Personality Modeling in Language Models

This field asks whether language models show stable personality-like tendencies, how those tendencies relate to human traits, and whether models can simulate individuals or populations. Researchers measure them with Big Five and other psychometric questionnaires, human-grounded behavioral tasks, interviews, and interaction histories; they also use prompting, fine-tuning, memory, and activation steering to shape persona expression. Benchmarks increasingly test behavior across situations and multiple turns, not just self-reports or fluent role-play. A central tension is that questionnaire scores and apparent persona fidelity may reflect prompt artifacts or response biases rather than behavior, while stronger steering can affect capabilities, safety, and consistency.