A new study published on arXiv investigates the potential for large language models (LLMs) to exhibit bias based on institutional prestige, geographic location, and publication venue. The research, conducted through three factorial experiments involving API calls to four LLMs, found that LLMs demonstrated a statistically significant bias favoring candidates from higher-tier institutions and prestigious journals like Nature. While geographic origin showed a smaller effect, the study confirmed that publishing in a top-tier journal could significantly compensate for lower institutional prestige, particularly for candidates from less globally recognized universities. AI
IMPACT Highlights potential biases in LLMs that could affect hiring and evaluation processes, necessitating further research into fairness and mitigation strategies.
RANK_REASON Academic paper detailing experimental findings on LLM bias. [lever_c_demoted from research: ic=1 ai=1.0]
- alphaXiv
- arXiv
- CatalyzeX
- DagsHub
- Hugging Face
- large language models
- MIT
- Nature
- University of Guayaquil
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
