The Structure, Inflation, and Malleability In Large Language Models’ Representations of Human Self–Other Personality Judgments.

We examined LLMs’ representations of human self- and other-personality judgments across five studies. LLMs reproduced a self-enhancement pattern but overestimated its magnitude. Targeted prompt design and empirical anchoring recalibrated these biases. These results advance the reliability, validity, and safety of LLMs in psychological modeling and human-centered applications.