It's one way to look at it, but consider that you need the noise source in case 1 and 0 are strictly equal, necessarily. You can't tell which one is the answer until you decided randomly.
Right, so the LLM needs some randomness to make that decision. The LLM performs a series of deterministic operations until it needs the randomness to make this decisions, there is no randomness within the LLM itself.