This paper addresses concerns about the reliability and trustworthiness of large-scale language models (LLMs). In particular, we focus on the tendency of LLMs to confidently produce hallucinations. The paper extensively reviews existing methods for quantifying uncertainty in LLMs and analyzes their characteristics, strengths, and weaknesses. We systematically categorize the various methods and present examples of their applications in chatbots, text applications, and robotics. Finally, we suggest future research directions for uncertainty quantification in LLMs.
