SummaryBot comments on ML research directions for preventing catastrophic data poisoning