Author: Hao Junhui Source: IT Times
On December 27, Baidu Create 2021 (Baidu AI Developer Conference) was held in the "Xilang" App. Xi'an is the first domestic meta-universe product released by Baidu, and Baidu said that it can accommodate 100,000 viewers to interact with the screen on the day of the conference. I have to say that Baidu's courage in this move is commendable, and Xi'an, which has been online for more than a month, is far from the meta-universe, and it is not satisfactory whether it is picture quality, characters or operation experience.
However, it is obviously not objective to judge Baidu's status in the metaverse era only by the "color" of Xi'an. Looking back at Baidu's development process in the past 10 years, although it missed the first wave of "ferry tickets" of the mobile Internet, Robin Li still showed a forward-looking attitude in more places, such as box computing, such as All In AI.
Dare to open a 100,000-person conference in the "Xilang" with a rating of only 2.3 in the App Store, Baidu may present a posture, a kind of "although tens of millions of people are going to die", just like the original plunge into AI. Or, Baidu is to define the starting point of an era in this way, Li Yanhong said in his speech at the conference, "the era of human-machine symbiosis is coming, and the creators will usher in the golden 10 years of Chinese intelligent intelligence". If this is true, perhaps 10 years later, someone will say that the 10 years that changed the history of artificial intelligence began with a Bug-filled Greek congress.
"The ideal is beautiful, the reality is bone"
I have to say that the experience of Xi'an is really "difficult to say in a word". No wonder that as soon as the concept of the meta-universe came out, all game developers were cheering and cheering, creating a beautiful virtual world with humanized operation and smooth interaction, which is indeed the strength of game companies. But Baidu is not without confidence in knowing the difficulties.
The bottom line is the "Baidu brain". As the integrator of Baidu AI technology accumulation and industrial practice, Baidu Computer includes flying propeller deep learning platform, Kunlun chip and core AI technologies and platforms such as speech, vision, knowledge graph, natural language processing, etc., with nearly 1400 AI open capabilities and daily calls exceeding 1 trillion times.
Robin Li defines Baidu Brain as the "AI toolbox" and the technical "big base" of social and industrial intelligent transformation, and its core essence is easy to understand: AI technology is becoming more and more complex, but AI technology development and application is getting easier and easier, by creating an AI production platform, truly releasing the productivity of AI, "using technology to make the complex world simpler"
From the perspective of "Baidu Brain", Xi'an is more like Baidu's demonstration garden and developer's test field. Baidu can not only apply its own AI underlying technology, present it to users, but also make it a social place for developers, after all, in the first science fiction novel "Avalanche" that proposed the concept of "metaverse", Metaverse (metaverse) was originally only a virtual world developed and played by geeks themselves. But Greem should become a space for developers' imaginations to land, and when a virtual world becomes more and more like the real world, then every experiment in Hiram may become a rehearsal to change offline life.
Of course, for the current Xi'an, these are just beautiful visions, at least in terms of product form, it is far from undertaking the grand goal of creating a "meta-universe platform", Baidu to the game company to dig the wall is still necessary.
Look up at the sea of stars with AI
Leaving aside The still concept stage of Greece, baidu's information revealed at the AI developer conference is very critical.
First, the increasing maturity of Baidu's flying propeller platform. Previously, Robin Li has introduced Flying Propeller on many occasions, which is China's first independently developed, feature-rich, open source and open industry-grade deep learning platform, which integrates core frameworks, model libraries, development kits and tool components, with chips underneath and applications on top, equivalent to "operating systems in the intelligent era".
Such an instrumental productivity platform is self-evident in promoting the AI of traditional industries.
According to Wang Haifeng, chief technology officer of Baidu, Feipao took the lead in the industry to realize the core framework of "dynamic and static unification", taking into account the flexibility of scientific research and development and the efficiency of industrial development, which can support developers to develop deep learning models conveniently and efficiently. Especially in terms of scientific computing API, Baidu Flying Propeller supports applications such as quantum computing, life sciences, computational fluid dynamics, and molecular dynamics.
In fact, AI empowers these basic science fields that have extremely high requirements for computing power, which are more worthy of attention. In the past, only a few scientific research institutions had the ability to carry out scientific research projects, but even these institutions faced the dilemma of lack of cross-border talents, understanding both professional and AI, and could withstand lonely scientific research talents. However, on the flying paddle platform, as long as you have some of the most basic computer operation foundation, have a certain text understanding ability, even if you don't know anything about the algorithm, you can design and generate AI models through the flying propeller.
For example, Baidu's ecological company Baitu Shengke, using its own biocomputing platform and large-scale pre-training model, greatly reduced the prediction time of typical protein structure, from the current industry benchmark of 30 minutes to less than 5 minutes, through the big model, can shorten the cycle of new drug research and development, reduce research and development costs, improve the accuracy and efficiency of target discovery.
For example, Baidu has laid out research directions such as quantum algorithms, quantum artificial intelligence, and quantum architecture, and created a 100-metric sub-platform. This is the first quantum computing cloud platform in China to provide one-stop service from applications to quantum processors, lowering the threshold for quantum computing learning and application, and promoting the application of quantum computing in the fields of chemistry, finance, and materials.
Compared with AI empowering traditional industries, Baidu Brain's layout in the field of basic research shows the yearning for the star sea in the innate genes of a technology company.
At this year's Create 2021, Li Yanhong had a dialogue with Ouyang Ziyuan, the first chief scientist of the astrochemistry and China's lunar exploration project and academician of the Chinese Academy of Sciences, to some extent, and Liu Cixin, who is also a fellow of Yangquan in Shanxi, looked up at the same starry sky, "Whether human beings are the only ones in this universe, it is the ultimate exploration of human beings for their own sense of meaning and belonging, and human beings strive to hold up the torch of civilization and rush to the vast universe through the scientific and technological capabilities they have created. In the end, it is to prove that I am no longer alone in this universe. ”
Previously, Baidu and China reached a strategic cooperation on lunar exploration and space projects, and Baidu's AI technology will be applied to China's deep space exploration fields such as lunar exploration and planetary exploration. Robin Li said that the perception and control capabilities of artificial intelligence have played a very important role in deep space exploration, which can help robots in deep space exploration to learn the external environment independently and make decisions and judgments autonomously.
The "Wild Hope" of Intelligent Transportation
Another important piece of information revealed by Baidu at the conference is its "wild hope" in the field of intelligent transportation.
Baidu, which has always been known as the "domestic autonomous driving Whampoa Military Academy", finally entered the substantive car manufacturing link in 2021, and on March 2, Baidu and Geely's joint venture company Jidu was officially established and began to build cars independently. According to Robin Li, who will become an "automotive robot" in the future of driverless cars, Jidu's automotive robots have now entered the intelligent driving and intelligent cockpit development stage of simu car (software integrated simulation car), and it is expected that the first concept car will be announced in the first half of 2022, and the first automotive robot will be delivered in mass production in 2023.
In Robin Li's definition, car robots have L4-level automatic driving capabilities and can "move freely"; they can accurately identify human-vehicle interaction and speech semantics, and cars and people can "communicate naturally"; they can learn and iterate on their own, and they can "grow themselves".
Judging from the "IT Times" reporter's Baidu's self-driving platform "Radish Run" experienced by shanghai Apollo Park, in the driverless state, Baidu's ROTOTAXI can turn, brake, and avoid, and the cost of the car continues to decline. According to Robin Li, in the past quarter, "Radish Run" provided 115,000 times of service, which made Radish Run the world's largest provider of self-driving travel services, and Baidu is expected to expand its services to 65 cities in 2025 and 100 cities by 2030. Compared with Baidu's previous announcement of its goal of expanding to 30 cities within three years, this is a nearly doubled increase.
Full of optimism about the future, it is the biggest feeling of Robin Li's speech. In addition to the acceleration of the "radish run" Kaicheng, Robin Li judged that Within 5 years, China's first-tier cities will no longer need to restrict purchases and restrictions; within 10 years, basically the congestion problem can be solved.
This prediction is based on Baidu's current exploration in the field of intelligent transportation. The so-called intelligent transportation refers to the deep integration of artificial intelligence, 5G communication, cloud computing and other technologies into the transportation field from the automatic driving, and solves traffic problems through "smart cars" and "smart roads".
At present, Baidu has landed practice in several cities in China. For example, Baidu's AIR intelligent road system deployed in more than 300 intersections across Yizhuang has the same perception, decision-making and control capabilities as Baidu Apollo technology. Cameras, lidar, millimeter wave radar allow intelligent road systems to "see six roads"; and Kunlun chip, Apollo, flying paddle deep learning platform blessing road edge computing platform, is the intelligent road system "brain", it can continue to learn from itself, continue to evolve; for example, hebei Baoding core area in the past two years the average annual growth of motor vehicles 7%, congestion index through some road section information control optimization fell by 6%; Guangzhou Huangpu District, the average number of times each car encountered red lights, from three or four times to about 1 time. Baidu found through the simulation of the combined graph neural network that if all intersections in a city are intelligent and city-level regional information control optimization, the traffic efficiency can be improved by 15%-30%.
It is only the problem of traffic congestion, often not just a technical problem, the reason why the city is called an organism, is because of its own development and non-objective factors, and compared with the "smart car", the "smart road" involved in the industry is more complex and diverse, belongs to the field of non-pure technology can change. Robin Li's optimism can be seen as a beautiful vision of an engineer, but whether it can be realized or not, the decision is not necessarily in the hands of Baidu.
But in any case, for Baidu, which always adheres to "changing the world with science and technology" and "making the complex world simpler", its investment and determination in AI are worth looking forward to. According to the financial report of the third quarter of 2021, Baidu's total revenue was 31.9 billion yuan, and the R&D expenditure was 6.2 billion yuan, an increase of 35% year-on-year, and the R&D cost accounted for 19.43%. Such a high proportion of R & D investment, can only be shoulder to shoulder With Huawei, Huawei's total revenue in 2021 is 634 billion yuan, of which R & D expenditure is 124 billion yuan of R & D expenditure, accounting for 19.56%.