Should a proper large language model be able to generate arguments for and against any side of any debate?