AI voice unicorn Yunzhisheng hit the sci-tech innovation board, and its core-making power and market planning power attracted attention

 Not long ago, Yunzhisheng formally submitted the IPO prospectus materials to the Shanghai Stock Exchange and entered the inquiry stage smoothly.

In the summer of 2012, in the domestic speech recognition circle, the news of Huang Wei "going away" to start a business spread like wildfire. As a technically capable person in the field of speech recognition, Huang Wei previously served as a senior researcher at Motorola for a long time, helping the company launch the world's first voiceprint recognition mobile phone; later, he served as a core executive of the Shanda Innovation Institute, and established a reputation for a while The voice branch.

Gao Guang's resume, Huang Wei's "where" undoubtedly aroused the attention of the industry. It didn't take long for a speech recognition company named "Yunzhisheng" to be formally registered.

In Huang Wei's own words, even though he has made a lot of achievements in the field of voice, when Yunzhisheng was just established, the company's specific direction was not completely determined, but it only saw the rapid development of mobile devices. However, what he never doubted is that in the future smart devices, "sound" will definitely become a very important way of interaction, which is promising in various business scenarios.

Slogans like "18 months is the time window" and "6 months is the line of life and death for startups" are not unfamiliar to the slightly "impetuous" technology startup circle. In contrast, there are very few companies that truly rely on innovative technologies and continue to invest in various business application scenarios. At the moment, Yun Zhisheng has quietly passed his eighth birthday and has become an important player in the AI ​​voice segmentation industry.

Starting from the voice recognition track, to providing a complete intelligent voice solution, accompanied by the start of core manufacturing, the "cloud-end-core" product strategy...In the past eight years, Yunzhisheng’s technical, engineering and industrial capabilities have Continue to mature. Yunzhisheng has grown into an AI voice company with "full stack" capabilities. At this time, it may be the right time for Yun Zhisheng to exert his strength to the next target and sprint into the capital market.

Not long ago, Yunzhisheng formally submitted the IPO prospectus materials to the Shanghai Stock Exchange and entered the inquiry stage smoothly. In this regard, the industry has heatedly said that "the first AI voice on the Sci-tech Innovation Board" may be born soon. Under the public's expectations, where is Yunzhisheng's hard power in technology? How high is the commercial moat of the enterprise? With the blessing of capital, what changes will Yunzhisheng's future development have?

Eight years of AI voice accumulation from unicorn to impact on science and technology innovation board

In the 1980s, the American futurist Alvin Toffler’s scientific and technological masterpiece "The Third Wave" was officially published. The book boldly predicted the digital age with computers and digital technology as the core driving force, and set off the scientific and technological circles Optimistic expectations of the "information society".

Passion is put into practice, and the future has arrived. After 2010, the digital economy has grown rapidly, and the world has officially entered the mobile Internet era. Among them, the new infrastructure revolution represented by 5G and artificial intelligence has become the most important source of power in the digital economy.

It is worth noting that during this period, not only were there a wide variety of networked devices, but also the networking scenes continued to produce new breakthroughs. Users can access the network through common hardware such as mobile phones and computers, and all kinds of home appliances, home appliances, and even medical and automotive hardware can be transformed into network interfaces to enrich the business scenarios of networking.

In this context, AI vision and voice tracks were born and gradually became an important part of the domestic artificial intelligence market. Compared with images, the voice interaction process and interaction logic are more complex, which also determines the complexity and difficulty of commercial application of its products, and users have higher expectations for product performance. On the other hand, the image also has a more fixed scene application than the voice track. Therefore, the commercial realization of the image field is more stable.

There is a voice on the market saying that for voice companies, survival means more testing the team's ability to "market planning". At this level, Yunzhisheng's technological accumulation and commercial application path may be worthy of reference by the industry.

In terms of intelligent voice interaction products, Yunzhisheng can currently provide cross-hardware platforms, cross-application scenarios, and cloud-core integrated IoT voice interaction products. At the same time, it can also face industry-level customers, covering "perception", "cognition", "Decisions" and other different levels of intelligent voice tool products. In the field of smart IoT solutions, the company mainly provides integrated smart IoT solutions for specific scenarios such as hotels, communities, residences, and hospitals, helping many enterprise-level customers to achieve interconnection between users and devices, and between devices and devices. interactive.

Regarding the development logic of Yunzhisheng in the AI ​​voice track, Huang Wei concluded, “Because we are at the forefront of the industry, not in the research room, we know what our customers need better. If this technology is to meet the needs of scenarios, What must we do."

Two minority choices to upgrade business liquidity

After the precise positioning of the track, the sustainable realization of the company also requires strong innovation capabilities. In this regard, Yun Zhisheng's two technical choices must be mentioned. Among them, the first introduction of DNN in 2012 can be regarded as Yun Zhisheng's first "minority" attempt.

The so-called DNN refers to the "deep neural network algorithm", which can successfully increase the previous speech recognition rate significantly. It has been a popular topic in the field of machine learning in the industry and academia in recent years. However, the verification of model correctness is complicated and requires more time and money to test, which is also the main feature of DNN.

For Yun Zhisheng, who had just entered the initial stage of entrepreneurship, the cost of trial and error for the enterprise can be imagined. But at this moment, Huang Wei still made the decision to enable DNN. In his view, with the increase in computing power and the expansion of the sample database, DNN is the most effective way to improve AI "IQ", and higher costs also mean higher and faster returns.

In the later commercial practice, Yun Zhisheng's technical prediction was also proved to be correct and efficient. After choosing DNN, the speech recognition accuracy rate of Yunzhisheng products has been greatly improved, and soon reached the standard for commercial use. From the current choices of most companies, DNN has also become a mainstream choice.

The second minority choice of Yunzhisheng is reflected in its layout of artificial intelligence voice chips. But compared to choosing the DNN algorithm, Yunzhisheng's "core-making" road has experienced greater doubts.

Since all core artificial intelligence technologies are computationally intensive, they have a rigid demand for hardware computing power. While providing sufficient computing power, the edge-side chips of the Internet of Things also need to consider functions, power consumption, cost, Many factors such as reliability. Therefore, without the ability to provide sufficient computing power from the end-side and the relatively cheap underlying hardware support, it will be difficult for the company to have the best cloud-side intelligent solutions, and there is no way to talk about rich commercial returns.

As Huang Wei said, "Not making a chip is a dead end." This is not a question of good or better, but a question of "to be or not to be". If you think this thing must be done, then just do it.

Despite the controversy, it seems that this time the core-making decision has become an important reason why Yunzhisheng is different from other AI voice companies and shapes the technological moat.

Public information shows that since 2015, Yunzhisheng has started to deploy artificial intelligence voice chips and began to develop uDSP processors and DeepNet IP technology. On this basis, it was the first to deliver artificial intelligence voice chips-"Swift" in 2018. In 2019, we will successively launch the car-grade chip "Snow Leopard" and the second upgraded version of the chip "Hummingbird" series for the home furnishing field, and start the research and development of the chip "Dolphin" with "image + voice" multi-modal interaction function, continuing to lead Same industry...

According to the prospectus disclosed by Yunzhisheng, the company's operating income in 2018 has increased by three times compared with 2017, and its main sources are reflected in the chip and hardware fields. And if it does not personally get involved in chip manufacturing, Yunzhisheng may not be able to empower the industry on a large scale and at low cost, and it will not be able to truly step into the small circle of technology and enter the large circle of industry.

Summarizing the company’s development past, if we want to characterize every “minority choice” of Yunzhisheng, perhaps it is — always stand from the perspective of practical problem solving, stand on the height of user experience, stand on the actual solution to user needs Direction.

Multiple business application scenarios, expecting capital to be empowered

With the blessing of its own chips, Yunzhisheng has rapidly strengthened its ability to commercialize the technology. In specific commercial applications, the company has successfully developed two series of intelligent voice interactive products and intelligent IoT solutions, and has gained widespread recognition in the home and medical fields.

In residential and hotel scenarios, Yunzhisheng uses air-conditioning voice modules as a breakthrough, and through in-depth cooperation with leading home appliance companies such as Gree, it has expanded and formed an IoT voice interaction product series covering dozens of IoT devices. , The company further upgraded its single product to a comprehensive solution, and delivered it to hotels, communities and other scenarios.

In addition, it has also established a joint venture with Shimao Group to deliver smart IoT solutions in batches to many hotels under Shimao Group, bringing rapid growth in sales revenue to the company.

In the hospital scenario, in 2016, Yunzhisheng cooperated with Peking Union Medical College Hospital to take the lead in implementing medical record transfer solutions in China, and it was quickly promoted in the industry. It has been used in more than 100 hospitals and more than 500 Home hospitals are used in testing. In addition, Yunzhisheng has also built a large-scale medical knowledge map based on the needs of smart medical care, and launched medical record quality control systems, smart follow-up products based on this, and built and improved medical smart IoT solutions to expand its influence in the hospital scene .

Focusing on the fundraising project of Yunzhisheng preparing for the Science and Technology Innovation Board this time, it once again reflects the company's continuous and in-depth development goals for technical capabilities, engineering capabilities, and industrial capabilities.

According to the prospectus, Yunzhisheng intends to issue no more than 20.25 million shares and plans to raise about 912 million yuan for use in artificial intelligence technology mid-stage construction projects, artificial intelligence chip research and development platform construction projects for edge computing of the Internet of things, and artificial intelligence Technical smart hospital solution construction project and supplementary working capital.

In addition to the capital blessing, the favorable policies seem to foretell Yunzhisheng's future development opportunities. In the "Recommendations of the Central Committee of the Communist Party of China on Formulating the Fourteenth Five-Year Plan for National Economic and Social Development and Long-Term Goals for 2035" recently released, expressions such as "Science and Technology Self-reliance as a Strategic Support for National Development" have once again triggered the industry Hot discussion. In addition, the "recommendation" also pointed out that it is necessary to target frontier fields such as artificial intelligence, quantum information, integrated circuits, life and health, and brain science, and implement a batch of forward-looking and strategic national major scientific and technological projects.

Some insiders commented that technology, engineering, and industry are compounded, which together constitute Yunzhisheng's AI full-stack capabilities. It is not difficult to imagine that if it successfully enters the capital market, Yunzhisheng will continue to maintain its technology and application leadership, further consolidate its core competitiveness, and promote the further development of the domestic AI voice track.

Help the machine install eyes, "Dingna Automation" completed the B round of 100 million yuan financing

 Dingna's short-term goal is to become the world's leading provider of intelligent visual inspection solutions.

Machine vision is an important front branch of the artificial intelligence industry. Machine vision gives machines the ability to "see" and "cognize" by simulating the human visual system, which is the basis for machines to understand the world.

The global machine vision market has been growing steadily since 2015 and is expected to reach US$25 billion by 2022, with a compound annual growth rate of 22.1%. China's machine vision market has entered a stage of rapid development since 2015 , with a compound growth rate of 24.9% from 2015 to 2022, which is higher than the global level.

Suzhou Dingna Automation is a company that can provide machine vision inspection solutions for manufacturing customers . Recently, Dingna completed a 100 million yuan B round of financing jointly invested by Source Code Capital, Yuanhai Minghua, and Xiaomiao Capital . The Series B financing will be used to expand production capacity, meet customer needs, introduce high-end experts and talents, develop new products, and maintain the competitiveness of technologies and products.

Dingna was established in 2010. The company is committed to 3D vision ultra-high-precision measurement technology, deep learning complex defect detection technology, artificial intelligence object recognition technology and robot vision intelligent grasping technology, and launches a comprehensive software and hardware integration Automatic intelligent manufacturing equipment currently covers subdivision scenarios such as 3C, automobiles, 5G, display panels, and semiconductors.

"So far, Dingna has 10 years of experience in the industry. The past 10 years can be divided into three stages. From 2010 to 2017, we are mainly doing machine vision system modules , and we have been exposed to nearly a thousand application scenarios ; from 2017 -In 2019, we made an overall solution based on visual inspection; in 2020, some of our projects will begin mass production, and most of our target customers are from the 3C industry.” said Dingna CEO Qin Yinghua.

Qin Yinghua recalled that the cooperation between Dingna and US consumer electronics giants began in 2013, but initially focused on "indirect cooperation". " We provide vision systems to third-party equipment integrators, and then they provide them to each other as a whole." Since 2017, Dingna has become a direct partner of each other.

"We are facing the quality system of the other side's entire parts supply chain, such as glass, metal parts, plastic parts, etc. We have to conduct quality inspections for the production links of these parts." Qin Yinghua said.

Qin Yinghua compared Dingna's attempts in the 3C field in the first few years to "from 1 to 99", but now it is "from 99 to 99.99". "The step from 99 to 99.99 looks relatively small, but it is the most difficult step. Because 1-99 is mainly iterative, there may be no way to achieve mass landing, and a small number of machines need to be put into the customer’s production line. Verify whether the technology is stable and reliable." Qin Yinghua said. "But the last step from 99 to 99.99 means that a large number of equipment will be launched for mass production."

Qin Yinghua said that the entire process needs to systematically optimize and iterate on equipment hardware and software algorithms. What customers need is a stable and reliable equipment that has been repeatedly tempered, not a single industrial camera or a piece of the latest algorithm code.

In the first 7 years of the company's establishment, Dingna completed the accumulation from 1 to 99. "We have done many projects in non-3C industries and have accumulated rich experience in many industries. However, it is not easy to gain the trust of customers at the top of the pyramid . It took Dingna a full 3 years. From our indirect service to the customer, Established contact with them and upgraded to a "direct partner", one is because the technology has been recognized by the other party, and the other is because we insist on being customer-centric and can quickly respond to the other party's requirements and meet customer needs."

"The other party was worried that our company's volume was too small and the risk was not easy to control, so initially we only cooperated on small projects, such as visual inspection related to smart watch parts." Qin Yinghua said that Dingna successfully completed the task. Gaining the trust of each other, while the company expanded in a planned way, the ability to undertake projects was stronger, and the two parties gradually began to dock tasks related to smartphone parts.

Qin Yinghua said that in the past two years, the American technology giant has consciously cultivated its domestic supply chain. At the same time, as a whole, the market size of machine vision inspection equipment in the field of consumer electronics and related display panels, chip manufacturing and other core components will be close to 3 billion yuan in 2019, and will reach nearly 4 billion in 2020. Consumer electronic products have a short life cycle and a large demand, so in the next few years it will drive the rapid growth of machine vision market demand.

According to Qin Yinghua's plan, Dingna's short-term goal is to become the world's leading provider of intelligent visual inspection solutions, and the long-term goal is to establish a firm foothold in the consumer electronics field and penetrate into display panels, automobiles and other fields in the future. Currently, Dingna has self-developed standardized modules and productizable solutions in the process of serving customers across industries.

In terms of team, Dingna has a professional team of 300 people, mainly technical R&D personnel. Dingna Automation invests 15% of its revenue in research and development every year. The company already has more than 80 patents and software copyrights. At present, the company has branches in Shenzhen, Zhejiang and Silicon Valley in the United States, and is accelerating the layout of its R&D headquarters in Suzhou Park.

Artificial intelligence can only "optimize", while humans can "evolve"

 

Consciousness is the function of the human brain

Editor's note: Artificial intelligence is booming under the feed of massive amounts of data, but is it truly intelligent? The author of this article Douglas Rushkoff believes that artificial intelligence is not. Although artificial intelligence is superior to the human brain in computing and other aspects, artificial intelligence is still unable to compare with the neurons of the human brain. How the human brain generates awareness is still a mystery and needs to be solved by later scientists. The original title The Difference Between Optimizing and Evolving.

Image credit: Paul Burns / Getty Issues

Artificial intelligence is not life.

Because they will not evolve. Although artificial intelligence may iterate and optimize, but that is not evolution. Evolution is random mutation in a specific environment. In contrast, machine learning is aimed at a specific, pre-programmed purpose. It may be complicated, but not as complicated as evolution, weather, ocean, or nature. Complex systems, such as traffic lights that can be seen everywhere in cities, can direct the actions of people and cars. In contrast, complex systems, such as transportation circles, can spontaneously form flows through the interaction of many participants. The machine has many complicated parts and processes, but there is no higher order; the complexity of life is born out of them.

No matter how many names people give to new functions of computers that sound like neurological terms, computers are not aware. Some people are keen to imagine what computers will look like in the future. They like to use terms such as "fuzzy logic", believing that these programming techniques can make computers operate closer to human thinking. Fuzzy logic means that computer programs can handle values ​​other than 1 or 0 and can represent them as 1 or 0. This is the whole meaning of fuzzy logic. It can be seen that in the face of real uncertainty, fuzzy logic is not fuzzy at all. It just reduces the roughness and complexity of the real world to binary that can be processed by the computer.

Likewise, neural networks are not like the human brain. A neural network has layers of nodes, and its way of learning new knowledge is to analyze hundreds of cases input from the outside world. Instead of describing what a cat looks like to the computer, we feed it hundreds of pictures of "cats" until it can determine the common and distinguishable features of "cats" in these pictures. However, humans can generalize the category of "cats" only by seeing a photo of a cat. How does the human brain do it? We are not so sure.

We cannot say exactly what it means to be a thoughtful, conscious human being, but this should not be considered a shortcoming of the human brain. Human thinking is not computer-based, and reality itself is only information. The human brain has intelligence, which is a unique ability of the brain; there are a lot of data in reality, but if there is no human consciousness to reflect them, then the human brain and these data are meaningless. We cannot downgrade human consciousness to primitive processing power. For example, the human body can certainly lift weights, but we can never lift heavy objects as lightly as a crane. Similarly, our calculation speed cannot be compared with supercomputers. But the value of human beings is far more diverse than mere "usefulness". Using technological intervention or directly replacing the human brain with technology to strengthen an index that is conducive to employment will only forget other potentially more important values, the most important of which is consciousness itself.

Science has shown that consciousness is a completely incalculable quantum state in the tiniest structure of the brain-microtubules. In the human brain, there are billions of these microtubules, and each microtubule has many active points. If all the computer chips in the world are integrated into a computer, this computer will also be in the human brain. Defeated in front of the complex.

Only computer developers will act as if consciousness is simple enough to be replicated by machines. Real neuroscientists are still confused about the impossibility of self-awareness in a bunch of neurons. This conclusion contradicts reality.

This does not mean that we will quickly deny the existence of consciousness. Consciousness is not an illusion of human brain caused by DNA in order for people to have survival instincts. We do not live in a virtual world. Our consciousness is real. With the development of physics up to now, even physicists have admitted that the reason for the existence of consciousness is more sufficient than the existence of objective reality: quantum theory believes that it may not exist before we observe objective reality. In other words, the universe is a bunch of possibilities, until someone's consciousness approaches it and sees it in a certain way, then it can condense into what we call reality.

Searching for the origin of consciousness is a bit like searching for the smallest cosmic particles. It is more like a doctrine from the perspective of mechanistic science, rather than a true reflection of how the brain works. Whenever we find a final determinant, such as a gene, we will later discover that it is determined by something else. Once we discover a certain pathogenic bacteria, we will grasp its characteristics, discover the most suitable environment for its growth, and even discover the mechanism that causes it to cause disease. The only way to solve the problem of the origin of consciousness is through first-hand experience. In addition, when researching, we should maintain respect for the world in which we live and the other people who share the whole world with us.

In this sense, we know the existence of consciousness because we know how it feels. There is a cup of coffee on the kitchen table, we can see it, animals or computers can also see it, but humans are different, humans also know what it feels like to see a cup of coffee on the table. Choosing to look at that cup and noticing it is the uniqueness of human consciousness. Computers can’t do this. They can only see everything within the camera’s range. They don’t “pay attention”, they don’t have the focus of their sight. Know the real direction.

To understand what it feels like to look at a cup of coffee, you need to construct your mind and self. Currently, only humans can do this, because humans are humans and computers are just objects.

Douglas Rushkoff's new book "Human Team". Image source: Internet

This article is section 58 of Douglas Rushkoff's new book "Human Team".