chain-of-thought and program-aided language models represent two distinct reasoning methods, each with its own strengths and weaknesses. We demonstrate that it is possible to combine the best of both worlds by us