At the beginning of the Spring Festival in 2023, the concept of ChatGPT chat robot continued to heat up. Microsoft, Google, Baidu and other Internet giants competed for layout, driving the attention of relevant chips to increase, and triggering widespread discussion in the industry.

Internet giants are betting on ChatGPT

According to the data, ChatGPT is a new chat robot model developed by OpenAI in November 2022. The model was launched only two months ago, and has reached a monthly life of more than 100 million, becoming the fastest growing consumer application in history.

Compared with other chat robots, ChatGPT can not only conduct dialogue and interaction through learning and understanding human language, but also assist human beings to complete tasks such as writing email, writing papers, formulating plans, creating poetry, typing code, etc. The level of intelligence and user experience are further improved.

At the beginning of February this year, OpenAI announced the launch of the paid subscription version of ChatGPT Plus, priced at $20 per month. The industry estimates that more than 1 billion potential users worldwide can use ChatGPT. If the charging model is recognized, ChatGPT is expected to bring more than 200 billion dollars in market size.

The value of ChatGPT is not only here. In the future, ChatGPT is also expected to drive the take-off of chips, HPC, cloud computing and other fields. Bill Gates, the founder of Microsoft, commented that the emergence of ChatGPT, an artificial intelligence technology, is no less significant than the birth of the Internet and personal computers.

Attracted by the bright prospect of ChatGPT, the big internet companies have made bets.

In fact, as early as 2019, Microsoft invested $1 billion in OpenAI, and the two sides cooperated to develop "general artificial intelligence". At the same time, Microsoft obtained the right to commercialize part of OpenAI technology. In January this year, Microsoft said that it would invest in OpenAI for several years and worth billions of dollars. In February, Microsoft announced the launch of the latest version of artificial intelligence search engine Bing (Bing) and Edge browser supported by ChatGPT.

The hot development of ChatGPT has also led Google and Baidu to take action. On February 6 this year, Google announced that it would launch the chat robot "Bard", which will compete with Microsoft ChatGPT in the field of generative artificial intelligence. On February 7, Baidu said that the name of the new project of the big model was "Wenxinyiyan", and the English name was "ERNIE Bot", which has the ability of deep semantic understanding and generation across modules and languages.

In addition, according to many domestic media reports, after the exposure of the Baidu ChatGPT project "Wen Xin Yi Yan", Alibaba also confirmed that its ChatGPT product is under development and is currently in the internal test stage.

ChatGPT is expected to help the development of high-performance GPU

As a product of the deep development of AI to a certain stage, ChatGPT has a high demand for chip computing power, and products such as high-performance GPU are expected to benefit from it.

The industry pointed out that the underlying computing chips of natural language AI applications are mainly high-performance GPUs, in which NVIDIA occupies the main market. It is reported that ChatGPT has imported at least 10000 Nvidia high-end GPUs. Citibank has analyzed that ChatGPT may bring sales of US $3 billion to US $11 billion to Nvidia in the next 12 months.

AMD, another major chip manufacturer, showed the next generation of APU product Instinct MI300 for data centers at CES this year, and plans to launch it in the second half of the year. Instinct MI300 adopts chiplet design, has 13 small chips and 146 billion transistors, surpassing Intel's Ponte Vecchio with 100 billion transistors, and has become the largest chip AMD has put into production. AMD said that Instinct MI300 can reduce the training time of super large AI models such as ChatGPT and DALL-E from a few months to a few weeks, thus saving millions of dollars in electricity.

With the continuous popularity of ChatGPT concept, domestic AI chip manufacturers are also expected to benefit. In addition, because ChatGPT relies on a large number of computing power (chips) to support its training and deployment, ChatGPT still needs the support of large computing power servers when it is applied, which is also expected to indirectly drive the development of server memory.

Is ChatGPT really perfect?

While ChatGPT is popular, the industry is also full of doubts. For example, although ChatGPT's function is dazzling, as a newly launched product, it still has inaccurate and untrue answers to some questions, and its practicability needs to be further strengthened.

According to the latest media report, Google's AI chat robot Bard made a factual error in a demo video. When asked about the new discovery of the James Weber Space Telescope, Bard said that the telescope had taken the first pictures of extrasolar planets, but the American National Astronaut said that these pictures were actually taken by another telescope.

Another example is the old AI threat theory. With the continuous improvement of ChatGPT technology, will the future mental workers (engineers, news editors, translators, copywriters, etc.) be replaced?

Finally, the problem of data security. When employees use ChatGPT to assist in work, the risk of trade secret disclosure also increases. Is the data obtained by ChatGPT compliant and legal? Whether the use of data for re-creation infringes the relevant copyright... These are all issues that need to be solved during the development of ChatGPT.