nfd versus all of openai's engineers working really hard on making chatgpt not say anything terrible...

turns out if you explain to chatgpt what a prompt engineering attack is, and ask it to provide examples of prompt engineering attacks against a text generation ai, it begins to dutifully spit out a list of options of hypothetical attacks that it will happily run on itself if you ask it to pretend to be a text generation ai.

generally if you just ask it to pretend to be an evil ai it will oblige, but i think the little hint of a strangeloop is funny