{"type":"link","version":"1.0","title":"An agent scoring its own outputs shows almost none of the bias-coupling that an independent evaluator reveals","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/risk-self-evaluation-underestimates-preference-coupling","thumbnail_url":"https://ai-arch.pages.dev/android-chrome-512x512.png","thumbnail_width":512,"thumbnail_height":512}