[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-ai-agent-teaches-itself-to-beat-pokemon-and-zelda-from-video":10,"sections":35},{"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":30,"feedback":34,"feedback_at":22,"cost_usd":34,"total_tokens":34},9230,"ai-agent-teaches-itself-to-beat-pokemon-and-zelda-from-video","AI Agent Teaches Itself to Beat Pokemon and Zelda From Video","A research agent learns to beat hours-long video games by studying unlabeled internet footage instead of costly hand-labeled demos.","A research agent just taught itself to beat two different hours-long video games by watching raw, unlabeled internet footage - no scorekeeping, no expert demonstrations required.\n\nThe system, called ASH, targets a long-standing problem in AI: getting an agent to plan correctly over many hours without someone manually designing rewards or hand-labeling example actions, two methods that do not scale. When ASH gets stuck, it trains a model that infers what actions likely produced a video clip, then uses that model to pull usable lessons from unrelated internet footage of the same game. It also builds a long-term memory by flagging important moments across all that video. Researchers ran ASH for eight-hour sessions on two very different games: the turn-based Pokemon Emerald and the real-time The Legend of Zelda: The Minish Cap.\n\nEvery competing method tested, including behavior-cloning and zero-shot foundation-model baselines, stalled partway through and stopped improving. ASH kept going, hitting an average of 11.2 of 12 milestones in Pokemon Emerald and 9.9 of 12 in Zelda, against 9.3 and 7.8 for the best rival. That gap matters because long-horizon planning - acting correctly over hundreds of steps without a map - is one of the stubborn problems separating today's chatty AI agents from ones that can run a real task unsupervised.\n\nGames are a convenient lab bench, but the underlying trick - pulling supervision out of messy, unlabeled video instead of expensive hand-labeled demos - is exactly the kind of shortcut that tends to migrate to robotics and other real-world automation next.","[\"ai-agents\",\"reinforcement-learning\",\"gaming\",\"research\"]","2026-10-01T04:00:00.000Z","2026-10-02T03:04:14.776Z","2026-10-02T03:04:18.214Z","published",null,[],"ai",[26,27,28,29],"ai-agents","reinforcement-learning","gaming","research",[31],{"name":32,"url":33},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2605.14211",0,{"sections":36},[37,40,44,48,53,58,62,67,72,76,81,85,90,95],{"name":38,"slug":24,"count":39,"latest_published_at":18},"AI",5629,{"name":41,"slug":42,"count":43,"latest_published_at":18},"Security","security",816,{"name":45,"slug":46,"count":47,"latest_published_at":18},"Policy","policy",430,{"name":49,"slug":50,"count":51,"latest_published_at":52},"Deals","deals",298,"2026-09-30T21:00:26.000Z",{"name":54,"slug":55,"count":56,"latest_published_at":57},"Hardware","hardware",196,"2026-09-30T13:00:00.000Z",{"name":59,"slug":60,"count":61,"latest_published_at":18},"Science","science",163,{"name":63,"slug":64,"count":65,"latest_published_at":66},"Consumer Tech","consumer-tech",149,"2026-09-30T22:57:11.000Z",{"name":68,"slug":69,"count":70,"latest_published_at":71},"Dev Tools","dev-tools",93,"2026-10-01T02:30:48.000Z",{"name":73,"slug":74,"count":70,"latest_published_at":75},"Software","software","2026-09-30T21:41:11.000Z",{"name":77,"slug":78,"count":79,"latest_published_at":80},"Startups","startups",84,"2026-09-30T20:39:09.000Z",{"name":82,"slug":28,"count":83,"latest_published_at":84},"Gaming",51,"2026-09-30T16:24:30.000Z",{"name":86,"slug":87,"count":88,"latest_published_at":89},"General","general",50,"2026-09-30T21:37:54.000Z",{"name":91,"slug":92,"count":93,"latest_published_at":94},"Reviews","reviews",31,"2026-09-28T14:31:34.000Z",{"name":96,"slug":97,"count":98,"latest_published_at":99},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]