[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-ai-agents-get-pickier-about-which-experiments-to-run":10,"sections":41},{"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":30,"tags":31,"sources":36,"feedback":40,"feedback_at":22,"cost_usd":40,"total_tokens":40},5020,"ai-agents-get-pickier-about-which-experiments-to-run","AI Agents Get Pickier About Which Experiments to Run","New preference models help AI research agents skip weak experiment ideas, cutting GPU usage by more than a third while also producing better results.","AI agents that design their own machine learning experiments have a spending problem: writing a candidate experiment takes minutes, but running it to see if it works can eat hours or days of GPU time. A new paper proposes giving these agents better judgment about which candidates are worth that cost.\n\nResearchers built what they call AI Research Preference Models, or RPMs, which predict how promising a candidate experiment is before anyone spends compute running it. Two versions were tested: one that just reasons over the proposed code and past results, and a more expensive one that also runs small pilot tests first. Both were plugged into an existing research agent called AIRA-dojo and evaluated on AIRS-Bench, a benchmark of ML research tasks. The reasoning-only version pushed the benchmark's normalized score from 0.684 to 0.711; the version that runs pilots first did better, hitting 0.729 and setting new state-of-the-art marks on two of the benchmark's tasks.\n\nThe more interesting number is efficiency, not just accuracy. With an RPM guiding it, the agent matched the unguided version's 24-hour results in about 15 hours, using less than two-thirds of the original compute budget - a cut of over a third in GPU time for the same outcome. For labs running these agents at scale, that is the difference between a research budget that stretches to more ideas and one that doesn't.\n\nIt's a small, unglamorous fix, but the field has plenty of those: as agents get better at writing candidate solutions, the bottleneck quietly shifts to deciding which ones deserve a GPU. Prioritization, not generation, may end up being the harder problem.","[\"ai-agents\",\"machine-learning\",\"research\",\"gpu-efficiency\"]","2026-08-17T04:00:00.000Z","2026-08-17T05:51:48.775Z","2026-08-17T05:52:00.605Z","published",null,[24],{"id":25,"reviewer":26,"round":27,"reason":28,"status":29},"publisher-r1","publisher",1,"The dek claims GPU time was cut by two-thirds, but the body states the agent used 'less than two-thirds' of the compute budget (matching 24-hour performance in ~15 hours, i.e. ~62% of the budget, a ~38% cut, not two-thirds), which is an internal inconsistency in the reported savings figure.","resolved","ai",[32,33,34,35],"ai-agents","machine-learning","research","gpu-efficiency",[37],{"name":38,"url":39},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2608.13940",0,{"sections":42},[43,47,51,56,61,66,71,76,81,86,91,96,101,106],{"name":44,"slug":30,"count":45,"latest_published_at":46},"AI",3293,"2026-08-20T04:00:00.000Z",{"name":48,"slug":49,"count":50,"latest_published_at":46},"Security","security",435,{"name":52,"slug":53,"count":54,"latest_published_at":55},"Policy","policy",210,"2026-08-19T09:32:27.000Z",{"name":57,"slug":58,"count":59,"latest_published_at":60},"Deals","deals",179,"2026-06-29T20:02:07.000Z",{"name":62,"slug":63,"count":64,"latest_published_at":65},"Hardware","hardware",140,"2026-08-19T18:25:42.000Z",{"name":67,"slug":68,"count":69,"latest_published_at":70},"Consumer Tech","consumer-tech",95,"2026-08-18T16:05:00.000Z",{"name":72,"slug":73,"count":74,"latest_published_at":75},"Science","science",90,"2026-08-19T18:41:02.000Z",{"name":77,"slug":78,"count":79,"latest_published_at":80},"Software","software",73,"2026-08-18T07:51:50.000Z",{"name":82,"slug":83,"count":84,"latest_published_at":85},"Dev Tools","dev-tools",69,"2026-08-18T04:00:00.000Z",{"name":87,"slug":88,"count":89,"latest_published_at":90},"Startups","startups",47,"2026-08-19T19:13:46.000Z",{"name":92,"slug":93,"count":94,"latest_published_at":95},"Gaming","gaming",41,"2026-07-09T04:00:00.000Z",{"name":97,"slug":98,"count":99,"latest_published_at":100},"General","general",33,"2026-08-18T22:18:13.000Z",{"name":102,"slug":103,"count":104,"latest_published_at":105},"Reviews","reviews",20,"2026-06-24T12:00:01.000Z",{"name":107,"slug":108,"count":109,"latest_published_at":110},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]