I was about to say "Using calibrated models", then I found this interesting paper:
Calibrated Language Models Must Hallucinate
https://arxiv.org/abs/2311.14648
https://www.youtube.com/watch?v=cnoOjE_Xj5g