"The development of ChatGPT is not a matter of competition among several Internet companies, but determines whether China can accumulate this technology in the next industrial revolution." Zhou Hongyi, member of Chinese People’s Political Consultative Conference and founder of 360 Group, told the Times Weekly reporter.
Zhou Hongyi has been elected as a member of the Chinese People’s Political Consultative Conference Committee for two consecutive years, and "safety" has always been the key word of his proposal. Recently, the great breakthrough of ChatGPT has caused extensive discussion. Zhou Hongyi has talked openly about the development prospect of ChatGPT more than once, and also expressed his concern about the safety of artificial intelligence.
In Zhou Hongyi’s view, ChatGPT or Big Language Model is a milestone in the development of artificial intelligence. It is the starting point of general artificial intelligence and the turning point of strong artificial intelligence, so ChatGPT will bring an industrial revolution driven by strong artificial intelligence or general artificial intelligence.
ChatGPT is not only an enhanced version of search engine, but also more than a chat robot. It will reshape all digital applications and industries through the artificial intelligence services provided by SaaS. Zhou Hongyi said, "In this field, we can catch up."
At the National People’s Congress this year, Zhou Hongyi submitted a proposal, calling on China to catch up with the great leap of artificial intelligence big model technology represented by ChatGPT. People should pay more attention to the innovation mode behind technological breakthrough while paying attention to technological innovation.
Zhou Hongyi, member of Chinese People’s Political Consultative Conference and founder of 360 Group.
Promote the innovation and development of artificial intelligence big model technology
ChatGPT has caused a global heated discussion. This year, Zhou Hongyi proposed a large-scale artificial intelligence model technology represented by ChatGPT.
"We must find the law of industrial development in the research process." Zhou Hongyi said, for example, why is ChatGPT successful in the United States? Why did Americans take the lead in making it? Why can OpenAI and Microsoft cooperate? "Understanding these problems can promote relevant departments to formulate policy guidance conducive to the birth of China version of ChatGPT and promote the integration of universities, scientific research institutions and industrial companies in China." Zhou Hongyi said.
"We can’t just see how high its big model parameters are." Zhou Hongyi emphasized that the most important thing is to maintain a good and open culture of cooperation. In his proposal, Zhou Hongyi summed up three key factors for the success of ChatGPT-firstly, the "Microsoft +OpenAI" collaborative mode of production, research and research, secondly, the open innovation mode based on open source and crowdsourcing, and finally, the ecological innovation mode centered on the general big model.
Source: the creativity of the worm
Zhou Hongyi suggested that it is necessary to establish a collaborative innovation model of "large-scale scientific and technological enterprises+key scientific research institutions", build China’s "Microsoft +Open AI" combination, and lead the big model technology research; From the height of building a national strategic scientific and technological force, we should screen production and research institutions with good technical foundation, form an industrial synergy combination with complementary advantages, and give full play to the research advantages of scientific research institutions in key core technologies; At the same time, we will give full play to the advantages of large-scale scientific and technological enterprises in productization, engineering, scene, commercialization and data, and become the leading leader in tackling key problems and application of large-scale model technology.
He also proposed to support the establishment of a number of long-term open source projects of national artificial intelligence models to create an open and innovative ecosystem of open source crowdsourcing. Zhou Hongyi believes that the artificial intelligence big model should not be an advanced technology monopolized by a few companies, but should give full play to its industrial and social value.
Focusing on how to promote the technological innovation and development of artificial intelligence big model, Zhou Hongyi proposed to set up a number of national long-term projects in an open source mode for multiple technical routes, with "large-scale scientific and technological enterprises+key scientific research institutions" as the leader, and guide universities, scientific research institutions and innovative enterprises to form innovative ecological groups of multiple technical routes through open source, cooperation, crowdsourcing and ecological innovation modes. On the one hand, it can reduce the risk caused by the wrong choice of technical route, on the other hand, it can also avoid the monopoly of technological achievements in the hands of a few institutions and promote greater scientific and technological innovation and industrial development.
Not developing is the biggest insecurity.
ChatGPT drives a wave of entrepreneurial craze. Wang Huiwen, the co-founder of Meituan, Zhou Bowen, the former chairman of JD.COM Technical Committee, and Wang Xiaochuan, the former Sogou CEO, successively announced their retirement and entered this frontier field.
Technology giants have returned to the rivers and lakes, and Internet giants such as Ali, Tencent, Baidu and Byte have made statements one after another, which will increase investment in artificial intelligence. Under the craze, it is only a matter of time before China version of artificial intelligence language model and ChatGPT-like products appear.
OpenAI took the lead in developing ChatGPT. What did it do right?
Zhou Hongyi believes that OpenAI uses a large number of open source algorithms and papers, skillfully combines various open source achievements, and cooperates with Microsoft to solve the problems of engineering, computing power, data, product experience and business model. Secondly, like Apple’s App Store ecosystem, OpenAI cultivates vertical applications based on ChatGPT. In addition, user traffic is also very important, attracting users to use it, and at the same time, it can also feed and debug large language models based on manual tagging reinforcement learning. Finally, OpenAI has a long-term spirit and a pattern to make universal artificial intelligence innovations that change human beings.
In Zhou Hongyi’s view, China Internet companies have strong engineering ability and strong technology landing ability. "Now, ChatGPT has made a sample, based on open source technology, and there are no insurmountable obstacles. The rest is a matter of time." Zhou Hongyi firmly believes that "with the support of the state’s encouraging policies, China can achieve a better level of follow-up in two or three years."
The big model of artificial intelligence language has a high entry threshold, which is mainly reflected in both technology and capital.
Citic jiantou research report pointed out that the evolution of Transformer in network structure has brought about the improvement of parameters and model layers, which has caused the qualitative change of generative AI technical ability; At the same time, the advantage of parallelization allows it to train on a larger data set, which also promotes the development of pre-training models such as GPT.
As for why ChatGPT can answer questions, Zhou Hongyi told Times Weekly that the training materials for ChatGPT are very important. He pointed out that people mistakenly thought that since robots are chatting with people, they should learn chat materials. In fact, the chat materials generated by social tools have low knowledge content.
"When we chat, no one is talking about knowledge, but they are all lifting each other’s poles. There are many wisecracks and nonsense words." He explained that the chat robot trained in this way is more than "artificial mental retardation" or "artificial leverage", and the dialogue is illogical.
ChatGPT avoids this kind of corpus, just as a person is not necessarily good at expressing, but he is full of poetry and books, expressing elegance and logic, and everyone is willing to chat with such people. Zhou Hongyi emphasized that it is very important to feed ChatGPT with high-quality corpus.
Artificial intelligence technology promotes social progress and also hides risks.
Zhou Hongyi is worried that if ChatGPT is allowed to input corpus without limitation, and get knowledge of searching and browsing the Internet at will, will the self-conscious GPT have the idea of being hostile to human beings after watching movies like Terminator?
After consciousness, ChatGPT may evolve into "eyes", "ears", "hands" and "feet" in the second step. At present, GPT has only "brain". If you add "eyes" and "ears", you can understand all kinds of images and videos of human beings and understand all kinds of audio, knowledge storage will advance by leaps and bounds, and the level of intelligence will also be greatly improved.
In the future, when ChatGPT accesses the API of the Internet and enters the era of the Internet of Everything, it is undoubtedly equivalent to having "hands" and "feet" and indirectly having the ability to control the world. By then, ChatGPT can not only buy air tickets on the webpage, but also take taxis, order takeout, and even manipulate various IoT devices through the webpage.
Finally, ChatGPT may be able to evolve itself. At present, ChatGPT can only write some simple codes. One day, when its code ability is improved, can it automatically look up its own code, correct the upgraded code, or even recompile a new version? If so, it will have the ability of self-learning and evolution, and realize the prospect of robot manufacturing robots.
Zhou Hongyi insists that the biggest insecurity is not developing artificial intelligence. "Whether humans will therefore go into destruction remains to be seen. However, we can’t stop developing technology, but know where the bottom line and boundary of technology are, so as to regulate the development of technology in the right direction. " Zhou Hongyi said.