[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-new-training-method-teaches-gui-agents-from-their-own-actions":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},7540,"new-training-method-teaches-gui-agents-from-their-own-actions","New Training Method Teaches GUI Agents From Their Own Actions","A two-stage self-teaching method lets AI agents learn multi-step app navigation from their own trial runs, beating prior methods on two benchmarks.","A new training technique lets AI agents that operate apps learn from their own actions instead of relying purely on canned demonstrations.\n\nThe method, called GUI-SD-v2, extends an approach known as on-policy self-distillation from single-step screen recognition to full multi-turn app interactions, where an agent has to plan, act, and remember context across many steps. It works in two stages. First, it trains the agent to follow privileged hints, essentially extra context about the screen state, by comparing rollouts done with and without those hints. Second, it uses a self-teacher model to selectively pass along reasoning and memory cues, helping the agent decide what to do next and what to carry forward to later steps. In tests on two mobile-agent benchmarks, AndroidWorld and MobileWorld, the method beat both prior self-distillation baselines and other state-of-the-art approaches on task success rates.\n\nThe interesting part is not the leaderboard win. Earlier self-distillation methods worked well for grounding, basically teaching a model to point at the right button, but multi-turn tasks like booking a flight and adding it to a calendar need memory and judgment across many screens, which is where most agents fall apart. This is a direct attempt to close that specific gap rather than another marginal grounding tweak.\n\nCode and training data are promised for public release, which will let outsiders check whether the benchmark gains survive contact with apps messier than a lab setup.","[\"ai\",\"ai-agents\",\"automation\",\"research\"]","2026-09-24T04:00:00.000Z","2026-09-24T05:16:41.986Z","2026-09-24T05:16:47.566Z","published",null,[],"ai",[24,26,27,28],"ai-agents","automation","research",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.27307",0,{"sections":35},[36,39,43,48,53,58,63,68,73,78,83,88,93,98],{"name":37,"slug":24,"count":38,"latest_published_at":18},"AI",4387,{"name":40,"slug":41,"count":42,"latest_published_at":18},"Security","security",720,{"name":44,"slug":45,"count":46,"latest_published_at":47},"Policy","policy",380,"2026-09-23T22:53:43.000Z",{"name":49,"slug":50,"count":51,"latest_published_at":52},"Deals","deals",220,"2026-09-23T22:00:00.000Z",{"name":54,"slug":55,"count":56,"latest_published_at":57},"Hardware","hardware",173,"2026-09-23T23:42:10.000Z",{"name":59,"slug":60,"count":61,"latest_published_at":62},"Science","science",135,"2026-09-23T22:45:49.000Z",{"name":64,"slug":65,"count":66,"latest_published_at":67},"Consumer Tech","consumer-tech",116,"2026-09-24T00:51:49.000Z",{"name":69,"slug":70,"count":71,"latest_published_at":72},"Software","software",85,"2026-09-23T20:00:00.000Z",{"name":74,"slug":75,"count":76,"latest_published_at":77},"Dev Tools","dev-tools",79,"2026-09-22T22:21:13.000Z",{"name":79,"slug":80,"count":81,"latest_published_at":82},"Startups","startups",66,"2026-09-23T17:28:38.000Z",{"name":84,"slug":85,"count":86,"latest_published_at":87},"Gaming","gaming",45,"2026-09-22T15:35:06.000Z",{"name":89,"slug":90,"count":91,"latest_published_at":92},"General","general",43,"2026-09-21T23:48:56.000Z",{"name":94,"slug":95,"count":96,"latest_published_at":97},"Reviews","reviews",27,"2026-09-22T13:00:00.000Z",{"name":99,"slug":100,"count":101,"latest_published_at":102},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]