Occupational Prompting Reveals Cultural Bias in Large Language Models

The paper presents a study on the influence of occupational prompting on the responses of open-weight large language models (LLMs) to value-survey questions, extending previous research that focused on nationality-based cultural prompting. By analyzing responses from models prompted with various occupations such as accountant, teacher, engineer, and nurse, the authors found that while responses remained within a predominantly Western cultural region, distinct occupational identities introduced specific biases in value expression. This research highlights the importance of understanding how professional roles can shape the output of LLMs, providing a new framework for evaluating cultural biases in AI systems.