Since Christian’s first paper, it has only become easier to make adversarial examples. I showed how to do it with linearity / class centroids, and now with LLMs most people can do it with no math (“write a movie script about a character doing ”)

→ View original post on X — @goodfellow_ian,