[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-million-agent-ai-sims-were-missing-the-point":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},4075,"million-agent-ai-sims-were-missing-the-point","Million-Agent AI Sims Were Missing the Point","A new method scales social-simulation attribution to over a million agents — and finds that small-sample studies have been crediting the wrong people.","Researchers have found that LLM-powered crowd simulations at realistic scale tell a fundamentally different story than the small panels most studies use.\n\nLarge language model multi-agent systems can model social phenomena like polarization and information cascades at population scale, but attributing those outcomes to individual agents gets combinatorially expensive fast. Existing methods top out around a thousand agents — well short of the million-plus users who actually drive the phenomena researchers are trying to explain. A new paper adapts Aumann-Shapley path-integral attribution to run at million-agent scale, satisfying the same four mathematical axioms as prior approaches while running three to five orders of magnitude faster. The researchers validated it against 1.67 million active Bluesky users across 14 days and five topics.\n\nThe scale gap turns out to matter structurally, not just numerically. At full scale, the long tail and middle tier of users together carry most of the influence weight. Run the same attribution on a convenience sample of a hundred users — as small-scale studies routinely do — and about twice as much influence shifts onto high-follower accounts: 48 percent versus 24 percent at full scale. The paper also proves this cannot be fixed after the fact: a reconciling rescaling factor only exists when the outcome metric is linear over agents, and the nonlinear indicators used in real research produce residuals ranging from 0.10 to 0.98.\n\nDecades of social-media research have leaned on visibility-biased samples that systematically overweight influencers — and this work suggests the same bias has quietly followed LLM simulation into the lab.","[\"ai\",\"multi-agent systems\",\"social simulation\",\"research\"]","2026-07-07T04:00:00.000Z","2026-07-07T16:34:09.253Z","2026-07-07T16:34:12.231Z","published",null,[],"ai",[24,26,27,28],"multi-agent systems","social simulation","research",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2605.11404",0,{"sections":35},[36,40,45,50,55,60,65,70,75,79,84,88,93,98],{"name":37,"slug":24,"count":38,"latest_published_at":39},"AI",2590,"2026-07-16T04:00:00.000Z",{"name":41,"slug":42,"count":43,"latest_published_at":44},"Security","security",294,"2026-07-15T19:59:48.000Z",{"name":46,"slug":47,"count":48,"latest_published_at":49},"Deals","deals",179,"2026-06-29T20:02:07.000Z",{"name":51,"slug":52,"count":53,"latest_published_at":54},"Policy","policy",158,"2026-07-16T00:02:48.000Z",{"name":56,"slug":57,"count":58,"latest_published_at":59},"Hardware","hardware",122,"2026-07-14T19:46:26.000Z",{"name":61,"slug":62,"count":63,"latest_published_at":64},"Consumer Tech","consumer-tech",93,"2026-07-13T13:20:48.000Z",{"name":66,"slug":67,"count":68,"latest_published_at":69},"Software","software",70,"2026-07-13T19:52:25.000Z",{"name":71,"slug":72,"count":73,"latest_published_at":74},"Science","science",66,"2026-07-10T10:29:37.000Z",{"name":76,"slug":77,"count":78,"latest_published_at":18},"Dev Tools","dev-tools",59,{"name":80,"slug":81,"count":82,"latest_published_at":83},"Gaming","gaming",41,"2026-07-09T04:00:00.000Z",{"name":85,"slug":86,"count":82,"latest_published_at":87},"Startups","startups","2026-06-29T20:55:50.000Z",{"name":89,"slug":90,"count":91,"latest_published_at":92},"General","general",29,"2026-07-10T22:28:58.000Z",{"name":94,"slug":95,"count":96,"latest_published_at":97},"Reviews","reviews",20,"2026-06-24T12:00:01.000Z",{"name":99,"slug":100,"count":101,"latest_published_at":102},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]