initial prompt: we're going ot make two levels of llm: supervisor and question answerer. There is a static question that the question answerer is prompted to anwswer. the supervisor llm is given the correct answer. when the question answerer answers, the supervisor looks to see its score out of 10 independent attempts. then the supervisor can change the prompt for the next quesiton answering round, trying to improve teh quesiton answer's score. then the updated prompt with the static question appended is fed to the answer again, and the iteration continues