{"type":"link","version":"1.0","title":"Self-monitoring still active when a model commits to its answer is the strongest latent signal of a wrong answer","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/lingering-self-monitoring-at-answer-time-predicts-wrong-answers","thumbnail_url":"https://ai-arch.pages.dev/og/lingering-self-monitoring-at-answer-time-predicts-wrong-answers.png","thumbnail_width":1200,"thumbnail_height":630}