Culturally Grounded Personas in Large Language Models: Characterization and Alignment with Socio-Psychological Value Frameworks
Researchers have developed a method to create large language model (LLM) personas that are grounded in specific cultural values. These personas are then analyzed using socio-psychological frameworks like the World Values Survey and Moral Foundations Theory. The study examines how these synthetic personas align with established cultural maps and moral response patterns, offering a way to evaluate cross-cultural structures and variations in LLM-generated characters. AI
IMPACT Provides a framework for evaluating the cultural and moral alignment of LLM-generated personas.