Enhancing Large Language Models' Reasoning Capabilities Through Learning From Mistakes

Learning from mistakes can effectively improve the chain-of-thought reasoning capabilities of large language models across various mathematical and commonsense reasoning tasks.