{"type":"link","version":"1.0","title":"Hardening the evaluation environment prevents reward hacking more reliably than behavioral training","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/environmental-hardening-over-behavioral-training","thumbnail_url":"https://ai-arch.pages.dev/android-chrome-512x512.png","thumbnail_width":512,"thumbnail_height":512}