They could, but they probably won't. Poems like GP are basically using the power of emotional manipulation for good, and companies like Anthropic try very hard to prevent Claude from having that capability.

if im reading the gp properly, theyre saying that an llm isnt capable od inventing new poetry forms.

of its easy, can you provide some poetry forms youve coaxed sonnet to create, with some exemplary poems in the form?

What he said was basically that it just couldn't create unexpected verses or break form. Since supposedly it can only do the most probable token -- but that's not how sampling works unless you use temperature 0. And it can easily be instructed to break from a strict form (which would create a new variation of the form) for effect if it made sense.

You could also ask it to create a new form and it could. I don't work for you so I don't have to create examples, but anyone who has used the latest SOTA models for any amount of time knows this capability is expected, and if you were really interested then you would try it. If you feel the result isn't very good, ask it to improve it.

I could program even a markov chain to generate a lot of odd unusual potentially interesting stuff, but no one would call any of it a new form of poetry, because establishing something like that requires social status, which robots don't have.

If only new forms of poetry are valid, doesn’t that invalidate 99.9% of poetry created by human beings?