“Martin Casado states that Chinese AI models are trained on copyrighted materials from sources like LibGen and movies, as copyright law is not a constraint for them.”