Why wasn't ChatGPT born in China?

Whether it is ChatGPT, which will be launched on November 30, 2022, or the largest and most powerful version officially released by OpenAI recently. The multi-modal pre-training model GPT-4 has really made the general public feel the powerful power of AI artificial intelligence, announcing the arrival of the era of strong artificial intelligence. Amid this bustling excitement caused by artificial intelligence, we cannot help but think further about a question: Why, after so many years, is it still the United States and not China that leads the industrial technology revolution?

#In 2021, China topped the list with 27.6% of the number of papers published at the world's top AI conferences, followed by the EU and the UK with 19.0%. The United States ranked third with 16.9%. Regardless of the number of AI papers published or the number of patent applications, China ranks among the best in the world. It stands to reason that China’s research in the field of AI is both extensive and in-depth. Landmark AI applications like ChatGPT should be Just be born in China. Although natural language understanding and generation is a peak in the field of artificial intelligence research, if you want to climb and conquer it, you need continuous investment of high-quality resources, and research institutions such as Alibaba Damo Academy, Huawei, and Tsinghua University are also doing similar things. . Unfortunately, it was OpenAI from the United States that took the lead in releasing the powerful ChatGPT. Today we are trying to objectively analyze the reasons why we have not become the initiators of a new generation of technological revolution, and what we should do in the future to become leaders rather than followers of future technology.

Cause analysis

Maybe it’s because of the Internet in China in the past twenty years With the rapid development of the Internet, the market value of some domestic Internet giants can compete with American Internet companies internationally. This has given us many illusions, mistakenly believing that we can also compete with the United States in technology. However, Is this really true?

I believe you can often see in the news that in order to curb China’s development in recent years, the U.S. government has continuously wielded sanctions against China’s high-tech industry. For example, it imposes a technological blockade on the Chinese company Huawei and prohibits American companies from selling technology and equipment to Huawei on the grounds of national security. For a time, China's high-end chip industry was in dire straits and faced a crisis of technological stagnation. While we lament the shamelessness of the U.S. government, we also lament that we did not prepare the underlying technology research and system construction in advance. The ChatGPT and GPT4 released by OpenAI this time have taught us another lesson. More and more people realize that the technological gap between China and the United States goes beyond high-end chips, cutting-edge artificial intelligence technology, powerful computing power networks, and many other high-tech technologies. There is a big gap with the United States in all aspects. The slap of the US government to suppress the development of China's high-tech industry has finally completely awakened some technical people who still had illusions about the United States. Whether it is high-end chips or cutting-edge artificial intelligence and other core technologies, they cannot buy or obtain them. They can only rely on themselves. Come out one step at a time. Of course, this also strengthens our determination to achieve domestic substitution of all technology categories related to the United States.

Excessive focus on business models and neglect of underlying technology research

In the past 20 years, China’s Internet industry has been in a In the stage of rapid development, according to CNNIC (China Internet Network Information Center) report, as of December 2022, the number of Internet users in my country has reached 1.067 billion. As the number of people online on the Internet continues to rise, various business models continue to be verified and developed in this hot land. Business forms such as mobile payments, e-commerce logistics, takeaway taxis, and short videos have profoundly changed the daily life of ordinary people, bringing great convenience and entertainment to everyone, while also improving the efficiency of the entire society. But at the same time, major Internet companies seem to be immersed in the cycle of how to develop new business models and maintain existing business structures, and almost all their energy is focused on business involution. Therefore, the time and resources invested in underlying technology research will be relatively squeezed or even completely eliminated. Ultimately, the probability of being stuck by other people's technology will be greatly increased, while the possibility of producing disruptive innovative technologies will be greatly reduced.

Comparing the technological development routes of China and the United States, we can find an interesting phenomenon. In China, the development of technology mostly revolves around business. That is to say, when a business model is launched, if you want to support such a business scale, what kind of technology needs to be developed, then what kind of technology needs to be researched and developed, Promote the forward development and evolution of technology through business model innovation and business development.

But in the United States, it seems to be just the opposite. The development and progress of a technology often gives rise to new ones. business form. Take OpenAI, the company that launched ChatGPT, as an example. It was originally a non-profit technical organization that brought together all the big guys in the AI ​​field. Its original intention was to make artificial intelligence develop in a friendly direction. Its core purpose is to "achieve security." Artificial General Intelligence (AGI)” to benefit humanity. After continuous iterative upgrades, the GPT model finally came up with ChatGPT, which is expected to become the basic base of future artificial intelligence application technology. Therefore, there may be a large number of upper-layer applications built based on ChatGPT in the future. It is no exaggeration to say, ChatGPT or general large model may be the operating system in the field of artificial intelligence. It is immeasurable how many new business models and business forms will be born in the future, this technological revolution is a bottom-up disruptive innovation that will have a profound impact on the future industrial form.

And we focus too much on the innovation of the upper-layer application business model, thereby neglecting the research on the underlying technology. We seem to have gradually lost our way in the e-commerce promotions and short video entertainment again and again. The superficial excitement cannot conceal the lack of underlying core technology. In fact, just thinking about it makes me feel cold. At present, cutting-edge technologies such as Internet technology and artificial intelligence technology seem to be based on other people's technical bases. The mobile payments, e-commerce logistics, etc. that we were once proud of are just the upper-layer applications of these underlying technologies. If you are kicked out of your job, you really won't be able to play anymore. So the underlying technology is really important. It can be said that without the support of the underlying technology, all upper-level applications are just castles in the air.

Lack of long-termism

Perhaps the domestic competition pressure is too great, both individuals and companies are eager to Invest money today and see the results and benefits tomorrow. If there is no visible return in the project in the short term, it is actually difficult for the average company to make continuous resource investment, let alone basic technology research that cannot see a clear return cycle. The impetuous atmosphere, endless technological gimmicks, and superficial excitement will only make everyone keen to follow the trend, but not many people can really calm down and conduct long-term technical research. How many companies can have the courage and strength of Mr. Ma to invest 1 billion every year for 10 years, and finally invest in Alibaba Cloud.

Speaking of this, I don’t know if you still remember the technological terms such as blockchain, VR, and metaverse. They were greatly affected when they first appeared. attention and enthusiastic pursuit of capital. But after the craze has passed, how many companies are still continuing to iterate products and conduct in-depth optimization research on technology? It cannot be said that there are none at all, but there are definitely very few. Rather than pursuing high technology, they are pursuing tools on how to make money quickly. It doesn’t matter whether it’s VR, Metaverse or ChatGPT. To them, it’s just a technical name.

But in fact, artificial intelligence applications such as ChatGPT and GPT4 are not born out of the blue. They are gradually developed iteratively based on GPT, GPT2 and GPT3. Among them How much long-term efforts of top artificial intelligence experts and continuous investment of large amounts of money have been expended. OpenAI burned nearly 100 million US dollars in the three or four years before receiving investment from Microsoft to come up with a prototype based on the first generation model of GPT. At that time, it was still far from the effect of ChatGPT today. Basic technology research is not something that can be completed overnight. It requires continuous trial and error, a unique vision of future technology development trends, and even more persistence in long-termism. However, we are gratified that Alibaba and Huawei have invested more than 100 billion in technology research every year. We believe that there will be new technological breakthroughs in the near future.

Lack of technical system support

# is like high-end chip manufacturing, which involves chip design, chip manufacturing and packaging testing It's all a complex set of processes that are indispensable. Similarly, the general large model is also a complex system engineering, and the training of the general large model is inseparable from the support of massive high-quality training data, excellent deep learning algorithms, and powerful computing resources. The success of ChatGPT is due to high-quality annotation data. In order to obtain high-quality training data, OpenAI hired more than 40 contractors for data annotation, which is almost unimaginable in China. In addition, based on the average number of independent visitors of ChatGPT in January 2023, which is 13 million, the corresponding chip requirement is more than 30,000 NVIDIA A100 GPUs. Without huge computing power support, it is difficult to train high-quality general-purpose large models.

In other words, the training of general large models requires a complete artificial intelligence technology system to support it, otherwise even with advanced algorithms, But how to obtain high-quality training data and where to find huge training computing resources are all problems that must be solved before training general large models, otherwiseWithout high-quality data and computing power support, it is difficult for a clever woman to make a meal without rice.

How to break the situation

General large models are an important development direction of artificial intelligence in the future, and we must increase resources Only by investing your efforts can you have a chance to occupy a place in the competition of cutting-edge artificial intelligence technology in the future. So what should we do to break the situation? I think we can start from the following aspects.

In terms of data

Although the current Chinese online Internet data is considered massive, the data quality is generally not high, and the data island problem is serious. If we want to train a general large model based on Chinese data, we must manage Chinese online data to improve the quality of Chinese online data, solve the problem of data islands, and promote data sharing and interconnection.

In terms of computing power construction​

Coordinate the computing power centers in various regions in conjunction with the National East-West Computing Project , forming a national general computing power infrastructure, accelerating the construction of a new national-level integrated computing power network system, and providing a steady stream of powerful computing power support for general large model training. Only in this way can we meet the computing power requirements required for large model training, and at the same time prepare computing power resources for the arrival of the era of strong artificial intelligence.

Consciousness aspect

And other people’s Of course we must try our best to make up for the technical gap. However, what we need to pay attention to is that we are making progress, and others are also making progress. Therefore, we cannot just catch up with others. We must have a longer-term vision, have accurate judgments about the future, and carry out advance technical verification and industrial layout to prepare for the next time. Prepare for the technological revolution and truly become a leader rather than a follower of future technological development.

Summary

The development and progress of science and technology cannot be achieved overnight. It may require the tireless pursuit of generations of technicians. Although we have made great progress in many fields, we must also be clearly aware that we still have a lot of gaps to make up in the fields of computers, Internet technology, and artificial intelligence technology. Therefore, less exposure of Internet celebrities and celebrities, more diligent and pragmatic publicity by scientific and technological workers, and efforts to create a social atmosphere where technology is first and technology first are the social environment foundation for the vigorous development of the technology industry in the future. I believe that if we can be less impetuous and more pure, maybe the next technological revolution will be born in China. Finally, I would like to share with you something that Mr. Xiama once said. I think it is very helpful for us to think about future technological development.

I have always felt that there is something wrong with the sentence "filling in the blanks", either because Europe and the United States are advanced, or we are To be filled. In fact, today we should not be integrating with something, adapting to which country's standards, or filling any gaps. Today we have to think about how to integrate with the future, how to adapt to future standards, and how to fill the gaps in the future. We must understand what the future is. How, and what kind of system we want to build, and then look at what others do. If we always repeat other people's words and discuss the themes set by others, we will not only lose the present, but also miss the future.

The above is the detailed content of Why wasn't ChatGPT born in China?. For more information, please follow other related articles on the PHP Chinese website!