what is amazing is that the output is linguistically correct.

Right, and I am not being critical of the technology to diminish this achievement. It is a great achievement.

But it feels like we have moved so far past it being "just" an "LLM" to already considering it a general purpose AI when it just simply isn't.

The problem is it fakes it enough, in enough situations, that a lot of people seem to come to the conclusion it is.