Lijia Ma’s research takes a deep dive into the process of LLM-powered search engines and how they craft intelligent chat responses. Read the full paper here.

• Analysis of Bing Chat’s preference in content selection.
• A focus on readable, structured, and predictable text by LLMs.
• Bing Chat’s text preferences appear intrinsic to the underlying language models.
• RAG technology shows higher similarity in website citations than conventional search engines.

This insightful exploration into the cognitive processes of search engines built on LLMs provides key understandings into how these advanced technologies may mold the future of information search and dissemination.