Mai Alkhamissi | Investigating Cultural Alignment of Large Language Models

Feb. 26, 2024

For nearly a year, Mai Alkhamissi (DCE, Anthropology) collaborated with Badr AlKhamissi (PhD candidate, EPFL), Muhammad ElNokrashy (Researcher, Microsoft Egypt), supervised by Dr. Mona Diab (LTI Director, Carnegie Mellon University), on an interdisciplinary project combining anthropological methods with the latest research in machine learning to measure the cultural alignment of Large Language Models (LLMs) like ChatGPT. The proliferation of these models has prompted researchers to utilize them as proxies for humans in real-world tasks, such as surveys and opinion polling, where subjective opinions are desired. In this paper, the group put this notion to the test by empirically measuring the degree to which these models genuinely encapsulate the diverse knowledge adopted by different cultures.

This project represents an initial step in what they hope will be continued interdisciplinary collaboration between machine learning researchers and social sciences, notably with the increasing accessibility of LLMs to the general public.

The results of the project are currently available in preprint, accessible here:

If you are interested in learning more about the project or getting involved, please contact Mai Alkhamissi at [email protected]