Oliver Daniels comments on Scalable End-to-End Interpretability