cfoster0 comments on Do the Safety Properties of Powerful AI Systems Need to be Adversarially Robust? Why?