And before someone says the Western models are copying others - there’s a difference between fair use of publicly visible information to train a model, which is like how humans learn, and creating tens of thousands of fake accounts and maliciously avoiding detection to violate the terms of a private service, which is more like a cyberattack.