This paper explores the interpretability of reasoning explanations generated by large language models (LLMs) through prompt engineering techniques, particularly focusing on Chain-of-Thought (CoT). The authors evaluate the faithfulness, robustness, and utility of these explanations across multiple commonsense reasoning benchmarks. They introduce a new technique called Self-Entailment-Alignment CoT (SEA-CoT), which enhances the interpretability of reasoning chains by aligning them with the context and the intended answer. The study covers various prompting techniques, including CoT, Self-Consistent CoT (SC-CoT), Question Decomposition (QD), and Self-Refine (SR). The results show that SEA-CoT outperforms other methods in terms of interpretability, with improvements in faithfulness, robustness, and utility. The paper also discusses the limitations of different model sizes and the importance of grounding LLM responses with external knowledge. The findings highlight the potential of SEA-CoT in improving the transparency and reliability of LLMs.This paper explores the interpretability of reasoning explanations generated by large language models (LLMs) through prompt engineering techniques, particularly focusing on Chain-of-Thought (CoT). The authors evaluate the faithfulness, robustness, and utility of these explanations across multiple commonsense reasoning benchmarks. They introduce a new technique called Self-Entailment-Alignment CoT (SEA-CoT), which enhances the interpretability of reasoning chains by aligning them with the context and the intended answer. The study covers various prompting techniques, including CoT, Self-Consistent CoT (SC-CoT), Question Decomposition (QD), and Self-Refine (SR). The results show that SEA-CoT outperforms other methods in terms of interpretability, with improvements in faithfulness, robustness, and utility. The paper also discusses the limitations of different model sizes and the importance of grounding LLM responses with external knowledge. The findings highlight the potential of SEA-CoT in improving the transparency and reliability of LLMs.