{"type":"link","version":"1.0","title":"The gap between a proxy reward and the true objective traces a predictable curve in KL distance from the start","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/reward-model-overoptimization-follows-a-predictable-curve-in-kl","thumbnail_url":"https://ai-arch.pages.dev/og/reward-model-overoptimization-follows-a-predictable-curve-in-kl.png","thumbnail_width":1200,"thumbnail_height":630}