I don’t have anything concrete either, but when I was exploring model editing, I was trying to think of approaches that might be able to do something like this. Particularly, I was thinking of things like concept erasure ([1], [2], [3]).
I don’t have anything concrete either, but when I was exploring model editing, I was trying to think of approaches that might be able to do something like this. Particularly, I was thinking of things like concept erasure ([1], [2], [3]).