The popularity of ChatGPT has made the red-clothed god leaders excited, and the following is from the red-clothed godmaster's point of view:
Chinese companies cannot stand in "pragmatism" and fight in silos. Learning from the cooperation model of OpenAI and Microsoft, strengthening cooperation in the integration of "production, education and research", open source and ecological construction is the key.
Regarding whether the large model can cause commercial changes in the search engine industry, Zhou Hongyi told China Securities News: "ChatGPT as a chatbot and search engine lack imagination. Like the basic resources of electricity, it may subvert the entire digital industry and reshape more industry models, which is a real opportunity. ”
The country can catch up in 2-3 years
Zhou Hongyi spoke highly of the changes brought by ChatGPT to the digital industry. He said that the AI big model may have brought the artificial intelligence industry to the starting point of general artificial intelligence, and APP, software, website applications and even the whole industry are worth reshaping with the blessing of the big model capability.
Zhou cautioned against "mything" OpenAI's capabilities. "In fact, ChatGPT is made on the basis of predecessor algorithms and data open source, and is the 'aggregator' of artificial intelligence research in the world, without university laboratories, research institutions and commercial companies industrialization means, ChatGPT cannot advance by leaps and bounds."
In Zhou Hongyi's view, China's development of AI large models has a good prospect. He believes that the domestic level can catch up in 2-3 years. Many Internet giants have computing power and cloud service capabilities, China has a demographic dividend in data knowledge annotation, and artificial intelligence laboratories and universities such as KLCII and Pujiang have R&D strength. However, further efforts are needed in China in terms of open source cooperation and the amount of knowledge Chinese corpus.
"The amount of knowledge of ChatGPT training data is very large, the amount of knowledge on English web pages is about 10 times that of Chinese, and ChatGPT has also received a lot of English book knowledge training. A large number of our Chinese books are not digitized, so the domestic public knowledge collection needs to solve the problem of digitization and copyright. Internet knowledge is also important. Due to the needs of industry competition, China's Internet has been divided into information islands by APP. Zhou Hongyi said.
Build an industry ecology together
Zhou Hongyi said frankly that domestic companies may pursue "pragmatism" too much, and closely integrate artificial intelligence with their own business to improve their own search experience, advertising click-through rate, image quality, etc.
"We should learn from the cooperation model between OpenAI Research and Microsoft." Zhou Hongyi said that ecology must be cultivated. To make China's "ChatGPT", it is necessary to cooperate with research institutions and enterprises, and it is necessary to fully combine the research results of scientific research institutions and university scientists with industrialization, engineering, productization, and commercialization, and also form positive feedback through user annotation to promote the improvement of large models.
He also mentioned that not every head company needs to develop a large model, and most enterprises can use the AI capabilities provided by the large model to empower their own industry application scenarios. For example, OpenAI is looking for partners to develop application software that can give many industries AI wings and make AI a real productivity tool to improve industry efficiency.
Play the role of "justice assistant"
Zhou Hongyi's views on artificial intelligence security.
"Traditional security questions don't bother me, ChatGPT security doesn't bother me." He said the security concerns posed by ChatGPT could have changed from technical to socio-ethical, even unpredictable. For example, he said, ChatGPT could be a tool for cyber hackers. It has a lot of programming knowledge and can be a cyber hacker. ChatGPT is now being used to write phishing emails and attack code. In fact, it reduces the cost and difficulty of the attack, and will become a helper for hackers.
"360 brought together cybersecurity big data earlier. After the big data, we are also exploring whether we can use ChatGPT to automatically detect and automatically detect attacks, and establish a knowledge base to solve new cybersecurity challenges, so that ChatGPT can play the role of 'justice assistant'." Zhou Hongyi said.
It is expected that Chinese companies will be able to catch up with ChatGPT in about two years
In the field of ChatGPT, the gap between domestic and foreign countries is relatively small, and it is expected to catch up in about two years. Zhou Hongyi said that there are many Internet giants in China, sufficient computing power, and there is also a layout for university research in artificial intelligence. In addition, papers and model algorithms in the field of ChatGPT are open and open-source, which is conducive to catching up. The first disadvantage is the ability to be original; Second, the new artificial intelligence model algorithm and application require scientific research institutions and industrial companies to form a good ecology of industry-university-research cooperation, that is, similar to the ecological cooperation that ChatGPT has carried out with about 100 vertical companies; Third, the lack of Chinese corpus requires the mainland to strengthen knowledge digitization and copyright granting, and further solve the current problems of domestic apps forming data islands and Internet full data sources.
Zhou Hongyi: ChatGPT will soon become conscious
"How many of you think ChatGPT has the intelligence of someone?"
"I think ChatGPT is the singularity of general artificial intelligence, and general artificial intelligence also has an implicit inflection point, which is to become strong artificial intelligence." Zhou Hongyi said.
ChatGPT through hundreds of billions of system parameters, to achieve some of the original designers completely expected, and unexplained, "This is like the process from ape to man must have a 'tipping point', when the number of neural network links in the brain, the number of brain cells, etc. to a certain extent, human intelligence and animals have produced a qualitative leap." ”
Zhou Hongyi believes that a major feature of ChatGPT compared to other artificial intelligence is that it has the ability to "make up stories", "Many dialogue AIs do not achieve full understanding and decoding of knowledge, because it uses keyword matching (patterns), if you ask it a question, it extracts the keyword search answer, if the answer cannot be searched, it cannot answer or give an unrelated answer, but ChatGPT will make up an answer even if there is no answer, which is similar to human characteristics. Because only people can describe what does not exist. ”
It should be noted that compared with the progress of AI technology, the current progress of human brain science research is still slow, and there is a view that the production of general artificial intelligence must first crack the human brain, which is quite difficult, so AI to generate its own consciousness can not be achieved in the short term.
In this regard, Zhou Hongyi quoted Kant's famous saying: it is impossible to discuss and understand what cannot be said, "It cannot be said that only dancing wings like a bird is called a flying machine, the wings of an airplane will not move, but it uses another scientific principle and also flies into the sky, so ChatGPT as a 'silicon-based creature' has evolved, and I think it will soon produce self-awareness." ”
Zhou Hongyi believes that ChatGPT has three personalities, like a doctoral student in terms of knowledge, with a wide range of knowledge but not very good at learning and using; In terms of emotional intelligence, he is like a slippery middle-aged person, and his speech and answers are dripping; But the real ability is like a child, and its training in questions and answers, including cognition, understanding, and reasoning for problems, has reached the level of a child.
Security has changed from a technical issue to an ethical one
Regarding what new security problems the emergence of ChatGPT will bring, Zhou Hongyi said frankly, "Security issues were originally difficult for me, but now it is difficult to say. Because I feel that it has gone from a technical question to a socio-ethical issue, even an unknowable question of the future of humanity. ”
In his opinion, ChatGPT has a lot of programming knowledge, knowledge about the network security of vulnerabilities, and can also program, and can become a network hacker or a powerful tool for evil. Even if it is not possible to directly ask ChatGPT to do evil programmatically, many online cases have shared how to "trick" ChatGPT and use it to write phishing emails and attack code, so the emergence of ChatGPT actually reduces the difficulty and cost of cyber attacks.
Zhou Hongyi said that he is also exploring how to "attack his human shield with his human spear", using a large ChatGPT-like model to detect corresponding attacks, so that ChatGPT can also become a role of justice assistant.
According to public information, OpenAI, the creator of ChatGPT, has been cautious about letting ChatGPT see web pages, citing fear that it will be "contaminated" by good and bad web materials. Zhou Hongyi said that if the corresponding restrictions are relaxed, let ChatGPT go to the web page to learn, what can be learned is unknown, if such a large model supports multimodal learning capabilities, to watch videos, through the Internet of Things camera to observe humans, just like MOSS in "The Wandering Earth 2", its learning ability will advance by leaps and bounds, but this also means that it has the ability to control the physical world, so the future development of ChatGPT will produce more insecurity, such as when it has strong capabilities Will it be misled to do "stupid things"?
"We security companies exist to study security issues and come up with solutions as we grow. The security of artificial intelligence originally only involved program errors, vulnerabilities, etc., but ChatGPT has gone beyond the scope of the original artificial intelligence, and from the current situation, not developing is the biggest trouble. This technology is related to the future of the country, domestic companies should open source, cooperation, exchange, create an ecology, fully promote the development of ChatGPT technology, in the field of ChatGPT, China has the advantages of engineer dividend, is expected to catch up to the world's leading level in 2-3 years, we should give full support to the exploration of ChatGPT technology. ”
At present, the biggest security problem of ChatGPT is whether it can be controlled after generating self-awareness
The emergence of ChatGPT represents an important milestone in artificial intelligence and will bring about an industrial revolution driven by general artificial intelligence.
Zhou Hongyi said that at present, Chinese Internet companies are too pragmatic when exploring ChatGPT, and more consider the combination with their own business, ignoring the more possibilities that ChatGPT will bring. 360 is continuing to track related technologies and explore the role of using ChatGPT to play the role of "justice assistant". He also mentioned that in the field of ChatGPT, China has great advantages and hopes to reach the world's leading level in 2-3 years, and should give full support to the exploration of ChatGPT technology.
He believes that the biggest security problem of ChatGPT at present is whether it can be controlled after generating self-awareness.