Chain-of-Thought Prompting Enhanced by GPT-4 for Improving Reasoning Capabilities in Large Language Models

The rapid evolution of artificial intelligence has led to significant improvements in complex NLP tasks such as arithmetic and reasoning. Recently, many studies have focused on these challenging tasks to enhance model performance. In this paper, we explore enhancing these capabilities in large language models through chain-of-thought prompting. By first identifying incorrect responses, we then leverage GPT-4's advanced abilities to annotate the rationale, facilitating further inference based on chain-of-thought prompting. Furthermore, adopting the selfconsistency approach ensures a more robust inference process, improving the overall accuracy of model outputs. Experimental results demonstrate that our method can significantly enhance the arithmetic and reasoning skills of large language models, offering a scalable method to refine AI problem-solving with minimal manual effort.