2月6日，据外媒报道，李飞飞等研究人员用不到50美元和16张英伟达H100 GPU，耗时26分钟训练出s1-32B人工智能推理模型。据外界分析，该模型基于预训练模型微调，成本未涵盖前期“烧钱”部分，且只在特定测试集上超过o1-preview，未超过“满血版”o1和DeepSeek-R1。在“白菜价”的背后，李飞飞团队重点研究了如何以简单方式实现“测试时拓展”，即模型在推理时多次迭代优化结果，还构建 ...

SEOUL, Jan. 26 (Xinhua) -- The Democratic People's Republic of Korea (DPRK) conducted a sea-to-surface strategic cruise missile test on Saturday under the supervision of its top leader to beef up its ...

A Long March-3B rocket carrying a test satellite for communication technology blasts off from the Xichang Satellite Launch ...