Cao Lei, Vice President of Tencent Cloud: The larger the model "understands" the industry, the faster it can be implemented

Recently, Cao Lei, vice president of Tencent Cloud and head of intelligent manufacturing and smart energy, said in an exclusive interview with a reporter from China Electronics News that the explosion of generative artificial intelligence technology represented by large models will accelerate the process of new industrialization. The large model must "understand" the industry better in order to land more "quickly".

Generative AI will accelerate the process of new industrialization

In the stage of industrialization driven by informatization, many industrial manufacturing enterprises are in the process of informatization from scratch, or may be in the process of moving from a weak foundation to continuous enrichment of informatization. At that time, the "new" of new industrialization referred more to tools and methods.

Today, with the continuous development of ICT technology, especially the explosion of generative AI technology represented by large models, it has brought new significance to the new industrialization. At this stage, the "new" of new industrialization means that the core competitiveness of enterprises needs to be redefined in the new era.

"This 'new' can be the innovation of technical capabilities from follow-up learning to independent research and development, the innovation of the supply chain from complex and changeable to full control, or the innovation of the business model from selling equipment and products to providing services. Cao Lei said that it has now reached the stage of practicing new industrialization.

Cao Lei believes that digital technology, especially cloud technology, big data, 5G, artificial intelligence and other IT technologies are more closely integrated with the core competitiveness of enterprises, and the competitiveness is formed and strengthened by improving the breadth of industrial chain connections and the speed of information flow, and digital technology can even be transformed into industrial product elements to create direct benefits for industrial enterprises. As a high-potential and high-value tool and method, artificial intelligence technology is accelerating the process of new industrialization.

Manufacturing companies should try more large-scale model applications

"At present, there are relatively few cases of traditional industrial enterprises applying large models to generate actual business value, because the closer to traditional industries, the deeper the industrial threshold, and the application of large models requires more thinking and attempts. Cao Lei said frankly.

Tencent Cloud helps manufacturing enterprises accelerate new industrializationHe pointed out that the most important principle for AI technology to play a role is machine learning the knowledge and laws hidden behind a large amount of data, and data density and abundance affect the effect of machine learning. The degree of digitalization of each business domain of R&D, production, supply, marketing and service of domestic industrial enterprises determines whether the model can be generated and whether it is effective.

Industrial enterprises have different business characteristics, and their digital investment in different links is also different. For example, electronics manufacturing companies pay more attention to production efficiency and product quality, and AI quality inspection scenarios are of high value, so enterprises are willing to invest a lot of costs and resources in the construction of digital workshops, collect data by deploying a large number of sensors, and then model them. The introduction of large models can effectively improve production capacity on the one hand, and improve the accuracy of quality inspection on the other hand, so such enterprises are willing to try related applications.

Another example is that aircraft manufacturers have a lot of detailed data that needs to be digitized. Throughout the manufacturing process, when a component changes, different types of data associated with it change. If measured in A4 paper, this data could pile up a ten-story building. If a large model can be introduced, the process of collecting, proofreading and verifying these data will be greatly simplified, and the accuracy of the data can also be effectively improved, the error rate will be infinitely close to zero, and the large model also has the ability to learn and combine, which is much more efficient than manual operation.

"At this stage, only some shallow applications can be seen, such as digital humans, knowledge graphs, etc., and deeper applications need to be further explored. Cao Lei said that traditional manufacturing enterprises also have urgent demands for large models, but they have not yet reached the stage of landing.

Cao Lei believes that everyone has not been able to clearly see the quantifiable input-output ratio brought by the application of large models, which leads to a cautious attitude towards the introduction of new technologies. This is also a direction that industrial enterprises that have mature artificial intelligence technology applications and technology companies that have advanced the application of artificial intelligence technology should work together to explore.

The large model must "understand" the industry better in order to land more "quickly".

Talking about the landing of large models on the industrial side, Cao Lei gave his own thoughts. First, the model needs to "learn" industry knowledge, and the knowledge of the general large model on the market comes mostly from the public data of the Internet, the advantage of these data is breadth, but the strength of professional data in specific fields is insufficient, and model manufacturers and experts in specific fields need to work together to collect and integrate data, so that the large model "understands" the industry.

Second, the "bigness" of the large model makes it have a strong generalization ability, but it also means that the cost of use is relatively high. In order to obtain better model results, the "inference" process often requires "result feedback" and then "retraining" of the model. This means that for enterprises, not only do they need manufacturers like Tencent to provide model product support, but they also need professional cloud services to help them "reduce costs and increase efficiency" and maximize operational benefits.

Third, the application of large models has a great impact on the traditional business processes of enterprises. Although the application of large models cannot directly replace one person to assume the entire job role, it will directly improve the efficiency and effectiveness of the specific work content of this job role, and will also affect the collaboration method and efficiency between job roles, which means that the organizational process and assessment mechanism and even performance returns will be redesigned.

In addition, the implementation of large models on the industrial side also needs to consider many aspects such as computing power support, data ownership and security, capital and strategic determination support, and the introduction of high-end talents, which requires more comprehensive thinking and a meticulous and thoughtful work path. Cao Lei said.

Cao Lei believes that with the proposal of new quality productivity, the industrial manufacturing industry has made it very clear that the core problem that needs to be solved at present and in the future is that "the development path must change", and the key lever of this "change" focuses on digital technology and even artificial intelligence. "In the face of fierce market competition, what Tencent needs to do is to practice its internal skills, polish the base of the large model, and then carry out industrial application. In the future, we will also consider cooperating with different types of customers, and may explore aspects such as model privatization. Cao Lei said. Ministry of Industry and Information Technology: Establish and improve the standard system for artificial intelligence to empower new industrialization

From "volume" training to "heavy" inference, enterprises have "core" requirements for deploying large models

Author丨Song JingEditor丨Zhao Chenmei Editor丨Maria Producer丨Lian Xiaodong