because an ordinary program is deterministic, LLM is probablistic + it has some synthetic-reasoning ability