On the afternoon of September 3, the three-day 2022 World Artificial Intelligence Conference (WAIC) officially ended. As a summit representing the trend of the global artificial intelligence (AI) industry, compared with the previous four wonderful “Double Horses” (Jack Ma, Musk) to have a dialogue on the same stage, the “AI Four Little Dragons” (Shangtang, Yuncong, Megvii) gathered together. , Yitu), Mr. Kai-Fu Lee talks about AI, which is hard to see in the past two years.
Due to the epidemic, Shanghai WAIC was postponed for two months this year, and under the comprehensive promotion of the Metaverse economy, Shanghai has made AI technology a “metaverse” concept, and made the computing power chip the “traffic password” of the conference. , more than 200 companies participated in the exhibition, and more than 100 forums discussed more extensive topics, involving cloud computing, big data, privacy computing, biomedicine, rule of law ethics, brain-computer interface and other fields.
At the same time, during this year’s WAIC, there were two internal and external incidents, which added more heated discussion space to the participating guests:
Second, the World Metaverse Conference held in Beijing was held two weeks ago, and the China International Service Trade Fair (“Service Fair”) was also held in Beijing on August 31. Therefore, everyone began to pay attention to the first-tier cities in Beijing and Shanghai. Metaverse”, chips and other high-tech industries.
However, the concept of “metaverse” is vague. How to closely integrate AI + metaverse and realize commercialization seems to be an important issue.
“We pay more attention to the development value and increment of the ‘Metaverse’ (it) to the entire industry, rather than the technological development of the ‘Metaverse’ itself. JD.com is still a relatively pragmatic company, and we will not be a Metaverse community or games. This kind of thing is about retail-related industrial scenarios.” Cao Peng, senior vice president of JD.com, told Titanium Media App.
A bank executive even said bluntly at a forum round table: “‘Metaverse’ is just a show, just a hot publicity.”
Chips and robots are the “traffic password” of AI
“From a big logic point of view, I think some things are irreversible once they happen. Although there is news that the license can be re-licensed, we found that the entire industry chain is tense.” Black Sesame Intelligence Chief CMO (Marketing Officer) ) Yang Yuxin said at the Huatai Securities sub-forum that the challenges and opportunities brought about by this incident are obvious: the opportunity is that car companies will have obvious demands for localization alternatives in the future.
“After all, it is startup companies that are picking AI chips in China. Compared with large companies, the time, capital and energy invested can be counted. So I think this is a coexistence of opportunities and challenges. But in general, we may have to put more emphasis on I’m so excited.” Yang Yuxin said.
Titanium Media App entered the first floor of the WAIC exhibition hall and found that in addition to Alibaba, Tencent, Huawei and other big companies, domestic AI/GPU chip design startups with large computing power have become important highlights of the conference:A number of domestic AI chip companies such as Tianshu Zhixin, Biren Technology, Suiyuan Technology, Hanbo Semiconductor, Denglin Technology, Cambrian, Horizon, and Aixin Yuanzhi have exhibited the latest technologies and products, advanced process chip products “A dazzling array of things”, and each company is also “competing for beauty”.
September 3,Suiyuan TechnologyReleased high-performance AI-accelerated cluster server products: CloudBlazer POD, which has built-in “Suisi” 1.0 and 2.0 AI chips. It is understood that Suiyuan products are mainly implemented in three major business directions: pan-Internet, traditional industries (finance, transportation, electric power, medical care, industry, etc.) and new infrastructure.
At the same time, downstairs at the Suiyuan press conference, the AI vision chip company “Hanbo Semiconductor”, which was established less than 4 years ago, released the new Hanbo unified computing architecture, the new data center (cloud) AI inference card Zetian VA10, and edge AI inference acceleration Kazaitian VE1, and VastStream extended version of Hanbo software platform. In addition, Hanbo also demonstrated the performance parameters of the domestic 7nm cloud GPU chip SG100, which is used in image rendering, video, “metaverse” and other fields.
Earlier on the first day of the WAIC conference, GPU chip system solution provider “Denglin Technology” released the “Denglin Hanhai” ecological partner program, providing hardware server/industrial computer manufacturers, algorithms, Internet, smart cities, transportation, finance, energy Provide system solutions to other industries, reduce customer migration and import costs, and fill the gaps in technology and products in the domestic high-performance GPU field.
In addition, Titanium Media App also noticed at the booth that Denglin is developing products such as Goldwasser L/XL, a computing card for data centers, and Goldwasser UL, an edge computing module that directly benchmarks against NVIDIA’s Jetson series.
Denglin said that its first-generation GPU products are being commercialized, and the second-generation products will enter the market in the first half of next year.
Suiyuan and Hanbo’s competitors, “Tianshu Zhixin”, are not to be outdone. The company’s CEO disclosed to Titanium Media App and others that Tianji’s first 7nm cloud AI training chip “Tiangai 100”, the cumulative product order amount has exceeded 230 million yuan.
It is reported that Tianshu Zhixin was established at the end of 2015, and officially launched the 7nm cloud general GPU (GPGPU) chip design in 2018. Its first 7nm cloud training product “Tiangai 100” was released in April last year. At WAIC this year, Tianshu Zhixin demonstrated many application scenarios of Tiangai 100, including the Internet, intelligent security, intelligent computing center, and intelligent medical care. The 7nm general-purpose GPU (AI) reasoning chip “Intelligent Armor 100”, which will be released in the fourth quarter of this year, also made its first public appearance and was rated as one of the eight “Treasures of the Town Hall” in WAIC 2022.
The so-called general-purpose GPU means that traditional GPU products cannot meet the computing performance requirements of AI models and algorithms. GPU companies have launched general-purpose GPU products with higher computing power, namely GPGPU.
Opposite the Tianshu Zhixin booth is Biren Technology, which participated in WAIC for the first time.This time, Biren showcased the first general-purpose GPU BR100 and BR103 chips of TSMC’s 7nm process released a month ago, as well as related accelerator cards and servers. It is reported that the BR100 series of general-purpose GPU chips have become one of the eight “treasures of the town hall” selected by WAIC. The characteristic is that the computing power reaches one trillion calculations per second, setting a global record for the computing power of general-purpose GPU chips. Currently, Ping An Technology , Inspur Information, Universal Data, etc. to establish cooperation.
AI vision chip company Aixin Yuanzhi displayed its three-year-old chip products at WAIC. Titanium Media App has learned that the company’s two generations of multiple AI vision processing chips have achieved large-scale commercial use. The latest products are expected to be released at the end of this year as soon as possible, and have formed a series of product matrix covering the high, medium and low-end markets.
In addition to start-up companies, there is also a listed company on the Science and Technology Innovation Board in the field of AI chips, Cambrian.At this year’s WAIC, Cambrian showed its chips and accelerator cards for cloud, terminal, edge, and car. At the beginning of this year, Cambricon released a new training accelerator card MLU370-X8, equipped with Siyuan 370. At present, it has achieved in-depth cooperation with some leading Internet customer scenarios, and its cloud products have formed a certain income scale in Alibaba Cloud and other companies.
According to the financial report, in the first half of 2022, the Cambricon cloud product line achieved revenue of 130 million yuan, while edge products continued to be shipped, and the cumulative shipments over the years have exceeded one million pieces.
Cambricon founder Chen Tianshi revealed that through the holding company Cambricon Xingge, a new generation of cloud Al training chip MLU590, L2+ autonomous driving and parking integrated chip SD5223, and L4 high-end autonomous driving multi-domain fusion platform SoC SD5226 are being developed. He believes that the intelligent computing center in the cloud will become the core infrastructure for autonomous driving research and development. “In the next few years, we will see that the L2+ autonomous driving level will be rapidly popularized and will exist for a long time.”
In addition, robots have also become another highlight of this WAIC.This year, the exhibitors of quadruped robots have changed from Yushu and Youbixuan of WAIC last year to Hangzhou Yunshen Technology.
Someone at the scene stepped on the four-legged robot dog of Yunshen Technology. It can stand up quickly, and can even get up on its own after it falls over. According to Qixinbao, the major shareholder of Yunshen is Zhu Qiuguo, who is an associate professor and doctoral supervisor of Zhejiang University, and is engaged in research on humanoid robots, bionic robots, and machine intelligence.
Dachang cautiously pursues the “metaverse”, and AI continues to be implemented in the industry
A highlight of this year’s WAIC is “Metaverse”.
In the middle of the exhibition hall, the organizer specially set up a “WAIC Metaverse Exhibition Area”, where Alibaba, Baidu, Huawei, NetEase, Kuaishou, SenseTime and other major manufacturers have displayed some VR/AR, digital human and other applications under the concept of “Metaverse”.
However, the big Internet companies seem to be more cautious in promoting the concept of “metaverse”.
Li Qiang, vice president of Tencent, said that Tencent looks at the metaverse more from the integration of data and reality, focusing on the concept of full-real interconnection, rather than from a pure virtual perspective. Currently, Tencent is in cloud computing, blockchain, digital twins, There are layouts in many fields such as quantum computing and holographic projection. “Although it has made a lot of reserves, Tencent now takes a cautious and pragmatic view on the Metaverse.”
Wu Yunsheng, vice president of Tencent Cloud, said bluntly, “Tencent will not set up a dedicated team to be responsible for the Metaverse.”
Ma Jie, vice president of Baidu, said that based on AI technology, Baidu will improve the infrastructure of the Metaverse and provide a platform for the innovation of the Metaverse. In an interview with the media, Ma Jie said that due to too much investment in research and development, Baidu Xiyang is not profitable. “We believe in the metaverse, it takes time, and there are still many technical problems in the infrastructure. We don’t want to make a simple stack, we have to overcome the technical problems inside.”
“Based on the thinking of the future supply chain, JD.com proposed the industry metaverse.” At this year’s WAIC, Cao Peng, senior vice president of JD.com, talked about JD.com’s metaverse layout in an interview with Titanium Media App. In his view, the deep integration of the metaverse and the real economy will also bring new opportunities to existing business models and industrial forms.
In fact, virtual activities seem to have become one of the important scenarios for the implementation of the “metaverse” concept. Liu Bai, head of NetEase Yaotai, said that the repeated epidemics and the explosion of the “metaverse” have further catalyzed the demand for virtual activities, and NetEase Yaotai, an immersive activity platform, has already applied NetEase Cloud Music IPO Conference, Huatai Securities , Peking University Guanghua School of Management, Alibaba Cloud, Huawei Cloud, Naixue’s Tea and other institutions provide online event support.
Kuaishou’s “Metaverse” is closer to the field of virtual live broadcast.At the WAIC booth, Kuaishou demonstrated to the Titanium Media App how its virtual studio assistant forms a “digital virtual human” and synchronizes actions through cameras and algorithms. According to Kuaishou, the exposure of its virtual human anchor support plan will exceed 10 billion.
However, a person in the relevant industry told Titanium Media App that the current laws and regulations related to “virtual human” are not perfect, the review mechanism is more stringent, and both algorithm review and manual review are required.
In contrast, another group of companies, such as Megvii Technology, have tried their best to distance themselves from the “metaverse” boom.
Tang Wenbin, the company’s co-founder and CTO, said, “Frankly, we are not experts in the ‘metaverse’, we don’t really understand it… But I think the word (metaverse) has appeared in the past two years. Inside, it may be a key change.”
Zhou Jingren, vice president of Alibaba Dharma Academy, highlighted his achievements in “big model”. Zhou Jingren believes that through multi-modal pre-training large models and other core AI technologies, the appearance of virtual digital people is close to that of real people, and they can talk and communicate like ordinary people. Multi-modal large models can imitate the process of human construction of cognition. , is expected to become the cornerstone of the next generation of AI algorithms, and may have a wide range of application scenarios in digital humans or “metaverse”.
In addition, Titanium Media App also noticed that companies are looking for new footholds for AI technology this year:
This year, Baidu will focus on strengthening the application of large models and the commercialization of autonomous driving; Megvii will increase the positioning and popularization of its AIoT (Intelligent Internet of Things); Cloud Cong Technology has begun to expand vehicle-road collaboration scenarios; Fourth Paradigm focuses on AI decision-making capabilities Deeply digging into the digitalization of vertical industries; Yuntian Lifei’s WAIC this year puts more emphasis on large-scale algorithm production capacity; iFLYTEK has begun to emphasize the business of smart cars.
At the same time, the AI company Geling Shentong exhibited at the Service Trade Fair for the first time released a new metaverse form – the big universe, which can provide the ability to perceive human movements and postures in large scenes, and perceive six degrees of freedom game equipment. .
The AI big data company “Hehe Information”, which is entering the process of listing on the Science and Technology Innovation Board, demonstrated the recognition ability of Dingwen at this WAIC. It is reported that this solution is to use the complex scene text recognition and natural language processing (NLP) capabilities of deep learning to flatten Zhong Dingwen from the concave state and translate it into simplified characters without manual intervention. The words that are together are correctly distinguished after being automatically processed by the “AI sentence segmentation” function.
Hehe Information, founded in 2006, has more than ten years of experience in the field of intelligent text recognition. Wang Lechen, the relevant person in charge of Hehe Information, told the Titanium Media App and others that the shape of the bell ding is small and dense, and the bronze ding surface of its carrier is curved, reflective, and uneven, making it extremely difficult to identify. Starting from a tripod, this technology can be widely used in the fields of identifying different cards, bills, and handwritten document characters.
It is worth mentioning that AI is also gradually being applied to the field of new drugs for protein synthesis. At the WAIC Biocomputing Forum,The first protein design workbench in China was officially released, which will accelerate the large-scale development and implementation of protein design work.
Dr. Miao Hongjiang, head of XLab, introduced, “The protein world has a huge space and rich functions. The possible sequence of a 100-amino acid protein is 20 to the 100th power – 1.3 × 10 to the 130th power. The obtained natural proteins are only a very small subset of them. There are still many functional proteins in the unknown space that may solve challenges such as medical health, energy and environmental protection that have not been touched.”
Currently, Shanghai is promoting AI technology legislation. On August 30, the General Office of the Standing Committee of the Shanghai Municipal People’s Congress issued a draft of the “Shanghai Regulations on Promoting the Development of the Artificial Intelligence Industry (Draft)” for comments, which is the first provincial-level local regulation in the field of artificial intelligence. The comment period is until September 13.
This draft outlines Shanghai’s various measures to promote the development of the AI industry, including providing financial support, nurturing entrepreneurs and start-ups, strengthening the focus on certain market segments, and establishing industry supply chains. Moreover, Shanghai also allows AI companies to continuously achieve technological innovation under loose management conditions. “Encourage market entities to actively innovate, clarify that relevant departments should formulate regulatory rules and standards that adapt to the characteristics of AI, and establish a list of minor violations that will not be punished.”
Ding Botao, deputy director of the Shanghai Academy of Social Sciences, pointed out that the draft for commentsProvides a source of power for the development of Shanghai’s AI industry。
Since the birth of AI in 1956, AI technology has continued to evolve. It has undergone multiple rounds of changes and has evolved from a single technology product to a scenario application. Since intelligence is the main feature of future Internet integration and innovation, AI+ industry solutions have become an important starting point for the intelligence of the 47 trillion digital economy.
In the past five years, with the WAIC held in Shanghai, the entire AI technology industry has also changed from the initial “altar” boom to more universal, calm and rational, and all enterprises are using AI.
Lin Nianxiu, deputy director of the National Development and Reform Commission, mentioned in the interpretation report of “Made in China 2025″ released in 2016 that the use of AI technology to develop “Internet +” will accelerate core technological breakthroughs and industrial development, enhance core competitiveness, and cultivate and expand emerging industries. business format. At the same time, Internet + AI has a basic supporting role in improving the level of intelligence in various fields of economy and society.
AI has never been more accessible than it is today. There are both source innovation and industrial application, presenting an intelligent digital world that blends virtual and real for everyone. We have never needed AI as urgently as today.
The “metaverse” itself is not new. With the rapid popularization and application of new-generation information technologies such as AI and cloud computing, any object in the world can be interconnected, perceptible, measurable, communicable, and controllable, making people and things smarter and more accurate. The ternary integration of human society, the information world, and the physical world has brought information services into the era of inclusive computing and networking.
Qiu Wenjin, deputy director of the Shanghai Municipal Development and Reform Commission, said that “Metaverse” is an important driving force for promoting the transformation of the new generation of information technology, and it is one of the main tracks for strategic competition in the future industrial field. “Metaverse” will bring major changes to production and lifestyle in the future, and will lead and drive the comprehensive upgrade of industries such as chips, terminals, and software.
According to statistics, by 2025, the domestic “Metaverse” industry market scale will be between 340 billion and 640 billion yuan, and the international market will be even larger. Zhang Ping, an academician of the Chinese Academy of Engineering, said that “Metaverse” has all the characteristics of future killer applications.
As developed countries have deployed the “metaverse” industry, many giants such as Meta, Microsoft, and NVIDIA in the United States have explored the metaverse, involving new infrastructure, spatial computing, social experience and other fields. Therefore, China hopes to exert its efforts on the “metaverse” to lead the world like AI technology and achieve overtaking.
Among them, Shanghai mentioned in the 2022-2025 action plan for cultivating a new “Metaverse” track that by 2025, the scale of “Metaverse” related industries will reach 350 billion yuan, driving the city’s software and information service industry scale to exceed 15,000 100 million yuan, and the scale of the electronic information manufacturing industry exceeded 550 billion yuan.
Beijing earlier released a two-year “Metaverse” innovation development plan covering 2022-2024, requiring districts to build technological infrastructure at the city level, cultivate and introduce more than 100 Metaverse ecological chain enterprises, and land Build more than 30 typical application scenarios of “Metaverse+”, promote the formulation of a number of Metaverse-related standards, and promote their application in different fields such as education and tourism.
But the problem is that the “metaverse” should not be on paper, nor should it be in a “concept” state for a long time, but should commercialize the “AI+metaverse” related technologies and deeply cultivate them in the industry.
Then, with the long-term reserve of underlying technologies and the high support of top-level design, when will China‘s AI+ metaverse era come? I think China may have to wait for a long time for technological growth, especially for key underlying technologies such as knowledge graphs and natural language processing. Both will reach maturity in 5 to 10 years.(This article was first published on Titanium Media App, by Lin Zhijia)Return to Sohu, see more