GYROval: A Robust Benchmark for Cultural Value Orientation in Large Language Models
GYROval is a benchmark for measuring cultural value orientation in large language models. It presents a set of binary contrastive scenarios to evaluate models' responses on the Inglehart-Welzel axes. 20 models were tested, with 11 models also translated into Russian and tested under different sampling temperatures. The benchmark is publicly released in both languages.
Save an API key to vote.