Joseph Miller comments on Normalizing Sparse Autoencoders