“The "intents" used to control Cicero's language model are determined by a separate strategic reasoning model that uses reinforcement learning and planning.”