A new research paper explores how graph feedback mechanisms can influence consensus and clique formation within populations of open-weight language models. The study, which tested models ranging from 1.1B to 32B parameters, found that while retained partner-label evidence is crucial, homophilous threshold-similarity routing can lead to fragmentation. Conversely, bridge-seeking routing, especially with available memory, demonstrated a greater ability to repair fragmentation and achieve behavioral consensus. The Qwen2.5-32B model, in particular, showed a strong tendency to reach stable consensus with retained history, unlike threshold-similarity routing which failed to achieve consensus in numerous settings. AI
IMPACT This research could inform the design of more cohesive and collaborative multi-agent language model systems.
RANK_REASON The cluster contains a research paper published on arXiv detailing findings about language model populations.
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
