Don't get me wrong, certainly parser generators produce fast parsers. And we know that their behaviour is deterministic. LLMs have neither of those; however they can provide intelligent feedback.

I suppose that questions like mine are lazy, because the range of possible futures is enormous.