Nathan Helm-Burger comments on AI #33: Cool New Interpretability Paper