Aidan Ewart comments on Sparse Autoencoders: Future Work