Reproduce “Enhancing Chain of Thought Prompting in Large Language Models via Reasoning Patterns” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yufeng Zhang, Xuepeng Wang, Lingxiang Wu, Jinqiao Wang
This paper introduces Pattern-CoT, a method that enhances Chain of Thought prompting by selecting in-context demonstrations based on their underlying reasoning patterns rather than semantic similarity. The approach consistently improves performance across various reasoning tasks and LLM architectures, demonstrating that explicit pattern-based selection leads to more robust and interpretable reasoning.
Key insights
- LLMs are influenced more by the structural patterns and templates of reasoning within demonstrations than by the factual correctness of the examples themselves.
- Selecting demonstrations based on explicit reasoning patterns can effectively mitigate noise introduced by semantic-based selection methods.
- The method offers enhanced interpretability, allowing for a clearer understanding of how Chain of Thought functions and guides LLMs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.