I think they're more than that, but I also think they're bullshit generators. Informed bullshit generators even.

Which is quite useful to do some tasks: writing fiction (it now wrote at least 6 characters in multiple rpg campaigns I participate in) and probably others (I used it to write tests data, but in fine I had to revalidate it by hand. Could do better.)

I do not see LLMs be more than that in the near future though. More informed, more accurate bullshit generators? Yes probably.

And I'm not saying it isn't impressive and very close to human behavior. I still think we're missing a part in our models though, and until we find it, breakthrough will be limited to improvements on accuracy and information.