[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-tiny-proxy-models-can-plan-training-order-for-bigger-ai":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},5143,"tiny-proxy-models-can-plan-training-order-for-bigger-ai","Tiny Proxy Models Can Plan Training Order for Bigger AI","A new controller called LogFloor uses small model training paths to sequence which skills big models tackle first, cutting compute by an average of 56%.","A small model's training run can tell a much bigger model exactly which skills to fix first, and in what order, without the big model ever repeating that search itself.\n\nResearchers built LogFloor, a closed-loop controller that tracks a model's weakest skills as it trains and steers each round toward whatever is currently the biggest bottleneck. Tested on five bAbI reasoning skill slices with Qwen2.5-1.5B, this approach cut the tokens needed to hit target performance floors by 56.2% on average. The bigger claim is transferability: a 70-million-parameter \"scout\" model's training path, replayed three times, got eight separate 12-billion-parameter models across every skill floor, saving 30.9% of tokens by pair mean and up to 39.4% in pooled training cost. On a separate MMLU-based test, the same frozen scout path worked across all eight of those 12B runs.\n\nData mixtures borrowed from small proxy models are already common practice in large-model training. This work suggests the order in which a model resolves its weaknesses is also a reusable, transferable asset, not just what data it sees. The team's own ablations back that up: flattening the scout's path into a single static mixture, or running the phases in reverse order, wiped out most of the gains, while just knowing which skills were bottlenecks was only partly helpful on its own.\n\nThe results are still confined to bAbI-style skill tests and one model family, so it is a promising lab result, not yet a production training recipe.","[\"ai training\",\"llm efficiency\",\"curriculum learning\",\"machine learning research\"]","2026-08-18T04:00:00.000Z","2026-08-18T06:51:41.082Z","2026-08-18T06:51:52.919Z","published",null,[24],{"id":25,"reviewer":26,"round":27,"reason":28,"status":29},"editor-r1","editor",1,"The dek claims 'cutting compute by up to 56%', implying a max\u002Frange, but the body only reports a single 56.2% average figure across five skill slices with no range given — reword the dek to say 'an average of 56%' or similar so it matches the body's actual figure.","resolved","ai",[32,33,34,35],"ai training","llm efficiency","curriculum learning","machine learning research",[37],{"name":38,"url":39},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2608.14936",0,{"sections":42},[43,47,51,56,61,66,71,76,81,85,90,95,100,105],{"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":18},"Dev Tools","dev-tools",69,{"name":86,"slug":87,"count":88,"latest_published_at":89},"Startups","startups",47,"2026-08-19T19:13:46.000Z",{"name":91,"slug":92,"count":93,"latest_published_at":94},"Gaming","gaming",41,"2026-07-09T04:00:00.000Z",{"name":96,"slug":97,"count":98,"latest_published_at":99},"General","general",33,"2026-08-18T22:18:13.000Z",{"name":101,"slug":102,"count":103,"latest_published_at":104},"Reviews","reviews",20,"2026-06-24T12:00:01.000Z",{"name":106,"slug":107,"count":108,"latest_published_at":109},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]