[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-simulated-users-beat-gpt-55-at-training-ai-agents":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},11098,"simulated-users-beat-gpt-55-at-training-ai-agents","Simulated Users Beat GPT-5.5 at Training AI Agents","A new 9B-parameter simulator mimics real user quirks better than Claude Opus 5, and agents trained on it generalize better than ones trained on GPT-5.5.","A new AI project trains agents against a simulated human instead of a cooperative AI assistant pretending to be one.\n\nResearchers built MIMESIS, a 9B-parameter model trained on real human conversations plus labeled behavioral patterns, designed to act like an actual user instead of an overly cooperative chatbot. It scored a SOUL-Index of 65.7, beating the strongest frontier model tested. Against Claude Opus 5, the previous best simulator, it scored 13.4 points higher on behavioral fidelity and cut the \"Turing distance\" gap by 3.6 points. The team then froze MIMESIS and used it as a training partner for a separate agent, teaching it through repeated back-and-forth reinforcement learning.\n\nMost agent training today just points one assistant LLM at another, so the \"user\" ends up unrealistically polite, explicit, and samey, and agents learn to handle a type of human that barely exists. Agents trained against MIMESIS beat ones trained against GPT-5.5 across eight environments and nine simulators the system had never seen, a generalization result most training setups don't get to claim. The researchers also added a coaching layer that turns the simulator's private reasoning into written feedback for the agent, squeezing more signal out of each conversation than a simple pass-fail reward would.\n\nThe benchmarks are still simulator-graded, so the real test is whether these agents hold up against actual, more unpredictable humans.","[\"ai agents\",\"llm training\",\"reinforcement learning\",\"ai research\"]","2026-10-09T04:00:00.000Z","2026-10-10T05:58:11.951Z","2026-10-10T05:58:17.306Z","published",null,[],"ai",[26,27,28,29],"ai agents","llm training","reinforcement learning","ai research",[31],{"name":32,"url":33},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.09484",0,{"sections":36},[37,40,44,49,54,58,62,67,72,77,81,86,91,96],{"name":38,"slug":24,"count":39,"latest_published_at":18},"AI",6804,{"name":41,"slug":42,"count":43,"latest_published_at":18},"Security","security",934,{"name":45,"slug":46,"count":47,"latest_published_at":48},"Policy","policy",486,"2026-10-08T22:40:11.000Z",{"name":50,"slug":51,"count":52,"latest_published_at":53},"Deals","deals",474,"2026-10-08T22:00:00.000Z",{"name":55,"slug":56,"count":57,"latest_published_at":18},"Hardware","hardware",232,{"name":59,"slug":60,"count":61,"latest_published_at":18},"Science","science",194,{"name":63,"slug":64,"count":65,"latest_published_at":66},"Consumer Tech","consumer-tech",181,"2026-10-08T23:26:35.000Z",{"name":68,"slug":69,"count":70,"latest_published_at":71},"Startups","startups",117,"2026-10-08T16:45:00.000Z",{"name":73,"slug":74,"count":75,"latest_published_at":76},"Software","software",114,"2026-10-08T17:57:01.000Z",{"name":78,"slug":79,"count":80,"latest_published_at":18},"Dev Tools","dev-tools",106,{"name":82,"slug":83,"count":84,"latest_published_at":85},"General","general",66,"2026-10-09T04:46:11.000Z",{"name":87,"slug":88,"count":89,"latest_published_at":90},"Gaming","gaming",58,"2026-10-08T20:08:45.000Z",{"name":92,"slug":93,"count":94,"latest_published_at":95},"Reviews","reviews",34,"2026-10-08T14:00:22.000Z",{"name":97,"slug":98,"count":99,"latest_published_at":100},"How-To","how-to",8,"2026-10-05T09:00:00.000Z"]