@kirk @pluralistic Oh, this goes so much deeper. The steps you would take to tune such a model would be nearly identical, other than the choice of source material and base model, to creating a salacious video generation model tuned to a particular person.
That feels quite wrong. So why does my hypothetical not feel as wrong? I think it has to do with consent. Cory explicitly consented to giving the speech in that form, with his voice, to an audience. He also consented to giving me the text.
Thus, the act I would hypothetically be creating is a recreation of something he has already done for another audience, and has publicly shared that performance in one medium (text). In the... other case... what's being created is explicitly not consented to, categorically, and would be objected to by most.
I wonder if there's some semi-novel right here. Probably not entirely new, but a scope of which has heretofore not been considered. There are likeness rights, but South Park can create a cartoon of anyone they want, and that doesn't run afoul of US law (I'm sure at one point they got sued but I don't recall details and I'm too lazy to look them up today, so 'citation needed'). It's probably context specific, perhaps related to the bounds around defamation.
This has probably been given a better treatment by legal scholars. Playing with the idea is fun, though.
(Sorry to use you as a hypothetical subject, Cory)
