"I think theory is great, uhm, its just I dont do it". Classic. I'm stealing that.
@michaelgoodale1589
10 күн бұрын
His metaphor is all mixed-up, syntax (and semantics/combinatorial operations more broadly) are the engine which allows languages whereas superficial facts about word frequencies are the equivalent of the number of wheels.
@GraczPierwszy
9 ай бұрын
where can I watch it in full?
@atheistbushman
9 ай бұрын
kzitem.info/news/bejne/qJ1j3ZOkcoB7fHo
@kennethmarshall306
6 ай бұрын
It didn’t help me understand what Hinton is saying. I am prepared to accept that there are basic genetic structures in the brain which govern all languages. Is Hinton denying that?
@federicopiersigilli4677
6 ай бұрын
Given what he said about Chomsky's theory, it is good he doesn't do theory
@practicaltheory
9 ай бұрын
I remember when I was in college thinking Noam was a legit intellectual, but he's a gatekeeper.
@DJWESG1
9 ай бұрын
Large language models imo reinforce chomsky , as they borrow so heavily from the esoteric world of the past.
@a_soulspark
4 ай бұрын
it's an interesting point. but LLMs don't borrow the grammatical structure that Chomsky argues we're born with, at least not directly. they learn it all from the data ("experiences"), and if LLMs can do it, then it disproves that an a priori linguistic structure is necessary to understand language (though it could still exist for humans, we don't know)
@DJWESG1
4 ай бұрын
@@a_soulspark which is why large language models can never be a true a.i, they are just a pale immitation of a crude network. its systems theory in action, thats all.
@Joeyjojoshabbadoo
4 ай бұрын
Excuse me, but wtf do these people know about language? They just write computer code, that's all they do. Maybe Chomsky's wrong, maybe he's not. It's not like these guys would have any idea. They have their little vision of how the brain works, with basically no idea whether it's correct. And on the basis of their preferred guess, they build their AI systems. That's why they nurture the ideas about the brain that they do, because it's useful to them when building their code architecture. Not because it actually resembles the human brain. That part is immaterial. But Geoffrey Hinton is not satsified with that. He wants to have it all....
@a_soulspark
4 ай бұрын
you'd be surprised how much we can learn about language from just computer simulations of it! keep in mind Geoffrey Hinton really is interested in understanding how the brain functions and how it differs from current AI. e.g., even though he's one of the co-authors of the 1986 backpropagation paper, he nowadays admits it's likely not how the brain works, based on what neuroscience now says, and is interested in exploring other possible explanations. (so yes, even by doing backprop, with no need for a pre-programmed grammatical structure as per Chomsky, language models learn language)
@Joeyjojoshabbadoo
4 ай бұрын
@@a_soulspark Yeah, how do you figure? How would you ever learn anything about how the human brain works, based on the workings of computer software that runs on computer chips that couldn't be anything less like a brain, and those same AI systems were developed and predicated on visions of how the brain works that were never in any way proven or confirmed as accurate, given how it's pretty much not at all technically understood how the brain works, and is merely wondered at? To the best of my layman's understanding, AI systems and machine learning and neural networks and the like are very deliberately designed to be able to execute human-like tasks, with that express objective in mind, even though they're totally not human at all, and they don't need to be. They exist to be able to simulate what the brain can do, not to replicate how the brain does it, notwithstanding any misplaced claims to the contrary. And as such do not significantly or quite likely don't at all resemble the workings of the human brain, until otherwise proven, which one would not expect to ever be the case for obvious reasons. And G Hinton's preening triumphalism isn't a good look. Not too cool snickering and taking potshots at Noam Chomsky.
@JrocTheReal
2 ай бұрын
What a bland way to disregard Chomsky's claims. It's almost as if he knows he will pass soon, and is starting to role out the propaganda to cover up his work. Seems like he doesn't want to have any discussions with Chomsky about these issues, instead criticize his viewpoints. Anyone can recognize at this point that language models are not to be taken seriously in the domain of human interpretation and consciousness. Language models are simply comlex code that can present us "Human like" responses. To draw the conclusion that language models help our understanding, then make the connections for us and elaborate on it. Don't just say 'Chomsky crazy, language models are helpful with understanding neuroscience" then never go into detail. And he is incorrect, most language is internal, not meant to convey or describe our surroundings verbally, but internally. Seems like the guy is taking shots at Chomsky without ever gaining any context as to his actual positions. The guy doesn't even sound like he is invested or articulated enough on the subject. Chomsky debated a biologist on this, live, after a lecture and the biologist is asking questions similar to what Hinton is calling fact. Go watch that interaction. That was decades ago, why is Hinton coming out of nowhere with contentions that were debunked so long ago????
@jedenzemenemsow
10 күн бұрын
Thanks for your comment, do you have a link to Chomsky's debate with the biologist? I would like to learn more about this, cheers.
Пікірлер: 16