"LLMs are actually pretty deterministic, so there is no need to do more than one attempt with the exact same data."

Is this true? I remember there being a randomization factor in weighing tokens to make the output more something, dont recall what

Obviously I'm not an Ai dev