ChatGPT set off the craziest round of market value surge in the history of AI stocks.
Since the Spring Festival, ChatGPT concept stocks have started to run wild. In just half a month, the value of ChatGPT concept stocks such as Haitian Ruisheng and Cape Cloud has increased by nearly 140 billion.
Such an explosive effect is due to the huge potential of commercialization shown by ChatGPT. You know, before this, whether it is Baidu, which has invested more than 100 billion in AI in ten years, or the AI four tigers trapped in hardware, they are all repeating the story of the difficulty of commercialization of AI.
The emergence of ChatGPT has turned AI from a productivity enabler directly into a tool for creating productivity. With the launch of the subscription model, ChatGPT has become the first consumer application that is directly monetized with AI technology at its core.
This paper holds the following core ideas:
1. ChatGPT is a beneficiary of AI technology iteration. In the past, due to limited technical capabilities, AI only had functions such as analysis, judgment, and prediction. Now that the technology paradigm has been upgraded, AI has the ability to create content, and its scope of application and applicable scenarios have been greatly extended, laying the foundation for the popularization of ChatGPT on the consumer side.
2. ChatGPT is expected to establish a sustainable business model. In the past, AI was used as a tool to improve efficiency, and it was difficult to directly generate revenue for the company. But ChatGPT can produce content by itself, which is essentially a production tool. From the past, production tools are often more likely to extend new business models and even change the business order.
3. The value of ChatGPT is similar to the generator of the second industrial revolution. In the second industrial revolution, electricity was not used on a large scale for 30 years after it was discovered. It was not until the generator converted electricity into mechanical energy that the electricity revolution was detonated. Today, ChatGPT is becoming a "generator" in the field of AI - it is directly transforming AI into productivity, and is expected to accelerate the development of the AI industry.
／01／The "legacy" left by three ups and downs
Before ChatGPT appeared, AI experienced three ups and downs.
In 1964, a machine called STUDENT was able to prove math problems, setting off the first wave of AI. At that time, the AI community believed that according to such a development speed, AI could replace humans. But the winter soon came, and in the early 1970s, almost all institutions stopped appropriating AI.
Insufficient hardware performance stalls AI. At the time, the limited memory and processing speed of computers were not enough to solve any practical problems, and AI was stuck in theory. Although this upsurge in AI is not practical, many world-class inventions have emerged in terms of algorithms, including the prototype of enhanced learning (the core idea of Google’s AlphaGo algorithm), perceptrons (the basis of deep learning models), etc., providing a basis for subsequent AI research base.
In 1980, Japan spent $850 million to create a computer with supercomputing power and human intelligence. Computer performance has improved dramatically, allowing AI to create expert systems. Expert systems can simulate the knowledge and experience of human experts to solve problems in specific domains. If it plays a role in the production of enterprises, the expert system jointly developed by DEC and Carnegie University can use AI to select the most suitable system components for technicians and assemble minicomputers.
Practical application drives a new upsurge. In 1988, the amount of investment in AI was three times that of 1984. However, problems soon appeared in the expert system. It can only be applied in a very narrow field, and the development and maintenance costs are high. It is not economical for enterprises to use it commercially. The expert system soon declined, and the winter came here. Although the second upsurge stopped at commercialization, it pushed artificial intelligence from theory to application.
The Internet era has brought the third upsurge to AI. The data explosion and the maturity of big data technology have brought AI into the era of deep learning. Deep learning enables AI to analyze and distinguish the feeding based on big data. The most well-known case is that AlphaGo combined the chess records learned by human experts in the game, and performed reinforcement learning in playing chess with itself and finally defeated Li Shishi.
AlphaGo has pushed AI to a new level. In 2016, when AlphaGo was born, the amount of financing for AI companies in my country alone doubled directly, and the amount of financing exceeded 100 billion a year. But looking back, most of the investment was "in vain". The four little dragons of AI have collectively fallen into the quagmire of high investment and low income.
The difficulty of commercialization is still the main problem of AI development at this stage. As an efficient tool, AI is difficult to directly generate revenue for the company, and its value is more likely to be embedded in products and packaged for sale. As a result, many companies have turned into a hardware company, such as a Four Tigers company, whose hardware revenue accounted for more than 70%, and the proportion of hardware revenue continued to expand.
Although the problem of commercialization has not yet been solved, deep learning has greatly optimized the accuracy, computing efficiency, generalization and other indicators of the AI model, and established the technical route of "big model + small scene model", laying the foundation for the emergence of ChatGPT. Base.
Looking back at the history of AI development, except for the first upsurge that stopped at hardware performance, the last two upsurges were essentially troubled by the difficulty of commercialization. So, can ChatGPT standing on the shoulders of previous waves end the AI winter?
／02／The difference between missiles and bows
In the eyes of the international academic community, the emergence of ChatGPT is an epoch-making product. The difference between it and the common AI before is almost the difference between missiles and bows. This difference is mainly reflected in the fact that ChatGPT makes AI directly a tool for creating productivity from a productivity enabler.
Before the emergence of ChatGPT, all AIs belonged to decision-making AI, that is, analysis, judgment, and prediction were performed based on existing data, and were mainly used in auxiliary decision-making of recommendation systems and risk control systems, such as information flow recommendations, automatic driving, etc.
In essence, analytical AI cannot be counted directly as productivity, but rather as an enabler to improve productivity. For example, analytical AI helps e-commerce companies deeply explore the relationship between users and items, and accurately push products and stores to users, driving the increase in e-commerce transaction volume. The role of a supporter makes the value of AI subordinate to various industries, and its commercial value as an industry alone is limited. Therefore, in the development of AI, commercialization has repeatedly been difficult to implement.
ChatGPT is a representative product of generative AI. Generative AI can be directly used as a production tool. Creation is the core of generative AI. Through algorithmic learning, not only the analysis and judgment of traditional AI can be realized, but also the creative functions beyond the capabilities of traditional AI can be realized to generate knowledge and creativity. Content, such as outputting text answers, game code, and so on.
The innovation of production tools often leads to new business models. So it can be seen that ChatGPT is gradually opening up the commercialization of AI. To some extent, ChatGPT is the first consumer application that is directly monetized with AI technology as the core. At present, ChatGPT has launched a membership subscription service. Each user has a limited number of free answers. Only users who have signed up for membership can enjoy unlimited times. Serve.
From the current point of view, the emergence of ChatGPT is more the result of upgrading the technical paradigm. For a long time, artificial intelligence has been dominated by small models. Small models can be understood as special-purpose model optimization algorithms and precision to solve vertical scene problems.
For example, a smart speaker uses a small model. Its system contains several Agents, one for chatting, one for poetry generation, one for code generation, one for marketing copywriting, and so on. If new functions need to be added, a new Agent needs to be trained.
This small model greatly limits the extensibility of AI technology, so that AI can only analyze and distinguish under a specific Agent, and the fragmentation of Agent makes it difficult to comprehensively generate content. Therefore, under the small model, if the user's problem exceeds the scope of the existing Agent, it will change from artificial intelligence to artificial mental retardation.
The emergence of large models has accelerated the popularization and application of AI. The big model can be understood as there is only one Agent behind it to solve all the problems of the users. The parameters of the large model are larger, and it can enable AI to comprehensively carry out machine learning of various modules, and finally generate new content comprehensively.
From the past, a lot of business value comes from the innovation of productivity. The same is true for ChatGPT, whose potential commercial value is largely due to the improvement of productivity.
／03 ／ChatGPT, the "generator" of the era of computing power
Production tools play a leading role in the means of production and will react against the changes and development of productive forces. In specific industries, the innovation of production tools often brings new industrial opportunities.
Taking the second industrial revolution as an example, the vigorous development of the electric power industry began with the birth of the generator. As early as 1831, British scientist Faraday discovered the phenomenon of electromagnetic induction. But in the next 30 years, electricity was not widely used.
Until the 1870s, the generator came out. The generator is essentially a new production tool, which can convert electrical energy into mechanical energy, so that electricity can start to drive machines to become a new energy source that supplements and replaces steam power. Subsequently, electrical products such as electric lights, trams, electric drills, and electric welding sprung up like mushrooms after rain. Human beings gradually began to step into the "Electrical Age".
AI, like electricity, is essentially a productivity boost. ChatGPT is similar to a generator, a production tool that directly transforms AI into productivity.
In practical terms, just as generators replace steam engines to provide new power sources for machines, ChatGPT demonstrates the potential to liberate repetitive mental (knowledge blue-collar) work.
In practical applications, ChatGPT is being more widely used in various business scenarios just like the electricity in the past. For example, the US version of "Today's Headlines" BuzzFeed has announced that it will cooperate with OpenAI to write articles. Microsoft CEO Nadella also said that he will commercialize OpenAI tools and integrate artificial intelligence tools such as ChatGPT into all Microsoft products. In the process of being widely used, ChatGPT is expected to stimulate AI to derive more business models, drive a paradigm shift in production relations, and even reconstruct the world business order.
From this point of view, ChatGPT may be playing the role of a "generator" and become a singularity that detonates the development of the AI industry.