There is a difference in being wrong and not being useful. I agree GPT-3 and the like are of little to no value to linguistics but that does not mean they are not useful or interesting in their own regard.

Large models, not just language ones, show we can brute force our way through. What they defend is that we should strive for more elegant and less brute-y ways to do it. No problem with that. At the same time, since we can bruteforce our way, its interesting to see how far it takes us.

Throughout time, there has always been two or more sides to advancements, AI is no exception, and generally this confrontation is good.

Some decades ago, symbolist approaches used to laugh at NNs. Maybe Chomsky & co are right and will see these approaches return. Maybe not. Time will tell.