Bogdan Ionut Cirstea comments on [Linkpost] Generalization in diffusion models arises from geometry-adaptive harmonic representation