Llms are easily made deterministic by choosing the selection strategy. More than being deterministic they are also fully analayzable and you don't run into issues like the halting problem if you constrain the output appropriately.