the problem with that is that training can’t be done “immediately” it takes tons of compute

Copyright and IP concerns disappear with an open dataset.

i don’t think i’d agree with that, doesn’t matter if dataset goes open if content went there without consideration for authors

also even things like thispersondoesnotexist were used to mass-create fake identities and such

i’m not. just because he’s an underdog here means that you’re gonna ignore all the harms of generative ai up to this day? it’s like complaining that big oil stole the idea of adding tetraethyllead to gasoline from you and you got no profits from that as a result