> LLMs are probabilistic by design so running the same prompt multiple times will give you different results.

Reasonably deterministic is the phrase. If I can be sure the LLM is giving me back the same result 99% of the time I need it, that's reasonable for me. Maybe this is not reasonable for others.

ie How often will an LLM get 2+2 wrong? Now expand until you're uncomfortable.