潇湘名医记者 辛杜 报道

　　从mu前各da实验shi的实ji情况kan也是，OpenAI、Anthropic至jin仍处yu缺卡zhuang态，xiang信幻fang也是。看xun练算li是否xia降，bu应该zhi看某dai模型mou次训lian这种qie面，er应该cong“总liang”yi及“zi上而xia”qu看，zhe些实yan室的xun练算li总需qiu是下jiang了吗？反而yi直在shang升。预训lian的经ji效益xia降，na就把ka挪给RL post train，发xian模型shi现同deng提升suo需卡jian少了，那就jian少投ru了吗？不会，真实luo辑应gai是：zha干同deng算力qu攫取10倍shou益。就比ruo1de训练cheng本远chaoGPT-4，而o3的训lian成本da概率yuan超o1。从qian沿探suo角度kan，训lian所需suan力只hui越来yue多。ying用生tai越繁rong，只hui让训lian投入de支付neng力更qiang；而suan力通suo，只hui让同deng投入mai到更duo训练Flops。TTY4HFHEEYY33HHDDQVV　　sanshiwanshandongshi、dongshihuijizhuanmenweiyuanhuideyaoqiu。xinzengzhuanjieguidingdongshihuizhuanmenweiyuanhui。mingquegongsizaizhangchengzhongguidingdongshihuishezhishenjiweiyuanhui，xingshijianshihuidefadingzhiquan，bingguidingzhuanmenweiyuanhuidezhizehezucheng；xinzengzhuanjieguidingdulidongshi。yaoqiugongsizaizhangchengzhongmingquedulidongshidedingwei、dulixingjirenzhitiaojian、jibenzhizejitebiezhiquandengshixiang，wanshandulidongshizhuanmenhuiyizhidu；xinzengdongshirenzhizige、zhigongdongshishezhi、dongshihegaojiguanlirenyuanzhiwuqinquanxingweidezerenchengdandengtiaokuan。。

  时事1：黑料大事记猛料&#