The article by Nikhil Sharma, Q. Vera Liao, and Ziang Xiao explores the effects of Large Language Models (LLMs) powered search systems on information diversity and selective exposure. It presents the results of two experiments investigating whether LLM-powered conversational search systems increase selective exposure compared to conventional search and how opinion biases in LLMs affect the user’s views. Findings reveal that LLM-powered search systems tend to increase biased information querying, with opinionated LLMs reinforcing this bias. The results carry significant implications for the development of LLMs, conversational search systems, and policies governing these technologies.
Publication date: 9 Feb 2024
Project Page: https://doi.org/10.1145/3613904.3642459
Paper: https://arxiv.org/pdf/2402.05880
