epistemic meristem comments on AI #33: Cool New Interpretability Paper