{"type":"link","version":"1.0","title":"Preference training on whole correct-versus-incorrect answers stops improving reasoning as synthetic problems scale","author_name":"AI Archs","author_url":"https://ai-arch.pages.dev","provider_name":"AI Archs","provider_url":"https://ai-arch.pages.dev","url":"https://ai-arch.pages.dev/n/whole-answer-preference-pairs-stall-on-reasoning-data","thumbnail_url":"https://ai-arch.pages.dev/og/whole-answer-preference-pairs-stall-on-reasoning-data.png","thumbnail_width":1200,"thumbnail_height":630}