Researchers from the University of Guayaquil and MIT conducted three experiments to investigate whether large language models (LLMs) discriminate against applicants based on their name ethnicity, institutional prestige, or geographic location. The studies used four LLMs and made a total of 4,320 API calls across five professional domains. The results show that LLMs tend to favor candidates from prestigious institutions over those from less-prestigious ones, with a statistically significant institution-tier gradient effect found in the first study. However, when controlling for geographic location, the prestige effect is more pronounced than the country-of-origin effect. Furthermore, publishing in a top journal like Nature can compensate for low institutional prestige, but this 'rescue effect' varies depending on the candidate's background. The researchers used the Neutrosophic Bias Index (NBI) to quantify the results and found that LLMs exhibit elevated evaluation inconsistency for low-prestige profiles. --- Why it matters: These findings are important because they highlight potential biases in large language models, which can have significant implications for hiring processes and candidate evaluations. Understanding these biases is crucial for developing fairer and more transparent AI systems. Source: https://arxiv.org/abs/2608.18107