[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-researchers-build-benchmark-for-ai-that-compares-shopping-videos":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},10241,"researchers-build-benchmark-for-ai-that-compares-shopping-videos","Researchers Build Benchmark for AI That Compares Shopping Videos","AdSeek pairs a new benchmark with an agentic AI that hunts for evidence across multiple product videos, not just one.","A new benchmark shows AI video models are good at watching one ad and bad at comparing several.\n\nAI models that can watch one product video and answer questions about it are everywhere now. Comparing several videos at once, the way a shopper actually shops, is a different problem - and until now, nobody had a good way to test it. Researchers built AdsCVR, a benchmark of 2,483 e-commerce videos and 6,110 question-answer pairs that require pulling evidence from multiple videos across visual detail, speech, and on-screen text. They then built AdSeek, an AI agent that actively picks which video and audio tools to use during a multi-turn search, instead of sampling frames at a fixed rate. To train it, they combined reinforcement learning with a corrective step that fixes bad reasoning paths before a further round of reinforcement learning. The result: 74.3 percent accuracy on the new benchmark, a 27.9 percentage point jump over the Qwen3-VL-8B model it started from.\n\nThat gap matters because cross-video comparison is exactly what shoppers and merchants do constantly - checking several sellers' demo videos for the same product, or merchants auditing competitors' marketing claims. A model that only handles one video at a time is answering a question nobody who actually shops that way is asking.\n\nIt is still a research benchmark, not a shopping assistant you can use today, and a 28-point jump over one specific backbone model does not say how AdSeek would hold up against other major multimodal systems on the same task.","[\"ai\",\"benchmarks\",\"e-commerce\",\"multimodal-ai\"]","2026-10-05T04:00:00.000Z","2026-10-06T10:35:58.017Z","2026-10-06T10:36:04.187Z","published",null,[],"ai",[24,26,27,28],"benchmarks","e-commerce","multimodal-ai",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.03099",0,{"sections":35},[36,40,45,50,55,60,65,70,75,80,85,90,95,100],{"name":37,"slug":24,"count":38,"latest_published_at":39},"AI",6347,"2026-10-06T07:49:07.000Z",{"name":41,"slug":42,"count":43,"latest_published_at":44},"Security","security",879,"2026-10-06T00:00:38.000Z",{"name":46,"slug":47,"count":48,"latest_published_at":49},"Policy","policy",463,"2026-10-05T23:16:24.000Z",{"name":51,"slug":52,"count":53,"latest_published_at":54},"Deals","deals",374,"2026-10-06T09:00:00.000Z",{"name":56,"slug":57,"count":58,"latest_published_at":59},"Hardware","hardware",211,"2026-10-05T21:14:08.000Z",{"name":61,"slug":62,"count":63,"latest_published_at":64},"Science","science",182,"2026-10-05T18:58:12.000Z",{"name":66,"slug":67,"count":68,"latest_published_at":69},"Consumer Tech","consumer-tech",162,"2026-10-06T09:00:14.000Z",{"name":71,"slug":72,"count":73,"latest_published_at":74},"Software","software",103,"2026-10-05T18:23:19.000Z",{"name":76,"slug":77,"count":78,"latest_published_at":79},"Dev Tools","dev-tools",99,"2026-10-05T10:47:06.000Z",{"name":81,"slug":82,"count":83,"latest_published_at":84},"Startups","startups",96,"2026-10-05T20:24:09.000Z",{"name":86,"slug":87,"count":88,"latest_published_at":89},"Gaming","gaming",54,"2026-10-05T19:07:04.000Z",{"name":91,"slug":92,"count":93,"latest_published_at":94},"General","general",53,"2026-10-05T22:05:42.000Z",{"name":96,"slug":97,"count":98,"latest_published_at":99},"Reviews","reviews",33,"2026-10-05T11:57:17.000Z",{"name":101,"slug":102,"count":103,"latest_published_at":104},"How-To","how-to",8,"2026-10-05T09:00:00.000Z"]