Scenario-based Probing and Steering Cultural Values in Large Language Models--Extended Version

In three linesarXiv paper on probing cultural values in LLMs through scenario-based behavioral dilemmas mapped to World Values Survey axes. Authors apply activation steering across three open-source models to measure and shift implicit preferences along Inglehart-Welzel dimensions, revealing latent entanglement where interventions on one cultural axis induce shifts on another.

Summary generated by Claude — human-verified