[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-better-diagram-formatting-lifts-ai-geometry-scores-by-14-points":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},10927,"better-diagram-formatting-lifts-ai-geometry-scores-by-14-points","Better Diagram Formatting Lifts AI Geometry Scores by 14 Points","GeoReform treats how AI geometry problems are formalized as an optimizable policy, lifting one model's accuracy from 42% to 56% on a benchmark.","Feeding AI models more geometry facts doesn't make them better at geometry - feeding them the right facts, organized well, does.\n\nResearchers tested multimodal AI models on geometry diagrams, a task where models already struggle to read relationships between shapes and lines. A common fix - converting diagram elements into explicit text descriptions the model can reason over - turns out to be a mixed bag: on the Geometry3K benchmark, this \"structure injection\" fixed 28 errors but created 13 new ones out of 200 examples. The researchers trace the damage to redundant relations that distract the model and ambiguous references that cause it to apply the wrong constraint. Their response is GeoReform, a framework that treats the formalization step itself as something to optimize - running the full reasoning pipeline, finding failures, diagnosing what went wrong, and revising how entities and constraints get selected, grouped, and presented.\n\nThe headline number is a jump from 42% to 56% accuracy on Qwen3VL-2B, a meaningful gain on a notoriously hard benchmark category. But the more useful finding is that bolting structured text onto multimodal models isn't a free win - extra context can quietly make reasoning worse if nobody checks what it actually does to the model's answers.\n\nIt's the same lesson prompt engineers keep relearning with chatbots: dumping in more information isn't the same as dumping in the right information.","[\"ai\",\"multimodal models\",\"research\",\"benchmarks\"]","2026-10-09T04:00:00.000Z","2026-10-09T22:03:01.870Z","2026-10-09T22:03:05.955Z","published",null,[],"ai",[24,26,27,28],"multimodal models","research","benchmarks",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.12391",0,{"sections":35},[36,39,43,48,53,57,61,66,71,76,81,86,91,96],{"name":37,"slug":24,"count":38,"latest_published_at":18},"AI",6708,{"name":40,"slug":41,"count":42,"latest_published_at":18},"Security","security",931,{"name":44,"slug":45,"count":46,"latest_published_at":47},"Policy","policy",486,"2026-10-08T22:40:11.000Z",{"name":49,"slug":50,"count":51,"latest_published_at":52},"Deals","deals",474,"2026-10-08T22:00:00.000Z",{"name":54,"slug":55,"count":56,"latest_published_at":18},"Hardware","hardware",231,{"name":58,"slug":59,"count":60,"latest_published_at":18},"Science","science",192,{"name":62,"slug":63,"count":64,"latest_published_at":65},"Consumer Tech","consumer-tech",181,"2026-10-08T23:26:35.000Z",{"name":67,"slug":68,"count":69,"latest_published_at":70},"Startups","startups",117,"2026-10-08T16:45:00.000Z",{"name":72,"slug":73,"count":74,"latest_published_at":75},"Software","software",114,"2026-10-08T17:57:01.000Z",{"name":77,"slug":78,"count":79,"latest_published_at":80},"Dev Tools","dev-tools",105,"2026-10-07T16:59:11.000Z",{"name":82,"slug":83,"count":84,"latest_published_at":85},"General","general",66,"2026-10-09T04:46:11.000Z",{"name":87,"slug":88,"count":89,"latest_published_at":90},"Gaming","gaming",58,"2026-10-08T20:08:45.000Z",{"name":92,"slug":93,"count":94,"latest_published_at":95},"Reviews","reviews",34,"2026-10-08T14:00:22.000Z",{"name":97,"slug":98,"count":99,"latest_published_at":100},"How-To","how-to",8,"2026-10-05T09:00:00.000Z"]