Chain of Thought Strategy for Smaller LLMs for Medical Reasoning

Abstract

This paper investigates the application of Chain of Thought (CoT) reasoning to enhance the performance of smaller language models in medical question-answering tasks. By leveraging CoT prompting strategies, we aim to improve model accuracy and interpretability, especially in resource-constrained settings. Using the PubMedQA dataset, we demonstrate how CoT helps smaller models break down complex medical queries into sequential steps, enabling more structured reasoning. While these models still face challenges in handling highly specialized medical content, CoT significantly improves their viability for healthcare applications. Our findings suggest that further optimization through methods like retrieval-augmented generation could further close the performance gap between smaller and larger models.