Noa Nabeshima comments on Efficient Dictionary Learning with Switch Sparse Autoencoders