Exploring the Impact of Explainability in Large Language Model (LLM) Applications on User Experience

Abstract

Due to the "black-box" nature, explainability has long been a significant research topic in machine learning. Researchers have been committed to explaining model principles behind models and their scope of influence and decision-making to experts and technical practitioners. However, with the increasing popularity of the Large Language Model (LLM), more general users interact with these applications, bringing new challenges for explainability. This study explores the impact of LLM explainability on trust and satisfaction, revealing that both are significantly influenced by the degree and presentation of explainability. Moreover, trust and satisfaction vary across different risk scenarios. The study further evaluates the pros and cons of different explainability strategies, offering practical insights for the design of LLM applications.

Ähnliche Arbeiten