Hallucination Detection and Mitigation in Large Language Models

Research Assistant at the Crowd Dyanamics Lab, UIUC, Guide: Prof. Hari Sundaram

Summary: This project aims to use causal inference and activation analysis to understand and mitigate LLM hallucinations.

• Utilizing causal inference and ensembling strategies to improve LLM reasoning on Legal and Medical domain data.
• Using LLM circuit analysis to analyze the impact of toxicity on LLMs and to mitigate its propagation to hallucinations.