news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

Smart Things
Author: Li Shuiqing
Edit Heart

In 2024, it is quite exciting to count the breakthroughs that China has made in the field of large models:

Baidu Wenxin Yiyan, Zhipu GLM-4, etc.Domestic ModelIn the global authoritative testBeating GPT-4; Dark Side of the Moon and other three domesticLarge model unicornbreak through20 billion yuanValuation "mark";"Domestic Sora" dark horseAishi Technology, Shengshu Technology, etc. have updated the duration of AI video generation to6 seconds, 14 seconds, 30 seconds; Wanka cluster breaks out of Moore thread, etc.Domestic GPU dark horse, supporting the training of trillion-parameter AI models...

All of these are related to Beijing, my country’s “first AI city”.

In July this year, the relevant person in charge mentioned at the 2024 Global Digital Economy Conference Artificial Intelligence Forum that as of then, 166 large-scale model products had been registered nationwide, includingBeijing has 71, accounting for nearly half of the countryAccording to public information, Beijing has 20% of the country’s large-scale model unicorns.80%At the same time, these large models have been accelerated to land in many fields such as Internet, education, medical care, finance, and government.

▲2024 Global Digital Economy Conference Artificial Intelligence Forum

At the end of July, Zhidongxi counted the financing of over 10 million yuan by companies related to my country's large-scale model industry in the first half of 2024 and July. The results showed thatBeijingThe company has obtained43 strokesFinancing,ShanghaiandShenzhenThey are11 pens, 9 pens, showing a clear gap. ("Big Model Hot Money Surge: Hundreds of Millions of Yuan in Half a Year, the Highest Exceeding 50 Billion Yuan")

Various signs indicate that Beijing is becoming China's "first big model capital".

From the start to the lead, it took only a little over a year for Beijing's big model industry to break through. In May 2023, Beijing issued "Several Measures for Promoting the Innovation and Development of General Artificial Intelligence in Beijing", announcing 21 measures such as "Carrying out Research on Innovative Algorithms and Key Technologies for Big Models", becoming the first city in the country to fire the first shot in the big model.

More than a year later, Beijing has formed a big model "roadmap" that comprehensively covers the entire industrial chain of big models, AI computing power, data, applications, solutions, and the integration of industry, academia and research.In Beijing, you can not only see China, but also the world.

Behind every "first", "strongest" and "heavyweight" progress in Beijing, China's large-scale model industry has moved from catching up to leading. Why Beijing? How did Beijing occupy the C position in the "100 Model Wars"? Let's explore together.

1. Nearly half of the large models registered in China, and the top large models are made in Beijing

Entering 2024,Beijing is rich in large modelsThe strength of the company has been continuously confirmed.

At the Second Beijing Artificial Intelligence Industry Innovation and Development Conference in March, relevant officials from Beijing mentioned that the number of AI-related companies in Beijing is about2200, accounting for about40%；

At the 2024 Global Digital Economy Conference in July, Beijing announced that the number of large models that have been put online through filing reached71, close toNational total166half。

Earlier in October 2023, the "Beijing Artificial Intelligence Industry Big Model Innovation and Application White Paper (2023)" showed that my country1 billion parameter scaleAbout half of the above large model manufacturers are in Beijing, with a total number of122。

Star model products and companies "Made in Beijing" are appearing one after another in China's technology circle.

Technology giants have launched a "100-model war", with Beijing becoming the most cutting-edge innovation base.Major technology companies such as Alibaba, Tencent, Baidu, ByteDance, Huawei, 360, JD.com, Xiaomi, and NetEase have all announced the latest progress in large-model research and development and applications. Large-model e-commerce, large-model cars, and new AI mobile phones have become popular trends.

For example, in October 2023, Baidu launched the 4.0 version of the Wenxin Big Model, which is benchmarked against GPT-4. More than 10 AI native applications and big model hardware were subsequently launched. By the end of the year, the user scale had exceeded 100 million. This year, it has upgraded many applications such as AI programming assistant, digital human platform, intelligent customer service platform, and knowledge management platform.

The "Five Tigers of Entrepreneurship" of domestic large-scale models have emerged and become unicorns with a valuation approaching 20 billion yuan.Among them, Zhipu AI, Baichuan Intelligence, Dark Side of the Moon and Zero One Everything are all registered in Beijing., MiniMax is located in Shanghai.

Founded in 2019Zhipu AIIn January 2024, the company launched the next-generation large-scale base model GLM-4, with overall evaluation results close to GPT-4; in July 2024, it launched the "Chinese version of Sora" Qingying, which supports the generation of 6-second videos. The company received more than US$400 million in financing this year, and its latest valuation exceeded 20 billion yuan.

Founded in 2023 by former Sogou CEO Wang XiaochuanBaichuan Intelligence, has launched 12 large models including Baichuan-7B/13B, Baichuan2-7B/13B, Baichuan2-192K, Baichuan 4, "Pengcheng-Baichuan·Naohai33B", etc. The company just completed a 5 billion yuan round of A financing in July this year, and its valuation also exceeded the 20 billion yuan "mark".

Founded in 2023Dark Side of the MoonWith its unique "To C" strategy, Kimi has been built into an AI hit product with over 20 million visits per month. The company received over US$1 billion in financing in February 2024, setting a new record for the largest single financing in the domestic AI field. The latest valuation is as high as US$3 billion, which is over RMB 20 billion.

An AI 2.0 company incubated by Dr. Kai-Fu LeeZero One EverythingIn March 2024, it open-sourced the Yi-9B model and released the Yi Model API open platform. Recently, it has also been reported that it has completed a new round of financing worth hundreds of millions of dollars.

Some large-model "new seedlings" have not yet joined the unicorn camp, but have become the "kings" of the subdivided track after only being established for about a year.

In the popularAI video generation field, led by Wang Changhu, former head of ByteDance’s visual technologyAishi TechnologyAs early as January 2024, the overseas version of PixVerse, a Sora-like video generation product, was launched, led by Zhu Jun, a professor at Tsinghua University.Biodata TechnologyIn April 2024, Vidu, China's first original, fully self-developed long-duration, highly consistent, and highly dynamic video model, was launched, which is ahead of OpenAI's "futures" product Sora.

In the much-watchedAI Infrastructure (AI Infra) Field, founded by Yuan Jinhui, former founder and CEO of OneFlowSilicon-based flow, which just launched the one-stop large model API cloud service platform SiliconCloud in June. It is reported that its self-developed SiliconLLM large model inference engine generates more than 10 times faster than similar open source products.

Joint innovation between industry, academia and research has become a trend.For example, in the field of general large models, the Beijing Academy of Artificial Intelligence and China Telecom jointly released the world's first low-carbon, high-performance open source multi-language large model Tele-FLM in July 2024, and announced that they will release large language models with hundreds of billions, 600 billion, and trillion parameters in the future.

The application of large models in the industry has begun to show results.In June 2023, Beijing released the first batch of 10 typical application cases of large industry models.Baidu, Zhipu Huazhang, Institute of Automation, Chinese Academy of Sciences, iFlytek, Unisound, Institute of Scientific Intelligence, 4Paradigm, Haomo Zhixing, Xianyuan, Mianbi IntelligenceApplication cases from 10 technology suppliers were selected, coveringEnergy, medical care, finance, consumption, smart driving, government, real estateAnd other industry fields.

▲The first batch of 10 typical application cases of large-scale models in Beijing

In July 2024, the Beijing Industry Large Model Innovation and Application Competition was awarded, and 6 companies won the first prize, namelyRuilai Technology (government affairs), Zhongke Wenge (finance), Ruibo (industry upgrading), Shukun Technology (medical care), Zhongke Brain (culture and education), Zhongke Ruijian (smart city)。

In just one year, the number of Beijing's main large models has increased from40, growing to the present July 202471Major technology companies have laid out the entire industry chain of large-model computing power, algorithms, and applications. Dozens of unicorns and leading startups in niche sectors have surpassed their international counterparts. The integration of industry, academia, and research has promoted a surge in application results, and the country has taken the lead in forming an AI large-model cluster.

Beijing has achieved Chinese speed and Chinese strength in the development of the large model industry.

2. Four E-level intelligent computing centers launched, 36,000P intelligent computing power activates data elements

Unlocking Beijing’s “model” power requires the “key” of computing power.

As of December 2023, Beijing has formed12,000PThe computing power supply scale of 10 ...

What does this mean? 1P refers to a computing speed of 100 trillion times per second.100P is equivalent to the computing power of 50,000 high-performance computers. For example, if you want to locate a certain characteristic star in a star map of 200,000 celestial bodies, it only takes 100 seconds.It is easy to imagine what kind of "model" power 12,000P will bring to Beijing.

Entering 2024, Beijing's intelligent computing infrastructure construction remains strong, providing continuous energy for the development of the large model industry.

In January 2024, in Haidian DistrictBeijing Artificial Intelligence Public Computing Platform (Shangzhuang)The first phase of 500P computing power has been officially launched, providing inclusive computing power services for the development of the large model industry.

In March 2024, as the largest public computing platform in Beijing,Beijing Yizhuang Artificial Intelligence Public Computing PlatformAfter official launch, it can provide 3000P computing power, which is equivalent to the computing power of more than one million high-performance computers.

The district will issue "100 million computing power coupons + 100 million model coupons" every year to serve enterprises; looking to the future, the Economic and Technological Development Zone where the platform is located has planned a computing power scale of over 10,000P.

In May 2024, in Chaoyang DistrictBeijing Digital Economy Computing Center Project Infrastructure ConstructionConstruction has started and is expected to be completed by the end of 2024. After it goes into operation, it will provide a cumulative intelligent computing power of 2000PFlops.

▲Rendering of Beijing Digital Economy Computing Center

According to the latest "Beijing Computing Infrastructure Construction Implementation Plan (2024-2027)", Beijing plans to build4 E's(EFlops, hundreds of trillion floating point operations per second)High-level intelligent computing powerBy 2027, the new intelligent computing power will reach36,000P。

Beijing not only provides impetus for the development of the large model industry from the computing power level, but also provides sufficient "fuel" in terms of data elements.

Data from the Beijing Municipal Bureau of Economics and Information Technology shows that after the Beijing International Big Data Exchange obtained its license, as of November 2023, the exchange has issued 27 data asset registration certificates, with a data transaction scale of more than2 billion yuan。

Behind the scenes, Beijing took the lead in creating the first data infrastructure system pilot zone in the country, and released a plan and policy list for the creation of the data infrastructure system pilot zone; it introduced innovative incentive measures such as the first registration, first transaction, first entry into the table and first opening of data assets, and continued to release policy dividends.

In March 2024, Beijing will openThe first AI data training base in ChinaAt the same time, the "regulatory sandbox" system is used to ensure the training of large models. The AI ​​data training base and Beijing computing power infrastructure have been combined to greatly help companies in the field of large models to develop the "strongest brains".

It can be seen that by grasping the three major elements of models, computing power and data, Beijing has grasped the "bull's nose" of the development of the large model technology industry.

3. After 30 years of AI evolution, Beijing has become China's "first capital of large models"

How did Beijing gradually become the capital of AI? This story began more than 30 years ago.

The rise of Beijing’s AI industry has a long historical origin.The Chinese Academy of Sciences' research on artificial intelligence began in 1987, Microsoft Research Asia was established in 1998, and the famous Tsinghua Yao Class was founded in 2005... They have created batches of talents for the development of AI in Beijing. Along with the first wave of AI waves set off by SenseTime, Megvii, and AI Research Institute in the early 2010s, they have also become the mainstay of the development of the current large model industry in the 2020s.

With the maturity of AI technology industry represented by computer vision,2020 is the year of explosive growth for AI in China, and also the first major explosion of AI in Beijing.

According to the "2020 Beijing Artificial Intelligence Development Report" released by Beijing Academy of Artificial Intelligence, the scale of Beijing's artificial intelligence-related output value reached 170 billion yuan in 2019; the total financing of AI startups exceeded 300 billion yuan by the end of June 2020, and 32 artificial intelligence unicorn companies such as SenseTime and Megvii were spawned; the number of AI-related papers published and patent applications ranked first in the country, and a number of major original achievements with international influence were produced, ranking first in the country.

This requires Beijing to hit the right milestones in its AI programmatic policies.

In September 2017, the Zhongguancun Science Park Management Committee took the lead in issuing Beijing's first AI policy, the "Zhongguancun National Independent Innovation Demonstration Zone Artificial Intelligence Industry Cultivation Action Plan (2017-2020)". In December, Beijing issued the "Guiding Opinions on Accelerating Technological Innovation and Cultivating the Artificial Intelligence Industry in Beijing", which provided a policy basis and support for the development of Beijing's AI industry. In 2019, Beijing was approved to establish the first national new generation artificial intelligence innovation and development pilot zone, which also provided conditions for its AI explosion.

Taking advantage of the situation, with large-model applications represented by ChatGPT becoming popular around the world, 2023 will be the year of evolution for Beijing's AI industry.

1. Fire the first shot of the national AI big model policy, 100 billion yuan to support the investment and financing environment

In terms of the programmatic policies for big AI models, Beijing acted early, did not hesitate, and did not wait and see.

On May 30, 2023, Beijing took the lead in firing the first shot of the local big model policy.The Beijing Municipal Government issued two major policies, "Implementation Plan for Accelerating the Construction of an Artificial Intelligence Innovation Source with Global Influence" and "Several Measures for Beijing to Promote the Innovation and Development of General Artificial Intelligence", proposing 21 measures such as strengthening the collection of large model training data and the development of governance tools, setting the tone for the development of Beijing's big model industry.

In September 2023, Haidian District of Beijing took the lead in prototyping.The district has released the "Implementation Plan for Innovation-led Development of General Artificial Intelligence in Zhongguancun Science City" (referred to as the "Plan"), which proposes to build a core industry scale of 230 billion yuan and provide talent subsidies in many aspects such as talent settlement, financial subsidies, and public rental housing.

In April 2024, Beijing launched another major policy move.The National Development and Reform Commission and other departments further jointly issued the "Several Measures of Beijing on Accelerating the Development of the General Artificial Intelligence Industry" (hereinafter referred to as the "Measures"). The measures focus on improving the supply of intelligent computing power, accelerating the innovative application of large models, and other 10 measures. They also provide big model companies with a "gift package" of real guns and bullets from all aspects of computing power, models, data, applications and environment.

These "gift packages" include but are not limited to: the highest reward for AI technology research projects included in major national strategic tasks100 million yuan, the highest support for projects that create benchmark applications of artificial intelligence50 million yuan, investment in the next five years will exceed100 billion yuanOptimize the investment and financing environment through financial subsidies and issuanceComputing power coupons, model couponsetc. to reduce the operating costs of large-scale model enterprises.

2. Create 45EFLOPS intelligent computing supply, and domestic computing power supports trillion-level parameter large model training

Beijing maintains its previous strong investment in large-scale model infrastructure construction.

In April 2024, the Beijing Municipal Bureau of Economy and Information Technology and the Beijing Municipal Communications Administration issued the "Beijing Computing Infrastructure Construction Implementation Plan (2024-2027)" (hereinafter referred to as the "Plan"). The Plan proposes that by 2025, the scale of Beijing's intelligent computing supply will reach45EFLOPS; By 2027, realize intelligent computing infrastructure software and hardware productsFull stack autonomous controlThe overall performance has reached the domestic leading level, and it has the ability to build a 100% autonomous and controllable intelligent computing center.

Beijing also announced that it will create more than 200 industry benchmark cases driven by intelligent computing power, focus on building E-level intelligent computing power highland in Haidian, Chaoyang, Yizhuang, and western Beijing (Shijingshan and Mentougou), and give priority to accelerating the construction of two 10EFLOPS large-scale intelligent computing clusters.

But Beijing’s computing power map does not stop there.Beijing-Tianjin-Hebei-Mongolia Computing Power Supply Corridor, planning to build a super-large-scale intelligent computing cluster that supports the training needs of large models with trillions of parameters, so as to accelerate the commercialization of large artificial intelligence models.

It is worth mentioning thatAccelerating the promotion of independent control of core software and hardware products will be a key task.

In the plan, Beijing proposed that for enterprises that purchase independently controllable GPU chips to provide intelligent computing services, the implementation plan proposes a guarantee measure of providing support in accordance with a certain proportion of the investment amount, so as to accelerate the realization of independent and controllable supply of intelligent computing resources.

In May's "Beijing Implementation Plan for Accelerating the Construction of an Artificial Intelligence Innovation Source with Global Influence (2023-2025)", Beijing also clearly proposed to promote breakthroughs in domestic artificial intelligence chips. It will actively guide large-scale model R&D companies to use domestic artificial intelligence chips and accelerate the improvement of the localization rate of artificial intelligence computing power supply.

Beijing’s real investment in domestic computing power is also a “big deal”.

For example, in October 2023, the Beijing Municipal Bureau of Economy and Information Technology issued the "2023 Beijing High-end Industry Development Fund Implementation Guidelines (Second Batch)" last night, proposing to focus on supporting the first round of tape-out rewards for integrated circuit design products, and the annual reward amount for a single enterprise shall not exceed30 million yuan。

3. Build a nest to attract talents, mobilize the national scientific and technological forces to "form a bureau", and activate the government, industry, academia, research and funding ecology

The ecological integration of government, industry, academia, research and finance is the source of vitality for the development of Beijing's large model industry.

To build a nest to attract the phoenix, Beijing has allocated a number of plots of land in the capital, where every inch of land is valuable, to develop the AI ​​industry.

September 2023,Zhongguancun Artificial Intelligence Large Model Industry Cluster"The unveiling ceremony was held to build four major industrial parks, namely Wudaokou Artificial Intelligence Industrial Park, Peking University Ximen Artificial Intelligence Industrial Park, Zhongguancun West District Artificial Intelligence Industrial Park, and Tsinghua Science Park, covering an area of ​​670,000 square meters, which provides a broad physical space for the grand blueprint of Beijing's large model industry.

July 2024,Artificial Intelligence Innovation DistrictConcept released. Beijing plans to build an AI innovation district in the 53 square kilometers of southern Haidian, connecting 52 national key laboratories, 10 new R&D institutions, 106 national scientific research institutions, 37 top universities, 89 AI2000 global top scholars, 12,300 AI scholars, 1,300 AI companies, more than 90 big model-related units, and 37 registered big model units.

Through a series of partnership programs and alliances, Beijing has brought many AI startup spaces to life.

In May 2023, Beijing releasedGeneral Artificial Intelligence Industry Innovation Partner ProgramThe first batch of 39 members was announced that month, and the second batch of 63 companies was announced in July, including computing partners such as Alibaba Cloud and Beijing Super Cloud Computing Center, data partners such as Beijing Big Data Center, Beijing Academy of Science and Technology, and Beijing Zhiyuan Artificial Intelligence Research Institute, model partners such as Baidu, Zhipu AI, Alibaba Damo Academy, and Kunlun Wanwei, application partners such as Beijing Municipal Government Services Administration, Bank of Beijing, and Kingsoft Office, and investment partners such as IDG Capital, CDH Innovation, and Qi Ji Chuangtan.

July 2024,Zhongguancun Artificial Intelligence Industry AllianceThe alliance was established and launched the General Artificial Intelligence Industry Innovation Ecosystem Plan. The alliance members include Beijing General Artificial Intelligence Research Institute, China Academy of Information and Communications Technology, Peking University, Tsinghua University, Renmin University of China, Institute of Automation, Chinese Academy of Sciences, Zhiyuan Research Institute, Haitian Ruisheng, Zhipu AI, Tencent, CRRC, Baidu, Baichuan Intelligent, Zhongguancun Science Park Company and hundreds of other universities and institutions.

It can be seen that in order to embody the market-oriented, open and cooperative docking mechanism, Beijing has mobilized the investment, industrial and academic circles of most of China to jointly promote the development of the large model industry.

The results achieved through these actions are not just a show of strength, but a prototype for the whole country, and the results will feed back into the development of the national large-scale model industry and digital economy.

Conclusion: Beijing takes the center position in the “100 Models Competition” by creating a benchmark for the digital economy

From digitalization to intelligence, Beijing is once again at the forefront of the leap forward.

In 1994, China first connected to the Internet. Beijing took the lead in establishing the Municipal Information Office in 1997 and formulated a long-term three-step goal of "Digital Beijing" in 1999: from 2000 to 2010, the basic framework of "Digital Beijing" will be formed. From 2011 to 2020, the digital level of the capital will be improved. From 2021 to 2050, Beijing will be built into a modern international digital city.

Leading the trends of Internet broadband, high-definition digital television, and riding on the trends of e-commerce, digital logistics, and digital government, it has created the historic launch of the "Window of the Capital" website, the popularization of municipal transportation cards, the 2008 Digital Olympics, and the opening of 3G network services... According to official data, the added value of Beijing's digital economy reached 871.9 billion yuan in 2015, and by 2022 it had grown to 1.7 trillion yuan, becoming one of the main engines of economic growth.

As an important sub-topic of digitalization and a representative of new productivity, AI has not only taken root in the fertile soil of Beijing's digitalization, but also sprouted under the positive AI special policies. With a multi-pronged approach of policies, industries, research, capital, and applications, Beijing has become the leading artificial intelligence industry talent center in China and even the world. Behind China's first city for large models, Beijing's moat seems to be getting deeper and deeper.