A New Future for Digital Governance! The boom of domestic AI large models is coming

　　Exclusive article by China's well-off network

　　Text: "Xiaokang", reporter of China's Xiaokang Network, Li Huijun

　　Science and technology are the primary productive forces. With the rapid innovation of science and technology and productivity, domestic AI large models have entered a new stage of large-scale application, and artificial intelligence will become an important driving force to promote the high-quality development of science and technology and empower thousands of industries.

　　On March 18, Kimi Chat, the first intelligent assistant product launched by Beijing Dark Side of the Moon Technology Co., Ltd. (hereinafter referred to as the Dark Side of the Moon) that supports the input of 200,000 Chinese characters, attracted the attention of many technology fans. On March 22, Ali Tongyi Qianwen followed suit by announcing that the long document processing function of 10 million words would be open to everyone for free. With the gradual popularization of a series of domestic large model applications such as Kimi, domestic AI large models will usher in a new golden period of rapid development.

　　AI large model ChatGPT has become popular and has brought fire to the general artificial intelligence model, but the development of artificial intelligence technology is a continuous process that requires continuous technological research and development and innovation. Photo by Ning Ying

　　AI models have a lot to offer

　　With the rapid development of the digital economy, AI models have become a hot topic for a new generation of artificial intelligence.

　　Large AI models are often used to process large-scale data and generate complex outputs. These models learn patterns and patterns in natural language by pre-training massive corpus data, and perform well in generative tasks. Large models are commonly used in natural language processing, image recognition, speech recognition, and other fields to achieve more accurate prediction and decision-making, and are deep learning models with huge parameter numbers and computing power.

　　On October 9, 2023, Moonlight launched the world's first intelligent assistant product that supports the input of 200,000 Chinese characters to the public, which is the first time that Moonlight has been unveiled in front of the public, and it is also the first time that the name of Kimi Chat has been made public. Since then, the dark side of the moon has gradually faded into obscurity, only to resurface when a new round of financing of more than $1 billion was completed in February this year. It is estimated that no one would have thought that a month later, this AI rookie would once again expose a big surprise, and the fierce competition in the field of AI long essays "self-involution" to a new height that is difficult to achieve.

　　Compared with the previous 200,000 words long text limit, the new Kimi intelligent assistant supports 10 times the maximum length of the original text, and the larger capacity of long context processing capabilities will undoubtedly help Kimi users unlock more new use cases, including long-form text translation and understanding, auxiliary analysis of legal issues, sorting out dozens of invoices at one time, and quickly understanding API development documents.

　　In addition to the significant improvement of long text processing capabilities, the latest Kimi has strengthened its own context window, lossless memory function and multi-language support, which makes Kimi also perform well in a variety of use scenarios such as online search and information collection, data processing, coding and simulated dialogue, opening up a new situation for the application of large models in the "long text era".

　　In fact, not only Kimi, but also Baidu, Alibaba, iFLYTEK and other domestic related companies have also launched upgraded versions of AI models in recent years. Alibaba Cloud released the 100-billion-level parameter model Tongyi Qianwen 2.0, which has significantly improved its ability to understand complex instructions, create literature, general mathematics, and memorize knowledge. iFLYTEK upgraded the iFLYTEK Xinghuo cognitive model V3.0 and started the training of the Xinghuo large model with larger parameters. On November 10, 2023, China Telecom launched the upgraded 100 billion parameter large model Xingchen semantics, which focuses on solving the problems faced by tens of billions of parameter semantic models in the process of commercialization, such as illusions and multiple rounds of logical reasoning, mainly focusing on improving the ability of image and text generation, image and text understanding, and improving the ability of Chinese image understanding and generation by 30%.

　　The industry model has also attracted many companies to increase their layout. China Mobile, together with China Information Technology, TravelSky, China Aviation Oil and other central enterprises, released the "Jiutian Zhongqing Pedestal Model", which integrates the expertise of eight major industries such as communications, energy, steel, construction, and transportation, and will build a large industry model and create intelligent applications for enterprises "tailor-made". China Telecom has launched more than 10 industry models, covering education, government services, transportation, finance, tourism and other fields, among which the education model can reduce the workload of teacher evaluation by 70%, and the grassroots governance model can improve official document writing by 6 times.

　　In November last year, the 2023 World Internet Conference Wuzhen Summit was held in Wuzhen, Zhejiang, showcasing 15 internationally representative projects that won the "Leading Technology Award", including Baidu's key technology of knowledge-enhanced large language models, IBM's generative artificial intelligence watsonx and other large models, and AIGC's new achievements.

　　Huawei's Pangu Weather Model is the world's first AI model with an accuracy that surpasses traditional digital forecasting methods, and can complete global weather forecasts for the next 24 hours in just 1.4 seconds, more than 10,000 times faster than traditional forecasts. In addition, the intelligent cabin model and intelligent cabin development tool chain developed based on the Wenxin large model can enable the car to have higher-level intelligent driving capabilities and more powerful self-learning and memory functions, change the way people interact with the vehicle, and improve the user experience.

　　In China, the scale of the core industry of artificial intelligence in mainland China is growing, with more than 4,400 enterprises, and innovative achievements such as smart chips, development frameworks, and general large models are constantly emerging. From accelerating the research and development of new drugs and accurately forecasting the weather, to shortening the delivery cycle of factory products and improving office efficiency, more and more industry models are accelerating the implementation. Up to now, the artificial intelligence model developed in China has shown great application potential in the fields of smart mining, drug research and development, meteorology, government affairs, finance, intelligent manufacturing, and railway management. According to the "2023 China Generative AI Enterprise Application Research" released by CCID Consulting, it is expected that the adoption rate of generative AI in Chinese enterprises will reach 85% by 2035.

　　At present, the development of the mainland model is closer to the industrial end, and the current large model has been applied to manufacturing, energy, electric power, chemical industry, transportation and other industries and fields, and generative AI empowers thousands of industries, which will create unprecedented business value for enterprises.

　　A new future for digital governance

　　Technology is the permanent driving force that drives humanity forward. The new innovative force of AI large models has rapidly pushed people's fantasies of general artificial intelligence into reality. However, as generative AI shows increasing creativity and influence, how to standardize guidance and scientific governance has also become an important issue of the times.

　　Dong Jian, deputy director and senior engineer of the Information Technology Research Center of China Electronics Standardization Institute, said that the excellent information throughput, knowledge reserve and human-computer interaction ability of the large model have shortened the distance between humans and machines, and also made it outstanding in the empowerment of medical health, creative industries and cultural education, attracting a lot of attention from the investment community and the R&D community.

　　Dong Jian believes that as a general enabling technology, the large model can be combined with most industries to promote intelligent upgrading and give birth to new models, new formats and new industries. Referring to the development cycle of artificial intelligence driven by deep learning since 2017, the industrialization of large models is expected to go through three stages: technology explosion, application exploration and industrial maturity. In the application exploration period, after the explosion of technology, the industrialization process of its technology and products has accelerated, the breadth and depth of enabling applications have been strengthened, and it has been further integrated with society and culture, and its application in the fields of medical care, education, and the Internet may show an explosive trend. During this period, the practice of large models will greatly amplify the potential ethical issues of artificial intelligence. When the industry matures, large-scale model technology may enter a bottleneck period, and the application scenarios and business models of various industries empowered by it will tend to stabilize. With the increase in public acceptance, the focus of market exploration is expected to shift to high-end intelligent products such as brain-computer interfaces and humanoid robots, or collide and integrate with new technologies and industries.

　　Lu Quan, president of the Institute of Consumer Artificial Intelligence, said that at present, the commercialization of domestic large models is accelerating, and participants in general basic large models and industry large models are looking for optimal solutions in their respective tracks. He believes that the general model needs to do a good job in five directions of technical governance from the beginning, namely, security system, standard system, compliance inspection, illusion detection, and dynamic evaluation mechanism. Compared with general scenarios, the rules of subdivided scenarios are clearer, the data quality is higher, and the standards are easier to unify. At the same time, in order to ensure security and compliance, the large model in the industry domain even does not hesitate to lose part of the performance, such as restricting the large model to only generate knowledge content within the scope of a given document content.

　　In the future, with the continuous development of AI technology, AI will become more and more powerful, and the world needs to take precautions and work together to ensure that AI technology is more inclusive, safe, reliable, and controllable for the benefit of all mankind.

　　("Xiaokang", exclusive article of China's well-off network)

　　This article was published in the late April 2024 issue of Xiaokang