I think it’s useful to draw a Chomsky-esque distinction here between understanding and usefulness.

I think LLMs haven’t advanced our understanding of how human language syntax/semantics work, but they’ve massively advanced our ability to work with it.