Economic Observation Network reporter Qian Yujuan Zhou Yingmei
The World Conference on Artificial Intelligence (WAIC) concluded in Shanghai on July 8.
The popularity of ChatGPT and various large models allows AIGC to subvert the direction of innovation in the industry while truly breaking into the world of ordinary people, allowing intelligent dialogue to happen, and making one-second paintings appear in front of our eyes.
In the past half a year, generative AI, large models, computing power... It has become almost a global focus, and this is also the topic of discussion at the 2023 World Artificial Intelligence Conference held in the past three days, when industry stakeholders from all over the world gather in Shanghai.
After observing the display of relevant AI technology innovation achievements and interviewing people in the artificial intelligence industry to discuss the current development status and future trends, the reporter of the Economic Observation Network microscopic belonged to the top ten keywords of WAIC in 2023.
Keyword 1: Big model
As the secretary general of Qingdao Artificial Intelligence Industry Association, Jiang Dainan's trip to Shanghai WAIC is not alone. "The vice mayor led the team," he told reporters, adding that in addition to 30 leaders at the municipal and district levels, 20 local AI-related entrepreneurs came to participate in the event.
Jiang Dainan said that the first time a group came to WAIC was in 2021, when the most seen AI applications in the exhibition area were: robots and autonomous driving. After two years, as far as the eye can see, I see mostly large models.
After the explosion of ChatGPT, Li Tielei, who paid attention to high-tech projects under the State-owned Assets Supervision and Administration Commission, raised her expectations for WAIC this year.
On the afternoon of July 7, before HUAWEI CLOUD CEO Zhang Pingan officially unveiled the Pangu model in Dongguan, Huawei demonstrated at WAIC that the model penetrated into multiple industries such as government affairs, manufacturing, mining, meteorology, and pharmaceutical R&D, bringing different degrees of efficiency growth.
On the first day of the opening, in the crowded audience, Li Tielei entered the World Expo Exhibition and Convention Center. After a round, his original understanding of artificial intelligence was still at the level of neural networks and big data, and he refreshed his understanding of the domestic artificial intelligence industry and technological development, "(compared to overseas) is not bad." ”
Unlike the belated arrival of Pangu's large model, Baidu, Alibaba Cloud, and SenseTime have announced the progress of their respective large models through WAIC. Of course, the self-developed large model not only attracted technology giants to invest in it, the reporter saw in the exhibition area that artificial intelligence unicorn companies such as Yuntian Lifei and Fourth Paradigm have also brought "Tianshu" and "Style Theory" big models.
"The hottest word this year is the big model." Xiao Yanghua, a professor at the School of Computer Science and Technology of Fudan University, who has participated in WAIC for six consecutive years, affirmed the significance of this wave of AGI technological change to the development of the industry, but he also expressed hope, "Don't let the big model become a gorgeous 'fireworks show'." ”
World Artificial Intelligence Conference Source: Photo by reporter Zhou Yingmei
Keyword 2: Application landing
When many enterprise manufacturers actively embrace large models, the technological changes brought about by general artificial intelligence directly attract attention to the acceleration of application landing.
In WAIC's Huawei exhibition area in the Shanghai World Expo Exhibition Center, the reporter met Jiang Dainan's Qingdao viewing group, and they were listening carefully to Huawei's booth personnel explaining the actual landing of Pangu models in some industry scenarios.
At present, in the state of a hundred flowers of domestic large models, investor Hou Bo personally came to Shanghai to visit the exhibition, focusing on the application layer. He believes that "whether it can be applied and commercialized" has become the criterion for testing the success of large models.
Benny, head of the Hong Kong Hospital Authority, attended WAIC for the first time and focused on finding large models or artificial intelligence technologies that can be integrated with hospitals and healthcare systems.
"Mainland is developing rapidly in the field of artificial intelligence." When the reporter met Benny, he was holding some corporate brochures related to "AI + Healthcare" that he had seen, and he would also communicate with Huawei to explore the possibility of AI technology and large models being applied in the medical field in Hong Kong.
Technological change is attractive to thousands of industries, but "how to land in the scene" has become a new testing ground for manufacturers in the big model competition. It can be seen that not only Huawei has emphasized the promotion of AI technology "to reality" at WAIC, but also manufacturers including Alibaba, Tencent and SenseTime have made great efforts in industry empowerment.
Keyword 3: computing power gap
"A bunch of GPT." Qiu Min, an exhibitor who has participated in WAIC for four consecutive years, is a representative of domestic computing power, and after the boom in large models, many people think that the computing power supplier has ushered in a dividend period, but in his opinion, "large models must be able to be used, and there will be opportunities for computing power manufacturers later." ”
Dai Qionghai, counselor of the State Council, academician of the Chinese Academy of Engineering, chairman of the Chinese Engineering Intelligence Society, and dean of the School of Information Science and Technology of Tsinghua University, also pointed out that in terms of artificial intelligence computing power, if there is no AI chip, it is like "falling into the water of no source".
Large model parameters are calculated in hundreds of millions, and the computing power required, as well as the chips and intelligent computing that provide computing power support behind it, have attracted much attention from the industry.
Sun Yuanhao, founder and CEO of Transwarp Technology, said that some customers will have a trouble when trying Transwarp's self-developed large model "Wuya" in the financial field, "They need to buy computing power and buy GPU card reasoning." "In the short term, the supply of computing power will be somewhat tight with the US chip export restrictions.
However, Sun Yuanhao believes that this problem can be solved in the future, and only around the booth of WAIC, Starwarp Technology has gathered local computing power manufacturers such as Fengyuan Technology, Denglin Technology, Hanbo Semiconductor, Muxi, and Moxin Technology, which are showing their "muscles".
Founded 5 years ago, Fengyuan Technology has participated in WAIC for four consecutive years, and its exhibited Yunxiao intelligent computing cluster is equipped with a self-developed AI computing power chip "Qisi", which can support the scale of more than 100000 calories; Domestic chip company Hanbo Semiconductor has appeared in WAIC for the third time, and it released the 7nm full-featured GPU chip product SG100 on the opening day.
Keyword 4: ecological alliance
Talking about the development of domestic general artificial intelligence industry, Xiao Yanghua suggested using application ecology to drive the base model, he believes that manufacturers who invest in AI large models need to be clear, "The base model is our short board, but the application ecology is our strength." ”
In the view of Zhou Jingren, CTO of Alibaba Cloud, the research and development of large models with more than 100 billion parameters at every turn cannot be completed only by algorithms and stacked GPU computing power, which includes the underlying computing power, network, storage, big data, framework, model design, etc., which is a system engineering involving complex technology.
After gathering millions of AI developers through the Modai community, Zhou Jingren hopes that it can become "China's largest 'big model free market'", and calls on the industry to jointly build China's big model ecology on WAIC.
The reporter saw that during the period, Huawei also emphasized strengthening the Ascend AI industry ecosystem, while Tencent said that the development of AI large-model technology is inseparable from industry chain collaboration and ecosystem co-construction.
On the closing day of WAIC, China Mobile launched the "Nine Days" model and announced the upgrade of the "Nine Days Moon" artificial intelligence industry cooperation ecology.
In addition to the ecological alliance driven by large manufacturers, Xiao Yanghua also pointed out in an interview with the Economic Observer Network reporter that the current constraints behind domestic large-model innovation are lagging with overseas in terms of "time" and "computing power", and technological homogenization is serious.
Facing up to the gap and learning from each other's strengths, Xiao Yanghua felt that "most people still need to make up for lessons" at this stage, and he especially called on manufacturers to promote the establishment of alliances related to computing power, data and technology.
Zhang Lei, founder and CTO of Hanbo Semiconductor, who is on the side of computing power, also said that in the current AGI era, the industry is facing the challenges brought by the growth of computing power demand, and needs upstream and downstream enterprises to jointly build an ecosystem of domestic large models to seize industrial opportunities.
Keyword 5: Make computing trustworthy
Although he affirmed that the current Chinese manufacturers' larger models are still in the following stage, Xiao Yanghua still suggested that manufacturers think more about how to overtake on curves, for example, the data side can promote high-quality data in exchange for the quality of the base model.
"The quality of the dataset is still worrying." Sun Yuanhao recounted that once tutored his daughter's geographical knowledge, and asked a GPT, "Which river is the first tributary of the Yellow River?" The first answer "Huangshui" was wrong, and the second time I asked it did I get the correct solution "Wei River".
In Sun Yuanhao's view, the current large model has a high tolerance rate for To C dialogue, judging from the teaching consequences of primary school students, if it is used to make investment decisions, medical diagnosis and treatment, etc., errors often have serious consequences.
In addition to improving the accuracy of the dataset and training model to make the answer more accurate, Sun Yuanhao saw that more and more industry customers want to achieve personalized and privatized data deployment in large model training, and he provides tools that are conducive to customer training of large models for those who want to be the "shovel seller".
Zhang Tianyi, deputy general manager of the big security machine intelligence department of Ant Group, told reporters that with the in-depth application of artificial intelligence, it not only puts forward higher requirements for data, algorithms and computing power, but also poses more challenges to security, privacy and ethics.
At the Ant Group booth, logos such as "Trusted AI" and "Privacy Computing" were very conspicuous. Wang Wenqiang, the person in charge of the booth, participated in WAIC for three consecutive years, and he told reporters that the "Antjian" AI security detection platform unveiled last year can evaluate data types such as text and images, and this year's WAIC was upgraded to version 2.0, adding AIGC security and AI explainable capabilities.
In Zhang Tianyi's view, AI must achieve "universal" and "credible" in the future, "it must be one and two wings, and one is indispensable." "In fact, more and more manufacturers pay attention to the security and credibility issues in the process of large-scale AI implementation, and SenseTime advocates the development of AI ethics, and participates in the formulation of industry, national and even international standards related to data security, privacy protection, AI ethics and sustainable AI.
Keyword 6: Virtual reality
From the theme of each year of WAIC, we can see what is the hottest format in the field of artificial intelligence that year. The four words "Yuan Sheng Unbounded" in the theme of 2022 directly let the rookie "Metaverse" dominate the limelight, and the WAIC organizers directly brought a "Metaverse Planet" about the conference to the online "cloud audience" last year.
The reporter saw that although this year is also the same online, the theme of "generating the future" has pushed generative AI, AIGC and large models to the C position.
Hong Liang, general manager of the security electronics division of Changjiang Computer Group, came to the main exhibition hall on the first day of WAIC this year, but after he circled the venue, he did not see many manufacturers or displays of "holographic technology".
You know, a year ago, the main exhibition hall occupied a prominent position were XR equipment and platform manufacturers related to metaverse technology. This year, WAIC set the metaverse theme exhibition in the Zhangjiang Science Hall in Pudong New Area, not because this industry and application market services are no longer hot, the reporter learned that Shanghai Pudong New Area intends to optimize the meta-universe industry track, the International Telecommunication Union even launched the "Metaverse Focus Group" working group here, which is also the first time the organization has held metaverse-related activities in China.
Is the metaverse still hot? "The two-way rush of general artificial intelligence and the metaverse" is the key point of Yuan Yu, vice president, director and chairman of the Standards Association of IEEE (Institute of Electrical and Electronics Engineers), who believes that generative artificial intelligence technology will greatly accelerate the development of the metaverse industry.
Like testing automatic driving in various complex road conditions in the virtual world, it is the germination of a new paradigm for artificial intelligence to achieve training in the metaverse, in Yuan Yu's view, the future metaverse and artificial intelligence will be synergistic and accelerated, and the trend of integrated development of the two industries will intensify.
Keyword 7: Concrete robots
When Tesla CEO Elon Musk predicted at the opening ceremony of WAIC, there will be more robots than humans at some point in the future. Huang Xiaoqing, who is deeply engaged in robot innovation and research and development in China, "liked" agreed.
Source: Photographed by reporter Qian Yujuan
"Tesla is not only a car company, but also an artificial intelligence company." Under such a striking sentence on the Tesla booth, compared with smart cars, the most "eye-catching" is its humanoid robot Optimus, even if it is placed in a glass box, it can only be viewed from afar, but it is surrounded by people who go to observe and take pictures.
Not far from the Tesla booth is the booth of Cloudminds Robot founded by Huang Xiaoqing, the reporter saw that the same humanoid robot, Cloudminds Cloud Ginger some are wearing formal dresses, "welcome" at the booth, and will interact with the crowd and wave sideways; Others wore basketball uniforms and competed with the crowd on the spot for basketball throwing...
Source: Photographed by reporter Qian Yujuan
Huang Xiaoqing, who is also a representative of the embodied intelligent enterprise participating in WAIC, believes that the innovation of Cloudminds compared with Tesla is that it proposes a cloud intelligent robot and develops large model applications such as RobotGPT based on the cloud brain operating system.
However, "among the embodied robots that have been developed and can walk, the most complete may be Tesla's humanoid robot." Chen Jianyu, assistant professor at Tsinghua University and young scientist of Shanghai Zhizhi Research Institute, revealed in an interview with the Economic Observer Network reporter that the new version of the humanoid robot developed by Zhizhi Research Institute is also relatively complete, compared with Tesla's products with 40 degrees of freedom, only 6 degrees of freedom. In Chen Jianyu's view, embodied robots attach more importance to action ability than general artificial intelligence technologies such as large models, and include a wide range of technologies, including deviation ontology hardware, control and safety management, vision, multimodal fusion, and haptic.
Talking about the current technical problems of embodied robots, Chen Jianyu summarized the three aspects that have attracted the most attention: due to the high integration and complexity of humanoid robots, the construction of hardware ontology is highly challenging; In addition, the construction of the motor cerebellum, how to control the legs to walk and manipulate the arms to complete various tasks, has not been well solved; Finally, the brain, how to extend the current large model of only language functions to make behavioral decisions in the physical world, is a crucial part of the realization of embodied universal intelligence.
Source: World Artificial Intelligence Conference official
Keyword 8: AI venture capital
On the afternoon of July 6, before Xiao Yanghua left at the end of a sub-forum, a student ran over to exchange ideas about the stage of large-model entrepreneurship.
Seeing the market value miracle created by the overseas Midjourney team, Xiao Yanghua felt more and more strongly, whether it is BAT head players, small and medium-sized enterprises, even if it is just a few people's entrepreneurial team, "General artificial intelligence is pulling everyone to the same starting line." He himself leads a team of more than a dozen people to train large models, "now may be an era of 'ants' overturning 'elephants'."
Not only academic research and innovation are optimistic about the current generative AI entrepreneurship, Qiming Venture Partners partner Zhou Zhifeng shared about AI models and industrial development trends at WAIC, he especially mentioned that ChatGPT has rekindled the wave of AI 2.0, making AI once again a hot spot for entrepreneurship and investment.
In the past three years, under the influence of the black swan of the epidemic, the attitude of the investment side towards AI industry projects has indeed become more rational, and Zhou Zhifeng also pointed out that there are very few leading AI companies in the industry. Today, both entrepreneurs and investors see the opportunities and possibilities brought by generative AI, but there are also many challenges behind the opportunities.
"The path a startup has to take may be a death canyon." Zhou Zhifeng believes that the technology giants and large model companies on both sides of the canyon will continue to squeeze, and with the continuous dynamic changes of AIGC technology, entrepreneurs should think hard about how to find a golden channel out of the canyon of death.
Wang Xiao, who participated in the development of Baidu's first-generation search engine and is now the founder of Jiuhe Venture Capital, proposed at WAIC that AIGC will lead a wave of structural opportunities, "there is no need to squeeze into the track of large models", especially in different vertical fields, he suggested that entrepreneurs can look for landing opportunities and avoid direct competition with large manufacturers.
Keyword 9: AI reshapes the enterprise
During the WAIC period, whether on the way from the Expo Center to the exhibition center, or queuing for admission and lunch, reporters can always hear people discussing how generative AI can accelerate the "rush".
If going to the cloud is a key step for the digital transformation of thousands of industries, the generative AI technology behind the ChatGPT wave is considered to be an important engine for industry transition into the fast lane and achieve cost reduction and efficiency increase.
"98% of executives believe that generative AI will be able to reshape the overall operation of the enterprise." Zhu Hong, global vice president and chairman of Accenture Greater China, gave a set of company research data that about 40% of working hours in various industries will be assisted or replaced by artificial intelligence technology based on big language models.
Although AI cannot completely replace humans, the wave of industrial reshaping brought about by AIGC is already considered a visible trend. This time, Alibaba Cloud released the Wensheng diagram model on WAIC, making Tongyi Qianwen "family" another member. When it comes to this product, Zou Ling, a senior product expert at Alibaba DAMO Academy, gave an interesting scene - smart shopping guide.
As a consumer, Zou Ling wants to buy food for domestic pet cats, based on generative AI and large language models, she only needs to send a question on Taobao, and the efficiency of intelligent recommendation to obtain purchase decisions has been greatly improved.
In order to let more audiences feel the actual effectiveness of AI, the reporter saw that manufacturers including NetEase Education, SenseTime, Huawei and other manufacturers presented their own cases of reshaping education, design, manufacturing and other industries based on AIGC technologies such as large models at the exhibition.
Talking about the past few years, Zhu Hong believes that many enterprises are undergoing "extrusion" transformation in volatility, and in the next 10 years, they will have the momentum to actively promote digital transformation and achieve comprehensive reshaping, and the key catalyst is artificial intelligence.
Keyword 10: Dual carbon goals
When the wave of large-model innovation emerges in China, whether it is the continuous optimization and iteration of AI products or the attempt to land application scenarios, behind the demand for high-frequency computing power, there is also a problem that has aroused the attention of people from all walks of life in the industry: while developing intelligent technology, achieving the goal of dual carbon.
How to use innovative technologies such as artificial intelligence to provide intelligent and sustainable solutions for ESG? The reporter saw that the computing power manufacturers represented by Flameland Technology showed their liquid-cooled artificial intelligence system solutions, energy conservation and emission reduction, which has become an aspect of the artificial intelligence industry chain that those manufacturers that carry out technological innovation pay close attention.
Focusing on the integration of smart technology and sustainable development, Li Junfeng, investment partner of Sequoia China and president of Sequoia Carbon Neutrality Research Institute, also proposed at WAIC that "the two major supports for future energy technology innovation should be energy and computing power. In his view, with the continuous improvement of the level of computing power in mainland China and the continuous decline in costs, artificial intelligence will be widely used, which will accelerate the innovation of energy transformation technology to a certain extent.
(Head image source: Photo by reporter Zhou Yingmei)