You do realize that an LLM is an enormous decision tree? The prompt is the state and output tokens added to that are the subsequent state that's fed back into the machine. Running the math without deliberately adding any randomness would produce a deterministic output.