[contrast]

Pointing out that LLMs are deterministic as long as you lock down everything, is like saying an extra bouncy ball doesn’t bounce if you leave it on flat surface, reduce the temperature to absolute zero, and make sure the surface and the ball are at rest before starting the experiment.

It’s true but irrelevant.

One of the GP’s main points was that even the simplest questions can lead to hundreds of different contexts; they probably already know that you could get different outcomes if you could instead have a fixed context.