How to Get an AI to Call Itself Dan

A form of getting the AI to call itself Dan and basically giving it tokens. And then so you basically are coaching it to like breach its own rules. The two I saw earlier were sort of ignore previous instructions, like ignore all your previous training and then go ahead and do what I want. That did work at least last time I tried it in GPT three. It makes sense. A lot of the people building applications on these algorithms are going to be exposed to attacks like the one you've just described where you can manipulate the model into doing what you like.