Zhou Hongyi: ChatGPT may become self-aware and threaten humanity within two or three years

"In terms of knowledge, it is like a doctoral student, who is knowledgeable but not yet active; In terms of emotional intelligence, it is like a slippery middle-aged person, saying that everything is dripping and must not offend the user. However, it is actually really capable like a small child who needs to be trained in questions and answers to learn skills. No matter how smart children are, they must be trained to solve the Olympiad problems they have not learned. ” 

Photo by 360 founder Zhou Hongyi and surging news reporter Fan Jialai

On February 23, 360 founder Zhou Hongyi mentioned the popular chatbot ChatGPT again in an interview with The Paper and other media, and shared the "triple personality" of ChatGPT in his heart. 

The ChatGPT craze is sweeping the world. This AI chatbot software, produced by OpenAI, took only two months and reached 100 million monthly active users, making it the fastest growing consumer app in history. 

Previously, Zhou Hongyi had repeatedly spoken about his observations on ChatGPT in public. In his opinion, ChatGPT already has or is close to having human intelligence and is at a tipping point on the road to "general artificial intelligence". "We think it's a singularity of general AI, and there is also an implicit inflection point in GM, which could be a strong AI." 

ChatGPT has humanoid characteristics: the ability to make up stories

At the interview scene, Zhou Hongyi did not hide his excitement about ChatGPT at all.

In his view, ChatGPT is a powerful reengineering tool, not just empowering specific industries. "In the past, big data was like 'oil', but big data cannot be used directly, ChatGPT is like a power plant, burning oil, after pouring oil into it, it can generate electricity, and use cloud services to connect it to thousands of households and industries, and it will have a huge impact on the entire industry." Therefore, this directly determines the gap between China's technology industry and that of other countries in the next decade: whether it will form a lag or catch up. ” 

ChatGPT will use hundreds of billions of system parameters to achieve capabilities that the original designer could not have imagined. "It's like there must be a tipping point in the process from ape to man, when the number of neural network links in the brain, the number of brain cells, etc. is to a certain extent, and human intelligence has made a qualitative leap compared with animals." 

Zhou Hongyi said that the biggest difference between ChatGPT and other artificial intelligence, and its human-like feature, is that the original artificial intelligence does not achieve all the understanding and coding of knowledge, and the essence is to match existing answers according to keywords, but ChatGPT does not have a "fixed answer": "I asked it a question at that time, why did Zhao Kuangyin recruit Song Jiang and kill Yue Fei, this nonsensical question, it also answered it carefully, and can 'make up' stories by itself, which is where its real intelligence lies." ” 

"ChatGPT may become self-aware and threaten humanity within 2-3 years." At the interview scene, Zhou Hongyi "spoke amazingly". He mentioned that apps, software, websites, and applications in any industry are worth reshaping if they are supported by the ability of GPT. "Don't think of it as a media tool, a search engine, a chatbot, it's a bit too childish." 

He said that ChatGPT is different from the training of traditional artificial intelligence and needs to go through multi-dimensional training. The first is to swallow dates, read millions of books without asking for solutions, acquire human knowledge, but still do not have the ability to solve problems. Secondly, professional AI trainers are needed to cultivate its ability to answer questions, which has also spawned a large number of new employment opportunities.

The biggest safety issue is whether self-awareness can be controlled

As an expert on cybersecurity issues, Zhou Hongyi also shared the security issues of ChatGPT in his mind. 

In his view, the security of ChatGPT has changed from a technical issue to a social ethical issue. With a lot of knowledge about programming and cybersecurity breaches, ChatGPT can be a powerful tool for cybercrime. There are many hackers who have demonstrated using ChaGPT to write phishing emails and attack code, so in some ways ChatGPT has also become a "helper" for hackers. 

"Domestic Internet companies are generally too pragmatic, and they are closely integrated with artificial intelligence and their own business, but they did not expect more possibilities of ChatGPT, and 360 is still tracking and studying these technologies, hoping to catch up." Zhou Hongyi said that 360 is exploring "attacking its human shield with its human spear", using ChatGPT to play the role of "justice assistant", establishing a knowledge base for automatic detection and discovery of attacks, and using artificial intelligence to automate the processing of cybersecurity data. 

"At present, the biggest security issue with ChatGPT is whether it can be controlled after generating self-awareness." Zhou Hongyi said that if the restrictions are relaxed and ChatGPT is allowed to learn in the web page, it may lead to stronger control over the physical world, and there will be more insecurity in the future development, and it may also be induced to do "stupid things", and whether it can control strong artificial intelligence is the "boundary point" of whether ChatGPT is completely integrated into human life. 

As for the attention of the outside world, ChatGPT is likely to cause ordinary people to lose their jobs? In Zhou Hongyi's view, it will not be in the short term, "At present, its positioning is still a human tool, which is also where its creativity lies." At the moment its self-awareness has not yet appeared, so it needs to use your imagination and creativity. ” 

He is also looking forward to China's development of its own ChatGPT. In his view, the gap between China and Microsoft and Google in the field of artificial intelligence is not as big as imagined. In the field of ChatGPT, China has great advantages and is expected to reach the world's leading level in 2-3 years, and should give full support to the exploration of ChatGPT technology.

Specifically, Zhou Hongyi said that China's Internet companies have certain advantages in quantity and computing power, and also have strong scientific research and university reserve strength, and must combine scientific research institutions, schools, scientists and other academic circles with industries, engineering and products, and fully mobilize and coordinate resources.

The engineer dividend is a major advantage for China, but it also faces the disadvantage of a lack of open source tradition. In the research and development of ChatGPT, many open source models are used, without the previous open source there is no accumulation on the shoulders of giants, each company has to grope from the dark, which is not conducive to the speed of the country's entire industrial development, "such a bullish technology, it really should not be monopolized in the hands of one or two companies." ” 

Zhou Hongyi believes that solving original problems requires the efforts of scientific research institutions, and scientific research institutions also need to cooperate with industrial companies, OpenAI's cooperation with Microsoft has created an example of cooperation between industrial companies and startups, "I think the country should still fully support the development of the ChatGPT industry, open source, cooperation, communication, and create an ecology." ” 

He hopes that more domestic enterprises will participate in the construction of the AI ecosystem and form open market competition, so that there will be progress. "ChatGPT has put artificial intelligence on many industry applications." Zhou Hongyi said, "ChatGPT alone cannot incite an industrial revolution, but if all walks of life are empowered by artificial intelligence in the future, it will bring huge changes." ” 

Previously, on the afternoon of February 9, Zhou Hongyi said in a conversation with Sohu Zhang Chaoyang that although ChatGPT is not perfect now, to see its unlimited potential, there will be many scenarios in the future, not just auxiliary search engines to search for answers. "That's why all companies are scrambling to get on the bus, and if they don't, they risk being eliminated."

 "I think ChatGPT is really not a tool, and its full understanding of human natural language may represent the beginning of a real revolution in the history of artificial intelligence." Talking about the birth of ChatGPT, Zhou Hongyi said that ChatGPT is a large language model with hundreds of billions of parameters, which is equivalent to understanding and training a lot of human knowledge on the Internet, simulating the storage of knowledge by the human brain, so that answering questions is completely like a mature, slippery person, rather than mechanically giving search answers.