Photo/Zhang Jian (NBD)

April 12 (NBD) -- Chinese tech companies are facing an AI arms race. In less than 3 days, 4 domestically produced large models have made their debut.

On April 11th, Alibaba released a large language model "Tongyi Qianwen" at Alibaba Cloud Summit 2023 in Beijing, following SenseTime (SenseNova), Kunlun Wanwei, and 360 Information Technology., etc. 

With the arrival of a new round of AI boom, those who seek to profit from it are flooding in. However, the high entry threshold is also noteworthy. 

The larger the model, the higher the demand for computing power. At the Artificial Intelligence Large Model Technology Summit, Wu Fei, director of the Artificial Intelligence Research Institute of Zhejiang University, stated that the training threshold for ChatGPT is 10,000 NVIDIA V100 chips, which cost about 1 billion yuan. The training model computes 10 trillion times per second and runs for 3,640 days. 

Computer scientist and natural language model expert Wu Jun mentioned the training cost of ChatGPT during a live broadcast, saying, "The amount of electricity for training is able to power 3,000 Tesla electric cars for 200,000 miles each. It's very expensive." 

"I agree with Baidu's Li Yanhong. One or two companies in the country are enough," said Pan Yunhe, academician of the Chinese Academy of Engineering, at the aforementioned Artificial Intelligence Large Model Technology Summit. "We should aim at the places where it hasn't reached yet." 

On the other hand, ChatGPT, ChatGPT has attracted attention from multiple countries regarding security risks. Italy was the first to announce a ban on the use of the ChatGPT chatbot, and subsequently, many countries began to consider adopting stricter regulatory measures for ChatGPT. 

On April 11th, the Cyber Space Administration of China publicly solicited opinions on the "Administrative Measures for Generative Artificial Intelligence Services (Draft for Soliciting Opinions)". It proposed that before using generative artificial intelligence products to provide services to the public, security assessments should be submitted to the National Internet Information Office in accordance with regulations, and algorithm filing, change, and cancellation procedures should be followed.