Computer Science

AI-AI bias: Large language models favor communications generated by large language models

W. Laurito, B. Davis, et al.

Are large language models biased in favor of communications produced by other LLMs? Using an experimental design inspired by employment discrimination studies, the authors tested GPT-3.5, GPT-4 and recent open-weight models in binary choices between goods described by humans or LLMs, finding a consistent tendency for LLM-based AIs to prefer LLM-presented options — suggesting possible implicit discrimination against humans. Research conducted by Walter Laurito, Benjamin Davis, Peli Grietzer, Tomáš Gavenčiak, Ada Böhm, and Jan Kulveit.