The new industrialization engine accelerates the launch of the large model and actively opens up application scenarios

At present, all industries are accelerating the embrace of large models. Recently, under the guidance of the Industry and Information Technology Think Tank Alliance, Baidu Development Research Center jointly conducted research on the application of large models with a number of think tanks such as China Academy of Information and Communications Technology, China New Generation Artificial Intelligence Development Strategy Research Institute, and China Internet Network Information Center.

"There is a strong incentive to use big models in the steel industry." Zhongtianiron and Steel experts said that long research and development cycle, high production costs, low collaborative efficiency are the pain points and difficulties faced by the high-quality development of the steel industry, covering all aspects of research, production, supply and marketing services, the need to develop new quality productivity, restructure production management, and explore the application of artificial intelligence technology and large models in the steel industry has become the general trend.

With the help of large models, transforming expert experience and industrial mechanism into the advantages of manufacturing industry and digital competitiveness has become the direction for enterprises to explore. GCL technology experts said that they hope to establish their own large model of slicing, improve the line speed, cutting process and other parameters through model prediction, and increase the yield from the current 96% to 98% or even higher. He said, "The competitiveness of enterprises is low cost, high yield. At present, the method of improving the yield through manpower has reached the bottleneck, and the follow-up is looking forward to finding an improvement method through a large model."

An important area in which large models help new industrialization is to explore how to deposit the knowledge and experience in the minds of old experts and old technicians. China's manufacturing industry has many scenarios and large amounts of data. In particular, a lot of experience is mastered in the minds of old experts and old technicians, and has not been transformed into knowledge of the industry and enterprises. If the use of general large model + industry enhanced capabilities, so that each post, each employee behind, there is a virtual senior expert, it will be more efficient and economical to meet the massive personalized needs of enterprises, improve the overall competitiveness of enterprises.

Liu Gang, chief economist at the China New Generation Artificial Intelligence Development Strategy Research Institute, said, "Scenarios are the unique advantages of China's manufacturing industry, and accelerating the application of continuous optimization model capabilities may run out of a large model development path with Chinese characteristics."

According to the latest Statistical Report on the Development of the Internet in China released by the China Internet Network Information Center, as of now, the penetration rate of artificial intelligence represented by large models in China has reached 16.4%. From the perspective of the industrial chain links, the current large model application distribution shows a "smile curve" characteristics, fast at both ends and slow in the middle, that is, the knowledge-intensive and service-intensive links such as R&D, design and operation services at both ends of the industrial chain are relatively fast, and the intermediate links such as production and manufacturing are relatively slow. Judging from the current application progress, content generation, intelligent interaction, and information extraction are the common needs of the combination of manufacturing and large models.

He Xia, former chief engineer of the Policy and Economic Research Institute of the China Academy of Information and Communications, deputy director of the Artificial Intelligence and Economic and Social Research Center, said, "How to evaluate the application value of large models needs attention, not only the success is valuable, but also the exploration in the process is valuable." National projects with some research and development nature may adopt process subsidies; Industrial projects can consider post-subsidy ways to further guide and stimulate the willingness of traditional enterprises to apply large models."

In the survey process, companies said that the application of large model concerns, the biggest difficulty is the data. First, data quality. The application effect is strongly related to the acquisition, standardization and accumulation of data in the early stage, which requires time and capital investment to accumulate. The number of professionals also determines the collection of high-quality data, the construction of data sets and the application effect. Second, data security. Data is the core asset and core competence of an enterprise. In the process of large model training and application, how to ensure the safety of key data in the production process and keep core data within the domain is also the biggest concern and consideration factor for enterprises. To solve the problem of data quality, manufacturing enterprises and artificial intelligence technology enterprises need to jointly explore and strengthen cooperation in multiple dimensions such as talent training, data collection and data analysis. Tsinghua University "large model Security Practice White Paper" pointed out that "cloud platform services with its mature security defense system, can support a wide range of service needs", the current cloud platform has been able to ensure data security in the whole life cycle of the large model, protect enterprise sensitive data within the domain, protect data confidentiality and integrity.

Industry experts said that in order to better play the industrial enabling value of the large model, it is necessary to make greater breakthroughs in concept innovation, objectively look at the efficiency and security of the application of the large model, and break the misunderstanding of "large model privatization deployment is safe" and "prioritizing open source large model". With the application of data privacy protection technology and the improvement of enterprise data governance system, "connection" does not mean insecurity, but promotes the transformation of the large model from general intelligence to advanced productivity, and helps industrial transformation and upgrading. Blindly emphasizing privatization and customization, on the one hand, will greatly increase the operation and peacekeeping service costs of enterprises; On the other hand, the application mode of "manual workshop" and "construction team" will damage the efficiency and quality of the application of large models, restricting the quality and efficiency of enterprises and developing new quality productivity. (Reporter Li Shuichu)