Comment by foobarqux
14 days ago
"Tom Jones" isn't a strawman, Chomsky is addressing an actual argument in a published paper from Steven Piantadosi. He's using a pseudonym to be polite and not call him out by name.
> instead of even attempting to summarize the arguments for his position..
He makes a very clear, simple argument, accessible to any layperson who can read. If you are studying insects what you are interested in is how insects do it not what other mechanisms you can come up with to "beat" insects. This isn't complicated.
>The systems work just as well with impossible languages that infants cannot acquire as with those they acquire quickly and virtually reflexively.
Where is the research on impossible language that infants can't acquire? A good popsci article would give me leads here.
Even assuming Chomsky's claim is true, all it shows is that LLMs aren't an exact match for human language learning. But even an inexact model can still be a useful research tool.
>That’s highly unlikely for reasons long understood, but it’s not relevant to our concerns here, so we can put it aside. Plainly there is a biological endowment for the human faculty of language. The merest truism.
Again, a good popsci article would actually support these claims instead of simply asserting them and implying that anyone who disagrees is a simpleton.
I agree with Chomsky that the postmodern critique of science sucks, and I agree that AI is a threat to the human race.
> Where is the research on impossible language that infants can't acquire? A good popsci article would give me leads here.
It's not infants, it's adults but Moro "Secrets of Words" is a book that describes the experiments and is aimed at lay people.
> Even assuming Chomsky's claim is true, all it shows is that LLMs aren't an exact match for human language learning. But even an inexact model can still be a useful research tool.
If it is it needs to be shown, not assumed. Just as you wouldn't by default assume that GPS navigation tells you about insect navigation (though it might somehow).
> Again, a good popsci article would actually support these claims instead of simply asserting them and implying that anyone who disagrees is a simpleton.
He justifies the statement in the previous sentence (which you don't quote) where he says that it is self-evident by virtue of the fact that something exists at the beginning (i.e. it's not empty space). That's the "merest truism". No popsci article is going to help understand that if you don't already.
That's understandable but irrelevant. Only a few people have major interest in how humans think exactly. But nearly everyone is hang on the question if the LLMs could think better.
It's not irrelevant, it's the core of the disconnect: The problem is that everyone is arguing as if they passionately care about how humans work when, as you say, they don't care at all.
People should just recognize, as you have done, that they don't actually care about how the human language faculty works. It's baffling that they instead choose to make absurd arguments to defend fields they don't care one way or another about.
When Chomsky says that LLMs aren't how the human faculty works it would be so easy to tell the truth and say "I don't care how the human language faculty works" and everyone can go focus on the things they are interested in, just as it would be easy for a GPS designer to say "I don't care how insect navigation works".
There is no problem as long as you don't pretend to be caring about (this aspect of) science.
Is it polite to deprive readers of context necessary to understand what the speaker is talking about? I was also very confused by that part and I had no idea whom or what he was talking about or why he even started taking about that.
I searched for an actual paper by that guy because you’ve mentioned his real name. I found “Modern language models refute Chomsky’s approach to language”. After reading it seems even more true that Chomsky’s Tom Jones is a strawman.
> After reading it seems even more true that Chomsky's Tom Jones is a strawman.
Lol. It's clear you are not interested in having any kind of rational discussion on the topic and are driven by some kind of zealotry when you claim to have read a technical 40 page paper (with an additional 18 pages of citations) in 30 minutes.
Even if by some miraculous feat you had read it you haven't made a single actual argument or addressed any of the points made by Chomsky.
It’s certainly not a dense paper with careful nuanced derivations that you have to ponder to grasp. It’s a light read you can skim especially if you aren’t interested in LLM Trump improv and you are familiar with the general thought behind connectionism, construction grammar, other modern linguistic theories and, of course, universal grammar. The debate is as old as UG, but now with a new LLM flavor.
I don’t know which argument you expect from me. I read it and found nothing similar to “Stop wasting your time; naval vessels do it all the time.” So I concluded it’s a strawman. Being against a particular controversial approach in linguistics doesn’t mean being against science.
4 replies →