I think the Nightshade algorithm is less like a poison and more like antibiotics. It works because it’s rare.
If it stops being rare, it will stop working.

Follow

@jes5199 I think it’s excellent experimental work, and I hope it inspires more work in this direction, and the way it’s being presented (not necessarily by its creators) is completely unrealistic.

· · Web · 2 · 0 · 1

@jes5199 (A rule of thumb is that if you can gradient-descend your way into something, you can gradient-ascend your way out of it, and while this is not always true in practice, I think it at least complicates a lot of claims made for unbeatable adversarial this-and-that.)

@vruba indeed. the metaproblem is that there’s no way to tell if it’s working, so people may be happy to adopt some completely ineffective system

Sign in to participate in the conversation
Horsin' Around

This is a hometown instance run by Sam and Ingrid, for some friends.