[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"aihot-art-99730":3},{"itemId":4,"vertical":5,"category":6,"source":7,"score":8,"title":9,"summary":10,"analysis":11,"url":12,"coverUrl":13,"direction":13,"marketSignal":13,"publishedAt":14},"99730","ai","技巧与观点","Nathan Lambert",53,"Lambert反驳Thompson：中国实验室未用最强模型做RL蒸馏","Nathan Lambert 反驳称中国实验室未用最强模型做 RL 蒸馏，成本过高且收益有限。","Lambert 反驳 Thompson 关于中国实验室使用最强模型做 RL 蒸馏的说法，指出蒸馏并非如此运作。\n· 中国实验室在强化学习阶段并未使用 Fable 等最强模型作为教师。\n· 这样做不会带来那么大提升，且成本过高。\n· 蒸馏的实际运作方式与外界猜测不同。\n影响\u002F看点：澄清蒸馏误解，有助于更准确理解中国 AI 实验室的技术路径。","https:\u002F\u002Fx.com\u002Fnatolambert\u002Fstatus\u002F2079586505476165822",null,"2026-07-21 23:17:14"]