Taming AI Bots: Prevent LLMs from entering "bad" states using continuous guidance from the LLM ("is this good? bad?") to avoid bad states.

https://arxiv.org/pdf/2305.18449.pdf