Neel Nanda comments on Useful starting code for interpretability