Nick Jiang comments on Towards data-centric interpretability with sparse autoencoders