How Chinese Enterprises Respond to Competition in the AI Industry
ChatGPT, an artificial intelligence chat robot launched by the American company OpenAI, has become popular. It has harvested 100 million monthly active users within two months of its launch, making it the fastest-growing consumer application in history.
This speed broke the previous record of over 100 million monthly active users in 9 months set by TikTok, a short video social platform owned by ByteDance.
What kind of future does this "phenomenal" product herald?
Does China have the foundation to develop similar products?
competition has arrived
It has become a fashion to display screenshots of conversations between oneself and ChatGPT on social platforms.
It can communicate with human users in natural language, understand user intentions, and complete various content generation tasks such as writing papers and codes; it can also show the aura of "student master", passed the MBA exam at Wharton Business School, and passed the University of Minnesota passed the law school exam...
"ChatGPT is another innovation in the field of artificial intelligence, which will promote a new round of AI (artificial intelligence) model competition." Zheng Rong, chief technology officer of Yuanjian Information Technology Co., Ltd. believes that the biggest contribution of ChatGPT lies in the basic realization of the ideal model. The interface layer of the language pre-training model allows robots to learn to adapt to human expressions, instead of allowing humans to learn instructions that can command machines. This is a human-machine interface technology that is more in line with human expression habits.
The popularity of ChatGPT has brought a huge impact on both the industry and academia.
"Since 2016, the scale of my country's artificial intelligence industry has shown a rapid growth trend, but in 2019, my country's artificial intelligence investment and financing have fallen sharply." Song Yu, vice president of CCID Consulting Co., Ltd. believes that overall, the ebb of artificial intelligence development is mainly There is still a lack of technological breakthroughs to support the rapid development of the industry, and the emergence of ChatGPT has just brought new impetus to the development and investment of the industry.
The high-tech investment institution Ark Investment released a report predicting that by 2030, AI will greatly improve the work efficiency of knowledge workers, and the average work efficiency will increase by 140%.
"The new generation of artificial intelligence technology will likely greatly reduce the work intensity of mental workers. If all this becomes a reality, it will be another profound change in human society after the power revolution in human history got rid of heavy physical labor. The technological revolution.” Su Zhong, a senior technical expert at the Ali Research Institute and a standing member of the Artificial Intelligence Committee of the China Computer Federation, said in a recently published article.
"ChatGPT is more inclined to the huge advantage of the dialogue mechanism, which is a huge challenge to traditional search engines including Google. Its popularity will greatly promote the development of large-scale language model research led by OpenAI. There will be breakthroughs in artificial intelligence-related fields." Hu Kun, an associate researcher at the Institute of Artificial Intelligence of Beihang University, said that because ChatGPT has high requirements for computing power, it is more difficult for universities with limited computing resources. The difficulty of academic research has increased, and there is an urgent need to open up corporate resources and strengthen industry-university-research cooperation.
Grasp the layout
In this wave of AI competition brought about by ChatGPT, China is obviously not far behind.
In the "White Paper on the Development of Beijing Artificial Intelligence Industry in 2022" released by the Beijing Municipal Bureau of Economy and Information Technology on February 13, it is clearly proposed to support leading enterprises to create a large model that is benchmarked against ChatGPT, and strive to build an open source framework and an application ecology of a general large model .
The breakthrough of AI technology from pre-training to large model is the basis for the success of ChatGPT.
In traditional AI model training, the cost of labeling has always been an obstacle that hinders AI algorithms from being pushed to larger data sets; however, pre-training technology can train a large-scale deep learning model without relying on data labeling.
In June 2020, OpenAI released a large-scale language model GPT-3 with a parameter scale of 175 billion. ChatGPT was created based on an improved version of this model.
The ability, potential and rich application scenarios demonstrated by the large model have already inspired many companies and research institutions to join this competition.
In January 2021, Google Brain launched a super model with a scale of 1.6 trillion parameters.
In November 2020, Huawei launched the Chinese pre-training large model project, and in April 2021, the Pangu large model with a scale of 100 billion parameters was released.
In June 2021, the Enlightenment 2.0 large model, led by Beijing Zhiyuan Artificial Intelligence Research Institute and jointly developed by several institutions, has a parameter scale of 1.75 trillion.
Alibaba DAMO Institute launched the Chinese multimodal pre-training model M6 project in early 2020. In October 2021, the parameter scale of M6 expanded to 10 trillion, becoming the world's largest AI pre-training model at that time.
Su Zhong believes that on this technological track driven by data, algorithms and computing power at the same time, the technical teams of China and the United States have formed a "two-horse carriage" development model, constantly refreshing the boundaries of the scale of artificial intelligence capabilities .
In the field of AIGC (artificial intelligence content generation technology) high-quality papers, the number of China and the United States is almost the same, and it is significantly ahead of other countries.
"Artificial intelligence can become a breakthrough for my country's independent innovation. Although there is a certain gap between my country's artificial intelligence industry and the United States, the gap is not large." Song Yu said, for example, in the face of the increasingly large computing power demanded by large models, Several supercomputing centers and specialized artificial intelligence computing power centers have been established in China.
"On the whole, my country's current computing power has already surpassed the demand for artificial intelligence, providing a guarantee for independent innovation."
Data show that in recent years, my country's artificial intelligence industry has grown faster than the global level.
From 2017 to 2021, the scale of my country's artificial intelligence industry has increased by 2.6 times, accounting for 16.8% of the world's total.
From 2012 to September 2022, there will be about 1 million papers published in the field of artificial intelligence worldwide, of which 28% are from China.
The proportion of high-level paper output in China will increase from 20.36% in 2012 to 50.71% in 2021.
In terms of the number of patents, the cumulative number of AI patents granted globally is 250,000, of which China accounts for 60%.
face the challenge
ChatGPT is one step ahead, can Chinese companies catch up?
"ChatGPT is far superior to domestic large-scale models in terms of logic and completeness of answers, and ChatGPT is also ahead in response speed." Zheng Rong believes that this gap has something to do with data quality.
The evolution of AI needs to feed a large amount of data, and the quality of Chinese data on the global Internet is still significantly lower than that of English data.
"We have to find a way to do a good job of data complementarity between Chinese and English languages." Zheng Rong said.
The large model used by ChatGPT is extremely computationally resource-intensive, implying a large investment.
In 2019, Microsoft invested $1 billion in OpenAI, and another round of investment in 2021 has an undisclosed amount.
In January this year, Microsoft announced that it would invest "multi-year, billions of dollars" in OpenAI. The two companies did not disclose the specific amount, but according to US media, Microsoft will invest 10 billion US dollars in OpenAI.
Faced with such a threshold, Hu Kun suggested to integrate resources through policies.
"Most enterprises cannot afford the huge consumption of computing power and other resources caused by the development of artificial intelligence. It is recommended that the country integrate relevant resources at the policy level, such as establishing a unified computing platform, ending the state of fragmented and redundant construction of computing resources in various enterprises, and promoting all parties to reach an agreement. Some kind of agreement, joint construction, mutual benefit." Hu Kun said.
On the other hand, the threshold is also an opportunity to improve and catch up.
ChatGPT requires a lot of computing power and energy input, and the cost is high. Large-scale commercial use must improve technology and reduce costs.
For example, AI chips not only need to ensure high performance, but also achieve low power consumption under high computing power. In this regard, domestic chip manufacturers can make a difference.
Platform-based technology companies can also give full play to their comprehensive advantages, cooperate with research institutions, universities, etc. to form joint innovation entities, and accelerate the product transformation of the new generation of AI technology.
"In terms of system and mechanism, we must have enough patience for enterprises that carry out independent innovation." Song Yu said that at the national level, it is recommended to further improve the assessment mechanism for talents and institutions, and to have long-term goals and enough patience to allow enterprises to independently innovate. Innovation requires a trial and error tolerance mechanism.
The artificial intelligence content generation technology represented by ChatGPT is currently facing many ethical and security issues, such as discrimination, fraud, plagiarism, rumors, forged identities, etc., which need to be regulated during development.
Many experts said that because ChatGPT plays an extremely important role in the Sino-US technological competition, it is recommended to adopt the regulatory principles of encouraging innovation, inclusiveness and prudence for related technologies and application fields.
At the same time, strengthen industry guidance, especially in scientific research, life services, commerce, consumption and other application fields where ethics and safety are relatively guaranteed, and the development should be actively encouraged.
It is expected that in the new round of AI industry competition detonated by ChatGPT, Chinese companies can make great achievements.
(Economic Daily reporter She Huimin)