[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"tag-counterfactual":3},{"tag":4,"articles":9},{"id":5,"name":6,"slug":6,"article_count":7,"description_zh":8,"description_en":8},"9edc1311-d471-4428-b8e4-931fd5f75300","counterfactual",0,null,[10],{"id":11,"slug":12,"title":13,"summary":14,"category":15,"image_url":16,"cover_image":16,"language":17,"created_at":18},"f31f51ba-4445-4e43-9bda-31e70f53d42b","audio-language-models-arbitration-reversals-zh","音訊模型不是聽不懂","這篇論文指出，音訊語言模型常常已經抓到正確的聲音線索，卻在最後判斷時被文字蓋過去。","research","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780553877373-ux95.png","zh","2026-06-04T06:17:27.890159+00:00"]