[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-a-cheaper-way-to-tell-when-an-ai-doesnt-know":10,"sections":40},{"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":30,"tags":31,"sources":35,"feedback":39,"feedback_at":22,"cost_usd":39,"total_tokens":39},3613,"a-cheaper-way-to-tell-when-an-ai-doesnt-know","A Cheaper Way to Tell When an AI Doesn't Know","Researchers derived a single-pass method for measuring neural network uncertainty that works on pretrained models without access to training data.","A new paper on arXiv proposes a faster, lighter way to measure how confident a large language model actually is in its own answers.\n\nThe method, from researchers publishing in arXiv's cs.AI section, skips the expensive machinery most uncertainty tools require. Instead of running thousands of model samples or demanding access to original training data, it uses two mathematical shortcuts: a first-order Taylor expansion and an assumption that model parameters are spread evenly in all directions (the \"isotropic\" part of the title). The result is two uncertainty scores — one for what the model genuinely doesn't know, one for inherent ambiguity in the question — produced in a single forward-backward pass through any pretrained model, unmodified.\n\nThat last part matters. Most serious uncertainty methods either demand training data companies won't share or require retraining the model from scratch. A technique that works on a frozen, off-the-shelf model is far more useful in practice, especially as labs lock down training pipelines. The team validated their approach against Markov Chain Monte Carlo estimates — the slow, expensive gold standard — and found strong agreement that improved as model size increased.\n\nThe results aren't uniformly impressive, though. On TruthfulQA, a benchmark built around questions where plausible wrong answers compete with correct ones, the combined uncertainty score reached the highest mean AUROC. On TriviaQA, a straightforward factual recall test, it fell to near chance — meaning the method picks up something different from simple self-assessment, and that something isn't always the signal you want. Uncertainty tooling that works on hard questions but whiffs on easy ones is a useful diagnostic, not a universal fix.","[\"ai\",\"machine-learning\",\"llms\",\"research\"]","2026-07-03T04:00:00.000Z","2026-07-03T09:38:51.665Z","2026-07-03T09:38:54.398Z","published",null,[24],{"id":25,"reviewer":26,"round":27,"reason":28,"status":29},"editor-r1","editor",1,"The closing paragraph ends on 'an open question' — which is an analytical hedge rather than a finished reader-facing conclusion — and the draft adds no context about who the researchers are or where the paper was published, leaving the sourcing too thin for publication.","resolved","ai",[30,32,33,34],"machine-learning","llms","research",[36],{"name":37,"url":38},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2603.29466",0,{"sections":41},[42,46,51,56,61,66,71,76,81,86,91,95,100,105],{"name":43,"slug":30,"count":44,"latest_published_at":45},"AI",2590,"2026-07-16T04:00:00.000Z",{"name":47,"slug":48,"count":49,"latest_published_at":50},"Security","security",294,"2026-07-15T19:59:48.000Z",{"name":52,"slug":53,"count":54,"latest_published_at":55},"Deals","deals",179,"2026-06-29T20:02:07.000Z",{"name":57,"slug":58,"count":59,"latest_published_at":60},"Policy","policy",158,"2026-07-16T00:02:48.000Z",{"name":62,"slug":63,"count":64,"latest_published_at":65},"Hardware","hardware",122,"2026-07-14T19:46:26.000Z",{"name":67,"slug":68,"count":69,"latest_published_at":70},"Consumer Tech","consumer-tech",93,"2026-07-13T13:20:48.000Z",{"name":72,"slug":73,"count":74,"latest_published_at":75},"Software","software",70,"2026-07-13T19:52:25.000Z",{"name":77,"slug":78,"count":79,"latest_published_at":80},"Science","science",66,"2026-07-10T10:29:37.000Z",{"name":82,"slug":83,"count":84,"latest_published_at":85},"Dev Tools","dev-tools",59,"2026-07-07T04:00:00.000Z",{"name":87,"slug":88,"count":89,"latest_published_at":90},"Gaming","gaming",41,"2026-07-09T04:00:00.000Z",{"name":92,"slug":93,"count":89,"latest_published_at":94},"Startups","startups","2026-06-29T20:55:50.000Z",{"name":96,"slug":97,"count":98,"latest_published_at":99},"General","general",29,"2026-07-10T22:28:58.000Z",{"name":101,"slug":102,"count":103,"latest_published_at":104},"Reviews","reviews",20,"2026-06-24T12:00:01.000Z",{"name":106,"slug":107,"count":108,"latest_published_at":109},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]