Comment by jll29

Here's another techniqe that gets you (even) more out of LLMs: run each prompt several times and compare the various outputs. It is not uncommon for models to contradict their own advice, and also you will get additional insights not included in previous runs.

This is due to the fact that LLMs are statistical processes that rely on pseudo random numbers in chosing what to say and how to say it to a substantial degree.