I have been doing some roleplay with a a LLM (ChatGPT) to help develop a script. The AI in questions started giving unusual responses, so I started analysising them with another chat. I did some experiments (as suggested by the analysis chat) and aske…