Monday, October 14, 2024

Hinton on Chomsky

The following videos are listed in reverse chronological order. For each video, I include the spots on video where Hinton talked about Chomsky (referring explicitly to Chomsky).

First Reactions (Interview with Adam Smith)
October 8, 2024

(5:20) “So there is a whole school of linguistics that comes from Chomsky that thinks that its complete nonsense to say that these things understand. That they don’t process language at all in the same way as we do. I think that school is wrong. I think it’s clear now that neural nets are much better at processing language that anything ever produced by the Chomsky school of linguistics. But there is still a lot of debate about that, particularly among linguists.”

Professor Geoffrey Hinton is presented with the UCD Ulysses Medal
University College Dublin
April 15, 2024

(13:40) “Linguists were misled for several generations by someone called Chomsky, who actually also got this prestigious medal. So it doesn’t last. He had a crazy theory that language is not learned. And he managed to convince lots of people of that. On the face of it, it’s just nuts, language is obviously learned. And now these big neural networks learn language. And they didn’t need any innate structure. They just started from random weights and a lot of data. And Chomsky is still saying but this isn’t really language. This doesn’t count. It’s not right. And lots of statisticians and cognitive scientists also said ‘You’ll never learn in a big net like this.’ So Chomsky never really had a theory of meaning, it was all about syntax. And if you think about meaning, there’s two very different theories of meaning. There’s the structuralist theory, that the symbolic AI people believed in, and most linguists believe in, which is that the meaning of a word comes from its relationships to other words.”

Romanes Lecture, University of Oxford
“Will digital intelligence replace biological intelligence?”
February 29, 2024

(6:03) “Chomsky managed to convince his follows that language wasn’t learned. Looking back on it, that’s just a completely crazy thing to say. If you can convince people to say something that is obviously false, then you’ve got them in your cult. I think Chomsky did amazing things, but his time is over.”

(13:22) “Many people, particularly people from the Chomsky school, argue they’re not really intelligent, they’re just a form of glorified auto complete that uses statistical regularities to pastiche together pieces of text that were created by people.”

CBMM10 Panel: Research on Intelligence in the Age of AI
November 20, 2023

(38:30) “So there’s this crazy guy at MIT called Chomsky who has been claiming it’s all innate. We know now that doesn’t have to be the case. And Chomsky’s whole view of language is kind of crazy when you look back on it because language is about conveying meaning. Its’ about conveying stuff. And Chomsky kind of ignored that aspect of it. It is as if you wanted to understand what a car is, and for all of us, understanding a car would mean – a large part of it would be understanding how the engine works that makes it go, but you can imagine someone saying ‘no’, ‘no’. The thing about cars is to understand why it is you get three-wheeled cars and four-wheeled cars, but you never get a five-wheeled car. And that is what we need to understand about cars. And that seems to me like Chomsky’s theory of language. He wanted to understand why certain syntactic constructions aren’t possible. And as far as I can see, he did everything he could to avoid the basic issue of how does language mean. And I think these large language models have put an end to that. Not in Chomsky’s mind, but in more or less everybody else’s mind.”



No comments:

Post a Comment

Note: Only a member of this blog may post a comment.