[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"topic-zh-direct-opd-rl-":3},{"cluster":4,"timeline":16},{"id":5,"slug":6,"title":7,"pinned":8,"status":9,"summary":10,"category":11,"language":12,"created_at":13,"merged_into":14,"article_count":15,"first_seen_at":13,"last_updated_at":13},"09cb40d0-8c27-4d70-b18e-7600fd44a0b2","direct-opd-rl-","Direct-OPD 讓弱模型 RL 成果可重用",false,"active","48.3% 提升到 62.4%。Direct-OPD 把弱模型 RL 產生的政策變化抽出來，再轉給更強模型，減少重跑強模型 RL 的成本。","research","zh","2026-07-07T06:32:32.575361+00:00",null,1,[17],{"id":18,"slug":19,"title":7,"summary":10,"category":11,"image_url":20,"cover_image":20,"published_at":21,"is_canonical_seed":22},"af1699b9-52dd-414b-a951-32e02a760b75","direct-opd-weak-to-strong-distillation-zh","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1783405977424-4py2.png","2026-07-07T06:32:31.331+00:00",true]