[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-one-agent-many-video-questions-far-fewer-passes":10,"sections":35},{"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":30,"feedback":34,"feedback_at":22,"cost_usd":34,"total_tokens":34},10985,"one-agent-many-video-questions-far-fewer-passes","One Agent, Many Video Questions, Far Fewer Passes","A new research agent answers a video's full question set in one shared pass, cutting reasoning rounds by 86 percent without losing accuracy.","A new AI agent answers every question about a video in one pass instead of starting over each time.\n\nVAMR, described in a new arXiv paper, is a video-understanding agent built to handle multiple questions about the same long video through a single shared tool-use trajectory, rather than relaunching video search for each question separately. At each step, a persistent policy model pulls visual evidence for one or more unanswered questions at once and submits answers once there is enough evidence, while a layered memory system keeps a shared understanding of the video alongside separate evidence for each individual question. The system is trained with supervised fine-tuning first, then a method called question-horizon policy optimization, which credits each question for the specific rounds that actually helped it, even though different questions finish at different times. On the LVBench benchmark, VAMR reached 62.1% accuracy, 4.3 points above a prior method called VideoARM, while cutting reasoning rounds by 85.9% and processed frames by 61.4%.\n\nMost video-understanding research still tests one question at a time, which doesn't match how people actually use these systems - firing off a batch of questions about the same lecture, meeting recording, or security footage. VAMR's trick of sharing exploration across questions matters less for the accuracy bump and more for the compute savings, since fewer frames and fewer reasoning rounds translate directly into lower cost at scale.\n\nThe gains are measured on three benchmarks against one named rival, not an independent evaluation, so how well this holds up on messier, real-world footage is still an open question.","[\"ai-agents\",\"video-understanding\",\"research\",\"benchmarks\"]","2026-10-09T04:00:00.000Z","2026-10-10T00:42:12.101Z","2026-10-10T00:42:17.773Z","published",null,[],"ai",[26,27,28,29],"ai-agents","video-understanding","research","benchmarks",[31],{"name":32,"url":33},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.11171",0,{"sections":36},[37,40,44,49,54,58,62,67,72,77,82,87,92,97],{"name":38,"slug":24,"count":39,"latest_published_at":18},"AI",6708,{"name":41,"slug":42,"count":43,"latest_published_at":18},"Security","security",931,{"name":45,"slug":46,"count":47,"latest_published_at":48},"Policy","policy",486,"2026-10-08T22:40:11.000Z",{"name":50,"slug":51,"count":52,"latest_published_at":53},"Deals","deals",474,"2026-10-08T22:00:00.000Z",{"name":55,"slug":56,"count":57,"latest_published_at":18},"Hardware","hardware",231,{"name":59,"slug":60,"count":61,"latest_published_at":18},"Science","science",192,{"name":63,"slug":64,"count":65,"latest_published_at":66},"Consumer Tech","consumer-tech",181,"2026-10-08T23:26:35.000Z",{"name":68,"slug":69,"count":70,"latest_published_at":71},"Startups","startups",117,"2026-10-08T16:45:00.000Z",{"name":73,"slug":74,"count":75,"latest_published_at":76},"Software","software",114,"2026-10-08T17:57:01.000Z",{"name":78,"slug":79,"count":80,"latest_published_at":81},"Dev Tools","dev-tools",105,"2026-10-07T16:59:11.000Z",{"name":83,"slug":84,"count":85,"latest_published_at":86},"General","general",66,"2026-10-09T04:46:11.000Z",{"name":88,"slug":89,"count":90,"latest_published_at":91},"Gaming","gaming",58,"2026-10-08T20:08:45.000Z",{"name":93,"slug":94,"count":95,"latest_published_at":96},"Reviews","reviews",34,"2026-10-08T14:00:22.000Z",{"name":98,"slug":99,"count":100,"latest_published_at":101},"How-To","how-to",8,"2026-10-05T09:00:00.000Z"]