[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-new-benchmark-tests-if-ai-can-ask-the-right-math-question":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},4471,"new-benchmark-tests-if-ai-can-ask-the-right-math-question","New Benchmark Tests If AI Can Ask the Right Math Question","MIRA-Math isolates a narrow but revealing skill: whether a model knows what it doesn't know and can ask for exactly that missing fact.","A new benchmark wants to measure something most math tests ignore: whether an AI can identify what it's missing and ask for it.\n\nMIRA-Math, introduced in a paper posted to arXiv, contains 2,310 problems drawn from 22 mathematical families — algebra, probability, signal processing, Markov chains, and more. Each problem has a unique answer, but the model is deliberately given an incomplete picture. One necessary fact is withheld. The model must request that fact in natural language, within a strict budget, then use the returned information to produce an exact answer. A constrained responder checks whether the request actually matches the missing fact; if not, it declines.\n\nThe benchmark surfaces something prior evaluations flatten over: request success and answer accuracy are separable. A model can identify the right missing fact and still botch the underlying computation. Conversely, it can fail to surface the right request and never reach a correct answer. That two-axis view is more diagnostic than leaderboard scores that only reward final outputs — and it points at a gap between knowing math and knowing what you don't know.\n\nMost existing math benchmarks hand models a complete problem statement and grade the answer. Interactive benchmarks add tools and retrieval but rarely isolate information-requesting as a distinct, testable skill. MIRA-Math's constrained, deterministic setup makes it reproducible in a way that open-ended dialogue evaluations often aren't — generators, verifiers, prompts, and run metadata are all being released publicly.\n\nThe research arrives as labs race to claim stronger reasoning on benchmarks like MATH and AIME. Whether frontier models can strategically seek out what they're missing — rather than hallucinate a substitute — is a quieter question. Now there's at least a formal way to check.","[\"ai\",\"benchmarks\",\"math\",\"llm\"]","2026-07-09T04:00:00.000Z","2026-07-09T04:46:37.509Z","2026-07-09T04:46:40.472Z","published",null,[],"ai",[24,26,27,28],"benchmarks","math","llm",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.07391",0,{"sections":35},[36,40,45,50,55,60,65,70,75,80,84,88,93,98],{"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":79},"Dev Tools","dev-tools",59,"2026-07-07T04:00:00.000Z",{"name":81,"slug":82,"count":83,"latest_published_at":18},"Gaming","gaming",41,{"name":85,"slug":86,"count":83,"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"]