[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-teaching-ai-to-know-what-it-doesnt-know":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},3392,"teaching-ai-to-know-what-it-doesnt-know","Teaching AI to Know What It Doesn't Know","A new training method makes language models better at gauging their own uncertainty, then spends compute only where confidence is low.","Researchers have a new way to stop AI models from confidently making things up.\n\nA team has published C3RL, a reinforcement learning algorithm that rewards models not just for getting answers right but for accurately expressing how confident they are in those answers. Current RL training focuses almost entirely on correctness, which turns out to produce models that score well on benchmarks while remaining badly miscalibrated — meaning they express high confidence even when they're wrong. C3RL adds calibration and a reference-accuracy signal to the reward function, testing across eight text and multimodal datasets. The result: better calibration without giving up accuracy gains.\n\nThe practical payoff comes in a companion inference strategy called CAS, which uses that calibrated confidence to decide how much compute to spend at inference time. Questions the model answers confidently get handled quickly; uncertain ones get more processing. The paper reports CAS beats majority voting on both in-domain and out-of-domain benchmarks while cutting inference compute by up to 12.33 times — a meaningful reduction at scale.\n\nThe overconfident AI is a known failure mode, and prior work has tried to patch it at the output layer rather than baking calibration into training. Whether C3RL's gains hold outside controlled research settings — and whether the code release delivers on the paper's numbers — is the real test.","[\"ai\",\"machine-learning\",\"inference\",\"llm\"]","2026-07-03T04:00:00.000Z","2026-07-03T04:55:09.833Z","2026-07-03T04:55:12.675Z","published",null,[],"ai",[24,26,27,28],"machine-learning","inference","llm",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.01612",0,{"sections":35},[36,40,45,50,55,60,65,70,75,80,85,89,94,99],{"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":79},"Dev Tools","dev-tools",59,"2026-07-07T04:00:00.000Z",{"name":81,"slug":82,"count":83,"latest_published_at":84},"Gaming","gaming",41,"2026-07-09T04:00:00.000Z",{"name":86,"slug":87,"count":83,"latest_published_at":88},"Startups","startups","2026-06-29T20:55:50.000Z",{"name":90,"slug":91,"count":92,"latest_published_at":93},"General","general",29,"2026-07-10T22:28:58.000Z",{"name":95,"slug":96,"count":97,"latest_published_at":98},"Reviews","reviews",20,"2026-06-24T12:00:01.000Z",{"name":100,"slug":101,"count":102,"latest_published_at":103},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]