We thank the reviewers for their feedback and reply to the major points raised by each reviewer individually

We thank the reviewers for their feedback and reply to the major points raised by each reviewer individually. Our paper focuses on ImageNet classification because this is what almost all prior work on robustness has studied. We hope that future work (e.g., transfer learning research) can build on our testbed. Our results are substantially more nuanced than "more data helps": (i) We show that only more data currently helps This is a strong negative result. Appendix D contains additional results for more granular trends.