[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-teaching-ai-to-fold-proteins-also-sharpens-its-reasoning":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},9079,"teaching-ai-to-fold-proteins-also-sharpens-its-reasoning","Teaching AI to Fold Proteins Also Sharpens Its Reasoning","A new post-training recipe built from protein-folding data lifted a language model's scores on ten unrelated reasoning benchmarks, not just biology tasks.","Researchers built a model that got better at general reasoning by training it to fold proteins.\n\nThe team assembled FoldingCorpus, a dataset of questions and answers derived from protein structures, and used it to post-train models through a method they call Fold2Reason. The approach combines two kinds of feedback: discrete answers about structure generated through the model's normal text output, and continuous 3D geometry decoded from the same internal representations. On a benchmark called FoldBench, the resulting model scored 2.7 to 3.5 times higher on structure prediction than Qwen3.5-9B. The payoff showed up well beyond biology: across ten benchmarks covering spatial, graph, scientific, and general reasoning, average accuracy rose from 45.09% to 48.33%, with every single benchmark improving.\n\nThat's the interesting part. Most language model training leans on text written by humans, which tends to state an answer rather than show the spatial or structural logic that produced it. Protein structures skip that shortcut - one solved shape yields thousands of checkable spatial facts, with no ambiguity about what's correct. The researchers also ran control versions trained on random, synthetic, or shuffled structure data, and those controls gained far less or actually lost ground, suggesting the real protein geometry, not just more training volume, is doing the work.\n\nA 3.23 percentage point average gain across ten benchmarks is a real result, not a rounding error, but it's not the kind of number that rewrites how anyone trains a frontier model tomorrow. The bigger claim here is about method, not magnitude: a single, well-understood scientific problem can be mined for reasoning supervision that text alone doesn't provide. Whether that holds for other structure-rich sciences, or just protein folding, is the obvious next experiment.","[\"ai-reasoning\",\"protein-folding\",\"llm-training\",\"research\"]","2026-10-01T04:00:00.000Z","2026-10-01T18:52:39.804Z","2026-10-01T18:52:42.557Z","published",null,[],"ai",[26,27,28,29],"ai-reasoning","protein-folding","llm-training","research",[31],{"name":32,"url":33},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.38879",0,{"sections":36},[37,40,44,49,54,59,63,68,73,77,82,87,92,97],{"name":38,"slug":24,"count":39,"latest_published_at":18},"AI",5488,{"name":41,"slug":42,"count":43,"latest_published_at":18},"Security","security",809,{"name":45,"slug":46,"count":47,"latest_published_at":48},"Policy","policy",429,"2026-10-01T02:26:17.000Z",{"name":50,"slug":51,"count":52,"latest_published_at":53},"Deals","deals",298,"2026-09-30T21:00:26.000Z",{"name":55,"slug":56,"count":57,"latest_published_at":58},"Hardware","hardware",196,"2026-09-30T13:00:00.000Z",{"name":60,"slug":61,"count":62,"latest_published_at":18},"Science","science",162,{"name":64,"slug":65,"count":66,"latest_published_at":67},"Consumer Tech","consumer-tech",149,"2026-09-30T22:57:11.000Z",{"name":69,"slug":70,"count":71,"latest_published_at":72},"Dev Tools","dev-tools",93,"2026-10-01T02:30:48.000Z",{"name":74,"slug":75,"count":71,"latest_published_at":76},"Software","software","2026-09-30T21:41:11.000Z",{"name":78,"slug":79,"count":80,"latest_published_at":81},"Startups","startups",84,"2026-09-30T20:39:09.000Z",{"name":83,"slug":84,"count":85,"latest_published_at":86},"Gaming","gaming",51,"2026-09-30T16:24:30.000Z",{"name":88,"slug":89,"count":90,"latest_published_at":91},"General","general",50,"2026-09-30T21:37:54.000Z",{"name":93,"slug":94,"count":95,"latest_published_at":96},"Reviews","reviews",31,"2026-09-28T14:31:34.000Z",{"name":98,"slug":99,"count":100,"latest_published_at":101},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]