AI RESEARCH
UGID: Unified Graph Isomorphism for Debiasing Large Language Models
arXiv CS.AI
•
ArXi:2603.19144v1 Announce Type: cross Large language models (LLMs) exhibit pronounced social biases. Output-level or data-optimization--based debiasing methods cannot fully resolve these biases, and many prior works have shown that biases are embedded in internal representations.