Signal Critical Acclaim Orientation in Large Language Models: Evidence from Film Preference Elicitation

Summary

This study elicited film preferences from eight large language models spanning four model families to test whether LLMs systematically reproduce evaluative hierarchies of cultural products. The authors note that prior research on cultural bias in LLMs has produced competing expectations. Models might mirror the popularity signals common in internet text, or they might instead reproduce forms of prestige embedded in critical discourse. The study's results favor the latter explanation. The tested LLMs systematically oriented toward critical acclaim rather than popularity signals when expressing film preferences. In other words, the models' preferences tracked critics' judgments more closely than mass popularity.