news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

Yesterday, the "Beijing Action Plan to Promote "Artificial Intelligence +" (2024-2025)" (hereinafter referred to as the "Action Plan") was released. Beijing Youth Daily reporters learned that the city will rely on the capital's advantageous industry resources and scientific and technological innovation capabilities to organize and implement a number of comprehensive and benchmark artificial intelligence major application projects in five fields, including robotics, education, medical care, culture, and transportation. By the end of 2025, the city will implement 5 benchmark application projects that benchmark the world's leading level, organize 10 national demonstration application projects, and promote a number of commercial application results with broad application prospects, striving to form 3-5 advanced, usable, and autonomously controllable basic large-scale model products, 100 excellent industry large-scale model products, and 1,000 industry success cases.

Lin Jianhua, deputy secretary of the party group and deputy director of the Municipal Development and Reform Commission, introduced that the core output value of the city's artificial intelligence industry will exceed 250 billion yuan in 2023. In the first half of this year, 7,348 PFLOPS of public intelligent computing were added, with a total scale of over 20,000 PFLOPS. At present, the city has registered and launched 82 large models, accounting for more than 40% of the country (183 models), ranking first in the country. In order to seize the opportunity, take advantage of the trend, and accelerate the application of large models of artificial intelligence, the "Action Plan" came into being.

Organize and implement "benchmark demonstration" in five major areas

In terms of highlighting demonstration and leadership, the Action Plan proposes to rely on the capital's advantageous industry resources and scientific and technological innovation capabilities, organize and implement a number of comprehensive and benchmark major application projects in five fields, including robotics, education, medical care, culture, and transportation, promote breakthroughs in core technologies of large models, enhance artificial intelligence engineering capabilities, improve the scientific and technological level and service quality of key industries, and form a new ecology for large model industry applications. At the same time, around 10 industry segments, including scientific research exploration, government services, industrial intelligence, financial management, spatial computing, digital marketing, judicial services, radio and television media, power security, and content security, support the city-level industry authorities, relevant districts, industry application companies, and large model companies to work together to break through the common difficulties in scene implementation and explore standardized, replicable, and popularizable large model industry application implementation paths.

The field of "artificial intelligence + education" will focus on gathering Beijing's high-quality teaching cases, handouts, teaching methods and other teaching knowledge, and cultivate an interdisciplinary and cross-grade education big model platform; "artificial intelligence + medical care" will build a "Beijing Doctor" medical big model platform that integrates medical supervision mechanism innovation, balanced benefits of hospital and doctor scientific research results, and intelligent upgrades of hospital information systems; "artificial intelligence + culture" will gather high-quality cultural and tourism data such as literary works, historical buildings, cultural relics, and scenic spot information to support the cultural big model service platform and provide citizens and tourists with personalized, efficient and convenient cultural and tourism services; "artificial intelligence + transportation" will generate simulation data such as roads, vehicles, pedestrians, and weather based on big models, accelerate autonomous driving simulation training, optimize the integrated vehicle-road-cloud-network technology route, and build an intelligent, efficient and safe urban transportation network.

In terms of commercial applications, Lin Jianhua mentioned that the city will start with small incisions and real scenarios, focus on industry hotspots and social concerns, give full play to the enabling role of large-scale model technology innovation in industry applications, and rely on industry enterprises to cultivate a batch of large-scale model commercial applications. Tang Jianguo, chief economist of the Municipal Bureau of Economy and Information Technology, introduced that the city focuses on subdivided application scenarios in industries such as education, medical care, and culture, supports large-scale model enterprises, system integration service providers, and industry users to carry out large-scale model API interface calls, cloud model tuning, and model privatization deployment, and encourages the development of large-scale model applications such as intelligent bodies and intelligent assistants; supports large-scale model enterprises and equipment manufacturers to work together to tackle key problems, and carry out embedded large-scale model components and intelligent system research and development applications around new hardware terminals such as computers, mobile phones, home appliances, and automobiles.

The city will regularly solicit joint R&D platforms

The Action Plan also mentions the need to build a joint R&D platform and plan to build a number of joint R&D platforms for artificial intelligence application scenarios. By integrating industry resources and high-quality data, opening up industry scenario requirements, and building a joint R&D environment, it will attract and gather advantageous innovation teams to jointly promote the application of artificial intelligence in the industry and the implementation of the industry.

"At the Zhongguancun Forum in April this year, the collection of joint R&D platforms for artificial intelligence application scenarios in Beijing was launched, and the construction of the first batch of platforms is currently being organized. The platform will serve as a bridge connecting industry users, enterprises, universities and research institutions, providing integrated services such as technology research and development, scenario verification, results transformation, and talent training, and opening up the 'last mile' of artificial intelligence application implementation." Han Jian, director of the Information Technology Department of the Municipal Science and Technology Commission and the Zhongguancun Management Committee, introduced the construction of the city's joint R&D platform and the conditions for platform application. The industry users who take the lead in applying need to have a high industry influence in their respective industries and accumulate a certain amount of high-quality industry data. Through their accumulated high-quality industry data, core business scenarios and computing power and other resources in related industries, they support large-model innovation enterprises to carry out model tuning and product verification. The relevant application results are applied first within the leading unit, which can effectively form a demonstration effect and quickly radiate to the entire industry, which has a strong driving and demonstration effect on promoting the implementation of new products and new applications. The fields involved include but are not limited to government affairs, finance, industrial upgrading, medical care, cultural tourism, education, smart cities, etc. The Municipal Science and Technology Commission and the Zhongguancun Management Committee will organize industry experts to comprehensively consider the industry influence, application prospects, data governance and other aspects of the applicants and provide support based on merit.

"In the next step, the Municipal Science and Technology Commission and the Zhongguancun Management Committee will regularly solicit joint R&D platforms across the city. According to the principle of 'mature one batch and start another batch', support policies such as scientific and technological planning projects, finance, talents, parks, intellectual property rights, exchanges and cooperation will be provided to the platform co-construction units to promote the construction of platform application scenarios, complete the transformation of scientific and technological achievements, and create demonstration applications for the large model industry." Han Jian introduced.

Support the application of basic large models in various industries

The Action Plan proposes six safeguard measures around the core elements of application development such as computing power, data, algorithms, funds, platforms, and talents, as well as key links and difficulties. Among them, in terms of resource guarantee, it is clear that the city will build and operate the Beijing computing power interconnection and operation service platform to provide convenient and ubiquitous computing power support for enterprise large-scale model training and user unit large-scale model deployment. Relying on the Beijing Data Basic System Pilot Zone, create a safe and reliable data space to guide enterprises and institutions to open and gather high-value industry data. Build a data training base to provide computing power, data, development tools, open source communities and other resources for large-scale model training. Promote data classification and grading management and "regulatory sandbox" mechanism. Support the promotion and application of basic large models in various industries and fields, encourage the use of self-controlled basic large models as the base to accelerate the training of vertical large models in subdivided industries, and improve the large model application tool chain. Encourage open source high-parameter self-controllable basic large models, support the construction of model and data set hosting cloud service platforms, and promote developer sharing and collaboration.

"We must fully adapt to the current rapid evolution of large model technology, and not only grasp the rapid iteration of basic models, but also promote the rapid iteration of application scenarios." Lin Jianhua mentioned that the implementation period of the "Action Plan" is this year and next year, which is actually one and a half years. This is precisely because we need to speed up the implementation of applications. At the same time, as the intelligence level of basic models continues to improve, application models and paths must also be rapidly iterated and expanded. We will continue to summarize and improve relevant support measures in the future. In the next step, the Municipal Development and Reform Commission will work with relevant units to do its best to interpret policies and provide supporting guarantees. It will selectively support a certain proportion of the benchmark demonstration application scenarios specified in the "Action Plan". Through dynamic reserves and rolling implementation, we will plan a batch, mature a batch, and implement a batch, so as to promote the innovative application of large artificial intelligence models in this city with more diverse and rich scenario construction. Text/Our reporter Wu Wenjuan