{"type":"link","version":"1.0","title":"Distilling a strong model's traces raises both single-shot and many-sample accuracy, where small-scale RL trades one for the other","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/distillation-preserves-exploration-diversity-not-just-greedy-accuracy","thumbnail_url":"https://ai-arch.pages.dev/og/distillation-preserves-exploration-diversity-not-just-greedy-accuracy.png","thumbnail_width":1200,"thumbnail_height":630}