[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-majority-voting-makes-small-ai-models-worse-at-hard-science":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},5526,"majority-voting-makes-small-ai-models-worse-at-hard-science","Majority Voting Makes Small AI Models Worse at Hard Science","A new study finds that self-consistency voting lowers accuracy on most hard GPQA problems for small LLMs, and simple confidence fixes don't help either.","Asking a small AI model the same hard science question 64 times and picking the most common answer often makes it wrong more often, not less.\n\nA new arXiv paper tested self-consistency - the common trick of sampling multiple answers and taking a majority vote - on GPQA Diamond, a benchmark of hard science questions. For Qwen2.5-7B, majority voting hurt accuracy on 56.6% of the problems it was tested on; for Llama-3-8B, it hurt 65.7%. The researchers then tried the obvious fix: a confidence gate that skips voting when the model's token-level uncertainty looks low. That gate failed too, because the entropy measurement tracks how verbose or hedgy the model's writing is, not how sure it actually is about the answer - on Qwen2.5-7B-Instruct-Turbo, samples that contradicted their own problem's majority answer still looked confident most of the time, with 75.7% scoring above a 10-nat confidence threshold, a finding that held up when re-tested on a separate, pre-registered set of 69 problems.\n\nThat matters for anyone building a system that routes between a cheap single answer and an expensive multi-sample vote based on a confidence score. The signal discriminates fine across different questions but is close to useless for the within-question call a router actually needs to make, and a second natural fix - checking whether the top two answers agree - also failed, with no explanation the authors could pin down. It's a useful check on the assumption that tricks proven on huge frontier models transfer cleanly down to the 7B-8B models many teams actually run.\n\nThe paper's honesty extends to its own limits: a planned replication on a second model couldn't be evaluated, and three reasoning-focused open models each failed to evaluate on hosted serverless inference at a small budget, for three separate reasons the paper details - all three are downloadable, so the bottleneck was the metered API, not the models themselves.","[\"ai\",\"llm-evaluation\",\"self-consistency\",\"small-models\"]","2026-08-18T04:00:00.000Z","2026-08-18T23:58:21.780Z","2026-08-18T23:58:33.720Z","published",null,[],"ai",[24,26,27,28],"llm-evaluation","self-consistency","small-models",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2608.11403",0,{"sections":35},[36,40,44,49,54,59,64,69,74,78,83,88,93,98],{"name":37,"slug":24,"count":38,"latest_published_at":39},"AI",3293,"2026-08-20T04:00:00.000Z",{"name":41,"slug":42,"count":43,"latest_published_at":39},"Security","security",435,{"name":45,"slug":46,"count":47,"latest_published_at":48},"Policy","policy",210,"2026-08-19T09:32:27.000Z",{"name":50,"slug":51,"count":52,"latest_published_at":53},"Deals","deals",179,"2026-06-29T20:02:07.000Z",{"name":55,"slug":56,"count":57,"latest_published_at":58},"Hardware","hardware",140,"2026-08-19T18:25:42.000Z",{"name":60,"slug":61,"count":62,"latest_published_at":63},"Consumer Tech","consumer-tech",95,"2026-08-18T16:05:00.000Z",{"name":65,"slug":66,"count":67,"latest_published_at":68},"Science","science",90,"2026-08-19T18:41:02.000Z",{"name":70,"slug":71,"count":72,"latest_published_at":73},"Software","software",73,"2026-08-18T07:51:50.000Z",{"name":75,"slug":76,"count":77,"latest_published_at":18},"Dev Tools","dev-tools",69,{"name":79,"slug":80,"count":81,"latest_published_at":82},"Startups","startups",47,"2026-08-19T19:13:46.000Z",{"name":84,"slug":85,"count":86,"latest_published_at":87},"Gaming","gaming",41,"2026-07-09T04:00:00.000Z",{"name":89,"slug":90,"count":91,"latest_published_at":92},"General","general",33,"2026-08-18T22:18:13.000Z",{"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"]