Splitting the output of OpenAI