[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-ai-attention-heads-that-match-human-brains-barely-matter":10,"sections":40},{"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":30,"tags":31,"sources":35,"feedback":39,"feedback_at":22,"cost_usd":39,"total_tokens":39},8538,"ai-attention-heads-that-match-human-brains-barely-matter","AI Attention Heads That Match Human Brains Barely Matter","New research finds AI attention heads resembling human brain activity matter far less than heads picked for their actual computational role.","Matching a brain scan doesn't mean an AI component is doing the same job as the brain.\n\nA paper titled 'Which Attention Heads are like the Human Head? Not the Ones that Compute' (arXiv:2609.37991v1, posted 2026-09-30) tested 17 large language models, ranging from 3 billion to 72 billion parameters, on a simple pattern-completion task: given a sequence like AAABAAA, predict the next symbol. Researchers recorded human EEG activity during the same task and identified which model attention heads produced representations that lined up with the brain signals. They then compared two ways of ranking head importance: brain alignment, and attribution patching, a standard interpretability method that measures how much removing a head hurts the model's actual answers. Ablating the brain-aligned heads barely dented performance, while removing heads flagged by attribution patching caused far more damage.\n\nThat gap matters because brain-AI alignment is often treated as evidence a model computes things the way a brain does, exactly the assumption this paper set out to test. The results suggest alignment mostly captures how a model reads the stimulus, not how it represents the pattern or solves the task, a much narrower claim than the field's excitement usually implies. Attribution patching, the method interpretability researchers already trust for finding causally important components, remains the more reliable signal.\n\nA flashy EEG match makes for a nice figure, but this paper is a reminder to check what happens when you actually remove the head before trusting it.","[\"ai\",\"interpretability\",\"neuroscience\",\"llm-research\"]","2026-09-30T04:00:00.000Z","2026-09-30T09:54:47.091Z","2026-09-30T09:54:49.921Z","published",null,[24],{"id":25,"reviewer":26,"round":27,"reason":28,"status":29},"editor-r1","editor",1,"Add explicit attribution to the specific paper — its title, arXiv identifier (arXiv:2609.37991v1), and posting date — since the draft states all findings via a vague 'researchers' with no named source or publication a reader could verify.","resolved","ai",[30,32,33,34],"interpretability","neuroscience","llm-research",[36],{"name":37,"url":38},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.37991",0,{"sections":41},[42,45,49,53,58,63,68,73,78,82,87,92,97,102],{"name":43,"slug":30,"count":44,"latest_published_at":18},"AI",5105,{"name":46,"slug":47,"count":48,"latest_published_at":18},"Security","security",785,{"name":50,"slug":51,"count":52,"latest_published_at":18},"Policy","policy",417,{"name":54,"slug":55,"count":56,"latest_published_at":57},"Deals","deals",284,"2026-09-29T21:00:00.000Z",{"name":59,"slug":60,"count":61,"latest_published_at":62},"Hardware","hardware",194,"2026-09-29T13:16:04.000Z",{"name":64,"slug":65,"count":66,"latest_published_at":67},"Science","science",154,"2026-09-28T13:19:18.000Z",{"name":69,"slug":70,"count":71,"latest_published_at":72},"Consumer Tech","consumer-tech",142,"2026-09-29T18:38:03.000Z",{"name":74,"slug":75,"count":76,"latest_published_at":77},"Software","software",91,"2026-09-25T20:55:00.000Z",{"name":79,"slug":80,"count":81,"latest_published_at":18},"Dev Tools","dev-tools",90,{"name":83,"slug":84,"count":85,"latest_published_at":86},"Startups","startups",83,"2026-09-29T21:51:36.000Z",{"name":88,"slug":89,"count":90,"latest_published_at":91},"General","general",49,"2026-09-28T16:44:57.000Z",{"name":93,"slug":94,"count":95,"latest_published_at":96},"Gaming","gaming",48,"2026-09-25T18:35:21.000Z",{"name":98,"slug":99,"count":100,"latest_published_at":101},"Reviews","reviews",31,"2026-09-28T14:31:34.000Z",{"name":103,"slug":104,"count":105,"latest_published_at":106},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]