Value Survey

Value surveys, particularly the World Values Survey (WVS), are increasingly used to benchmark the cultural and moral value alignment of large language models (LLMs). Current research focuses on developing datasets derived from WVS data to assess LLMs' ability to accurately reflect diverse cultural values and identify biases, often employing techniques like in-context learning and Wasserstein distance metrics to quantify alignment. This work is crucial for mitigating harmful biases and promoting fairness in AI, impacting both the development of more ethically sound LLMs and the understanding of cultural values themselves.