Do large language models judge
facial attractiveness like humans?

Human judgments of attractiveness are shaped by our evolved psychology and socioecology, and have been well-studied since the 1960s. In contrast, large language models’ perceptions of attractiveness have entirely different origins and are comparatively understudied, with only a handful of investigations in the literature. As large language models (LLMs) become collaborators in clinical and consumer aesthetics, the question of how these models assess attractiveness and how its facial evaluations are similar or different from ours becomes pressing.

 To help answer this question, our scientists are running an empirical research study comparing four frontier models against an existing dataset of human ratings of attractiveness. The findings will be published openly alongside our methodology.