Re:OpenAI's justification for why training data is fair use,...
by make3 on Thursday October 05, 2023 @01:41PM
(#37781560)
I don't think it fails (3) specifically, and the others are moot if it doesn't fail (3). 99.9999% percent of the time, it just straight up does not reproduce any concrete part of the piece.
It just reproduces some very very hard to quantify tiny fraction of the logic or idea of the document.
It's like saying someone is infringing copyright if it's able to recite a document they read, because it has had an unimaginably tiny effect on their general writing skills, and they're able to recite it if they're asked to.
It just reproduces some very very hard to quantify tiny fraction of the logic or idea of the document.
It's like saying someone is infringing copyright if it's able to recite a document they read, because it has had an unimaginably tiny effect on their general writing skills, and they're able to recite it if they're asked to.