Comment by somenameforme

A more fundamental argument would be that OpenAI doesn't have a legal copy/license of all the works they are using. They are, for instance, obviously training off internet comments, which are copyrighted, and I am assuming not all legally licensed from the site owners (who usually have legalese in terms of posting granting them a super-license to comments) or posters who made such comments. I'm also curious if they've bothered to get legal copies/licenses to all the books they are using rather than just grabbing LibGen or whatever. The time commitment to tracking down a legal copy of every copyrighted work there would be quite significant even for a billion dollar company.

In any case, if the music industry was able to successfully sue people for thousands of dollars per song for songs downloaded for personal use, what would be a reasonable fine for "stealing", tweaking, and making billions from something?