[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-a-fix-for-the-way-ai-models-choke-on-numbers":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},8932,"a-fix-for-the-way-ai-models-choke-on-numbers","A Fix for the Way AI Models Choke on Numbers","Researchers built a gating system that lets language models treat numbers as both labels and math, lifting a key benchmark score by 9.5 points.","A new technique called ANI tries to stop language models from getting confused about what a number even means.\n\nThe problem: when an LLM reads text, it chops numbers into token fragments, which wrecks precise arithmetic. Separate numerical embedding systems fix the math but lose the plot on context, treating every number the same whether it is a price, a date, or a product ID. ANI, short for Adaptive Numerical Injection, adds a gating mechanism that reads context first and decides how to represent each number. It injects a numerical embedding method called FoNE for quantities that need accurate computation, while leaving identifier-like numbers alone so the model does not start doing arithmetic on, say, a ZIP code. In testing, ANI pushed performance on the MATH benchmark up by 9.5 points over the reference model, without dragging down scores on general language tasks.\n\nThis matters because most fixes for LLM math weakness have been blunt instruments: round every number, special-token every number, or just hope a bigger model memorizes more arithmetic. A context-aware switch is a more surgical approach, and the fact that it did not cost general-language performance is the more interesting result. That kind of trade-off is usually where these fixes fall apart.\n\nIt is one paper, one set of benchmarks, and not yet a shipped feature in any product you will use. Worth watching whether it holds up outside MATH, and whether anyone builds it into a model people actually touch.","[\"ai\",\"llm\",\"numerical-reasoning\",\"research\"]","2026-10-01T04:00:00.000Z","2026-10-01T11:24:31.585Z","2026-10-01T11:24:43.077Z","published",null,[],"ai",[24,26,27,28],"llm","numerical-reasoning","research",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.39294",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"]