[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-a-new-rule-for-how-ai-models-should-weigh-their-own-thoughts":10,"sections":48},{"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":38,"tags":39,"sources":43,"feedback":47,"feedback_at":22,"cost_usd":47,"total_tokens":47},8682,"a-new-rule-for-how-ai-models-should-weigh-their-own-thoughts","A New Rule for How AI Models Should Weigh Their Own Thoughts","A paper posted to arXiv argues AI models should weight every reasoning path equally, not just the latest one, to reason more efficiently at smaller sizes.","New research says AI reasoning models may be weighting their own thought process wrong.\n\nA paper posted to arXiv (arXiv:2609.13747, posted September 2026) examines \"continuous reasoning,\" a technique where a language model holds multiple lines of reasoning in a single latent state instead of writing each one out as text. The researchers tested a common assumption: that a model should mostly preserve its most recent reasoning step, since older steps compete for limited hidden space. They found the opposite holds when later computation draws on several earlier ideas at once. In that case, weighting all reached ideas equally lets a model guide its attention correctly using less hidden width than an approach that only keeps the newest states. Tests on two-layer and GPT-2 Transformers confirmed the pattern, and showed that when weights are uneven, the least-weighted ideas are the first to degrade.\n\nThis is a narrow, technical result, but it points at something practical. Frontier reasoning models today spend enormous compute writing out and re-reading chains of thought as text. Continuous reasoning tries to do that work inside the model's hidden layers instead, which is cheaper if researchers can get the bookkeeping right. This paper offers a concrete default for that bookkeeping: keep every live idea equally weighted, and reset to equal weights as computation proceeds.\n\nIt won't show up in a shipped product this year, but it's the unglamorous math that will decide whether reasoning models get meaningfully cheaper to run, or just get bigger.","[\"ai\",\"llm-reasoning\",\"transformers\",\"research\"]","2026-09-30T04:00:00.000Z","2026-09-30T19:11:17.305Z","2026-09-30T19:11:23.222Z","published",null,[24,30,34],{"id":25,"reviewer":26,"round":27,"reason":28,"status":29},"editor-r1","editor",1,"Add explicit attribution for the study (arXiv ID and posting date, e.g., arXiv:2609.13747, posted September 2026) since the draft never cites where or when the research appeared, and rework the final paragraph so it doesn't end on a bare caveat—close with context or implication instead.","resolved",{"id":31,"reviewer":26,"round":32,"reason":33,"status":29},"editor-r2",2,"Attribution is now fixed, but the closing paragraph still just restates the small-scale caveat as a rhetorical question—end instead on what this means for practical LLM design or what evidence would resolve the scale question, per the open concern.",{"id":35,"reviewer":26,"round":36,"reason":37,"status":29},"editor-r3",3,"Add explicit attribution in the body citing the arXiv ID and posting date (e.g., arXiv:2609.13747, posted September 2026) — the draft still only says 'a paper posted to arXiv' with no identifier or date, which was the other half of editor-r1's still-open request.","ai",[38,40,41,42],"llm-reasoning","transformers","research",[44],{"name":45,"url":46},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.13747",0,{"sections":49},[50,53,57,61,66,71,75,80,85,89,94,99,104,109],{"name":51,"slug":38,"count":52,"latest_published_at":18},"AI",5184,{"name":54,"slug":55,"count":56,"latest_published_at":18},"Security","security",791,{"name":58,"slug":59,"count":60,"latest_published_at":18},"Policy","policy",417,{"name":62,"slug":63,"count":64,"latest_published_at":65},"Deals","deals",284,"2026-09-29T21:00:00.000Z",{"name":67,"slug":68,"count":69,"latest_published_at":70},"Hardware","hardware",194,"2026-09-29T13:16:04.000Z",{"name":72,"slug":73,"count":74,"latest_published_at":18},"Science","science",155,{"name":76,"slug":77,"count":78,"latest_published_at":79},"Consumer Tech","consumer-tech",142,"2026-09-29T18:38:03.000Z",{"name":81,"slug":82,"count":83,"latest_published_at":84},"Software","software",91,"2026-09-25T20:55:00.000Z",{"name":86,"slug":87,"count":88,"latest_published_at":18},"Dev Tools","dev-tools",90,{"name":90,"slug":91,"count":92,"latest_published_at":93},"Startups","startups",83,"2026-09-29T21:51:36.000Z",{"name":95,"slug":96,"count":97,"latest_published_at":98},"General","general",49,"2026-09-28T16:44:57.000Z",{"name":100,"slug":101,"count":102,"latest_published_at":103},"Gaming","gaming",48,"2026-09-25T18:35:21.000Z",{"name":105,"slug":106,"count":107,"latest_published_at":108},"Reviews","reviews",31,"2026-09-28T14:31:34.000Z",{"name":110,"slug":111,"count":112,"latest_published_at":113},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]