jcoglanMar 5
just idly remembering how a bunch of guys got incredibly mad when I relicensed *my own work* from MIT to Apache 2
Show threadjcoglanMar 5

the reason I did this is that Apache 2 gives the author greater protection from vexatious lawsuits while placing no real additional burdens on users

if someone can just strip the license off a published work you as an author have *zero* protection

Show threadjcoglanMar 5
i.e. this would be a really effective way to destroy FOSS by exposing authors to an enormous volume of legal risk they were previously shielded from
Show threadRuss Garrett
@jcoglan surely either the original author is an author of the AI-rewritten version and so the license has been violated, or they're not and so they have no liability.
Mar 5 at 3:15pmWeb
Show threadPhilMar 5
@russss @jcoglan IIUC the idea is that, since you cannot copyright AI generated code, you have no way to prevent it from being re-licensed by some 3rd party. You could license it Apache 2, and then someone could just copy it and slap an MIT or whatever license they want on it. AFAIK we still don't really know how these issues would be treated by our legal system, and I'm very much not an expert. But my basic understanding is that copyright is the only/main thing that really prevented people from doing that previously.
Show threadRuss GarrettMar 5

@psFried @jcoglan you can't license something which isn't copyrightable, though. Only the copyright holder can enforce a license violation.

IMO if the LLM's training set included chardet then there's a very good argument that this is a derivative work though.