{"type":"link","version":"1.0","title":"Self-judged preference training lifts open-ended answers but barely moves math, code and logical reasoning","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/self-rewarding-gains-skip-math-and-reasoning","thumbnail_url":"https://ai-arch.pages.dev/og/self-rewarding-gains-skip-math-and-reasoning.png","thumbnail_width":1200,"thumbnail_height":630}