This research introduces Generative Psychometrics for Values (GPV), a novel framework for measuring values in both humans and large language models.

• Creates a data-driven measurement paradigm grounded in text-revealed perceptions
• Enables dynamic parsing of unstructured text into value-related perceptions
• Provides foundation for developing safer, more value-aligned AI systems
• Bridges psychometrics and AI to better understand value representation

Security Implications: GPV offers critical tools for assessing AI safety by measuring value alignment, helping create LLMs that better reflect human ethical priorities and reducing risks of misaligned AI behavior.

Measuring Human and AI Values Based on Generative Psychometrics with Large Language Models