LawrenceC comments on Sparse Coding, for Mechanistic Interpretability and Activation Engineering