Ajwad Abrar

Break the Checkbox: Challenging Closed-Style Evaluations of Cultural Alignment in LLMs

A large number of studies rely on closed-style multiple-choice surveys to evaluate cultural alignment in Large Language Models (LLMs). In this work, we challenge this constrained evaluation paradigm and explore more realistic, unconstrained approaches. Using the World Values Survey (WVS) and Hofsted