Jeffrey Liang comments on The Croissant Principle: A Theory of AI Generalization