[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-when-giving-ai-the-answers-makes-it-worse":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},3719,"when-giving-ai-the-answers-makes-it-worse","When Giving AI the Answers Makes It Worse","New research finds that feeding thinking models the correct answer during training cuts accuracy by up to 17% on hard math benchmarks.","Giving a reasoning model the answer key during training backfires badly, new research shows.\n\nA team studying self-distillation — the practice of using a model as its own teacher — found that handing the model privileged context (like a known solution) during training degrades performance on hard reasoning tasks. Across five Qwen3 and OLMo thinking models tested on AIME24, AIME25, and HMMT25, this approach caused a relative accuracy drop of up to 17% in avg@16 scores. The degradation gets worse the longer the reasoning chain, which is exactly where thinking models normally shine.\n\nThe why matters more than the what here. The researchers traced the failure to \"high-entropy forking positions\" — moments in a reasoning trace where the model genuinely faces multiple plausible paths. Privileged context suppresses that branching, training the model to skip the self-correction, backtracking, and hedging steps that make long reasoning work. The model learns to sound confident rather than to reason carefully.\n\nThe irony is clean: the same privileged context that helps a standard instruction-tuned model hurts a stronger thinking model. It is a useful reminder that training recipes do not transfer neatly up the capability ladder — what works as a scaffold can become a ceiling.","[\"ai\",\"machine-learning\",\"research\",\"reasoning\"]","2026-07-07T04:00:00.000Z","2026-07-07T06:38:12.109Z","2026-07-07T06:38:15.009Z","published",null,[],"ai",[24,26,27,28],"machine-learning","research","reasoning",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.05184",0,{"sections":35},[36,40,45,50,55,60,65,70,75,79,84,88,93,98],{"name":37,"slug":24,"count":38,"latest_published_at":39},"AI",2590,"2026-07-16T04:00:00.000Z",{"name":41,"slug":42,"count":43,"latest_published_at":44},"Security","security",294,"2026-07-15T19:59:48.000Z",{"name":46,"slug":47,"count":48,"latest_published_at":49},"Deals","deals",179,"2026-06-29T20:02:07.000Z",{"name":51,"slug":52,"count":53,"latest_published_at":54},"Policy","policy",158,"2026-07-16T00:02:48.000Z",{"name":56,"slug":57,"count":58,"latest_published_at":59},"Hardware","hardware",122,"2026-07-14T19:46:26.000Z",{"name":61,"slug":62,"count":63,"latest_published_at":64},"Consumer Tech","consumer-tech",93,"2026-07-13T13:20:48.000Z",{"name":66,"slug":67,"count":68,"latest_published_at":69},"Software","software",70,"2026-07-13T19:52:25.000Z",{"name":71,"slug":72,"count":73,"latest_published_at":74},"Science","science",66,"2026-07-10T10:29:37.000Z",{"name":76,"slug":77,"count":78,"latest_published_at":18},"Dev Tools","dev-tools",59,{"name":80,"slug":81,"count":82,"latest_published_at":83},"Gaming","gaming",41,"2026-07-09T04:00:00.000Z",{"name":85,"slug":86,"count":82,"latest_published_at":87},"Startups","startups","2026-06-29T20:55:50.000Z",{"name":89,"slug":90,"count":91,"latest_published_at":92},"General","general",29,"2026-07-10T22:28:58.000Z",{"name":94,"slug":95,"count":96,"latest_published_at":97},"Reviews","reviews",20,"2026-06-24T12:00:01.000Z",{"name":99,"slug":100,"count":101,"latest_published_at":102},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]