Li Yanhong said at Wenxin Yiyan's press conference that he deeply understood the industrial opportunities brought about by the ChatGPT artificial intelligence wave. He mentioned that the era of large-scale models will generate three major industrial opportunities: general large-scale models, industry large-scale models, and AI applications.
These three industrial opportunities are a pyramid structure, and the top layer is a general-purpose model, which can only be achieved by a very small number of large companies.
The industry model is on the second level, and only a small number of companies can complete it. The most important thing in this level is the data element. Whoever owns the data will have the barriers to competition.
AI applications on the bottom layer mostly use the models of the upper two layers for further engineering and present them directly to users.
Let's take a closer look at these three opportunities.
The first opportunity is the general-purpose MAAS model, such as OPENAI's ChatGPT and Baidu's Wenxinyiyan . Before the emergence of ChatGPT, many researchers believed that this direction would not work, but OPENAI solved the problem from 0 to 1, and verified the feasibility with chatgpt. Now more and more large companies are beginning to join in the research and development, and the common large model R&D has become the core area of the new round of technological competition among countries.
General-purpose large models require huge computing power, data, and algorithm training as support, which is a high-cost activity that only super-big companies can afford. Microsoft behind OPENAI has invested tens of billions of dollars. In China, only companies like Baidu, Tencent, Ali, and 360 that have data, capital, and human resources can set foot in this field.
The general large model is similar to the operating systems in the Internet age, such as WINDOWS, Android and iOS operating systems, and is the foundation of artificial intelligence. In fact, there won't be too many of them at the same time, because they have high barriers and first-mover advantages. The competition will be fierce, and you will end up with only a few big companies, depending on who does it best.
At present, Baidu's Wenxinyiyan is the first company to launch a general-purpose model in China, but big companies such as Ali and Tencent must also follow up the technological revolution of artificial intelligence. I believe that there will be several general-purpose models in China in the future. Model.
The operation and service of the general large-scale model are all on the cloud, and the general large-scale model company is the service itself, the person who "sells the shovel".
Baidu's Wenxin Yiyan provides services to the outside world through Baidu Smart Cloud, while ChatGPT provides services to the outside world through Microsoft Cloud. This has high requirements for cloud computing, and language models, frameworks, chips, and applications must be highly integrated.
There are many investment opportunities in the direction of general large models, such as AI chips, computing power, cloud computing, optical communications, and large models themselves.
Correspondingly, the barriers to competition at this level are extremely high, not only reflected in artificial intelligence technology, but also have very high requirements for data and financial resources, such as AI chips, which provide computing power services for large-scale model training At present, only Nvidia's chips in the world can hold this heaven-defying computing power requirement.
The second opportunity is a large industry model, that is, a large model of a subdivided vertical industry . This is the middle layer between the generic big model and the enterprise. Based on their insights into the industry, they can invoke general large-scale model capabilities to provide solutions for industry customers.
A high-quality industry model is indispensable and of great significance, and it is the basis for the commercialization of artificial intelligence.
As an operating system, the general-purpose large model has good generalization ability and can be grafted to all industries and companies, but it also has two very obvious shortcomings: one is that the knowledge coverage is wide but the depth is not enough; The requirements are not precise enough.
The reason for these two shortcomings is that the general large model needs to process a large amount of data, resulting in insufficient knowledge of specific industries. At the same time, for different downstream enterprises, there are also differences in their needs and characteristics, and it is difficult for a general large-scale model to meet such individual needs.
Therefore, the emergence of large-scale industry models has filled this gap. Through in-depth research and analysis of industries, they closely combine the capabilities of general large-scale models with industry applications, and can provide more accurate solutions for specific industries. This is why industry large models are favored by more and more companies.
It is worth noting that the development and implementation of industry large models require corresponding technology and professional knowledge, and professional technical personnel are required to maintain and upgrade them. Therefore, in addition to large technology companies, there are also many small companies involved in the development and promotion of large industry models. This also provides more opportunities for SMEs
In order to solve these shortcomings, enterprises need to use professional large-scale industry models, which will deeply explore the knowledge of specific domains and have more precise matching capabilities.
For example, like HowNet, all the paper data on HowNet is stored in HowNet’s own server, and this part of the data is privatized data.
It is impossible for a general-purpose large model to have access to these data, but HowNet can use the general-purpose large model, combined with its own privatized data, to further train an industry-specific large model for writing papers.
In the future, just type in the title with one click, and it can help you write a very good paper.
Similarly, there are Flush, Douyin, Tencent’s game data, etc. Using Douyin’s industry model can generate videos that are more in line with Douyin users, and these videos will also be full of Douyin elements.
Therefore, industry giants + general-purpose large models will basically continue the original competition pattern.
For enterprises, having a professional industry model will become an important foundation for commercialization. By using these models, businesses are able to better meet customer needs and gain a higher competitive advantage.
The third industry opportunity is application development . Based on the existing ability to generate text, pictures, voice and video, applications that are unimaginable today can be developed. In the era of artificial intelligence and large models, this phenomenon-level application will also appear, but it is developed based on large models.
The most typical one is "Talking Tom Cat", which is a representative of this type of application. It is directly connected to chatgpt and presented to users. However, this application still continues the thinking of the mobile Internet, and it is necessary to truly make phenomenal applications , must break the original product thinking .
In addition to the general model and industry model, application development is also an important part of the artificial intelligence industry. Using artificial intelligence technology, various applications can be developed based on various forms of data such as text, pictures, voice and video. In the era of large models, these applications will be more powerful and advanced, bringing more convenience and innovation to people's lives.
In this field, there will be many opportunities for small and medium-sized companies and individuals.
We have witnessed that chatgpt can bring huge productivity, which also means that, like in the PC era, one person is a webmaster, and in the era of large models, it will continue to be staged .
For various applications, homogeneity and templating may also be more serious.
The development of the artificial intelligence industry is in the era of large models. General large-scale models, industry large-scale models and AI application development will all become important components of the future artificial intelligence industry.
For enterprises and individuals, in order to maintain a competitive advantage in the era of artificial intelligence, they must have a deep understanding of these industrial opportunities and actively invest in research and development and applications.