Uncertainty Quantification Towards Trustworthy Large Language Models

Abstract

Research talk:

Generative Artificial Intelligence (GenAI) has made notable advancements in fields such as computer vision, natural language processing, and drug discovery. However, despite these successes, challenges remain in improving the trustworthiness and reliability of GenAI systems. A significant issue is that the content generated by large language models (LLMs) may include errors or misinformation, limiting their practical use in real-world applications. My research focuses on developing strategies to quantify the prediction uncertainty of LLMs, which is a crucial step toward mitigating these challenges. Future efforts will aim to extend these techniques to multi-step reasoning processes within LLMs, enhancing their reliability further.