news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

Original title: Entering the "main battlefield" of industry, what are the chances of winning with large models?

Since the beginning of this year, my country's large model industry has developed rapidly, especially industrial scenarios have become a blue ocean for large model applications. The "China AI Large Model Industrial Application Index (2024)" released at the 2024 World Artificial Intelligence Conference and the High-level Conference on Global Governance of Artificial Intelligence shows that the accuracy of text generation in the industrial field of the country's top large models is already competitive, but the mathematical ability needs to be improved. How do large models perform in the "main battlefield" of industry?

Full chain application exploration

At present, my country has 41 major industrial categories, 207 medium categories, and 666 small categories, covering all industrial categories in the United Nations industrial classification. Among the 500 industrial varieties, more than 40% of my country's product output ranks first in the world, with unique advantages of being comprehensive, diverse, and large. In 2023, the overall scale of my country's manufacturing industry ranked first in the world for 14 consecutive years.

The huge scale of the industry provides fertile soil for the implementation of large industrial models.

The "Industrial Big Model Application Report" (hereinafter referred to as the "Report") released by Tencent Research Institute shows that my country's industry is in the stage of moving from digitalization to intelligence, and big models have become a key force in promoting industrial intelligence with their excellent understanding, generation and generalization capabilities, and are expected to expand new space for the integration of artificial intelligence and industry.

The report points out that the rise of big models is expected to bring a new paradigm of "basic models + various applications" to the industrial field. On the one hand, big models can deeply understand complex problems in the industrial field, understand and process massive data, and mine patterns and trends from them; on the other hand, big models will expand new scenarios for the application of artificial intelligence in the industrial field.

At present, the application exploration of big models has been carried out in the entire industrial chain. In the field of R&D and design, big models improve R&D efficiency by optimizing the design process; in the field of production and manufacturing, they expand the boundaries of intelligent application; in the field of business management, they improve the level of business management based on the assistant model; in the field of product services, they promote the intelligence of products and services based on interactive capabilities.

Solving problems and doing difficult things for the industry

Combining with actual application scenarios in production and life and empowering thousands of industries to increase efficiency is the inevitable development direction of big models. As Liu Yunjie, an academician of the Chinese Academy of Engineering, said, the way out for the development of my country's artificial intelligence industry lies in industry big models.

At present, many domestic technology companies have released industrial big model products. As the first big model positioned in the industry, Pangu Big Model has indicative significance. Zhang Pingan, executive director of Huawei and CEO of Huawei Cloud, introduced the industrial practice of Pangu Big Model in detail at the Huawei Developer Conference held in June, showing a sample of industrial AI "solving problems and doing difficult things".

In the field of industrial design, Pangu big models can be widely used in the fields of electronic products, automobile styling design, etc. For example, in automobile styling design, designers can interact with big models through dialogue, drawing, etc., improve creative inspiration, generate 3D automobile digital models, and adjust the style of the models, edit parts, and change colors. This can greatly shorten the design cycle that originally took 1-2 years.

In the field of architectural design, Pangu Model can generate a colorful and textured 360-degree real-life video of the building complex by simply inputting a black and white sketch of the design. It can also build a highly realistic 3D model of the building, shortening the conceptual design cycle of a complex building complex from weeks to tens of minutes.

In the steel industry, the Pangu model can also be used. In the past, every time the Baowu Steel Group hot rolling production line adjusted the type and size of the steel plate, engineers had to readjust more than 300 parameters of the seven-pass finishing mill, a process that usually took about five days. Now, the Pangu model can predict the optimal parameters, significantly reducing the adjustment time and improving the prediction accuracy and steel plate yield rate.

In addition, Caos launched the industrial big model COSMO-GPT, which has been successfully implemented in multiple application scenarios such as industrial indicator optimization, industrial information generation, and industrial question and answer. Supported by the iFlytek Spark cognitive big model technology base, Antelope Industrial Internet Company has created the Antelope Industrial Big Model based on the actual needs of industrial scenarios. The model has five core capabilities: industrial text generation, industrial knowledge question and answer, industrial understanding calculation, industrial code generation, and industrial multimodality, and has served many companies... The distinctive industrial big model products have built a diversified big model ecosystem and injected new vitality into industrial intelligence.

Three major challenges faced in implementation

It should also be noted that compared with consumer scenarios, the application of large models in industrial scenarios still faces some obstacles. The report analyzes that the application of industrial large models faces three major challenges: data quality and security, reliability, and cost.

First, industry involves a wide range of fields and has high requirements for data security. However, the current industrial data structure is diverse and the data quality is uneven. The data quality and security of industrial large models need to be further improved. Secondly, industrial production environments often involve complex process flows, high-precision operation control, and strict safety standards. Any model prediction or decision-making errors may lead to production accidents, quality problems, or economic losses. Industrial large models must also meet high reliability and real-time requirements. In addition, high costs limit the input-output ratio of industrial large model applications.

Despite various challenges, the development of industrial large-scale models is an inevitable trend.

The cost-reduction and efficiency-enhancing effects of industrial large-scale models are obvious. Zhang Pingan gave an example, blast furnace smelting is considered to be the most difficult application scenario for the implementation of artificial intelligence. The blast furnace is a 5,000 cubic meter "black box" with a maximum internal temperature of 2,300 degrees Celsius. The smelting process is "invisible and intangible" and highly dependent on manual experience. If the Pangu large-scale model is used, the "black box" can be turned into a "gray box" or even a "white box" to guide the precise control of the blast furnace. The consumption of coke can be reduced by 1 kg per ton of molten iron, reducing the cost by 3 yuan.

With the evolution of technology, the application of large industrial models will be faster and more stable.

The report believes that through the combination of industrial basic big models and industrial apps, we can respond to challenges in the industrial field widely and quickly, and promote the intelligent upgrade of various industrial scenarios. At the same time, with the development of new technologies such as intelligent bodies and embodied intelligence, big models will open up more application scenarios in the industrial field and improve production efficiency and safety. In addition, big model compression-related technologies will effectively reduce the number of model parameters and computing requirements, reduce training and deployment costs, make big models more suitable for resource-constrained environments, and accelerate their application and promotion in the industrial field. (Reporter Cui Shuang)