{"type":"link","version":"1.0","title":"Human-preference reward models reward persuasiveness, which a policy can satisfy without being correct","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/preference-rewards-conflate-persuasiveness-with-truth","thumbnail_url":"https://ai-arch.pages.dev/og/preference-rewards-conflate-persuasiveness-with-truth.png","thumbnail_width":1200,"thumbnail_height":630}