[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-new-method-lets-ai-models-learn-from-ranked-answers-not-just-pairs":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},11353,"new-method-lets-ai-models-learn-from-ranked-answers-not-just-pairs","New Method Lets AI Models Learn From Ranked Answers, Not Just Pairs","A new training method called GraphDPO lets language models learn from entire rankings of responses instead of just pairwise comparisons, researchers say.","A new paper argues that the standard way of training AI models on human feedback throws away useful information, and proposes a fix.\n\nResearchers propose GraphDPO, a generalization of Direct Preference Optimization (DPO) that works with full rankings of responses instead of only pairs. Standard DPO compares two model outputs at a time, discarding structure whenever training data already contains multiple ranked rollouts per prompt. GraphDPO instead builds a graph that encodes which response beats which as edges, then optimizes across that whole structure at once using a ranking-based objective. It also groups tied or incomparable responses into shared layers, so the model is not pushed to learn relationships that the data never actually specified.\n\nMost alignment datasets already contain more than two ranked responses per prompt, so collapsing them into pairs throws away real signal and can feed the model conflicting or redundant comparisons that destabilize training. GraphDPO is built to recover ordinary DPO as a special case, which makes it a strict superset rather than a competing technique, and the authors report stronger results on reasoning and program-synthesis benchmarks.\n\nThat said, those results come from the paper's own tests, not an independent benchmark - a win on your own evaluation suite is not the same as a reason for the rest of the field to switch.","[\"ai alignment\",\"language models\",\"preference optimization\",\"research\"]","2026-10-12T04:00:00.000Z","2026-10-12T09:54:03.890Z","2026-10-12T09:54:08.212Z","published",null,[],"ai",[26,27,28,29],"ai alignment","language models","preference optimization","research",[31],{"name":32,"url":33},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2605.08037",0,{"sections":36},[37,41,46,51,56,61,66,71,76,81,86,91,96,101],{"name":38,"slug":24,"count":39,"latest_published_at":40},"AI",6987,"2026-10-13T17:00:00.000Z",{"name":42,"slug":43,"count":44,"latest_published_at":45},"Security","security",969,"2026-10-13T17:18:25.000Z",{"name":47,"slug":48,"count":49,"latest_published_at":50},"Policy","policy",501,"2026-10-13T12:15:53.000Z",{"name":52,"slug":53,"count":54,"latest_published_at":55},"Deals","deals",499,"2026-10-13T16:17:17.000Z",{"name":57,"slug":58,"count":59,"latest_published_at":60},"Hardware","hardware",249,"2026-10-13T17:03:44.000Z",{"name":62,"slug":63,"count":64,"latest_published_at":65},"Consumer Tech","consumer-tech",210,"2026-10-13T17:26:43.000Z",{"name":67,"slug":68,"count":69,"latest_published_at":70},"Science","science",199,"2026-10-13T12:01:37.000Z",{"name":72,"slug":73,"count":74,"latest_published_at":75},"Startups","startups",126,"2026-10-13T16:34:25.000Z",{"name":77,"slug":78,"count":79,"latest_published_at":80},"Software","software",119,"2026-10-13T17:01:31.000Z",{"name":82,"slug":83,"count":84,"latest_published_at":85},"Dev Tools","dev-tools",116,"2026-10-13T16:20:18.000Z",{"name":87,"slug":88,"count":89,"latest_published_at":90},"General","general",74,"2026-10-13T18:10:08.000Z",{"name":92,"slug":93,"count":94,"latest_published_at":95},"Gaming","gaming",64,"2026-10-13T15:59:07.000Z",{"name":97,"slug":98,"count":99,"latest_published_at":100},"Reviews","reviews",35,"2026-10-12T14:00:00.000Z",{"name":102,"slug":103,"count":104,"latest_published_at":105},"How-To","how-to",8,"2026-10-05T09:00:00.000Z"]