Sparsity and interpretability? — AI Alignment Forum