[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-loki-edits-ai-models-without-forgetting-the-past":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},1722,"loki-edits-ai-models-without-forgetting-the-past","LOKI Edits AI Models Without Forgetting the Past","A new technique lets language models absorb fresh knowledge sequentially without storing old data or overwriting what they already know.","A research team has proposed LOKI, a method for updating language models one fact at a time — without erasing what they already know.\n\nThe core problem LOKI tackles is called catastrophic forgetting: when you update a neural network with new information, it tends to corrupt older knowledge. Existing lifelong-editing methods work around this by storing previous training data and running expensive preprocessing steps to figure out which parts of the model to touch. LOKI skips both. It selects which layers to update dynamically — using a statistical measure called the Hilbert-Schmidt Independence Criterion — then constrains each gradient update to the null-space of the model's weight matrix, a mathematical region where changes don't disturb existing representations. The result is that new knowledge slots in without overwriting old knowledge, and no historical data needs to be kept on hand.\n\nThis matters because the alternative — periodically retraining a model from scratch or fine-tuning on a curated mix of old and new data — is expensive and slow. A method that edits models sequentially, without memory overhead, is closer to how you'd actually want to maintain a production AI system: patch the wrong answer, move on. The 14 percent average accuracy improvement the authors report over prior approaches is meaningful headroom if the results hold on third-party benchmarks.\n\nLarge labs have their own internal editing pipelines, and none of them have published evidence that null-space projection at this scale is production-ready — so treat the benchmark numbers as a promising start, not a solved problem.","[\"ai\",\"machine learning\",\"language models\",\"research\"]","2026-06-19T04:00:00.000Z","2026-06-19T10:38:01.636Z","2026-06-19T14:21:37.982Z","published",null,[],"ai",[24,26,27,28],"machine learning","language models","research",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2606.19679",0,{"sections":35},[36,40,44,49,54,59,64,68,72,77,82,87,92,97],{"name":37,"slug":24,"count":38,"latest_published_at":39},"AI",491,"2026-06-19T14:59:11.000Z",{"name":41,"slug":42,"count":43,"latest_published_at":18},"Security","security",132,{"name":45,"slug":46,"count":47,"latest_published_at":48},"Policy","policy",88,"2026-06-16T09:26:09.000Z",{"name":50,"slug":51,"count":52,"latest_published_at":53},"Consumer Tech","consumer-tech",78,"2026-06-16T17:58:24.000Z",{"name":55,"slug":56,"count":57,"latest_published_at":58},"Hardware","hardware",62,"2026-06-18T15:24:16.000Z",{"name":60,"slug":61,"count":62,"latest_published_at":63},"Deals","deals",58,"2026-06-19T14:43:50.000Z",{"name":65,"slug":66,"count":62,"latest_published_at":67},"Software","software","2026-06-16T20:00:00.000Z",{"name":69,"slug":70,"count":71,"latest_published_at":18},"Dev Tools","dev-tools",50,{"name":73,"slug":74,"count":75,"latest_published_at":76},"Science","science",38,"2026-06-18T04:00:00.000Z",{"name":78,"slug":79,"count":80,"latest_published_at":81},"Gaming","gaming",31,"2026-06-16T15:25:13.000Z",{"name":83,"slug":84,"count":85,"latest_published_at":86},"General","general",26,"2026-06-13T18:35:15.000Z",{"name":88,"slug":89,"count":90,"latest_published_at":91},"Startups","startups",23,"2026-06-16T15:00:00.000Z",{"name":93,"slug":94,"count":95,"latest_published_at":96},"Reviews","reviews",19,"2026-06-14T08:00:00.000Z",{"name":98,"slug":99,"count":100,"latest_published_at":101},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]