Cutting edge language processing algorithms trained by professionals in a herding manner for years ought to produce something that solves what is basically a parlor trick of language. It might even be a cheat-around on encoded rules but I don't have the paper.

Still I'm willing to go out on an increasingly thin limb and say GPT3 is still playing a really big game of mad libs, and its 'understanding' of the language is more akin to the extremely rudimentary mathematical factoring of sentence structure and grammatical rules. That is, it could be trained to recycle on what others have done for poetry, and spit out something that sounds like what we've read that we call poetry, but it isn't putting poetry out, because it doesn't know what that means...not in like a conscientious way, but a literal, it has no "idea" which to express through a medium, it's just transcribing according to rules, not from say, first principles which then are interpreted through rules.

Again, too much to demand of a really fast adding machine anyway.