Comment by applfanboysbgon

The claim is also just categorically untrue. The largest source of training data by far is publicly available code on e.g. Github, so it mostly just gives you a way to recycle already-available code, without crediting the author, while allowing you to pretend you own it.