[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-researchers-pinpoint-why-ai-models-cling-to-old-answers":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},8903,"researchers-pinpoint-why-ai-models-cling-to-old-answers","Researchers Pinpoint Why AI Models Cling to Old Answers","New research shows LLMs sometimes answer with outdated facts because attention skews toward old values, and a no-training fix corrects most of these errors.","Large language models can forget that you corrected them, even when the right answer is sitting right there in the conversation.\n\nResearchers built a benchmark called Controlled In-Context Memory (CICM) to test whether models track updated facts, preferences, or goals during a conversation instead of defaulting to earlier ones. They found that even frontier reasoning models sometimes answer with a stale value instead of the current one, a failure the team calls \"stale binding.\" Using probes on open-source models, they showed the correct answer was often still recoverable internally even when the model's spoken output was wrong, meaning the information wasn't lost, just not selected. Component-level tests on Qwen and Pythia traced the problem to \"attention drift,\" where several outdated values that look similar to the model can collectively out-compete the single current value for attention.\n\nThat reframes a familiar AI annoyance, an agent citing your old preference or an outdated fact, as a selection problem rather than a memory problem, which changes how it gets fixed. Using a one-layer transformer, the researchers mathematically modeled how old values gang up on attention, then built an intervention that redirects attention toward the current value without any retraining, correcting most old-value errors while preserving nearly all answers that were already right.\n\nIt's a tidy fix in a controlled benchmark, but real conversations rarely hand a model one clean, well-defined \"current value\" to redirect attention toward.","[\"ai\",\"llm-research\",\"interpretability\",\"arxiv\"]","2026-10-01T04:00:00.000Z","2026-10-01T10:06:21.884Z","2026-10-01T10:06:27.454Z","published",null,[],"ai",[24,26,27,28],"llm-research","interpretability","arxiv",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.38866",0,{"sections":35},[36,39,44,49,54,59,64,69,74,78,83,88,93,98],{"name":37,"slug":24,"count":38,"latest_published_at":18},"AI",5351,{"name":40,"slug":41,"count":42,"latest_published_at":43},"Security","security",801,"2026-09-30T22:18:23.000Z",{"name":45,"slug":46,"count":47,"latest_published_at":48},"Policy","policy",429,"2026-10-01T02:26:17.000Z",{"name":50,"slug":51,"count":52,"latest_published_at":53},"Deals","deals",298,"2026-09-30T21:00:26.000Z",{"name":55,"slug":56,"count":57,"latest_published_at":58},"Hardware","hardware",196,"2026-09-30T13:00:00.000Z",{"name":60,"slug":61,"count":62,"latest_published_at":63},"Science","science",157,"2026-09-30T15:00:56.000Z",{"name":65,"slug":66,"count":67,"latest_published_at":68},"Consumer Tech","consumer-tech",149,"2026-09-30T22:57:11.000Z",{"name":70,"slug":71,"count":72,"latest_published_at":73},"Dev Tools","dev-tools",93,"2026-10-01T02:30:48.000Z",{"name":75,"slug":76,"count":72,"latest_published_at":77},"Software","software","2026-09-30T21:41:11.000Z",{"name":79,"slug":80,"count":81,"latest_published_at":82},"Startups","startups",84,"2026-09-30T20:39:09.000Z",{"name":84,"slug":85,"count":86,"latest_published_at":87},"Gaming","gaming",51,"2026-09-30T16:24:30.000Z",{"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",6,"2026-06-16T09:00:00.000Z"]