Generative AI is experiencing a debate about the prospect of landing.

In Hildesheim, a city in north-central Germany, the world's industrial giant Bosch has put generative AI technology to its production line. They synthesized more than 15,000 images of product defects using AI simulations, which were then used to inspect the quality of the motor stator production line. With the help of generative AI, Bosch was able to shorten the project duration by six months using an automated optical inspection model, thereby increasing the annual output value by six figures.

And across the Atlantic in the United States, OpenAI, the detonator of the generative AI concept, just released a controversial AGI roadmap in July. According to OpenAI's definition, the chatbot at this stage is only the first stage of AGI, followed by four development levels of reasoning, agent, innovation, and organization, and the fifth level of AGI will be achieved as soon as 2027. However, this roadmap has been widely questioned by all sectors of society, including Fortune and other media, because OpenAI only gives a conceptual development stage, but does not really endorse the future of AI application in the real world.

(Source: The Verge)

Not everyone agrees with using pure large language models to explore the endgame of AGI. In the real world, AI should have more possibilities, as Bosch did in the industrial sector.

There are many fixed tasks and scenarios on the production lines of industrial and manufacturing industries, so that generative AI can truly create output value, not just a topic. This is precisely the real demand of the real industry. Bain & Company mentioned in the 2024 Global Machinery and Equipment Report that 75% of advanced manufacturing companies give priority to the use of AI technology in their engineering R&D strategies, and more than 90% of machinery and equipment companies have already begun to collect and store production data, providing "nourishment" for the application of AI.

Furthermore, if an AI technology tries to take root in industrial and manufacturing scenarios, it naturally needs to have a corresponding industrial environment. And in China, which has a huge industrial landscape, companies that have mastered these AI technologies are supporting emerging applications.

McKinsey mentioned in the study that nearly 60% of the top use cases in the list of 21 new lighthouse factories announced in December 2023 use AI technology. 11 of the 21 lighthouse factories are from China, including CITIC Pacific Special Steel's Jiangyin plant in China, LONGi Green Energy's Jiaxing plant in China, and Hengtong Optical Fiber Technology's Suzhou factory in China, all of which have introduced AI technology.

While humans are worried that AI and large models will lead human social life to an unpredictable future, the industrial scene provides certainty and a broad future for players participating in this ecosystem.

The intelligent release of AI needs to carry scenarios

To understand the importance of scenarios for AI, we can start with two pieces of information about human intelligence.

There are many well-known stories about "wolf children" in the human world, which describe a phenomenon in which a human with an intelligent brain will not be activated when he is in the survival scene of a beast (such as a wolf) for a long time, so as to maintain a low level of intelligent performance.

United States psychologists Winthrop and Luella Kellogg conducted reverse experiments with apes and children more than ninety years ago to test the ability of other species to learn human intelligence, and wrote their conclusions into the book "Apes and Children: A Comparative Study of the Influence of the Environment on Early Behavior": "No matter how socialized and humanized training is carried out, there are clear limits to the degree of humanization that non-human species can achieve. ”

The two different stories reflect common philosophies: highly intelligent individuals who lack training will not unleash their cognitive and learning potential; Although the upper limit of relatively low intelligence is low, they can master the ability to deal with real-world problems through imitation and training, and reach a certain "degree of humanity", but it will be difficult to go one step further.

Therefore, when AI also begins to learn the logic of human work and infiltrate into real life, when we find that the learning ability and application level of AI are temporarily limited, and it cannot lead to the depth of AGI, choosing the right track and scenario to let AI create visible value first becomes a road worth trying. At this stage, AI has a certain level of intelligence, and after being trained to meet the needs of real-world scenarios, it can participate in production operations.

(Source: MakinaRocks)

MakinaRocks, a Korea AI engineering company that has deployed more than 4,000 AI models in manufacturing and industrial environments, believes that based on the diversity of industrial scenarios, industrial large language models can cover all levels of the manufacturing plant and will be able to manage and optimize the entire operational process. McKinsey pointed out that AI can fully participate in the optimization and upgrading of the industrial value chain and bring significant productivity improvements.

(Source: McKinsey & Company official website)

It is precisely because of the universality of AI technology that it can provide knowledge of any industry that AI technology is becoming a new pearl in the industry. In late July, the United States National Institute of Standards and Technology (NIST) announced a $70 million support program to support the United States Manufacturing Institute, which focuses on using AI technology to improve the resilience of American manufacturers.

In China, we are seeing a widespread trend of industrial companies embracing AI models. In the field of display panels, BOE launched a large display industrial model in December last year; In the field of new energy, Goldwind launched a large language model for the wind power industry in March this year, and LONGi Green Energy launched a production compliance video detection technology based on a multi-modal large model in May this year.

Even so, data released by the CCID Research Institute of the Ministry of Industry and Information Technology shows that as of 2023, the enterprise adoption rate of generative AI in China is still only 15%, and there is huge growth potential. In the practice of Chinese companies, industrial enterprises have benefited, and AI technology companies with deep technical background have also grown.

China's AI innovation companies have a unique path to development

When it comes to the scene of manufacturing, China cannot be avoided. In the context of a huge industrial and manufacturing industry, the possibilities are endless.

In a report earlier this year, Forbes magazine explored five industries that are highly compatible with AI. For example, the automotive industry will be one of the leading industries in the manufacturing industry to adopt AI technology. In China, where the development of new energy vehicles is in full swing, AI-driven intelligence and digitalization are almost the same as the automotive industry. As a result, many advanced technology use cases have emerged in China's automotive industry.

At the end of last year, Fortune magazine announced the 2023 Fortune Most Influential IoT Innovation List, and Innovation, a full-stack AI solution provider for Chinese enterprises, stood out among nearly 200 declared projects with the case of "Intelligent Automated Production Line of Li Auto Factory" and successfully made the list. By integrating AI technology into the automated production line of the new energy vehicle factory, the factory has improved its ability to detect potential problems, controlled the failure rate and production production losses, and shortened the effective technical man-hours by 70%.

This is not the only result. Since its establishment in 2018, Qizhi has accumulated profound technical capabilities in AI fields such as computer vision and machine learning, mastered the "MMOC artificial intelligence technology platform" and "AInnoGC industrial large model technology platform", and established industry solutions covering eight major fields, including iron and steel metallurgy, panel semiconductors, 3C high-tech, engineering construction, automotive equipment, energy and power, food and beverage & new materials, and intelligent manufacturing training.

(Source: Innovation Qizhi)

This just proves that China's diversified industrial scenarios are a hotbed for the implementation of AI technology. According to Bain & Company, the use of AI in the manufacturing industry is focused on three directions: reducing assembly defects and improving quality control, increasing productivity, and simplifying warehouse inventory management. By using the rich data sets and industrial experience accumulation in industrial scenarios, the integration effect of "AI + Manufacturing" surpasses the traditional automation technology and improves the ability of control optimization, thus achieving the purpose of optimizing the manufacturing process and improving quality control.

Wang Jinlin, a Chinese investor in Silicon Valley and founding partner of Foothill Ventures, once pointed out in an interview with Tencent Technology that the key to the AI application layer is data barriers. By analogy, the key to the application of industrial large models and industry generative AI is to output capabilities or content that meet the actual needs of industrial enterprises based on AI models with a reliable data base, and to escort industrial production processes.

In terms of data volume, the representative use cases all reflect deep links to the industry. For example, COSMO-GPT, the large industrial model of Haier Kaos, has more than 3 million high-quality industrial data, more than 3,900 mechanism models and more than 200 expert models, and has injected 562 industrial datasets. In the first half of the year, the Qizhi Kongming Industrial Large Model 2.0 (AInno-75B) was released, with a parameter volume of 75 billion, and the scale and performance progress also introduced multi-modal large model capabilities, and even output CAD or actions.

From the perspective of the ability of landing application scenarios, Bosch generated tens of thousands of sample images with double-digit real images of each type according to the different fault types of the stator. Generative AI greatly expands the potential of defect sample data to inform inspection on the production line.

(Source: Innovation Qizhi)

This approach of "getting an adequate picture of all defect types without actually producing a defective part" is also found in the use case of Ingenuity. Based on the AInnoGC industrial model, Qizhi can help vertical industries generate unique defect sample data and perform other processing such as segmentation and annotation, breaking through the limitation of insufficient industrial defect samples.

At present, in the automobile, LCD panel, iron and steel metallurgy and other industries, Qizhi has established a vertical intelligent detection and analysis program. With its proficient application of related technical solutions, Qizhi Innovation has been ranked as the No. 1 AI industrial quality inspection solution manufacturer in China in the "China AI-Enabled Industrial Quality Inspection Solution Market Share, 2023" report recently released by IDC.

Supported by industrial scenarios, industrial data, AI model capabilities, vertical industrial applications and other elements, the corresponding To B business has formed a business model with endogenous growth. In its annual report, Qizhi disclosed that the revenue of the AI-enabled manufacturing business segment in 2023 has reached 1.176 billion yuan, a year-on-year increase of 24.1%. At last year's performance communication meeting, its management also said that the company will transform to high-quality growth and achieve a turnaround as soon as possible on the basis of the company's healthy development.

AI innovation companies rooted in China's industrial market are entering an era of multi-point flowering. In this living environment with abundant "nutrients", the generative AI industry has also found a new "way to live" at this stage.

This article is from Xinzhi self-media and does not represent the views and positions of Business Xinzhi.If there is any suspicion of infringement, please contact the administrator of the Business News Platform.Contact: system@shangyexinzhi.com