Apparently you can avoid hallucination by basically reading the model’s mind instead of asking it questions.

https://arxiv.org/abs/2212.03827

Raises some ethical issues…