Original link: https://www.latepost.com/news/dj_detail?id=1744
Huawei, Tencent and Ali are here, so are Intel and Qualcomm; Tesla is here, with its 1:1 humanoid robot model; Nvidia didn’t make an appearance, but it’s also appearing everywhere, at speeches by Chinese competitors , in inquiries from potential customers.
This is considered a regular configuration of an AI conference, but among the nearly 180,000 people who came to Shanghai to participate in the 6th World Artificial Intelligence Conference (WAIC) from July 6th to July 8th, there were jewelry testing agencies, teams from the Public Security Bureau, and Employees from nuclear power companies and hospitals came to look for opportunities, and some even organized some elementary school students to visit, more than one group.
At a time when growth is becoming scarcer and certainty is rarer, the AI Big Model promises a rare new possibility. People from all walks of life gathered in Shanghai Pudong with their questions and confusion, and when they left, some questions were not answered.
“If the big model is the answer, what is the question?” This is the question of an exhibitor looking for entrepreneurial opportunities after coming to the exhibition. He used to work in ByteDance, and now he is preparing to start a large-scale model business, hoping to find some inspiration from the more than 30 large-scale models exhibited at this conference.
After shopping all morning, he thought of the last round of artificial intelligence craze he experienced. AlphaGo defeated Ke Jie in 2017, and the following year, WAIC was held in Shanghai for the first time. At that time, the largest variety of giant screens were displayed in the venue, showing the traffic flow or streets under the surveillance of the camera. AI companies have dubbed such systems “smart cities.”
Since then, the development of artificial intelligence has not been completely driven by the market, but mixed with government guidance and expectations. This boom in large models is no exception. There are officials at the opening ceremony and some large-scale forums, expressing their willingness to provide policy support for the development of the artificial intelligence industry.
Comparing the two waves, the entrepreneur has more questions: “CV (computer vision) has security, what will the big model have?” An imperfect chatbot helps itself answer customer service questions or do document refinement and summarization.
His question shows the other side of the big model craze: the big model is like the hammer of Thor, but if there is no suitable nail for it, it will be difficult to realize the expected huge commercial value.
People from all walks of life are here to see how big models can help themselves
After 9 a.m. on July 6th, before the WAIC opened the pavilion, the crowds had already occupied the cafes near the Shanghai World Expo Center and the World Expo Exhibition Hall. Those who arrived late could only squeeze into the nearby noodle shop to discuss AI and The big model, in the past, only at noon, the business here would get better.
The “scalper” also took his place near the entrance, asking in a low voice “Have you made an appointment?” According to official requirements, you will not be able to enter the venue without an appointment in advance. The scalpers said that they could handle the entry problem, “it will cost 400 yuan.”
AI practitioners were not the only ones who saw the exhibition. Staff members could be seen everywhere in the exhibition hall holding signs and leading teams to visit. The signs said “Jewelry National Inspection and Purchasing Team”, “Hongkou Branch of Shanghai Public Security Bureau”, “Great Wall Motor Purchasing Team” and so on.
In the large-scale model forum organized by Huawei, the media reporters who arrived late found that the reserved media seats had been seized. It was not because the media colleagues were too enthusiastic, but because there were too many people present. Two employees of the China National Nuclear Power Group are also sitting here. In the past few years, they have used artificial intelligence technology in the maintenance and repair of equipment. Now they care about the large model: “See if we can find some joint points.”
At the moment when the growth momentum is weakening, large models are one of the few bright spots. After the hype in the past six months, some people are worried about being replaced and subverted by AI, while others are determined to jump into the trend before being replaced. However, this industry-wide enthusiasm lacks a cognitive foundation for the time being, and most people still do not understand what a large model is and what it can do.
“Does a large model mean large computing power and take up more storage space?” A person who works in a hospital asked nearby staff after visiting the Tencent exhibition area. She knows the visual ability of artificial intelligence and can already help doctors see CT images, but she doesn’t know much about what large models can do for hospitals, and she is eager to know more.
Too much expectation formed too quickly often brings an equal amount of disappointment. At the exhibition site, many people gathered in front of the computers at each booth to experience the large models: “Write a brand story of creative coffee” and “Write a travel plan for elementary school students”. An exhibitor caught a glimpse of the “writing love letter” function displayed by Alibaba Tongyi Qianwen, and immediately complained: “What is this thing? Can you do something practical for me?”
Visitors gathered around the WPS booth to experience the large-scale application, the picture is from WPS.
Few companies are talking about reinventing OpenAI
After sending away a few people who failed to test the large model, an exhibitor from an Internet company sighed: “No way, the technology is still not good enough.”
The one who just left took a math problem to test the big model, let it calculate “how many five-digit numbers are different in each number”. After seeing the result, the other party raised his phone and said, “ChatGPT is more reliable.”
Before attending the conference, the staff member at the booth saw that his peers had made dozens of large-scale models in a few months, and he already had a feeling: making a large-scale model does not seem to have so many technical barriers, but it must be done well difficult. During WAIC, after he took the time to experience the large models of his peers, this feeling became stronger: “The effect is not much different, and there is a gap between them and ChatGPT.”
Just half a year ago, many companies claimed to be China’s OpenAI. However, in the booth and forum speeches of this conference, not many people have mentioned this goal.
The new narrative is: industry big model and “big model empowers thousands of industries”.
Huawei exhibited a large model of a mine.
Hu Houkun, the rotating chairman of Huawei, said at the opening ceremony of WAIC, “The goal of the large model is to serve different applications in different industries… in order to exert greater value.” Tang Daosheng, Senior Executive Vice President of Tencent Group, also expressed in a follow-up speech Similar views: “Industry large models are a better option for enterprises to embrace large models.”
Neither Tencent nor Huawei put computers on their booths for people to experience large models. Although Alibaba, which is next to the Tencent booth, put more than a dozen computers in the center of the exhibition area for people to experience, Zhou Jingren, CTO of Alibaba Cloud, also began to emphasize ecology instead of self-developed large models. He said in the WAIC forum: “Alibaba Cloud will take promoting the ecological prosperity of China’s large-scale model as its primary goal…let all walks of life enjoy the dividends of large-scale model technology.”
The change in wind direction is not that big companies give up research and development of larger models, but that they hope to find a feasible way to let this immature new technology play value first and generate income. Their logic is:
- The general-purpose large model (similar to the large model behind ChatGPT) is expensive to use, and the model parameters generally start at hundreds of billions. It takes a lot of resources to actually operate. ChatGPT and New Bing once drained the computing power of hundreds of thousands of GPUs accumulated by Microsoft. The company simply cannot bear it.
- Generic large models do not work well in specific scenarios. General large-scale models are generally trained based on public literature and network information, and the accumulation of professional knowledge and industry data is insufficient, resulting in insufficient accuracy of answers. “Once a company provides wrong information to the public, it may cause serious consequences.” Tang Daosheng said.
- Large industry models have smaller parameters and lower deployment costs. After targeted training, they can answer specific questions better. Moreover, large companies provide cloud MaaS (model as a service) for training or deploying large-scale model services for companies in various industries, which can also help them sell some cloud services first.
Smaller companies are more realistic. A chief scientist of an AI unicorn company said that they have been researching large language models since 2018, and in the first two years they also made an application for writing articles. Because no customers paid for it, the company has not increased its investment. After ChatGPT became popular, they also released a self-developed large model, but they do not plan to train a larger model for the time being, because customers feel that the cost performance is limited. After all, training a model with hundreds of billions of parameters will cost tens of millions of yuan.
However, even a large-scale industry model that is cheaper in theory and closer to implementation is still expensive to use. “LatePost” learned that a large-scale model with 6 billion parameters of a large-scale model start-up company in China is sold for one million yuan, and the price of 100 billion parameters is tens of millions of yuan per year-most of the cost is Chip computing power.
It is not sure what the big model can do, and the equipment is very active
What the big model can do and how much it can do is still being explored, but the consensus of most people is that the equipment must be prepared before panning for gold.
An obvious change in the WAIC venue this year is that the booths of domestic AI chip companies are larger and closer to the C position, which also attracts more attention: Suiyuan Technology, Tianshu Zhixin, Hanbo Semiconductor, Muxi Integrated Circuit, Denglin Technology, etc. The booths of chip companies are close to Tencent and Baidu, and the booth area is similar to these big companies. Huawei, which occupied the largest booth in the audience, separately opened a Shengteng ecological booth. Shengteng is a complete set of Huawei AI computing products including AI chips, MindSpore AI training framework and software services.
Every chip company’s booth was crowded with people, and customers who came to consult wanted to know how these chips performed and what they could do. One of the most common questions is: is there a replacement for the A100?
The booth of Chinese AI chip companies is larger than in previous years.
The A100 GPU, launched by Nvidia more than two years ago, is now almost standard for training large models. When ChatGPT was born, AI startups and tech giants everywhere rushed to buy the A100. The U.S. government’s export controls on AI chips such as the A100 have exacerbated their shortage in China, but at the same time, domestic chip companies have seen alternative opportunities. Zhang Dixuan, President of Huawei’s Ascend Computing Business, said in an interview, “In the past we were the ones looking for companies, but now many companies are looking for them.”
Different from companies that develop large-scale models and are still trying various final application scenarios, what chip companies show is much clearer and more intuitive: in the conspicuous positions of each chip booth, there are often various types of chips and AI chips equipped with them. They look like large chassis with rows of AI accelerator cards inserted in them. Each company will also use on-site computers to demonstrate the application effects of AI large models or AIGC with the support of their own chips: including dialogue robots, AI paintings, etc.
Nvidia did not set up a booth at WAIC, did not have a naming forum, and did not win any awards. In the special chip forum, Nvidia only sent one technical director. He was the last one to come on stage, after Qualcomm, AMD and Intel. But almost every chip company will compare the indicators of Nvidia A100 when promoting their products; when Zhao Lidong, CEO of Suiyuan Technology, gave a speech at the same forum, he started with the fact that Nvidia’s market value has exceeded one trillion U.S. dollars, which shows that Wall Street is using real money. Bet on the big opportunity of AI computing.
The Chinese government also pays more attention to supporting AI computing power than before: last year, a deputy mayor of Shanghai attended the leader’s speech at the special chip forum of the conference. The director of the Municipal Economic and Information Commission, and a deputy director of the Science and Technology Department of the Ministry of Industry and Information Technology.
In addition to chip companies, those struggling to sell “equipment” also include cloud computing vendors and data service companies, as well as headhunting companies and local parks.
It is difficult for most companies to buy GPU chips to train large models. A better way is to directly rent the computing power provided by cloud vendors. Microsoft Azure and Amazon AWS have both been on the main forum of WAIC this year.
“LatePost” learned that Appen, a data collection and labeling platform, has made a big bet this year, putting most of the annual exhibition budget on WAIC; its Chinese counterparts – the stock price has fallen within just one month Haitian Ruisheng, which has risen by more than 2 times, is the first time to participate in the meeting. The staff on the scene said that in addition to receiving successive potential customers, there are also “many shareholders who came to thank us”.
Various service providers in the entrepreneurial ecology are also looking for customers. A person from an exhibitor said that he met several waves of headhunters and investment recruiters from local parks in one morning, and received a stack of business cards. No matter what new opportunities you want to try, talents and business sites are costs that one group of companies have to pay, and they are new development opportunities for another group of companies and places.
Large-scale models drive general-purpose robots to become hot spots, unmanned vehicles, metaverse ebb
Among various relatively abstract applications and system solutions, robots are one of the rare application directions that can be “visible and tangible”.
In the past six months when the big model became hot, a concept called “Embodied AI” (Embodied AI) also gained attention. Simply put, embodied intelligence refers to the combination of artificial intelligence software and hardware to solve real-world problems. In May of this year, Nvidia CEO Jensen Huang said that embodied intelligence will be the next wave of AI. The typical representative of embodied intelligence is a robot, especially a general-purpose robot that can complete a variety of complex tasks in the same form of product.
At the WAIC site, there were more robot dogs and humanoid robots visible to the naked eye. The organizer said there were more than 20, compared with single digits in previous years.
The lively scene in the venue was a group of people “chaking the dog” around the robot dog, trying to push it down and ride it. The robot dog in Yunshen fell down while climbing the steps in a performance, and immediately aroused a burst of sighs among the onlookers: “It’s over, it’s over…it’s over”.
People watch the robot dog. Source: Visual China.
Practitioners are currently at a loss as to how to combine large models and robots. Some viewers asked the staff of the robot dog company Yushu Technology: “Will you use the large model on the robot dog?” The staff was stunned for a while, and said: “I haven’t found anything that can be done yet.”
The founder of a robot company told “LatePost” that the obvious application direction of large models in robots at this stage is to replace codes with natural language and directly input instructions to robots, so that robots have some “common sense” and can convey what people want to convey. The task is divided into various robot subtasks; however, the execution of subtasks needs to rely on the basic capabilities of the robot, such as navigation, planning, control, etc. The large model may be helpful, but it is not a substitute or subversion.
Companies developing humanoid robots are much more optimistic than robot dog companies who are hesitant to face large-scale models. DATA Technology released the “RobotGPT Industry Model”, claiming to “lead a new era of embodied intelligence”. They transported more than a dozen robots to the scene and asked them to form a line and dance the “Thousand-armed Avalokitesvara”.
A robot performing “Thousand-Handed Avalokitesvara” by Dadai Technology.
It met a new opponent at this AI conference. The booth is at Fourier Intelligence, which is opposite to Data. The company previously made intelligent rehabilitation equipment, and now it has released a humanoid robot, and also announced that it will “lead AI into the era of embodied intelligence.”
The most eye-catching humanoid robot is Tesla’s Optimus. Outside the red line protecting the robot, the gathered crowd scrambled to get closer and closer to take pictures and record it, as if they were worshiping a statue of a god, even though it was only a 1:1 model.
Visitors crowded in front of Optimus, Tesla’s humanoid robot.
Passers-by often ask the staff, “Can it move?” Some people even shake the phone up and down and back and forth, just to get a dynamic effect. “Although it is a model, it is shocking,” said a young man at the scene. He visually observed that the robot was 1.9 meters tall. This may be the halo brought by black technology. The official height of Optimus is 1.72 meters.
In previous years, such grand occasions belonged to unmanned vehicles and metaverses.
The 2021 AI conference will feel like another Shanghai auto show. Self-driving companies Inceptio Technology, Pony.ai and TuSimple will spend a lot of money to transport trucks that are 4 meters high and weigh nearly 10 tons to the booth. AutoX, SAIC Motor, Huawei Car BU, Baidu Apollo, SenseTime, and even Xinchi Technology, which makes automotive chips, all directly demonstrated cars equipped with their own technologies or products. WAIC also specially set up an unmanned vehicle experience area, and unmanned minibuses were used for the venue connection.
This year, the experience area and unmanned minibuses are gone. Only a few car companies such as SAIC Zhiji, Jidu, and Tesla still have cars on their booths. Among the “three heroes of trucks” two years ago, this year only TuSimple has displayed a perception kit wrapped in a plastic box. Almost no car companies came.
A comparison of Tucson’s future 2021 booth (top) and 2023 booth (bottom).
The intelligent driving forum of the conference was arranged on the last day of WAIC. When participating in the roundtable discussion, Cao Guangzhi, the co-founder of Yunji Zhixing, encouraged his peers to tide over the difficulties together, while laughing at himself: “Automatic driving is such a bright spring and snowy thing, how can we get involved like this? “
The Metaverse has a similar cold reception. Last year’s WAIC conference embedded the concept of Metaverse into the theme name, “Intelligently Connecting Everything, Yuansheng Boundless”. Liang Youberry, President of Meta Greater China, was invited to give a speech at the opening ceremony. The Oriental Pearl Tower and Wukang Building and other places have set up metaverse check-in points.
A year later, there are only a handful of XR (extended reality) companies left that use the metaverse as a pitch. Among the 10 official cutting-edge technologies, there are only 3 metaverse-related forums this year, which is one-fifth of last year. From the entrance of the main venue to the end, you can see some small booths of companies related to the Metaverse. Regardless of whether it is autonomous driving or metaverse, under the background of current trends and financing difficulties, most companies have reserved limited budgets for maintaining company operations.
When a new technology boom appears, there are often two evolution paths: one is that the new technology realizes its value and becomes a part of the infrastructure, and is no longer paid attention to, such as the Internet and recommendation algorithms. The other is that the new technology cannot realize its value in the short term, and then resources and limelight are taken away by the new craze. Now large models have become a new hot spot, but after the past rounds of technical hype, both insiders and outsiders have calmed down a lot. Those practitioners who really want to do something in this new opportunity actually hope that the enthusiasm and expectations of the public will be more pragmatic.
Wu Yunsheng, vice president of Tencent Cloud, said that Tencent has cooperated with customers in more than ten industries such as finance, cultural tourism, government affairs, media, and education to create more than 50 large-scale industry model solutions. As far as we know, none of these plans have been launched yet. He said that now is the initial stage of the development of large models.
Zhu Likun also contributed to this article.
This article is transferred from: https://www.latepost.com/news/dj_detail?id=1744
This site is only for collection, and the copyright belongs to the original author.