If they had won then copyright trolls like prenda law would have flooded the internet with AI crap and sued everybody that ever make anything resembling any of their outputs, claiming copyright infringement.

This ruling strongly throttles the ability of copyright trolls to use ML that way because the defendant can much more easily argue they are producing far too many works (and with to many ML-ish obvious errors) to be human made, thus no copyright protection and then you don’t even have to prove you didn’t copy it anymore (but of course you should still try to argue both when you’re the defendant).

Sidenote - derivative works can be protected separately if the addition itself holds creative height. Your copyright only covers your own addition.