Matthew A. Clarke comments on Are Sparse Autoencoders a good idea for AI control?