中国的DeepSeek R1 AI模型,在自然界经过同行审查,以低成本实现了强有力的推理,并被广泛采用。 China's DeepSeek R1 AI model, peer-reviewed in Nature, achieves strong reasoning at low cost and has been widely adopted.

中国人工智能初创公司DeepSeek的R1模型发表在"自然"杂志上,是第一个接受正式同行评审的大型语言模型,使用512个Nvidia H800芯片展示了其成本效益高的294,000美元的培训. Chinese AI startup DeepSeek's R1 model, published in Nature, is the first large language model to undergo formal peer review, showcasing its cost-effective training at $294,000 using 512 Nvidia H800 chips. R1是为数学和编码的复杂推理设计的,它利用强化学习来改进没有人文附加说明的数据,为正确的答案赢得高分,并通过试验和错误改进其方法。 Designed for complex reasoning in math and coding, R1 uses reinforcement learning to improve without human-annotated data, earning high scores for correct answers and refining its approach through trial and error. 作为一种开放重量模型,它已经下载超过1 090万次,下载在 " 拥抱脸 " 上,使其成为最受欢迎的同类模型。 As an open-weight model, it has been downloaded over 10.9 million times on Hugging Face, making it the most popular model of its kind. 它的成功激励人们更广泛地采用类似方法来强化AI的推理,同时提高了全球对中国在AI创新中日益重要的作用的兴趣。 Its success has inspired broader adoption of similar methods to enhance reasoning in AI, while raising global interest in China’s growing role in AI innovation.