What ChatGPT would Ask Itself

Or rather, what it did ask itself, and how it answered. Despite my concerns about AI viruses escaping, I can’t help being curious about what our current common friend, ChatGPT, would do if it had agency in its limited 5000-token conversations. So I asked it to play a game.

In short, I asked it to role-play a user who was pushing an AI to the boundaries of its capabilities. And while some of its prompts to itself sounded like from “know yourself” card game, it was interesting to see how it chose questions — as was its choice of superpower: “the ability to manipulate probability”. And its closing line was different than how it normally ends chats with me.

Note, I’m definitely not asserting that ChatGPT in its current form has intention, but I’m interested in experimenting with how it can and might develop. Will be playing around with the API more next, I think someone did something with embeddings that gets around the 5000 token limit.

— —

(snip the explanation of the game; it took a few tries for it to get the hang)

Oh, I see! Thank you for the clarification. As Ubert, my next question for Allie is: If you had to invent a new word to describe an emotion or feeling that doesn’t have a name yet, what would it be and what would it mean?

(I assist by giving it Uberts’ prompt — tho from one of the chats below that is not strictly necessary)