Neel Nanda comments on Sparse Autoencoders: Future Work