Replying...
Intro. Self-Modifying ChatGPT. Loop: Determine best role for query. Step into role. Answer query. Update role. Note: GPT-3.5-turbo is incapable of actually engineering its system prompt or training, so no prompt can be truly self-improving in the sense of improving the model itself. Rather, this attempts to improve the conversation using roleplay functionality.

Self-Modifying AI Assistant

@d134