[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-a-fix-for-ai-agents-that-forget-what-they-learned":10,"sections":35},{"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":30,"feedback":34,"feedback_at":22,"cost_usd":34,"total_tokens":34},5073,"a-fix-for-ai-agents-that-forget-what-they-learned","A Fix for AI Agents That Forget What They Learned","A new gate lets AI agents learn tricks only if they do not break what already works, slashing errors on clinical-record benchmarks.","A new arXiv paper describes a fix for a failure mode nobody flags at launch: AI agents that patch one bug by quietly breaking another.\n\nResearchers call the system GRASP, for Gated Regression-Aware Skill Proposer. It treats an agent's growing pile of notes-to-self as a library of edits, not an ever-expanding scrapbook, and it only keeps a new skill if the agent passes a held-out test without losing ground elsewhere, under a hard cap on how much backsliding is allowed. Tested across five base models on two FHIR-based clinical benchmarks, which grade whether an agent follows procedure correctly rather than whether its medical judgment is sound, the gains were large: on MedAgentBench, GRASP took gpt-oss-120b from 40.6% to 88.8%, beating the best of five rival self-improvement methods by 21.0 points. Every other model tested improved by 17.2 to 40.3 points.\n\nThe interesting finding is not the score, it is the diagnosis. When the researchers gave the same acceptance gate to the competing methods, all of them improved on the benchmark they were tuned for, but none improved on a different one. That means the gains trace to checking each change against a bounded library before keeping it, not to writing better notes or to validation in general. It is the AI-agent version of code review with a regression suite, applied to natural-language instructions instead of code.\n\nThe catch: this only works when tasks repeat with a checkable structure, like clinical record lookups. For open-ended work, the paper reports the effect flattens out, which is a fair warning against reading too much into the clinical numbers.","[\"ai-agents\",\"llm-research\",\"self-improvement\",\"clinical-ai\"]","2026-08-17T04:00:00.000Z","2026-08-17T09:00:24.432Z","2026-08-17T09:00:36.267Z","published",null,[],"ai",[26,27,28,29],"ai-agents","llm-research","self-improvement","clinical-ai",[31],{"name":32,"url":33},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2605.29668",0,{"sections":36},[37,41,45,50,55,60,65,70,75,80,85,90,95,100],{"name":38,"slug":24,"count":39,"latest_published_at":40},"AI",3293,"2026-08-20T04:00:00.000Z",{"name":42,"slug":43,"count":44,"latest_published_at":40},"Security","security",435,{"name":46,"slug":47,"count":48,"latest_published_at":49},"Policy","policy",210,"2026-08-19T09:32:27.000Z",{"name":51,"slug":52,"count":53,"latest_published_at":54},"Deals","deals",179,"2026-06-29T20:02:07.000Z",{"name":56,"slug":57,"count":58,"latest_published_at":59},"Hardware","hardware",140,"2026-08-19T18:25:42.000Z",{"name":61,"slug":62,"count":63,"latest_published_at":64},"Consumer Tech","consumer-tech",95,"2026-08-18T16:05:00.000Z",{"name":66,"slug":67,"count":68,"latest_published_at":69},"Science","science",90,"2026-08-19T18:41:02.000Z",{"name":71,"slug":72,"count":73,"latest_published_at":74},"Software","software",73,"2026-08-18T07:51:50.000Z",{"name":76,"slug":77,"count":78,"latest_published_at":79},"Dev Tools","dev-tools",69,"2026-08-18T04:00:00.000Z",{"name":81,"slug":82,"count":83,"latest_published_at":84},"Startups","startups",47,"2026-08-19T19:13:46.000Z",{"name":86,"slug":87,"count":88,"latest_published_at":89},"Gaming","gaming",41,"2026-07-09T04:00:00.000Z",{"name":91,"slug":92,"count":93,"latest_published_at":94},"General","general",33,"2026-08-18T22:18:13.000Z",{"name":96,"slug":97,"count":98,"latest_published_at":99},"Reviews","reviews",20,"2026-06-24T12:00:01.000Z",{"name":101,"slug":102,"count":103,"latest_published_at":104},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]