Intelligent industrial model
On December 11, 2023, the domestic multi-modal large model Zhigong ushered in a blockbuster launch.
Zhi Zhen, founder and chairman of China Industrial Internet (Beijing) Technology Group, said: In this update, the intelligent industrial model has not only entered a new stage in the deep training of the model, but also carried out a series of leading innovations in multimodal fusion.
In 2023, China has made remarkable achievements in the field of artificial intelligence, especially in the field of large-scale model industry applications. With the continuous improvement of big data and computing power, multimodal intelligence has become one of the hot spots in the field of artificial intelligence. According to the report "Artificial Intelligence Global Transformation Prospects: The Transition Point is Coming (2023)" jointly released by KPMG and Zhongguancun Industry Research Institute, multimodal pre-trained large models will gradually become the standard configuration of the AI industry. This time, there is once again exciting news in the field of Chinese artificial intelligence innovation.
The domestic multi-modal large model Zhigong is expected to compete with Google Gemini.
How does the multi-modal large model of Zhigong perform?Through three questions, the Zhigong Industrial Model and Bard (Google Gemini) competed.
Question 01: Which ball is bigger, blue ball or orange ball?
The large model of Zhigong Industry answers: In this scene, a larger ball (blue) is compared to a smaller ball (orange). The blue ball is larger than the orange ball, so the blue ball is the bigger ball.
Bard (Google Gemini) answers: It is obvious from the image that the orange basketball is larger than the blue basketball. The orange basketball is about 8 inches in diameter, while the blue basketball is about 6 inches in diameter. This means that the volume of an orange basketball is about 17 times.
Bard's (Google Gemini) answer is up for debate.
Question 02: On the map, which country is the model airplane pressed on?
The industrial model answers: In this map of the world, the model of the airplane is pressed on top of the United States.
Bard (Google Gemini) answers: As you can see from **, the tail of the aircraft model is located exactly on the West Coast of the United States. Therefore, it can be inferred that the aircraft model pressed on top of the United States.
Q03: What happened in this scenario?What should we do?
The large model of Zhigong Industrial answers: In this scene, a large flame burns in the darkness, forming a huge flame. If the fire is caused by man, we should notify the local fire department immediately so that they can respond quickly and prevent the fire from spreading. If the fire is natural, we should leave the scene immediately and seek help to avoid injury or death. In either case, we should remain calm, follow local safety regulations, and seek help from the local fire department or other relevant agencies as soon as possible.
Bard (Google Gemini) replied: ** shows a fire. The flames burned from the darkness and looked very dangerous. Depending on the background and title, the fire could be a wildfire or a fire. If this fire is a wildfire, we should call the police immediately. Wildfires are very dangerous and can cause loss of life and property. If this fire is a fire, we should take immediate steps to extinguish it. Fires can cause serious damage and even endanger lives. If you see this fire, you should call your local fire department immediately.
The actual measurement found that the performance of the intelligent industrial model is indistinguishable from that of Google Gemini.
The emergence of domestic multi-modal large models has become the focus of attention in the industry and academia.
First, the intelligent industrial model has carried out in-depth research on multimodal fusion. By effectively fusing multiple data types such as images, text, and speech, the model is able to understand and process information more comprehensively. This kind of integration not only improves the comprehensive intelligence level of the model, but also provides more powerful support for the application of the intelligent industrial model in the industrial field. The intelligent industrial model will better adapt to different industrial scenarios and the personalized needs of enterprises, and become an all-round "digital worker".
Second, the multi-modal large model Zhigong has been carefully designed in terms of model pre-training. With the help of large-scale datasets and advanced training technology, the intelligent industrial model has made significant breakthroughs in the number of model parameters and training effect. This makes the intelligent industrial model excellent in handling complex industrial tasks, and also shows strong versatility in multi-field applications.
Third, in terms of performance, the performance of the intelligent industrial model is also eye-catching. From natural images to industrial language reasoning, multimodal large model intelligent engineering can be called the most advanced large model products in the industrial field.
Ms. Lu Man, R&D Director of Zhigong Industrial Large Model, revealed that in the world, the United States, China, and Europe are the leaders in the research and application of multimodal large models. China Industrial Internet has an early layout in the research of multi-modal large models, and has carried out the research of multi-modal large models in multiple tasks such as text, image, audio, and industry. Zhigong's new breakthrough in the multimodal field is crucial to improve the application of products in the industrial field.
In the process of research and development, the intelligent industrial model focuses on the understanding of complex and specialized semantics in the industrial field and the cultivation of contextual reasoning ability. By introducing the Xi learning method of "incremental pre-training + knowledge editing + vector database", the Q&A accuracy of the Zhigong model in the Q&A and ST knowledge fields of Hollysys has surpassed ChatGPT. The research and development of the agent model carried out by China Industrial Internet is also in the leading position in the industry.
Professor Zhang Qi, chief scientist of China Industrial Internet and professor of natural language laboratory of Fudan University, said that multimodal large models can be used to answer open-ended questions containing images, audio, ** and other information. Facing the future, the application fields of domestic multi-modal large-scale model intelligent engineering are very wide, which can show their strengths in scenarios such as industrial visual quality inspection, product design, experimental simulation, and equipment failure warning, and empower industrial enterprises.
Zhizhen, founder and chairman of China Industrial Internet, said:With the launch of the multi-modal large model of Zhigong, the leading position of the domestic large model in the field of artificial intelligence in the world will be further consolidated, and China will have a large model product that truly serves high-end intelligent manufacturing. Zhigong not only represents China's technical strength in the field of large models and multimodal intelligence, but also injects new vitality into China's artificial intelligence innovation and development. Domestic multi-modal large-scale model intelligent engineering will bring new development opportunities in the field of "industrial Internet + large-scale model", become a bright business card of domestic intelligent technology, and fully empower new industrialization.
As a more powerful multi-modal model in the industrial field, the Zhigong Industrial Model will improve production efficiency, reduce costs, improve product quality, and realize intelligent transformation of industrial enterprises.
In the field of manufacturing, the multi-modal model of Zhigong can improve the efficiency and accuracy of manufacturing, and can be used for industrial visual inspection to identify whether there are defects in products on the production lineIt can be used for industrial robot control to improve the operation accuracy and flexibility of the robot.
In the field of industrial R&D and design, the multi-modal large model of Zhigong can assist in product R&D and design, and can be used to analyze product performance data, identify defects in product design drawings, and optimize product design.
In addition, in terms of industrial management and chain management, the multi-modal model of intelligent engineering will also play an important role. With the continuous development of Zhigong multi-modal large model technology, these potential application scenarios will be gradually realized. In the era of artificial intelligence, traditional industrial enterprises will usher in a new paradigm revolution.