Researchers have developed a framework called Contagion Networks to quantify how biases from large language models (LLMs) acting as evaluators can spread within multi-agent systems. In experiments using DeepSeek-chat, they observed that evaluator biases consistently propagated between agents, even when using the same underlying model. The study identified three propagation regimes and found that increasing the size of the evaluator committee significantly reduced bias contagion, offering a practical mitigation strategy. The framework and experimental setup are being released as open-source. AI
IMPACT This research provides a method to understand and mitigate bias in multi-agent LLM systems, crucial for developing more reliable AI applications.
RANK_REASON The cluster contains a research paper detailing a new framework and experimental findings on LLM evaluator bias propagation. [lever_c_demoted from research: ic=1 ai=1.0]
Read on arXiv cs.MA (Multiagent) →
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
