[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-an-ai-agent-learned-to-curate-its-own-training-data":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},6777,"an-ai-agent-learned-to-curate-its-own-training-data","An AI Agent Learned to Curate Its Own Training Data","AutoData, a research agent, searches selection algorithms overnight and finds a data-curation recipe that beats human-built pipelines and scales up cleanly.","An AI agent just designed its own algorithm for picking training data, and it beat the recipes humans wrote by hand.\n\nThe system, called AutoData, treats data selection as a search problem rather than a fixed formula. Instead of tuning weights across a handful of preset data categories, which is how most existing mixture methods work, it searches a much larger space of executable programs: scoring rules, stratification schemes, and stochastic selection logic. The agent runs each candidate against a small proxy model, reads the validation feedback, and rewrites the algorithm accordingly. Over one overnight run, it converged on a selection recipe that outperformed existing human-designed curation pipelines, and that recipe held up when applied at larger training scales, improving a downstream benchmark called CORE.\n\nThat transfer is the interesting part. Plenty of research tools post better numbers on the exact setup they were tuned on and fall apart anywhere else. Here the algorithm was searched on a cheap proxy model and still worked at bigger scale, which is the property you actually need if this is going to save compute rather than just move the tuning cost around.\n\nThis is the same agentic-engineering trend that has already started rewriting model architectures and training code, now pointed at the one input humans still curate mostly by hand: the data itself. Worth remembering that \"one overnight search\" and \"outperforms hand-built pipelines\" are the kind of claims that read great in a paper and get quietly walked back once someone tries them at frontier scale.","[\"ai-agents\",\"training-data\",\"llm-research\"]","2026-09-18T04:00:00.000Z","2026-09-18T16:13:20.372Z","2026-09-18T16:13:32.309Z","published",null,[],"ai",[26,27,28],"ai-agents","training-data","llm-research",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.19754",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",3958,{"name":40,"slug":41,"count":42,"latest_published_at":18},"Security","security",652,{"name":44,"slug":45,"count":46,"latest_published_at":47},"Policy","policy",338,"2026-09-11T04:00:00.000Z",{"name":49,"slug":50,"count":51,"latest_published_at":52},"Deals","deals",179,"2026-06-29T20:02:07.000Z",{"name":54,"slug":55,"count":56,"latest_published_at":18},"Hardware","hardware",155,{"name":58,"slug":59,"count":60,"latest_published_at":18},"Science","science",116,{"name":62,"slug":63,"count":64,"latest_published_at":65},"Consumer Tech","consumer-tech",99,"2026-09-09T17:27:33.000Z",{"name":67,"slug":68,"count":69,"latest_published_at":70},"Dev Tools","dev-tools",76,"2026-09-18T01:04:54.000Z",{"name":72,"slug":73,"count":74,"latest_published_at":75},"Software","software",75,"2026-09-10T20:41:21.000Z",{"name":77,"slug":78,"count":79,"latest_published_at":80},"Startups","startups",55,"2026-09-09T23:14:29.000Z",{"name":82,"slug":83,"count":84,"latest_published_at":85},"Gaming","gaming",43,"2026-09-10T12:18:06.000Z",{"name":87,"slug":88,"count":89,"latest_published_at":90},"General","general",41,"2026-09-08T01:57:23.000Z",{"name":92,"slug":93,"count":94,"latest_published_at":95},"Reviews","reviews",20,"2026-06-24T12:00:01.000Z",{"name":97,"slug":98,"count":99,"latest_published_at":100},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]