> I'm so worried about this case. Treating copyrighted training data as fair use, and letting models learn as a child might learn from a book or movie, is the best way to proceed. It widens the playing field for both development and disruption.

Which also opens the floodgates for bleeding GPL and other copylefted code to proprietary realm. Very convenient, yes.

I’m never OK with someone making a derivation engine which offers my GPL code to a closed source base.