Imo, I’d prefer a “contamination” approach where the strictest licenses in the training data applies to all outputs. I doubt such a rule would get through big business filters but it would maximize the public good and any country that does manage something like that would probably gain the most benefits from these companies.
The strictest license in the training set is definitely just the normal copyright protections, which is more strict than copyleft.
Edit: explanation, this is because everything is inherently copyrighted. You have rights to protections. So you’re forgoing some of those protective rights by licensing it out.
Imo, I’d prefer a “contamination” approach where the strictest licenses in the training data applies to all outputs. I doubt such a rule would get through big business filters but it would maximize the public good and any country that does manage something like that would probably gain the most benefits from these companies.
The strictest license in the training set is definitely just the normal copyright protections, which is more strict than copyleft.
Edit: explanation, this is because everything is inherently copyrighted. You have rights to protections. So you’re forgoing some of those protective rights by licensing it out.