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

[tailcalled]

Though if X and Y both require some shared limited resource, then the weaker adversarial robustness may essentially require the stronger one. 🤔