Although large language models (LLMs) possess extensive world knowledge and strong reasoning capabilities, and are widely regarded as excellent few-shot learners, they still exhibit significant ...