There is an error that many in the dispute are making…
Imagine that BORG-AI is an ai that was trained ONLY on GPL2 program-code…
Imagine that you use it to fill-in some functions in your codebase…
What sort of copyright-status should be on those??
I say they should be GPL2, and they should be considered derivative of the ENTIRE training-data-set.
That doesn’t mean I think that the BORG-AI should be a copyright holder, though!
I’m saying that there should be a NEW category, between uncopyrighted & copyrighted, and that the training-sets need to be segregated by license, so that derivatives CAN know what their legal licensing-status is.
GPL2, GPL3, BSD, LGPL2, whatever… it needs to be consistent within the training-data-set, so that the derivative of THAT module/expert can be having the same license, see?