{"type":"link","version":"1.0","title":"Retrieving external skills during on-policy RL bakes their benefit into weights, so the trained policy improves even with the memory removed","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/on-policy-skill-retrieval-internalizes-memory-into-weights","thumbnail_url":"https://ai-arch.pages.dev/og/on-policy-skill-retrieval-internalizes-memory-into-weights.png","thumbnail_width":1200,"thumbnail_height":630}