AI Scientist) 01:26.160
If you try to predict at the pixel level. Uh and it's because the real world is messy. There's a lot of things that that may happen, plausible things that may happen. Um and you can't really represent a distribution of all possible uh things that may happen in the future because
Yann LeCun (Chief AI Scientist) 01:44.360
it's basically an infinite list of possibilities and we don't know how to represent this. um efficiently. And
Yann LeCun (Chief AI Scientist) 01:51.040
so those those techniques that work really well for text or for sequences of of symbols do not work for real world sensory data. They just don't. They absolutely don't. And And so we need to invent new techniques. So, one of the things I've been proposing in one in which the the
Yann LeCun (Chief AI Scientist) 02:09.960
system learns an abstract representation of what it observes and it makes prediction in that abstract representation
Yann LeCun (Chief AI Scientist) 02:16.000
space. And this is really the way humans and animals function. And we we find abstractions that allow us to make predictions while ignoring all the detail the details we cannot predict.
Janna Levin (Professor of Physics and Astronomy) 02:26.400
So you really think that despite the phenomenal successes of these LLMs that they are limited and and their limit is quickly approaching. You don't think that they're scalable to this you know artificial general intelligence or a super intelligence.
Yann LeCun (Chief AI Scientist) 02:41.320
That's right. No, they don't and and you we see the performance saturating. So we see uh progress in in some domains like mathematics, for example. And mathematics and and code generation, you know, programming are two domains where the uh the the manipulation of symbols
Yann LeCun (Chief AI Scientist) 02:59.160
actually gives you something,
Yann LeCun (Chief AI Scientist) 03:00.440
right? As a physicist, you you know this, right? You write the equation and it actually kind of You You can follow it and it it's uh it drives your your thinking to some extent, right? I mean, you you drive it by intuition, but but the symbol manipulation itself actually has
Yann LeCun (Chief AI Scientist) 03:16.600
meaning. So, this type of problems, LLM's actually kind of handle pretty well, where the the reasoning really consists in kind of searching through sequences of symbols.
Yann LeCun (Chief AI Scientist) 03:25.360
But it's only there's only a small number of problems for which that's the case. Chess playing is another one. Um you search through sequences of of of moves that, you know, for a good one or sequences of derivations in mathematics that will produce a particular result, right?
Yann LeCun (Chief AI Scientist) 03:41.240
Um but in the real world, you know, in high dimensional continuous things where the search has to do with like how do I move my muscles to, uh, you know, grab this, uh, you know, grab grab this, this
Yann LeCun (Chief AI Scientist) 03:55.000
glass here. I'm not going to do it with my left hand, right? I'm going to have to change hand with this and and then grab it, right? You need to plan and have some understanding of what's possible, what's not possible, that, you know, I can just attract the glass, you know, by
Yann LeCun (Chief AI Scientist) 04:10.080
telekinesis or I can just I can't just make it appear in my in my left hand like this. I can't have my hand kind of cross my
Yann LeCun (Chief AI Scientist) 04:18.040
body. Like, you know You know, all of those intuitive things, we we learn them when we're babies um and and we learn, you know, how our body reacts to our controls and how uh you know, the the world reacts to to the actions we take. So So, you know, if I push this glass, I know
Yann LeCun (Chief AI Scientist) 04:38.200
it's going to slide. If I push it from the top, maybe maybe it's going to flip. Maybe not because the friction is not that high.
Yann LeCun (Chief AI Scientist) 04:45.120
If I push with the same force on this table, it's not going to flip. I You know, we have all those those intuitions that allow us to kind of apprehend the real world. But this is it turns out much much more complicated than manipulating language. We think of language as kind of
Yann LeCun (Chief AI Scientist) 05:00.960
the epitome of you know human intelligence and stuff like that. It's actually not true. Language is actually easy.