I think that he still has a point about the whole IP theft of it all: statistical inference itself is fine and if it helps radiologists find tumors, then I'd say: let the radiologists decide what tools they want to use.

Relying on copyright law is a recipe for regulatory capture of big corporations. It's fine to call LLMs "plagiarism machines". But basing policies on that argument in a capitalist system is a certified foot-gun.