LLMs, like every AI tech before them (you know, the ones that aren't called AI anymore because they actually do useful things and are now called "speech recognition" or "OCR" or "translation software" or "chess engines" or such in accordance with Tessler's Theorem), will wind up in a small ghetto of actual utility (my money is on translation being a big feature, though it will require human checking) once the hype has worn down, the actual cost of running one is tallied and people flee using them en masse as a result. When that day comes, it may actually help people.

That day hasn't come, and it won't come at all while Silly Con Valley techbrodouches keep trying to cram it uncomfortably into each and every one of humanity's collective orifices in a vain attempt to force us to like it.

Right now it's shit and it won't get better until techbrodouches are removed from it.