Caleb Biddulph comments on The Croissant Principle: A Theory of AI Generalization