Deng Pan

Deng Pan

Postdoctoral Researcher at the University of Notre Dame specializing in Responsible AI. My research integrates interpretability, safety, fairness and robustness in both traditional machine learning and modern large language models, aiming to build transparent and accountable AI systems that advance responsible and trustworthy AI.

On the Unstable Impact and Bias Propagation of Chain-of-Thought Reasoning in Large Language Models

Published:

Abstract: This work examines the stability and bias propagation characteristics of chain-of-thought (CoT) reasoning in large language models. Chain-of-thought prompting has emerged as a powerful technique for improving the reasoning capabilities of LLMs, but its impact on model behavior and potential for bias amplification remains underexplored. Our research reveals that CoT reasoning can exhibit unstable behavior across different prompts and contexts, and may inadvertently propagate or amplify existing biases in model outputs. We provide a comprehensive analysis of these phenomena and discuss their implications for the responsible deployment of LLM-based reasoning systems.

Status: Under review at Nature Machine Intelligence

Recommended citation: Pan, D., Germino, J., Ma, Y., Elizabeth, D., Chawla, N. (2025). "On the Unstable Impact and Bias Propagation of Chain-of-Thought Reasoning in Large Language Models." Under review at Nature Machine Intelligence.