Lee Sharkey comments on Examining Language Model Performance with Reconstructed Activations using Sparse Autoencoders