#non-determinism

Question 1: Single choice

You send the same prompt to an LLM chat app twice and get two different answers. Which statement best describes how to treat LLM output?

Variation only appears at the highest temperature setting; any lower setting gives identical answers1
The model computes exact results, so its arithmetic and code can be used without checking2
Outputs are sampled and can vary for the same prompt; treat them as a first draft to check and refine3
The model looks answers up in a stored table, so different outputs mean the model was updated in between4