{"slug": "graph-feedback-controls-consensus-and-clique-formation-in-open-weight-language", "title": "Graph Feedback Controls Consensus and Clique Formation in Open-Weight Language-Model Populations", "summary": "Researchers at arXiv studied convention formation in open-weight language-model populations using a naming-game protocol, finding that graph feedback controls consensus and clique formation. Threshold-similarity routing amplified fragmentation, while bridge-seeking routing often repaired consensus when memory was available. The results highlight the critical role of interaction graph design in multi-agent LM systems.", "body_md": "arXiv:2607.12077v1 Announce Type: new\nAbstract: Multi-agent language-model systems increasingly route local interactions, yet the runtime interaction graph is often treated as an implementation detail. We study convention formation in open-weight LM populations spanning 1.1B-32B parameters with a naming-game protocol. Restricted first-token scores over tokenizer-safe labels let us measure prompt-conditioned score-state distributions, construct state-similarity graphs, and separate sampled-label agreement from latent state-space consensus. Across controlled interventions, in the main open-weight repair grids, retained partner-label evidence is necessary but not sufficient: homophilous threshold-similarity routing deletes cross-basin exposure and amplifies fragmentation, while bridge-seeking routing often repairs fragmentation when memory is available. In a three-seed mixed four-model grid, threshold-similarity produces no final behavioral or state consensus in 189 setting-seed runs, whereas state-component and label-disagreement bridges recover final behavioral consensus in 14/18 retained-memory runs. Across homogeneous model populations, retained history generally shifts fragmented dynamics toward consensus; the clearest case is Qwen2.5-32B, which reaches stable behavioral and final state consensus in all 18 retained-history well-mixed settings, while threshold-similarity reaches neither form of consensus in 189 settings. Robustness over state thresholds, population size, and vocabulary size preserves the qualitative ordering, and early-window graph-energy features provide useful within-grid diagnostics.", "url": "https://wpnews.pro/news/graph-feedback-controls-consensus-and-clique-formation-in-open-weight-language", "canonical_source": "https://arxiv.org/abs/2607.12077", "published_at": "2026-07-15 04:00:00+00:00", "updated_at": "2026-07-15 04:22:37.788156+00:00", "lang": "en", "topics": ["large-language-models", "artificial-intelligence", "ai-research", "ai-agents"], "entities": ["arXiv", "Qwen2.5-32B"], "alternates": {"html": "https://wpnews.pro/news/graph-feedback-controls-consensus-and-clique-formation-in-open-weight-language", "markdown": "https://wpnews.pro/news/graph-feedback-controls-consensus-and-clique-formation-in-open-weight-language.md", "text": "https://wpnews.pro/news/graph-feedback-controls-consensus-and-clique-formation-in-open-weight-language.txt", "jsonld": "https://wpnews.pro/news/graph-feedback-controls-consensus-and-clique-formation-in-open-weight-language.jsonld"}}