I think I prefer the Emily Bender approach of asserting that no one should be allowed to train deep learning models at all. If you're going to claim some sort of authority over a technology you don't actually develop then you might as well go hard.