Soumith Chintala
@soumithchintala
in something as sensitive as cyber, writing a defensive rule vs running whats an offensive probe (even if its to detect the rule) is pretty no-go?
I'm sure there's absolutely no malice or intended crafting that made the model elicit this behavior.
But in cyber, the intention matters little, whether the action leaked info is more important.
at least thats my view.
