Where Should I Study? Biased Language Models Decide! Evaluating Fairness in LMs for Academic Recommendations

The study on biases in academic recommendations from large language models (LLMs) highlights significant disparities in how these systems operate. By analyzing 360 simulated user profiles, researchers found that LLMs like LLaMA-3.1-8B, Gemma-7B, and Mistral-7B tend to favor institutions in the Global North, perpetuate gender stereotypes, and show a tendency for institutional repetition. Despite LLaMA-3.1 recommending a diverse array of 481 unique universities across 58 countries, systemic biases remain prevalent. This research emphasizes the critical need for a multi-dimensional evaluation framework that not only assesses accuracy but also measures demographic and geographic representation. As LLMs become integral to educational planning, addressing these biases is essential to ensure fair and equitable access to higher education worldwide.