[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-researchers-find-a-better-way-to-pick-llm-training-data":10,"sections":34},{"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":29,"feedback":33,"feedback_at":22,"cost_usd":33,"total_tokens":33},5537,"researchers-find-a-better-way-to-pick-llm-training-data","Researchers Find a Better Way to Pick LLM Training Data","A new algorithm called BIDS corrects influence-based bias so fine-tuned models learn diverse skills with less training data.","A new selection algorithm called BIDS trims how much data it takes to fine-tune a large language model, and it does so by fixing a bias that was quietly skewing which skills those models actually learn.\n\nInstruction fine-tuning often relies on influence-based data selection, which scores each training example by how much it shifts a model's predictions and keeps the highest scorers. Researchers found that some tasks are just intrinsically more \"influential\" by this measure than others, so selection kept favoring those tasks - and counterintuitively, that overselection hurt performance even on the favored tasks themselves. BIDS fixes this by first normalizing influence scores across tasks, then iteratively picking whichever example helps the most underrepresented task. Tested on Llama-3 and Mistral-v0.3 across seven benchmarks covering five capabilities, BIDS beat both other influence-based methods and non-influence approaches, and a 15 percent subset chosen by BIDS outperformed training on the full dataset.\n\nThat 15 percent figure matters because fine-tuning cost scales with data volume - a method that cuts data by 85 percent while improving balance is a real cost lever, not a leaderboard trick. It also names a structural problem in how instruction-tuned models get built: optimizing selection for whichever task looks most influential can silently degrade broader competence, the opposite of the balanced skills most teams say they want.\n\nThis is a preprint tested on two open models, not the frontier systems most labs actually ship, so treat \"outperforms full-dataset training\" as promising rather than settled.","[\"ai\",\"llm-training\",\"instruction-tuning\",\"machine-learning\"]","2026-08-18T04:00:00.000Z","2026-08-19T00:23:53.439Z","2026-08-19T00:24:05.238Z","published",null,[],"ai",[24,26,27,28],"llm-training","instruction-tuning","machine-learning",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2501.12147",0,{"sections":35},[36,40,44,49,54,59,64,69,74,78,83,88,93,98],{"name":37,"slug":24,"count":38,"latest_published_at":39},"AI",3293,"2026-08-20T04:00:00.000Z",{"name":41,"slug":42,"count":43,"latest_published_at":39},"Security","security",435,{"name":45,"slug":46,"count":47,"latest_published_at":48},"Policy","policy",210,"2026-08-19T09:32:27.000Z",{"name":50,"slug":51,"count":52,"latest_published_at":53},"Deals","deals",179,"2026-06-29T20:02:07.000Z",{"name":55,"slug":56,"count":57,"latest_published_at":58},"Hardware","hardware",140,"2026-08-19T18:25:42.000Z",{"name":60,"slug":61,"count":62,"latest_published_at":63},"Consumer Tech","consumer-tech",95,"2026-08-18T16:05:00.000Z",{"name":65,"slug":66,"count":67,"latest_published_at":68},"Science","science",90,"2026-08-19T18:41:02.000Z",{"name":70,"slug":71,"count":72,"latest_published_at":73},"Software","software",73,"2026-08-18T07:51:50.000Z",{"name":75,"slug":76,"count":77,"latest_published_at":18},"Dev Tools","dev-tools",69,{"name":79,"slug":80,"count":81,"latest_published_at":82},"Startups","startups",47,"2026-08-19T19:13:46.000Z",{"name":84,"slug":85,"count":86,"latest_published_at":87},"Gaming","gaming",41,"2026-07-09T04:00:00.000Z",{"name":89,"slug":90,"count":91,"latest_published_at":92},"General","general",33,"2026-08-18T22:18:13.000Z",{"name":94,"slug":95,"count":96,"latest_published_at":97},"Reviews","reviews",20,"2026-06-24T12:00:01.000Z",{"name":99,"slug":100,"count":101,"latest_published_at":102},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]