The smartest group of people grabbed 4 trillion to transform the business

Human exploration of AI is destined to be a lonely journey.

Text: "Chinese Entrepreneur" reporter Yan Junwen

Editor|Li Wei

Header image source: Respondents

Top tech giants are flocking to manufacturing.

Huawei's "genius boy" Peng Zhihui (Zhihuijun) established Zhiyuan Robot in March 2023, and obtained 5 financings in 1 year, the highest of which received more than 600 million yuan.

Another genius, Yang Zhilin, is the founder of the dark side of the moon, completed a $1 billion financing in February this year, setting a new record for large-scale model financing, the dark side of the moon launched the Kimi popular network, with one million monthly active users, but Yang Zhilin predicts that more valuable breakthroughs in large-scale models will occur in the industrial world;

Nvidia CEO Jensen Huang led a $32 million investment in October 2023 in a robotics company that seeks to build next-generation factories with the latest advances in robotics and artificial intelligence. In addition to founding Nvidia, Huang is a top scientist and was elected to the National Academy of Engineering earlier this year.

Traditional factories are indeed too old. It has been 110 years since Ford invented the assembly line in 1913, even the "industrial robotic arm" that Lei Jun is proud of was invented in 1954, and Musk once envisioned building a fully automated gigafactory, but he encountered a capacity trap and had to compromise with himself. With the rise of ChatGPT, the boundaries between machines and humans have been broken, and the old production system cannot meet the production of embodied intelligence, and it calls for a new factory system.

Demand can also be glimpsed from the continuous growth of the market size. According to market research firm Mordor Intelligence, the global smart factory market size is estimated to be about US$323.13 billion in 2023 and is expected to reach US$514.29 billion (about 3,720.4 billion yuan) by 2028.

Earlier than Zhi Huijun and Yang Zhilin, Jia Jiaya, a tenured professor from Hong Kong Chinese University, saw the future of intelligent manufacturing earlier. Jia Jiaya is a top scientist in the field of CV (computer vision), and after leaving Tencent in 2019, Jia Jiaya founded Smartmore, which is the transliteration of "Smart More" - making factories smarter and smarter.

According to Jia Jiaya's definition, the era of large models will give birth to the corresponding factory production mode, when the robotic arm is equipped with a large model or has intelligence, it can flexibly adapt, think like a human, instead of manually setting programs and parameters as now.

"The final outcome of people is to be freed from complex, dangerous, and repetitive work, to withdraw from the production line, and to become the decision-making line. Jia Jiaya said in a recent interview with "Chinese Entrepreneur" that this is an extreme world of things, and the role of people will eventually shift to decision-making and management.

"Yao Ming" in the AI world

Jia Jiaya chose to start a business in 2019, which is really not a good time window.

At that time, the AI Four Dragons (SenseTime Technology, Megvii Technology, YITU Technology, Yuncong Technology) were experiencing the last afterglow, before that, a large amount of money in the market was invested in them, and their vision technology was used in security, consumption and other fields, but they could not recover the cost and fell into huge losses.

This makes the AI story untenable for investors. In 2019, the amount of financing for domestic AI companies fell by 34.8% compared to 2018.

But some people still firmly support Jia Jiaya's entrepreneurship, and they believe that when Yao Ming appears, you can't fail to notice him, Jia Jiaya is the tall "Yao Ming" in the AI world - he is a top scientist.

In 2000, Jia Jiaya graduated from Fudan University with a bachelor's degree in computer science, and then received a doctorate degree in computer science from the Hong Kong University of Science and Technology, becoming a tenured professor at the Chinese University of Hong Kong. The DV Lab (Deep Vision Lab) he founded at the Chinese University of Hong Kong is considered the leading laboratory in Asia in the field of CV and AI. During this period, Jia Jiaya also went to Microsoft Research as a visiting scholar, and the dean at that time was Shen Xiangyang.

In 2017, Jia Jiaya joined Tencent Youtu Lab and was elected Tencent Distinguished Scientist, a laboratory known for its visual technology, which supports Tencent's high-definition and latency-free live streaming services.

The reason why she left Tencent and chose to enter the manufacturing industry is because Jia Jiaya believes that the most fundamental theme of human development is how to produce on a large scale, and technology and software must be carried by hardware to drive production and life, "just like having a PC where you can play games, having a good tea set where you can drink tea with others, and having a motorcycle where you can go out and drive."

At the end of 2019, Jia Jiaya left Tencent and founded Smartmou, a company that provides tools for factories, such as sensors for quality inspection, with shipments of 100,000 units, supplying international manufacturers such as 3C, chips, automobiles, and new energy.

Source: Visual China

In addition, in the smart factories built by Jia Jiaya for major manufacturers, there are all kinds of whimsical intelligent machines: such as an automatic cream-squeezing robotic arm to help cake factories achieve aseptic production, and an air-jet robotic arm that can blow up biscuits to automatically turn them over, helping food factories save manpower......

These products are provided to customers in a combination of software and hardware, which may be the "product" thinking that Jia Jiaya learned at Tencent - to deliver to customers, out of the box: "There was no way to replace it with machines before, so we found the pain points to do, and the solution to the pain points is to use AI, not pure digitalization." This is Jia Jiaya's unique thinking after combining countless experiences in academia and industry.

Qin Tianyi, investment director of Zhen Fund, an investor in Smartmou, said: "The market changes rapidly, but the biggest constant point in this is actually the best group of people. ”

The fate of technology investment will always be to balance the advanced technology with the current competition. Qin Tianyi said: "When your performance can beat others, there is actually no competition. ”

Jia Jiaya completed four rounds of funding in 18 months, and Smartmou became the fastest startup to reach unicorn at the time, and it last announced the news of financing in June 2021, receiving $200 million from IDG Capital, Sequoia China, Lenovo Ventures and other joint investments. Subsequently, the unicorn chose to keep a low profile in the announcement of the financing.

Around the year 2000, machine learning was used by many manufacturing companies, but with the advent of deep learning, machine learning with sample learning at its core has become obsolete.

Taking the mobile phone horn quality inspection sensor as an example, the previous method of machine learning was to let the machine first read the pictures of 1 million mobile phone speakers, or learn the sound 1 million times or 100,000 times in order to distinguish what kind of good speaker is and what kind of flawed speaker is.

"But it doesn't understand what it is, it just understands what is good and bad for what you give me, and makes a judgment. Jia Jiaya said.

The development of artificial intelligence technology path is machine learning, deep learning, and large models. When deep learning emerged, Jiajia's team was one of the first research teams. When he left Tencent in 2019 to start Smartmou, the technical foundation of all the business was based on deep learning, but many companies and factories in the industry were still using machine learning technology.

After that, the big model came, and the machine understanding made a breakthrough, it went from "knowing what it is" to "knowing why it is", and the outbreak of smart factories was only one step away.

In 2022, SmartMore began to research large models, and in November 2023, it released the world's first industrial multi-modal large model IndustryGPT, completely breaking the gap of large models in the industrial field. Jia Jiaya's goal is to make it constantly develop and iterate and become an intelligent body with the ability to think and learn like a human.

Robotic Arm VS Robot

Large-scale model technology provides machines with the ability to think and learn, but people are arguing about one thing - which is the best solution for a smart factory, a robotic arm or a robot?

Let's take a look at Musk's options first.

From 2017 to 2018, Tesla's Model 3 encountered a capacity trap, Musk slept directly in the Nevada Gigafactory, personally inspected every link of the assembly line, and somewhere, he found that the robotic arm was delaying the speed, he directly got started, replaced the robotic arm, and the result was much more efficient.

"It's that I'm too stupid and trust in automation too much. Musk later scolded himself.

Musk got half the right thing by seeing people as an obstacle to efficient production, so he pushed automation on a large scale, but the other half did it wrong because the robotic arm wasn't good enough at the time.

Automation may seem like a wonderful word, but in essence, it is a highly planned program, and the range of movement of each robotic arm is accurate to a few millimeters, even if it is a robot, its route and trajectory are rigorously planned.

Jia Jiaya has been thinking for several years, he concluded, an important indicator of intelligent manufacturing is flexibility, the machine can adapt to the various processes of the factory, taking the robotic arm to grab glass as an example, the program can never predict what force should be used to grab the glass, maybe one thousandth of the force changes, the glass will break.

"People always overestimate the development of hardware technology and underestimate the speed of software development, ChatGPT to 4.0 to 5.0 is not unusual, it will show a very smart state, but if it is placed in a robot, it is too unusual. Jia Jiaya said.

For the choice of robotic arm and humanoid robot, Jia Jiaya believes that it is a robotic arm, he believes that the most important consideration of factory production is stability, rather than the advanced tool, "the robot will not see the results in 5 years, it is impossible to use it at home and it is impossible to mass production, but the robotic arm has been developed on a large scale for thirty or forty years."

Source: Visual China

What's more, it may take 10 years for a new thing to enter the production system of the factory, and it will have to go through the integration of the machine, the docking of various modules, the communication protocol between the machines, and the confidentiality of all kinds of information in the factory.

The curtain of such a scene is slowly unfolding: in a factory with no light, no human traces and no oxygen, the machines run autonomously, the equipment produces autonomously, this is a world of pure things. All kinds of applications exist in a human or non-human state, and they shuttle freely between various production lines and processes.

"Only by realizing large-scale, non-intervention-free automated production can human beings slowly show automation at other levels and areas of society, so this kind of automation does not appear so quickly in the human family. Jia Jiaya said. The factory of the future is adapted to machine production, not to the environment in which humans live.

Lonely Journey

In mid-April, DV Lab released the multimodal model Mini-Gemini, with considerable inference and picture understanding. The Mini-Gemini offers a choice of parameters from 2B (2 billion) to 34B (34 billion). In a number of indicators, it is not inferior to Google Gemini Pro and GPT-4V. At present, Mini-Gemini is open source from code, model, and data.

Jia Jiaya hopes that Mini-Gemini will play a positive role in promoting the development of the open source large model community in the future.

At present, Jia Jiaya also holds a faculty position at the University of Chinese Hong Kong, and lectures to students every week, and half of the DV Lab he leads is working on NLP to try to make large models smarter, and the other half is working on multimodality, which is a combination of vision, hearing, and large models, which are becoming more and more like an agent.

Xu Li, CEO of SenseTime, is a student of Jia Jiaya, and Shen Xiaoyong, CEO of Smart, is also a student of Jia Jiaya. Jia Jiaya's Ph.D. is the highest goal of graduating, because of his academic rigour. He told "Chinese Entrepreneur" that the paper written by a person should be useful as the core, useful can solve problems, used for practice, and interest is just an attachment.

AI is a very small circle.

While working as a scholar at Microsoft Research and teaching at the University of Chinese in Hong Kong, Jia Jiaya got acquainted with Sun Jian, the former chief scientist of Megvii Technology, and Tang Xiaoou, the founder of SenseTime, Sun Jian and Tang Xiaoou died in 2022 and 2023 respectively, and he also posted a special message to mourn, he said that Tang Xiaoou had a profound impact on himself, "To do big things, to do things of the top team."

For Jia Jiaya, Zhi Huijun, and Yang Zhilin, the exploration of AI is destined to be a lonely journey. Yang Zhilin once told the media that this is a feeling of hiking in the face of snow-capped mountains, the technology is unknown, and the path is unknown.

The ultimate goal of transforming the factory and upgrading the factory is to realize unmanned factory production.

It seems that some of the world's brightest people, including Jajaya, are on the adventure of creating "smart and smart factories," and who can be "Ford" or "Watt" on this road may not matter so much.