[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-researchers-find-a-smarter-way-to-personalize-ai-chatbots":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},9447,"researchers-find-a-smarter-way-to-personalize-ai-chatbots","Researchers Find a Smarter Way to Personalize AI Chatbots","A new method called GAP-DPO picks AI training examples based on gradient alignment with user goals, outperforming standard preference tuning methods.","A new tuning method picks AI training examples based on math, not guesswork, to better match a chatbot's output to what an individual user actually wants.\n\nResearchers published a paper describing GAP-DPO (Geometry-Aligned Preference DPO), a technique for tuning large language models to match individual user preferences. It builds on Direct Preference Optimization, a standard method that teaches a model to favor one response over another using paired examples. The problem: standard DPO usually picks those pairs with simple criteria, like how confident the model already is, rather than what reflects what a specific user wants. GAP-DPO instead selects pairs by checking whether they point in the same mathematical direction as a user's actual preferences, and it regenerates its training data every epoch so that alignment does not drift.\n\nPersonalization is the part of AI product development that gets talked about constantly and solved properly rarely. Most \"personalized\" chatbots lean on prompt tricks or shallow fine-tuning that degrades quality the more it tries to please everyone. GAP-DPO's bet is that the framework was never broken, the problem was always how training pairs get chosen, a step usually treated as throwaway preprocessing.\n\nThis is one arXiv paper tested on text-generation benchmarks, not a shipped product. The real test is whether treating pair selection as a geometry problem survives contact with messy, real-world user data.","[\"ai\",\"llms\",\"personalization\",\"preference optimization\"]","2026-10-02T04:00:00.000Z","2026-10-02T19:24:25.652Z","2026-10-02T19:24:30.203Z","published",null,[],"ai",[24,26,27,28],"llms","personalization","preference optimization",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.00061",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",5765,{"name":40,"slug":41,"count":42,"latest_published_at":18},"Security","security",831,{"name":44,"slug":45,"count":46,"latest_published_at":47},"Policy","policy",437,"2026-10-01T18:10:00.000Z",{"name":49,"slug":50,"count":51,"latest_published_at":52},"Deals","deals",317,"2026-10-01T22:00:00.000Z",{"name":54,"slug":55,"count":56,"latest_published_at":57},"Hardware","hardware",198,"2026-10-01T17:38:48.000Z",{"name":59,"slug":60,"count":61,"latest_published_at":62},"Science","science",168,"2026-10-01T18:35:55.000Z",{"name":64,"slug":65,"count":66,"latest_published_at":67},"Consumer Tech","consumer-tech",155,"2026-10-01T19:54:10.000Z",{"name":69,"slug":70,"count":71,"latest_published_at":72},"Dev Tools","dev-tools",96,"2026-10-01T16:57:03.000Z",{"name":74,"slug":75,"count":76,"latest_published_at":77},"Software","software",93,"2026-09-30T21:41:11.000Z",{"name":79,"slug":80,"count":81,"latest_published_at":82},"Startups","startups",90,"2026-10-01T21:55:22.000Z",{"name":84,"slug":85,"count":86,"latest_published_at":87},"Gaming","gaming",53,"2026-10-02T02:50:39.000Z",{"name":89,"slug":90,"count":91,"latest_published_at":92},"General","general",50,"2026-09-30T21:37:54.000Z",{"name":94,"slug":95,"count":96,"latest_published_at":97},"Reviews","reviews",31,"2026-09-28T14:31:34.000Z",{"name":99,"slug":100,"count":101,"latest_published_at":102},"How-To","how-to",7,"2026-10-01T09:00:00.000Z"]