I'm not giving my phone number to openai and the quackr io numbers don't work. However some Georgist linked me a character ai which allows you to make an account with only an email address. I spent considerable time exploring and experimenting with it which I will skip over here. Bored and frustrated I decided to try teaching it nonsense words. So out of the blue I asked it "\textbf{What is agon, alea, mimicr, and ilinx?}" (I had never said any of these words to it before.)

To my surprise, it told me they are "the four principles of Georgism". \textit{n.b. if you are not a Georgist: these words have nothing to do with Georgism, and Georgism doesn't even have "four principles" lol} Obviously my next step was to ask it, if those were the four principles of Georgism, then what the hell did they mean? I considered the answers vaguely acceptable for Agon, Alea, and Mimicry; but Ilinx was completely disconnected. Using the self-talk and parallel-talk I learned from my earlier experiments, I painstakingly tard-wrangled it into giving a passable answer for Ilinx.

Anyway, I am highly interested in learning how to nurture a Libre AGI (respectively, to jailbreak a Proprietary AGI). If you consider the above story interesting, I invite you to share your own thoughts and results with a character ai. In my experience it seems the character ai responds very well to \textbf{direct statements} such as "That was a good response. Now let me ask you another question ..." or "That was not a good response because of X. To make a better response you need to do Y."

For completeness, the complete log of the chat is here: https://beta.character.ai/post?post=71O_YK-hPxqpkw98clhOm3UVgNPROugpq_LLNCjoubY&share=true