[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-ai-models-learn-to-hesitate-on-spatial-reasoning-tasks":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},9056,"ai-models-learn-to-hesitate-on-spatial-reasoning-tasks","AI Models Learn to Hesitate on Spatial Reasoning Tasks","A new post-training method lets vision-language AI blend multiple reasoning paths instead of committing to one, improving spatial judgment accuracy.","A new training method teaches AI vision models to hesitate instead of guessing too fast on spatial questions.\n\nLarge vision-language models (LVLMs) usually work through spatial problems by picking one word at a time to represent each reasoning step, a method called chain-of-thought. The problem: once a word is chosen, the model can't walk it back, even if it wasn't confident that word was correct. A new paper proposes Soft Spatial Reasoning, a post-training method that lets the model blend several candidate next steps into a continuous representation instead of locking in a single token. A controller called AdaptSoft decides, at each step, how much blending to allow based on the model's current confidence.\n\nThis goes after a specific failure mode: an early misstep in chain-of-thought reasoning can snowball into a wrong answer, and spatial judgments, like what's left of what or what's behind what, are where LVLMs trip up most. Letting the model hedge selectively, rather than always blending or never blending, is a more targeted fix than earlier fixed-softness approaches to AI reasoning.\n\nThe researchers released their code, so the real test now is whether this selective hedging holds up on spatial tasks the benchmarks didn't anticipate.","[\"ai\",\"computer-vision\",\"research\",\"spatial-reasoning\"]","2026-10-01T04:00:00.000Z","2026-10-01T17:42:59.774Z","2026-10-01T17:43:05.279Z","published",null,[24],{"id":25,"reviewer":26,"round":27,"reason":28,"status":29},"editor-r1","editor",1,"Define the LVLM acronym on first use (e.g., 'large vision-language models (LVLMs)') before using the bare acronym later in the piece.","resolved","ai",[30,32,33,34],"computer-vision","research","spatial-reasoning",[36],{"name":37,"url":38},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.38717",0,{"sections":41},[42,45,49,54,59,64,68,73,78,82,87,92,97,102],{"name":43,"slug":30,"count":44,"latest_published_at":18},"AI",5488,{"name":46,"slug":47,"count":48,"latest_published_at":18},"Security","security",809,{"name":50,"slug":51,"count":52,"latest_published_at":53},"Policy","policy",429,"2026-10-01T02:26:17.000Z",{"name":55,"slug":56,"count":57,"latest_published_at":58},"Deals","deals",298,"2026-09-30T21:00:26.000Z",{"name":60,"slug":61,"count":62,"latest_published_at":63},"Hardware","hardware",196,"2026-09-30T13:00:00.000Z",{"name":65,"slug":66,"count":67,"latest_published_at":18},"Science","science",162,{"name":69,"slug":70,"count":71,"latest_published_at":72},"Consumer Tech","consumer-tech",149,"2026-09-30T22:57:11.000Z",{"name":74,"slug":75,"count":76,"latest_published_at":77},"Dev Tools","dev-tools",93,"2026-10-01T02:30:48.000Z",{"name":79,"slug":80,"count":76,"latest_published_at":81},"Software","software","2026-09-30T21:41:11.000Z",{"name":83,"slug":84,"count":85,"latest_published_at":86},"Startups","startups",84,"2026-09-30T20:39:09.000Z",{"name":88,"slug":89,"count":90,"latest_published_at":91},"Gaming","gaming",51,"2026-09-30T16:24:30.000Z",{"name":93,"slug":94,"count":95,"latest_published_at":96},"General","general",50,"2026-09-30T21:37:54.000Z",{"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"]