Joar Skalse comments on Towards Guaranteed Safe AI: A Framework for Ensuring Robust and Reliable AI Systems

[Joar Skalse]

I think the distinction between these two cases often can be somewhat vague.

Why do you think that the adversarial case is very different?