In November 2022, OpenAI's ChatGPT was born, with 1 million users in seven days and more than 100 million monthly active users only two months later, making it the shortest platform in the world to break 100 million users. Today, the number of ChatGPT visits has risen to 889 million. The popularity of ChatGPT in the world has led to a number of similar AI language models (or platforms) springing up like bamboo shoots after a spring rain, such as Baidu's Wen Xin word and Ali's Tongyi Thousand questions are the first domestic AI platform products.

Improved Internet-enabled ChatGPT-3.5 version of Microsoft NewBing with built-in Openai

So, while we marvel at the strength of foreign ChatGPT and the rapid growth of users, we will also be concerned about the gap between the current domestic AI language model and ChatGPT? On this issue, Baidu CEO Robin Li said at the geek park event that the gap between Wen Xin and the ChatGPT version is not obvious, and it may only be a month or two.

For this view, Wang Xiaochuan, founder of Sogou, joked that the media may be interviewing Robin Li from a parallel world, and he believes that the gap between China's current big model and ChatGPT can not be only two months. Wang Xiaochuan added that if we catch up with the current ChatGPT-3.5 level, we will probably be able to do so within this year, but if we want to catch up with the future GPT-4 and GPT-5, it will take about three years.

According to public data, OpenAI launched the first generation of GPT-1 in 2018 and GPT-2 in 2019. By 2020, the number of parameters has changed from 120 million to 175 billion. At present, the number of GPT-3.5 and GPT-4 we use will be at least 100 billion.

ChatGPT has its own advantages in terms of research and development time and the number of parameters. However, according to the data released so far, the number of parameters in Baidu Wenxin is about 260 billion, which is no less than GPT-3. Moreover, it has local advantages and has great potential for domestic development in the future.

Editor's comment: back to the gap between AI and ChatGPT in China, I think it is possible for us to catch up from behind, but two months is obviously a bit too optimistic. In addition to the growth of the number of parameters, it is also faced with many factors, such as computing power, landing application and policy, otherwise it will only be a flash in the pan like the previous VRAR and cryptocurrency.