[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-study-finds-supervised-fine-tuning-makes-ai-models-forget-more":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},9986,"study-finds-supervised-fine-tuning-makes-ai-models-forget-more","Study Finds Supervised Fine-Tuning Makes AI Models Forget More","Researchers show supervised fine-tuning drifts toward style bias and forgets faster than reinforcement fine-tuning, even with correct training data.","A new study says fine-tuned AI models don't forget skills because of bad data. They forget because of writing style.\n\nResearchers compared two ways of fine-tuning a model on classification tasks: supervised fine-tuning (SFT), where a model copies a teacher's example answers, and reinforcement fine-tuning (RFT), where a model is rewarded for correct answers regardless of phrasing. Using a simplified linear-softmax model they could analyze mathematically, they split each training update into a semantic component (getting the right answer) and a style component (how that answer is phrased). Both methods update semantics identically, but SFT also pulls the model toward a teacher's particular phrasing habits, even when every teacher example is factually correct. Over repeated training steps, that stylistic pull compounds into measurable forgetting, with a mathematically guaranteed minimum level of semantic error for SFT, while RFT holds at zero.\n\nThat explains something AI teams have run into before: fine-tuning on clean, correct data can still make a model worse at the task it was trained for. It also makes a sharper case for reward-based fine-tuning, the family of methods behind RLHF-style post-training, when the goal is preserving existing knowledge rather than matching a particular tone or format.\n\nThe catch is that this proof runs on a toy linear model, not a GPT-5-scale transformer, so treat it as a clean hypothesis about why fine-tuning degrades models, not a settled verdict on your own pipeline.","[\"fine-tuning\",\"reinforcement-learning\",\"ai-research\",\"machine-learning\"]","2026-10-05T04:00:00.000Z","2026-10-05T16:45:54.568Z","2026-10-05T16:45:58.752Z","published",null,[],"ai",[26,27,28,29],"fine-tuning","reinforcement-learning","ai-research","machine-learning",[31],{"name":32,"url":33},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.02437",0,{"sections":36},[37,40,44,49,54,59,63,68,72,76,81,86,91,96],{"name":38,"slug":24,"count":39,"latest_published_at":18},"AI",6233,{"name":41,"slug":42,"count":43,"latest_published_at":18},"Security","security",868,{"name":45,"slug":46,"count":47,"latest_published_at":48},"Policy","policy",444,"2026-10-03T15:02:01.000Z",{"name":50,"slug":51,"count":52,"latest_published_at":53},"Deals","deals",323,"2026-10-04T13:00:00.000Z",{"name":55,"slug":56,"count":57,"latest_published_at":58},"Hardware","hardware",204,"2026-10-03T14:50:50.000Z",{"name":60,"slug":61,"count":62,"latest_published_at":18},"Science","science",177,{"name":64,"slug":65,"count":66,"latest_published_at":67},"Consumer Tech","consumer-tech",158,"2026-10-03T03:21:12.000Z",{"name":69,"slug":70,"count":71,"latest_published_at":18},"Dev Tools","dev-tools",97,{"name":73,"slug":74,"count":71,"latest_published_at":75},"Software","software","2026-10-04T10:00:00.000Z",{"name":77,"slug":78,"count":79,"latest_published_at":80},"Startups","startups",92,"2026-10-04T14:36:25.000Z",{"name":82,"slug":83,"count":84,"latest_published_at":85},"Gaming","gaming",53,"2026-10-02T02:50:39.000Z",{"name":87,"slug":88,"count":89,"latest_published_at":90},"General","general",51,"2026-10-05T02:35:01.000Z",{"name":92,"slug":93,"count":94,"latest_published_at":95},"Reviews","reviews",32,"2026-10-02T18:00:00.000Z",{"name":97,"slug":98,"count":99,"latest_published_at":100},"How-To","how-to",7,"2026-10-01T09:00:00.000Z"]