Home > News content

Baidu lifeline AI

via:博客园     time:2019/6/3 17:07:24     readed:2049

Burning financial (ID: rancaijing) original

Author: Wang Lin

Editor: Allen

Departed from Hailong, Wang Haifeng took the position, Baidu's two eras — — search and AI (artificial intelligence), is accelerating the handover.

On May 17, Baidu Senior Vice President and President of Search Company announced the resignation of Shi Hailong; on May 31, Baidu announced the appointment of Senior Vice President Wang Haifeng as Chief Technology Officer of Baidu Group and continued to be the overall responsibility of AI technology platform system and basic technology system. people.

On the next one, the intention is obvious.

After missing many opportunities in the mobile Internet, Baidu hopes to directly transition from the PC era to the artificial intelligence era, and for this radical layout, Baidu AI has formed a layout with Baidu brain as the core, DuerOS, Apollo, and intelligent cloud. Among them, DuerOS is responsible for keeping the search, Apollo is betting on the future, and the intelligent cloud is partly responsible for realizing it.

However, the three carriages have their own problems. The DuerOS's small-sized speakers are subsidized to obtain the market. However, Baidu's IoT layout faces strong opponents such as Xiaomi and Ali. The automatic driving is in the foreseeable time and faces policy risks. Late in force, can only be killed in the stock market.

Baidu wants to “invest in growth”, but the capital market asks Baidu to make a profit. This requires Baidu to ensure continuous growth in performance and balance new business investment and returns. At present, Baidu is not doing well. On the day of the departure of Hailong, the first quarterly earnings report released by Baidu in 2019 showed that the quarter's revenue did not meet expectations, and the profit turned from profit to loss. This is Baidu's first quarterly loss since its listing in 2005. In the half month after the release of the earnings report, Baidu’s market value fell by nearly 30%.

Leave Baidu AI to prove that your time is running out.


Four key sir

Even after Lu Qi left, Li Yanhong publicly stated that he never said “All in AI”, but Baidu’s attention to AI has long been known.

The most direct manifestation of this emphasis is the adjustment of the company's organizational structure. After Lu Qi left in May 2018, Baidu's two organizational restructurings had two major moves: the search company transformed into a mobile eco-business group, while the smart cloud business unit upgraded to a smart cloud business group, one litre and one drop. Between, Baidu's transformation is determined.


After this adjustment, Baidu formed seven business groups, among which, the four groups of business focus are related to AI —— artificial intelligence business group is mainly responsible for all AI technology research and development of Baidu; intelligent driving business group Responsible for the intelligent driving of the car; the smart life business group focuses on the construction and operation of the DuerOS platform and ecology; the intelligent cloud business group carries the artificial intelligence to B business and cloud business.

This is highly consistent with the Baidu AI picture that Lu Qi painted. In the Baidu Developer Conference in 2017, Lu Qi pointed out that the development path of Baidu AI is platform and ecological, and the mode of obtaining commercial returns is “data + algorithm + software + hardware”, which achieves user value and then generates commercial value.

Specifically, the Baidu AI platform consists of two parts: Baidu Brain and Baidu Smart Cloud. The AI ​​platform is the foundation for developing the application layer. Baidu's AI capabilities will also be open to developers through the AI ​​open platform. On this basis, Baidu launched DuerOS, a voice interactive system and Apollo, an open platform for intelligent driving. The former provides integrated software and hardware solutions, which help automakers and autopilots quickly build their own autopilot systems.

Nowadays, the various departments that undertake the AI ​​business are independent and gain higher weight, which is enough to see Baidu's attention.

Not only that, but within Baidu, the department related to AI has also become “饽饽香;”. “Intra-company promotion, AI related must be faster. One edge department may have 10 individuals qualified to report, only one promotion place, but 10 people in the AI ​​department report, it is likely that 8 people will be promoted. ” Baidu middle-level employee Wang Da told the burning of finance.

In order to encourage employees to improve their skills, there will be some AI-related competitions inside, once or twice a year, and the bonuses are very high. ” A Baidu employee told the Finance.

More importantly, as Haidu search spokesperson's departure from Hailong, this seems to Wang Da is Baidu farewell to search, another performance of All in AI.


Baidu Organizational Structure Scheduling / Burning Finance

Today, Wang Haifeng, Li Zhenyu, Jing Hao and Yin Shiming have become the key people of Baidu AI. Among these four people, Wang Haifeng’s identity is more special. Wang Haifeng joined Baidu in 2010 and is now a member of Baidu Estaff, while Estaff represents Baidu's highest decision maker. At the same time, on May 31, Wang Haifeng was officially appointed as Baidu CTO. Prior to this, Baidu CTO was once vacant for 10 years. After this appointment, Wang Haifeng is the heavyweight of Baidu AI, who is the head of Baidu CTO, the head of AI technology platform system and the head of basic technology system and the dean of Baidu Research Institute.

Everyone else is at the vice president level. Li Zhenyu joined Baidu in 2007 and worked as the head of the AI ​​platform department. “Baidu Brain” was born in 2014. Jing Hao joined Baidu in 2014 and made great achievements on the DuerOS platform. Lu Qi believes that it is the best. Product Manager; Yin Shiming joined Baidu in 2016 and previously served as General Manager and Head of Ecosystem for Apple Greater China.

They will continue to be responsible for building the AI ​​Building on the land of Baidu, which is based on the Baidu brain.


a core

It is reported that in 2014, the “Baidu Brain” program was regarded as one of Baidu's decisive strategic plans for the next 5 to 10 years. For this reason, Baidu also recruited Google Brain project founder Andrew Ng to take charge of Baidu brain research and development. .

For five years, Baidu's brain has evolved from 1.0 to 3.0:


1. In the 0 stage, Baidu's brain mainly completes the basic ability building and core technology initial opening, namely calculation method + computing power + big data, which is the basic layer of Baidu brain, and is also recognized as the three pillars of AI in the industry.

2. In the 0 phase, Baidu's brain formed a complete technical system, including the complete technical layout including the basic layer, the sensing layer, the cognitive layer, and the platform layer, and opened more than 60 core AI capabilities.

3. In the 0 stage, Baidu's brain first proposed the concept of “multimodal deep semantic understanding” in the industry, forming a AI full stack technology layout from chip to deep learning framework, platform and ecology, and opening more than 170 AI capabilities.

Among them, multi-modal depth semantic understanding includes the semantics of data, the semantics of knowledge, and the understanding of images, video, sound, and voice. It allows the machine to understand, see, and understand the real world in depth. Good support for a variety of applications.

Take visual semantics as an example. The visual semantic technology first recognizes people, objects and scenes, and captures the behaviors and relationships between them. It forms semantic knowledge through sequential, digitized and structured methods, and finally combines fields and scenes for intelligent reasoning.

Today, Baidu's brain technologies include image and face recognition, speech semantics, reading comprehension, etc. All of these technologies have a common foundation at the bottom, namely paddlepaddle, similar to Google's TensorFlow. Paddlepaddle is a deep learning framework independently developed by Baidu from 2012. Now it has been updated to paddlepaddle3.0. With the support of this framework, developers can realize the rapid application platform of zero-based threshold.


Baidu brain four-layer technology architecture finishing / burning financial

The latest data from Baidu's AI open platform shows that Baidu's brain has opened 171 technical capabilities and achieved 24-hour rapid integration. The developer's usage exceeded 1.1 million.

Externally, Baidu's brain empowers all walks of life. For example, combined with Baidu's big data, deep learning and other technologies, it is possible to collect, transmit, store, and analyze the data of the charging pile equipment, so that equipment monitoring, fault diagnosis, and predictive maintenance can be realized.

Internally, Baidu's brain is the core engine that drives the two platforms of DuerOS and Apollo.


DuerOS: Biting the search

Search is a place where Baidu started. Today, although Baidu search is still the first in China, in the era of mobile Internet, users' search habits are directly carried out in one APP, which makes it difficult for Baidu search to achieve a centralized entrance effect. The search competition of Sogou, Weibo and other companies has given Baidu a lot of pressure.

“The future search technology is not limited to computers. It can be integrated directly into cars and household appliances, and even integrated into public facilities. ” Baidu’s middle-level employee Zhang Fei told Chen Finance, “But Baidu has never wanted to understand what the future search scenario looks like, and there is no thorough research on future payment scenarios by Ali”.

Until 2017, when Ali, Google, and Xiaomi fought smart speakers, Baidu made a force. “It’s already late, maybe at that time Baidu still didn’t want to understand what the ultimate goal was, just to see others do it themselves. & rdquo; Zhang Fei said.

At the Baidu AI Developers Conference in 2017, Baidu launched DuerOS. At that time, Baidu conveyed that DuerOS is Android in the AI ​​era. DuerOS includes two directions: the open platform of smart devices and the open platform of skills.

The intelligent device open platform is aimed at intelligent hardware vendors and provides a layered technical service solution. Currently, it includes three types: core access components, development kits, and reference designs, which can meet the needs of different types of vendors.

The skills open platform is targeted at content providers and service providers. In addition to its more than 100 native capabilities in its 10 primary categories, DuerOS also accesses a wide range of third-party resources and content such as music, audiobooks, news, entertainment and more.


DuerOS function displayed on Baidu AI Open Platform official website

In 2018, Jing Hao further elaborated DuerOS's business model: The Skill Store on DuerOS is similar to the App Store on iOS. Similar to the App Store, the Skill Store has two ways to monetize: build paid skills, or pay in-app.

For example, after DuerOS 3.0 is officially launched, users need to pay when downloading an application. This is a “paid skill”; when users buy an audio and video in an application like VIPKID and “Ke Shu Story” When the content is, it is "paid within skills".

Relying on DuerOS Baidu has launched a variety of intelligent hardware such as small speakers.In May of this year, Jing Hao publicly stated that smallness is the future search. The user originally turned on the smartphone or PC to get the information, but in fact, the user gets the information regardless of time and occasion. Sometimes we are in the car, sometimes at home.data-copyright=0

For this we see that Baidu is constantly adding code on the smart hardware side. In the price of smart speakers, Baidu's adjustment speed is amazing. From November 2017, the Raven of 1699 yuan was launched. In March 2018, the small price of 599 yuan was introduced at home. In June 2018, the small smart speaker of 89 yuan was launched. Baidu's adjustment of smart speakers was amazing. It can be seen that it does not count profits, does not hesitate to cost, and wants to establish a firm foothold in the smart speaker market.

At the same time, since 2018, the small series hardware family has grown stronger, and has introduced small-scale home, small-scale smart speakers, small-scale smart speaker Pro, small-scale voice car bracket, and small TV companion.

Not only that, Baidu has already reached cooperation with a number of companies, in order to create a smart home ecosystem with DuerOS as the core, to seize the entrance as much as possible.

But Baidu's advantage is not obvious. Unlike Xiaomi's investment in building eco-chain companies, Baidu and Ali are more inclined to cooperate with traditional home appliance companies to develop the IoT eco-layout. Gree announced that it will cooperate with Tmall in the IoT field, and Gree's home appliances will be connected to the Tmall Elf. Coupled with Ali's advantage in the field of e-commerce sales channels, its appeal to traditional home appliance companies is clearly stronger than Baidu.

Obviously, Baidu wants to get bigger search scenarios through the openness of DuerOS and enough ecological partners. “With more search scenarios, you can take more time for users and increase user stickiness. Is there any traffic that is connected to the same business as it is now? & rdquo; Zhang Fei said. But for now, "the reality is that the positioning of the small is a sound, far from the smart search."

Not only that, for Baidu, just doing the search well, can only keep the current market value, and want to return to the front line, Baidu needs to find something other than search.


Apollo: betting on the future

In addition to the search, Baidu Heavy Gold bet the driverless.

In April 2017, Baidu announced the “Apollo Project”, announcing the creation of an autonomous driving platform. It wants to be Android in the autonomous driving world and become the biggest matchmaker among developers, automakers, parts dealers, and new car companies in the autonomous driving industry.

According to Apollo's official website, Baidu already has 137 partners, including local governments, car companies, startups and other companies in the automotive industry chain.

Autopilot is a hope for Baidu. In the Baidu Developer Conference in 2018, Li Yanhong said that the world's first L4-class mass-produced autonomous bus "Apolon" was mass-produced, which indicates that Baidu did not stop at the stage of "PPT car-making". Today, Apollon has already produced 100 units, and they are sent to Beijing, Guangzhou, Shenzhen, Pingtan, Xiong'an and Tokyo, Japan.


But it is not so much the automatic driving is Baidu's hope, it is better to say that it is wild. Unlike Google's Waymo's own sensor kits, Baidu chose a more labor-saving method —— the hardware needed for autopilot (lidar, vision, radar, and computing platform hardware, etc.) Basically, they are handed over to partners, while they are mainly focusing on software.

This allowed Baidu to quickly achieve “production”, but it was also a problem. Yu Kai, the founder of the Horizon, said that in the future, to truly solve the application of artificial intelligence, it must be from software to hardware, software and hardware.

Yu Kai said that through the practice of the horizon, it is found that there is a separation between software and hardware. This separation is iterative at the speed of the week and month. “We are seeing a trend. In the autonomous driving industry, semiconductor manufacturers are not just doing hardware. They are actually moving more and more to the upper layers, and more and more are building software architectures. We also see another scenario where traditional players who used to only make software are now moving in the direction of soft and hard. For example, Google, which used to be mainly software, is now developing towards artificial intelligence hardware and software and hardware integration. ”

At present, Baidu does not have the ability to integrate software and hardware, and there is no determination to integrate soft and hard. The separation of hardware and software is likely to pose a safety hazard to the self-driving vehicles that are about to go on the road.

On the other hand, although Baidu's Apolon claims to have achieved L4 level autopilot, it is completely different to achieve L4 level autopilot on a closed campus and on an open road.

A Baidu Apollo platform partner said that the autopilot technology is now between 2.5 and 3.0, while the research is 3.5. This means that it takes a long time to realize an unmanned vehicle that can be mass-produced.


According to official data, the Apollo minibus in Beijing Haidian Park has a full load speed of 20km/h, but the test found that the maximum speed of the car is limited to 10km/h, except for the downhill slip. It reached 10km/h and the vehicle was not fully loaded, only less than 60%.

Some media reports that the cost of an Apollon is about 1 million. Although Baidu said that Apolon has customers paying for it, such a high cost, how to mass-produce and mass-market is a problem. It seems that Apolon is more like It is an experiment by Baidu.

In addition, there are two current business models for driverless driving. The first is unmanned taxis. For example, Waymo, a company owned by Google in the United States, adopts this model. The other is unmanned trucks, such as transporting on the road. Unmanned trucks, park logistics vehicles, and some special vehicles such as cleaning vehicles and campus vending trucks. However, the former is restricted by law and the latter is too small.

What's more, “traffic involves life safety, and the renewal of a transportation method has to be at least a hundred and eighty years. A company can't afford the cost of time, public opinion pressure, technical cost, and policy restrictions.” Zhang Fei said.

No one can tell when the automatic driving is in the end. In the opinion of a Baidu middle-level employee, “autopilot is more like an investment that has to be made to increase the stock price. ”For companies, the investment and output ratio must be considered. Therefore, Baidu unmanned vehicles recently sent out the split news, although Baidu responded to the news is not true, but many investors told the burning of finance, Baidu smart driving has begun to contact outside capital.

But Baidu seems to have missed the best time. “Now the talents are gone, the financing is already late, and it is best to do this in 2017. "A driver in the driverless sector said that according to the current situation of Apollo, the Baidu unmanned vehicle business is estimated to be between $5 billion and $10 billion.


Smart Cloud: Bear the realization

As one of the Baidu AI troikas, Smart Cloud is the only business that can generate revenue.

The “China Public Cloud Service Market (2018 second half) Tracking” issued by IDC, the world's leading consulting organization, shows that in the third quarter and the fourth quarter of 2018, from the overall market share of IaaS+PaaS, the top five cloud service providers respectively For Alibaba Cloud, Tencent Cloud, China Telecom, Amazon AWS and Baidu Smart Cloud.

Among them, at the PaaS level, Baidu Intelligent Cloud revenue grew by more than 410% year-on-year, the fastest growth among all vendors. The market share ranked fifth in the second half of the year, among which it ranked fourth in Q4, and has stabilized the first camp of domestic cloud service providers.


At the same time, in the fourth quarter of 2018, only Tencent Cloud and Baidu Smart Cloud achieved doubled revenue. Baidu Smart Cloud joined the single-quarter revenue “Billion Club” for the first time.

Baidu Vice President, Baidu Intelligent Cloud General Manager Yin Shiming explained Baidu's growth. “Our cloud is different. We are called Baidu Smart Cloud. We believe that the future of cloud computing will enter the era of Cloud2.0, not just a calculation-centric change, but a huge change in the mindset of artificial intelligence. ”

Specifically, Baidu Smart Cloud is an intelligent cloud computing service platform for the enterprise-level market, with ABC (A is AI, artificial intelligence; B is Big Data, big data; C is Cloud Computing, cloud computing). Provide solutions for intelligent transformation of various industries.

But the reality that Baidu Smart Cloud has to face is that its power is 7 years later than Alibaba Cloud and 3 years after Tencent Cloud.

From the global rankings, the top five public cloud vendors are AWS, Microsoft, Google, Alibaba Cloud and IBM. AWS maintains a leading position in the world with its unpredictable market share, while Alibaba Cloud enters. The world's top five only domestic cloud service providers.

In 2018, Amazon, Microsoft and Google together accounted for more than 60% of the market. On the public cloud, there are almost no other players.

This means that Baidu seems to be able to snatch the market cake from the top five by means of price wars. But in fact, whether it is Alibaba Cloud or Google Cloud, they are also playing a big price. Although Baidu Cloud follows, the effectiveness of this move has yet to be verified.

Fortunately, the 5G era is about to bring, and the cloud computing market may have a chance to reshuffle.


Baidu life AI

It is not difficult to find that most of Baidu's business model is based on to B.

The AI ​​itself does not contain a direct solution. Its goal is to replace humans with a variety of functions such as identification, cognition, classification and decision making. This makes AI must be integrated into all walks of life, and deep transformation of existing industries can create value. In other words, AI technology is “brain” and it is impossible to walk independently without limbs.


This “energy” attribute further leads to the current AI company's main business model is to provide technical products or services, and AI's customers are enterprise customers who have mastered the scene in the industry. “The problem with Baidu is that the business scenario has not been found. Baidu is not lacking in technology and lacks business leaders. ” An AI industry practitioner said.

Compared to the to C Internet company, which can quickly form a scale effect, to B is naturally a business that cannot be up. This means that in the short term, AI can't bring more cash flow to Baidu.

In Baidu, this company has invested heavily in AI in the past. In 2017, Li Yanhong publicly stated that Baidu has invested tens of billions in research and development in almost every past year. “This is about 15% of Baidu's total revenue. This ratio is definitely the number one in the top 500 in China, and no one can do it right. ”

But what the market wants is profit, what kind of chemical reaction can AI produce, and the capital market is still on the sidelines. In the 2018 Baidu AI developer's big household, the main meeting scene to the middle of the front row of the stage is the analyst's seat. From the perspective of the analyst area, the capital market is also very concerned about the strategic layout of AI after Baidu.


Baidu needs to balance —— to address the continued growth of revenue, but also to balance new business investment and returns.It can be understood that after Li Yanhong regained the Baidu steering wheel, he showed a more pragmatic and more succinct, completely different from Lu Qi's AI picture. This means that, in the next period, the responsible person of the Baidu AI platform may have to carry on more commercial and practical considerations.

However, the road to AI monetization is far away. If it can't bring explosive growth to Baidu in the life cycle of the enterprise, Baidu may no longer be able to return to its peak moment by relying solely on search and information flow.

* Some of the images in the text are from Visual China. At the request of the respondents, both Wenfei and Wang Da are synonymous.

China IT News APP

Download China IT News APP

Please rate this news

The average score will be displayed after you score.

Post comment

Do not see clearly? Click for a new code.

User comments

Related news