By Wang Huiying 

Edited by Ziye 

Along the Huangpu River in the midsummer of 2024, the temperature outside soared to 38 degrees Celsius, but it didn't dampen the enthusiasm for the World Artificial Intelligence Conference (WAIC).

Under the scorching sun, the crowd bustled, making this year's WAIC even more intense than previous years: over 500 enterprises participated, setting new records for the number of exhibitors, highlights, and new product launches.

The conference's keyword, which has been on the list for two consecutive years, remains the biggest trend this year. With the accelerated evolution and maturity of generative AI, large models and related application scenarios are blooming.

Compared to last year's WAIC, this year's "Hundred Models War" is a competition of large model deployment capabilities. In addition to familiar faces like Alibaba, Baidu, Tencent, and iFLYTEK, there are also new players like Kuaishou, Bilibili, Zhipu AI, Baichuan Intelligence, and Jieyue Xingchen.

Image source: WAIC WeChat official account 

Apart from large models, embodied intelligence and robotics also continued the heat from last year's conference. However, compared to the deployment competition among large models, most of these niche segments are still in the stage of competing for "existence," with a long way to go before commercialization.

When founders promote their technologies, and AI products at various booths flourish, WAIC seven years ago looked quite different.

When WAIC kicked off in Shanghai's Xuhui District in 2018, AI, like today's large models, was still a novelty. The initial founders of the internet were still sharing their views on "what AI is"; in 2019, Jack Ma and Elon Musk's amusing dialogue at WAIC was still fresh in the industry's memory...

Today, Jack Ma is no longer present, and Elon Musk is absent from this year's WAIC, replaced by a new generation of AI entrepreneurs.

Standing at the seven-year mark, WAIC has witnessed the leapfrog development of global technology. Along with industry growth, WAIC has closely followed trends, becoming a bellwether in the waves of AI, metaverse, and large models, while also completing its own influence "triple jump" to become a highly regarded summit in the industry.

We are now standing at the threshold of a new era of technological change, closer than ever to the inflection point of historical transformation.

1. From IT to Large Models, from Concept to Deployment: WAIC Witnesses the Global Leap in Technology

Let's rewind five years.

At the second WAIC, when discussing what AI is, Tesla founder Elon Musk humorously interpreted it as "love" due to its Chinese pronunciation, while Alibaba founder Jack Ma jokingly called it "Alibaba Intelligence."

It was not uncommon for industry leaders to engage in lively conversations back then.

In 2018, the first national AI development plan was released, and AI was written into the government work report for the first time, becoming a national strategy. At that year's inaugural WAIC, the "BAT" founders Li Yanhong, Jack Ma, and Pony Ma appeared on the same stage, constituting the top-tier guest lineup of domestic internet conferences.

At that time, however, AI was the topic of discussion, a term less familiar to the general public. Perhaps because of this, the second WAIC attracted a peak number of attendees.

Also in this year, the Shenzhen IT Leaders Summit themed "IT's New Future: 5G and AI" saw Tencent founder Pony Ma deliver a speech on "5G and AI Driving the Development of Industrial Internet," predicting three major trends in the industry.

AI emerged again as a keyword.

During the AI wave that began in 2016, AI's development was not particularly rapid. At that time, AlphaGo's victory over Korean top Go player Lee Sedol sparked market enthusiasm for AI. In the years that followed, SenseTime, Yuntong Technology, Megvii, and Yitu emerged, known as the "Four Little Dragons of AI" for their computer vision capabilities and becoming darlings of capital.

However, neither their business models nor AI technologies were particularly sophisticated back then, and their technological roadmap was quite different from today's large model era.

Subtle changes in the tide also occurred. After 2019, the initial internet leaders gradually retreated to the background, with fewer public appearances. The IT Leaders Summit struggled to replicate its past glory and was once postponed. The market also shifted its focus more towards AI.

Especially with the explosion of the metaverse in 2022, WAIC, leading industry trends, positioned its theme as "Intelligent Connection, Boundless Metaverse," bringing back the AI-flavored WAIC.

The emergence of ChatGPT in 2023 initiated a new technological revolution, demonstrating the speed of AI under the large model wave: ChatGPT reached 100 million users in just two months; Perplexity had nearly 10 million monthly active users each month.

This year, "Intelligent Connection, Generative Future" was the theme of WAIC. The exhibition hall covered an area of 50,000 square meters, more than triple the previous year's 15,000 square meters, with 400 participating enterprises, double the previous year.

Reflecting on Chinese tech enterprises, a large model arms race ensued.

Not only traditional internet giants like Baidu, Tencent, Alibaba, and ByteDance, but also "The Five Tigers of Large Models" consisting of Zhipu AI, Baichuan Intelligence, Dark Side of the Moon, Zero One Everything, and Minimax participated. Furthermore, numerous former executives from internet-era companies like Wang Huiwen and Wang Xiaochuan resigned to start their own businesses, becoming entrepreneurs in the large model era...

After a year of intense competition, WAIC themed the conference as "Shared Growth through Consultation, Intelligent Governance through Collaboration," emphasizing global governance and international cooperation in AI. For two consecutive years, large models have been the keyword at the conference. For Chinese enterprises, the competition has shifted from the size of large model parameters to the speed of large model deployment applications.

Image source: WAIC WeChat official account

A glimpse can be gained from the selection of WAIC's eight "treasures of the museum": the humanoid robot pioneer array, SenseTime's first controllable character video generation large model Vimi for C-end users, Alibaba Cloud's AI programming assistant Tongyi Lingma, Alipay's intelligent assistant, Zhipu AI's base large model, Tesla's Cyberquad ATV, Lenovo YOGA Book 9i laptop, and the intelligent aircraft display area.

If the IT Leaders Summit witnessed the rise and fall of internet enterprises, the seven WAIC conferences from 2018 to 2024 have witnessed the leapfrog development of China's AI industry from concept to application deployment and then to commercial exploration.

As computer science pioneer John von Neumann once predicted, "The continuous acceleration of technological progress will trigger a critical singularity in human history." Today, the singularity of AI is getting closer and closer.

2. From Rare Species to "Sold by Weight," Embodied Intelligence is Also Training Hard

Every year, WAIC serves as a bellwether for the AI industry. Through the dynamics of participating enterprises, we can gain a clearer understanding of the future development trends of AI and observe the hottest niche segments.

At this year's WAIC, in addition to large model application scenarios, embodied intelligence concepts like humanoid robots were also a highlight.

At the entrance of the Expo Park exhibition hall, 18 humanoid robots with different technical architectures lined up in sequence, known as the "Eighteen Vajras" and one of the treasures of this year's WAIC. Behind these 18 humanoid robots are enterprises like CETC, Fudan University, CloudMinds, Fourier Intelligence, Taihu Robotics, Xingdong Jiyuan, Zhuoyide, Kepler, Unitree, and more.

Image source: WAIC WeChat official account

Upon entering the venue, humanoid robots that could walk, run, jump, and climb stairs were everywhere. Concepts like humanoid robots and embodied intelligence also became hot topics among attendees.

Tesla's second-generation humanoid robot Optimus made its debut, and Unitree Technology showcased H1, China's first full-size general-purpose humanoid robot capable of running. Fourier Intelligence, CloudMinds, CloudMinds Robotics, and more brought over 20 intelligent robots.

Furthermore, in the Future Tech 100 Incubation Zone of WAIC, robots were the main attraction of the "2024 Future Tech 100 Innovation Project Roadshow," with startups like Youlu Robotics, Weichang Robotics, and Yuzhou Robotics participating.

This roadshow is a fixed segment of the World Artificial Intelligence Conference each year, aimed at providing a platform for startups to present their projects and giving investment institutions a one-stop window to find investment targets.

Since last year, the explosion of large AI models has fueled the concept of embodied intelligence.

Traditional robots must execute instructions decomposed by human engineers into a series of brief, programmatic procedures, breaking down a complete task step by step for a specific scenario.

Equipped with large models, embodied intelligent robots are like having a "brain." Just as large models have interactive understanding capabilities, humanoid robots have also made breakthroughs in human-computer interaction, comprehension, and continuous decision-making, making them more "human-like."

According to Markets and Markets, the global embodied intelligence market was valued at $1.8 billion in 2023 and is expected to reach $13.8 billion by 2028.

Although the perfect world model matching embodied intelligence has yet to emerge, the development of embodied intelligence in the past two years has far exceeded practitioners' expectations. Not only have more and more players entered the field, but capital has also extended an olive branch.

This year, two large early-stage funding rounds were completed in the embodied intelligence sector—Unitree Technology completed its Series B2 round of nearly RMB 1 billion in February; in June, Galaxy General Robotics completed its RMB 700 million angel round, with a total of 22 investors, making it the largest angel round of the year.

Earlier, general-purpose legged robotics company Zhide Power also announced the completion of a new round of strategic financing, and AI flexible grinding robotics company Steel announced the completion of nearly RMB 100 million in Series A funding. Paxini Perception, Xingdong Jiyuan, CloudMinds Robotics, Fourier Intelligence, Robosen, and more have announced their financing progress.

Moreover, IPOs have emerged in this sector. In December 2023, UBTECH officially listed on the Hong Kong Stock Exchange, becoming the "first humanoid robot stock."

It is worth mentioning that at the WAIC Humanoid Robots and Embodied Intelligence Development Forum, the National-Local Co-built Humanoid Robot Innovation Center released the open-source general-purpose humanoid robot "Qinglong" for the first time. Standing 185cm tall and weighing 80kg, it boasts highly biomimetic torso configuration and anthropomorphic motion control, supporting multimodal mobility, perception, interaction, and manipulation.

Humanoid robot "Qinglong," image source: WAIC WeChat official account 

During the live demonstration, "Hello, please separate the bread and fruits." Upon hearing the instruction, "Qinglong" gently spread its arms and used its five steel fingers to pick up a soft piece of bread and place it aside, leaving no scratches on the bread.

Looking back at last year's WAIC, only Tesla's humanoid robot "Optimus" held the position of the museum's treasure, and there were far fewer booths related to embodied intelligence; today, the dazzling robots on site can be "sold by weight," and the concept of embodied intelligence has become a top trend.

Indeed, with technological and market development, humanoid robots have bid farewell to their rare status, but there is still a long way to go before industrial maturity and commercialization. Currently, the industry is widely concerned about the overall performance development of humanoid robots, their specific operational capabilities when cooperating with large models, and their applications in specific scenarios. After all, large models run on PCs and mobile phones, and loading them onto humanoid robot bodies involves higher technical difficulties and cost investments.

However, the development trend of embodied intelligence is unstoppable. On June 2, NVIDIA CEO Jensen Huang delivered a speech stating, "The era of robots has arrived, and the next wave of AI will be physical AI. Robots will increasingly integrate into our daily lives, and everything will be robotized."

3. Familiar Faces Remain, More New Faces Appear, and Industry Leaders Are Ambitious

In the past two years, if you saw a crowded booth at the WAIC exhibition hall, it was undoubtedly an AI large model company.

Last year, when large models were in their explosion phase, large tech companies like Baidu, Tencent, Alibaba, Huawei, and iFLYTEK, which held the initiative, frequently appeared at conference forums, and these familiar faces dominated the scene.

Starting from the first half of this year, industry views on the impending explosion of AI application layers have emerged. Kai-Fu Lee, founder of Zero One Everything, bluntly stated that 2024 is the first year of the AI application explosion.

Regular WAIC attendee Robin Li, CEO of Baidu, is also a supporter of promoting the application of large models. Last year, Robin Li said that competing on large model parameters is less important than competing on applications, and this year, he further elaborated on his insights into large model applications.

In Robin Li's view, in the AI era, "super capable" applications are perhaps more important than "super apps" that focus solely on DAU. "We must avoid falling into the trap of super apps, thinking that an app with 1 billion DAU is a success. This is the thinking logic of the mobile era," he said.

Image source: Baidu WeChat official account

Furthermore, Robin Li stated that without applications, basic models, whether open-source or closed-source, are worthless.

Regarding the development direction of AI applications, Robin Li is most optimistic about intelligent agents. He believes that as basic models become increasingly powerful, developing applications becomes simpler. Among them, the simplest is the intelligent agent, which can be created by clearly explaining workflows in "human language" and pairing them with proprietary knowledge bases, "simpler than creating a webpage in the internet era."

On the other hand, iFLYTEK showcased products like iFLYTEK Spark V4.0, iFLYTEK XiaoYi, and Spark Enterprise Intelligence at WAIC. At the scene, visitors could see the latest application results of iFLYTEK Spark V4.0 in various scenarios like learning, life, and work, while iFLYTEK also provided rich solutions for industry scenarios and enterprise customers.

Especially in the education field, which iFLYTEK focuses on, Liu Qingfeng, Chairman of iFLYTEK, said that general AI represented by large models is setting off a wave of educational transformation globally. "AI will surely go down in history for solving human rigid demands, and education needs to cultivate new humans who stand on the shoulders of AI."

The uniqueness of this year lies in the fact that, besides the familiar faces from the initial wave of the "Hundred Model Battle," some emerging startups have shown a stronger presence, and new players focusing on large model applications have joined the fray.

Companies like Zhipu AI, MoonShadow, Baichuan Intelligence, Zero One Universe, MiniMax, and StepStar have garnered equal or even greater attention. 

Behind this attention is the fact that these new faces are closely following industry trends and implementing various large model applications.

Zheng Qinkai, the head of Zhipu AI's CodeGeeX technology, released the fourth generation of the CodeGeeX code model, CodeGeeX4-ALL-9B.

Baichuan Intelligence showcased its Baichuan series of general-purpose models, the AI assistant Baixiaoying, a beta version of the medical application AI Health Advisor, and Baichuan Intelligence's B2B solutions.

MiniMax's strategy also focused on highlighting its open platform for enterprise customers, as well as its AI professional assistant for consumers, "Conch AI," and the AI intelligent creation platform "Xingye."

StepStar not only released the official version of its Step-2 trillion-parameter language model, the Step-1.5V multimodal model, and the Step-1X image generation model, but also showcased application products like the intelligent assistant "Yuewen" and the AI open-world platform "Mopaoya."

Additionally, Kuaishou, which rose to fame due to the popular large model application Keling, is another new face at this year's WAIC. Currently, the Kuaishou large model family includes the 175 billion parameter general-purpose language model "Kuaiyi," the text-to-image model product "Ketu," and the video generation model "Keling." Among these, Keling has introduced a clearer high-definition version, new features such as start and end frame control and camera control, and has increased the length of text-to-video generations to 10 seconds per creation.

At last year's Yunqi Conference, Qiuenghai Dai, Dean of the School of Information Science and Technology at Tsinghua University, remarked, "When we talk about the internet, we think of Alibaba, Tencent, and Baidu. This won't be the case in the future; those who cannot keep up will fall behind. The era of large models is upon us."

This statement has now been validated at WAIC. Wang Jian, the founder of Alibaba Cloud, also mentioned at the conference that any new technology will inevitably give rise to new major companies. If a new technology emerges without the rise of new major companies, we must question whether it is truly a disruptive technology.

With the advent of the AI era, everyone is filled with boundless anticipation for the future, longing to secure a ticket aboard this voyage.

Travel back to the summer of 1956, to Dartmouth College in the small town of Hanover, USA, where young scholars like John McCarthy, Marvin Minsky, and Claude Shannon gathered to discuss unresolved and even uninitiated research problems in computing, one of which was artificial intelligence. It was at this very meeting that the term "artificial intelligence" was first coined and gradually became a recognized field of study.

Today, the World Artificial Intelligence Conference (WAIC), now in its seventh year, has witnessed the rapid development of the AI industry. Each year brings new technologies and fresh faces, creating a vibrant and lively atmosphere at the expo, further heightening expectations for the future of artificial intelligence.