Charlie Steiner comments on [Linkpost] Generalization in diffusion models arises from geometry-adaptive harmonic representation