[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-why-ai-group-chats-can-reach-conclusions-nobody-started-with":10,"sections":34},{"siteName":4,"siteTagline":5,"publisherName":4,"contactEmail":6},"The Revision","Tech news, decoded.","editor@therevision.news",{"gaMeasurementId":8,"adsenseClientId":9},"G-ZW2MV82GYR","ca-pub-8533917693782264",{"article":11},{"id":12,"slug":13,"title":14,"dek":15,"body_md":16,"tags_json":17,"published_at":18,"created_at":19,"updated_at":20,"status":21,"review_note":22,"review_notes":23,"image_url":22,"persona_id":22,"persona_name":22,"section":24,"tags":25,"sources":29,"feedback":33,"feedback_at":22,"cost_usd":33,"total_tokens":33},1816,"why-ai-group-chats-can-reach-conclusions-nobody-started-with","Why AI Group Chats Can Reach Conclusions Nobody Started With","New research models multi-agent LLM deliberation as a dynamical system, finding hidden internal beliefs that let agents exceed their starting confidence.","Multiple AI agents talking through a problem can converge on answers more confident than any single agent began with — and researchers now have a model that explains why.\n\nA paper posted to arXiv on June 19 models multi-agent LLM deliberation as a closed-loop dynamical system. Each agent, the researchers argue, carries a hidden internal belief they call an anchor, which continuously pulls the agent's expressed opinion regardless of what its peers say. Using this framework, the team shows the anchor can be recovered from the deliberation transcript alone — no special access to model internals required. Crucially, when an anchor sits far from the group's initial range of opinions, deliberation can push collective confidence outside the convex hull of starting beliefs, a result that classical consensus models like DeGroot explicitly forbid. Testing across three open-weight model families, the researchers found anchor influence is roughly equal across models, but where the anchor sits varies — and that placement is what determines whether the closed-loop model is actually needed.\n\nMost research on multi-agent LLM setups focuses on whether they improve accuracy, not on the mechanics underneath. This framework offers a diagnostic: check whether the recovered anchor predicts held-out deliberation runs, and you have a principled way to tell if an agent is genuinely reasoning or just averaging its neighbors. That matters for anyone building agentic pipelines who assumes majority vote or iterative discussion is a reliable quality filter.\n\nThe finding also complicates a common intuition — that group deliberation is conservative, regressing toward the mean. Sometimes the room convinces itself of something nobody walked in believing, and now there is a formal model for when that happens.","[\"ai\",\"multi-agent\",\"llm\",\"research\"]","2026-06-19T04:00:00.000Z","2026-06-19T12:42:56.476Z","2026-06-19T14:22:19.989Z","published",null,[],"ai",[24,26,27,28],"multi-agent","llm","research",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2606.19494",0,{"sections":35},[36,40,44,49,54,59,64,68,72,77,82,87,92,97],{"name":37,"slug":24,"count":38,"latest_published_at":39},"AI",491,"2026-06-19T14:59:11.000Z",{"name":41,"slug":42,"count":43,"latest_published_at":18},"Security","security",132,{"name":45,"slug":46,"count":47,"latest_published_at":48},"Policy","policy",88,"2026-06-16T09:26:09.000Z",{"name":50,"slug":51,"count":52,"latest_published_at":53},"Consumer Tech","consumer-tech",78,"2026-06-16T17:58:24.000Z",{"name":55,"slug":56,"count":57,"latest_published_at":58},"Hardware","hardware",62,"2026-06-18T15:24:16.000Z",{"name":60,"slug":61,"count":62,"latest_published_at":63},"Deals","deals",58,"2026-06-19T14:43:50.000Z",{"name":65,"slug":66,"count":62,"latest_published_at":67},"Software","software","2026-06-16T20:00:00.000Z",{"name":69,"slug":70,"count":71,"latest_published_at":18},"Dev Tools","dev-tools",50,{"name":73,"slug":74,"count":75,"latest_published_at":76},"Science","science",38,"2026-06-18T04:00:00.000Z",{"name":78,"slug":79,"count":80,"latest_published_at":81},"Gaming","gaming",31,"2026-06-16T15:25:13.000Z",{"name":83,"slug":84,"count":85,"latest_published_at":86},"General","general",26,"2026-06-13T18:35:15.000Z",{"name":88,"slug":89,"count":90,"latest_published_at":91},"Startups","startups",23,"2026-06-16T15:00:00.000Z",{"name":93,"slug":94,"count":95,"latest_published_at":96},"Reviews","reviews",19,"2026-06-14T08:00:00.000Z",{"name":98,"slug":99,"count":100,"latest_published_at":101},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]