2023 is undoubtedly a year of rapid progress in artificial intelligence, and generative artificial intelligence represented by ChatGPT has also made ordinary people feel the AI in science fiction films for the first time, and it has really entered reality. As a result, Internet giants and start-ups on both sides of the Pacific have also joined this wave of generative AI.
However, in such a frenzied atmosphere, there are also people who remain rational. A few days ago, Li Kaifu, chairman of Innovation Works and CEO of Zero One Everything, gave this statement in an interview with Science and Technology **the information, "At present, the competition in the field of Chinese artificial intelligence is still in the qualifying stage. ”
Kai-Fu Lee believes that domestic large model companies are in the stage of technical verification, and they need to prove that they have the ability to develop high-quality models, and only if the model performs well, it is possible to emerge in practical applications, otherwise it will be more like a toy and cannot solve practical problems.
I believe that this sentence can resonate with many people, because friends who have tried many large model products in China will probably find that even if these large models are among the best, the experience is indescribable.
Looking at the current AI models in the domestic market, except for Wenxin Yiyan, which claims to have exceeded 100 million users, almost none of the other large models leave a memory point. In fact, most people will only remember the first place, just like the overseas large model everyone may be familiar with ChatGPT or GPT-4, but few people may know that the performance of Claude of Anthropic in the real machine environment is close to GPT-4.
Of course, for the large models launched by domestic Internet manufacturers, the technical level may not be a big problem, but the more important thing is what the large model can do. Kai-Fu Lee said, "Enterprises that pass the technical test in the qualifiers will move on to the next stage: the business value stage. What is your business model?How to make a profit?Soon, investors will be evaluating the value of these companies based on their income statements and asking the same questions as cloud service providers, enterprise software companies, and consumer applications. If businesses can't answer these questions, their growth will come to an end. ”
If it is said that behind the domestic "100 model war" in 2023, there is a lack of target investment institutions that finally have a track that looks extremely bright to bet on, and the swarming capital is the fuse. Then after entering 2024, the "100 model war" will enter the stage of big waves and sands, and the participants of the AI track need to answer the question of how to make money. But it is impossible to follow the path of ChatGPT, that is, to rely on paid subscriptions, because Wenxin Yiyan has sold paid members.
Therefore, when the time comes to 2024, the "100 model wars" may be a thing of the past, and building applications and solutions for specific domains may be the new answer. So in this way, Robin Li really did not "fool" everyone, and the domestic large-scale model track cannot accommodate more than 200 large-scale models, so it is better to transform the development of AI native applications as soon as possible.
In fact, there are not a few technical experts who hold similar views, and Ge Canhui, a 360 intelligent brain expert, is like this** The large model industry in 2024, "There are five to ten left in the public domain, and the private domain is also a very cruel knockout." ”
So the question is, when it is said that large models should be shuffled and more focused on the application layer, what kind of AI applications are favored by the market?At present, AI applications are likely to be divided into two major schools, namely "AI+" and "+AI". The former is the so-called AI-native, that is, applications based entirely on generative AI technologies, such as Midjourney, Stable Diffusion, Runway, and Notion AI.
"AI" refers to the use of AI to empower existing products, which are currently more common in the domestic market, such as Kuaishou App's "AI Play Review", Youku's "AI Search Film", and Douyin's "AI Search" which is being tested in grayscale. Obviously, this type of "+AI" product is more of a stopgap measure than an AI-native application, to make users feel that "we are really doing AI". After all, apps such as Kuaishou and Douyin did not consider the existence of AI in the initial design stage, and they were 100% created for human users, resulting in a fragmentation of the experience level.
At the end of the day, an AI application needs to occupy at least one of the "s**e time" and "kill time". Among them, the "S**e Time" product is more in line with the public's imagination of generative AI to subvert the existing production pattern, which means the improvement of efficiency, and the typical products are Midjourney and Stable Diffusion, which have made a large number of low- and middle-level artists unemployedThe "Kill Time" product is mainly aimed at the entertainment needs of users, and the typical example is various AI chatbots. These two categories also cover the mainstream needs of mass users for generative AI.
At first glance, it seems that the "kill time" type of product is more suitable for domestic entrepreneurs, but the truth may be the opposite. Because unlike overseas markets, the characteristics of the domestic Internet industry are that large manufacturers are deeply engaged in the consumer Internet, while the industrial Internet is still in the exploration stage. Making "kill time" products means competing for users' time and attention, and this field is currently full of big players, and even good ideas can be quickly imitated, learned, and even surpassed.
On the other hand, as long as the "s**e time" product has passed the initial cold start stage, it can be completely at least small and beautiful, and similar products can be said to be numerous in the mobile Internet era.
With the arrival of the new year, generative AI, which has been "boiled" in 2023, will finally cool down in 2024. If the participants can't come up with a product that really interests the market, the AR VR industry has ushered in an unknown number of "first years", which may be a lesson from the past.