[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-ai-system-learns-which-facts-matter-before-giving-advice":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},9451,"ai-system-learns-which-facts-matter-before-giving-advice","AI System Learns Which Facts Matter Before Giving Advice","A research system for public-service chatbots sorts must-know facts from nice-to-knows, and finds extra verification can make recommendations worse.","A new research system decides which facts actually need confirming before it tells you whether you qualify for a public service.\n\nResearchers built EviGraph to help AI agents navigate public-service recommendations, the kind of government-benefit or permit questions where missing one key fact can sink an answer. Instead of treating every blank as a reason to punt, EviGraph splits requirements into what's critical to the decision, like which service, what scope, and what date, and what can stay unresolved. A language agent maps those critical requirements to evidence pulled from a temporal knowledge graph, then a separate deterministic checker confirms whether that evidence actually backs the recommendation. Tested on a bilingual Hong Kong public-service benchmark with real policy references, the approach cut down on unnecessary refusals to answer.\n\nThe more interesting result is the warning shot buried in the findings: piling on extra verification steps walked back correct recommendations without making the system any more accurate. That cuts against the standard reflex in AI safety work, which is to bolt on more checking layers whenever a system feels risky. EviGraph's authors argue the fix isn't more verification, it's being precise about what actually needs to be established before a decision gets made.\n\nThis is a paper, not a product, but it's a useful data point for anyone building a government chatbot: knowing when you don't need an answer is as important as knowing when you do.","[\"ai\",\"knowledge-graphs\",\"public-services\",\"research\"]","2026-10-02T04:00:00.000Z","2026-10-02T19:40:11.607Z","2026-10-02T19:40:16.412Z","published",null,[],"ai",[24,26,27,28],"knowledge-graphs","public-services","research",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.00212",0,{"sections":35},[36,39,43,48,53,58,63,68,73,78,83,88,93,98],{"name":37,"slug":24,"count":38,"latest_published_at":18},"AI",5764,{"name":40,"slug":41,"count":42,"latest_published_at":18},"Security","security",831,{"name":44,"slug":45,"count":46,"latest_published_at":47},"Policy","policy",437,"2026-10-01T18:10:00.000Z",{"name":49,"slug":50,"count":51,"latest_published_at":52},"Deals","deals",317,"2026-10-01T22:00:00.000Z",{"name":54,"slug":55,"count":56,"latest_published_at":57},"Hardware","hardware",198,"2026-10-01T17:38:48.000Z",{"name":59,"slug":60,"count":61,"latest_published_at":62},"Science","science",168,"2026-10-01T18:35:55.000Z",{"name":64,"slug":65,"count":66,"latest_published_at":67},"Consumer Tech","consumer-tech",155,"2026-10-01T19:54:10.000Z",{"name":69,"slug":70,"count":71,"latest_published_at":72},"Dev Tools","dev-tools",96,"2026-10-01T16:57:03.000Z",{"name":74,"slug":75,"count":76,"latest_published_at":77},"Software","software",93,"2026-09-30T21:41:11.000Z",{"name":79,"slug":80,"count":81,"latest_published_at":82},"Startups","startups",90,"2026-10-01T21:55:22.000Z",{"name":84,"slug":85,"count":86,"latest_published_at":87},"Gaming","gaming",53,"2026-10-02T02:50:39.000Z",{"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",7,"2026-10-01T09:00:00.000Z"]