For more quality content, please follow the wechat official account “AI Front” (ID: AI-Front).
Let’s start with a bit of teasing
First to ridicule the weather forecast, say good rain, Beijing, Beijing is not only the sun is shining today, however, is still hot, is really a long story (￣ _ ￣ | | |)
If meeting a meeting on a hot day is sour, queuing up to get in on a hot day is sour quadratic. However, it is this that shows the enthusiasm of developers and media for Baidu AI Developer Conference
The pageantry of the queue
Having made fun of it, let’s start to talk about business. What heavy news did Baidu AI developer conference bring today?
The show opened even cooler than last year, followed by the usual drop-dead handsome Robin Li.
Robin said the most common question he has been asked over the past year is: When will Baidu’s self-driving car go into mass production? In response to this question, Li said, “Building a car is different from making PPT. There is often a delay in delivery in building a car…” However, Baidu’s mass-produced driverless car arrived today as promised and did not disappoint the eagerly awaited audience.
Last year, Li boasted that the self-driving car would be mass-produced in 2018. This year, he gave a more specific date: July 2018, which is now. Subsequently, Robin Li connected the production line of Xiamen Golden Dragon and witnessed the 100th world’s first L4 level autonomous bus: Apollon officially rolled off the production line.
V.qq.com/x/page/v071…
According to Robin Li, Apolong unmanned buses will be put into: Beijing, Xiongan, Shenzhen, Fujian Pingtan, Country Garden, Wuhan, Hubei, Tokyo and other places, the era of automatic driving seems to be coming to us in advance. But Li said the evolution of self-driving cars is just beginning.
The year 2018 will be the first year for the commercialization of autonomous driving, comparable to the historic point when the Model T rolled off the production line, Li said. The biggest disruption in the AGE of AI will be the gradual simplification of the car’s complex components (the Apollon has no steering wheel, no driver’s seat), replaced by a lot of code. The Apollo platform is said to have 22W+ lines of code open so far, a six-fold increase from last year. Baidu’s improvement from 1.0 to 3.0 is more about software.
At the conference, Robin Li played an audio clip, which was the same as the previous demo on Google I/O, using Baidu voice AI to contact the attendees of the AI Developer Conference and ask them for relevant information.
During the demo, some people actually didn’t recognize whether it was AI or a real person, but, like Google’s demo, it was a recording, not a live demonstration.
Li said that conversational AI is fundamentally changing the way humans interact with each other, reconstructing work and life.
This ability has led not only to new products such as smart speakers, but also to whole new changes in some service industries. For example, baidu’s intelligent customer service once made hundreds of phone calls to developers attending the baidu AI Developer Conference. Many of the developers who received the call didn’t realize it was a robot he was talking to. The intelligent customer service always responds to the developer’s questions, including transportation, dining and conference guests. Even smart developers can “see through their identity” and respond tactfully when being “teased”.
According to Li, the language abilities perfected by humans over hundreds of thousands of years of evolution are increasingly being imitated by machines. Behind this is the continued massive investment in AI technology over the years. Baidu has brought together thousands of AI engineers, more than billions each year spending on research and development of high performance computing army consisting of one million servers that support levels billions of parameters, billions of vlsi neural network training samples and features, plus trillions “feed” of training data, to push ascension and iterative algorithm, to calculate force.
In addition, Robin Li also said: “Baidu AI will ‘equal empowerment’, in the PC era, Baidu’s mission is to make people the most equal and convenient access to information, we hope to continue to bridge the information gap through technological innovation. In the future, Baidu will bridge the ‘intelligence gap’ by opening up its accumulated AI technology, namely data, computing power and algorithms, so that people can have equal access to AI capabilities.”
However, software upgrade alone is far from enough. Robin Li said that the combination of software and hardware can accelerate the growth of Baidu’s brain computing power. Subsequently, Robin Li released China’s first cloud fully functional AI chip developed by Baidu: Kunlun.
Kunlun is a chip produced in China’s large-scale AI computing practice. The AI accelerator based on Baidu’s CPU, GPU and FPGA for 8 years has been developed over 20 iterations, which is considered another milestone in China’s AI chips.
Kunlun is by far the most powerful AI chip designed in the industry (260Tops at 100+ watt power consumption). It can efficiently meet both training and inference requirements. In addition to cloud requirements such as common deep learning algorithms, kunlun can also be adapted to natural language processing, large-scale speech recognition, autonomous driving, Computing requirements for specific terminal scenarios, such as mass recommendation.
Famous Chinese computer expert, the first batch of Chinese Academy of Engineering, as told by ni guangnan think baidu has always attached great importance to the AI, do chip can play to baidu accumulate in the field of AI AI technology advantages, and make the chip can be first in baidu own AI applied on platform and its application, through the application can promote the development of the chip, forming a virtuous circle.
Robin Li said that in the past 40 years of China’s reform and opening up, in our development process, for high-end chips, in fact, has been dependent on imports, which is the heart of our generation of IT people forever pain. That will change when we enter the age of artificial intelligence. The high-end chips independently developed by Baidu will meet the various needs of developers around the world, support hundreds of thousands of developers and benefit the society.
Ren Zhengfei once said in a speech: “Chips are urgent and cannot be urgent. It is not only about technology, equipment and consumables, but the stock market exaggerates too much in order to make money. To be steadfast, not to pursue in a bubble.” Now, domestic technology enterprises have laid out chips, and even many enterprises have good products. However, to make chips that catch up with the world level, not only Baidu, is no small challenge for any enterprise. We look forward to the day when China’s independently developed chips catch up with the United States.
In October 2016, “Baidu Brain 1.0” came out of the blue. In less than two years, “Baidu Brain” has realized the comprehensive opening from the initial establishment of the foundation to more abundant capabilities. Baidu Brain 3.0 has covered the technology, platform, ecological landing full stack layout.
After Lu Qi leaves Baidu, Baidu AI is in full charge of wang Haifeng. Baidu Brain 3.0 is already capable of multi-modal deep semantic understanding, and has opened up 110+ capabilities for developers to use, according to the report.
Dr. Wang Haifeng, senior vice president of Baidu, introduced that the core technology breakthrough of Baidu Brain 3.0 is, in a word, multi-mode deep semantic understanding. Not only to make the machine hear and see clearly, but also to understand the meaning behind it, so as to better support various applications. For example, during the World Cup, baidu AI can “understand” matches like human commentators through semantic processing of structured information.
In addition, Dr. Wang haifeng also introduced the upgrade of paddlePadd3.0, which provides EasyDL, AutoDL and other platforms for developers to use on the basis of the original.
According to reports, Baidu brain 3.0 chip into the technical system for the first time, it makes Baidu brain with a more complete integration of software and hardware ability, driving baidu brain computing power explosive growth. The AI chip will also be deeply combined with the PaddlePaddle deep learning framework independently developed by Baidu to promote the rapid development of the AI industry ecology.
Jing Kun, general manager of Intelligent Living Group, then brought an update on DuerOS.
The conversational AI market in China has exploded. According to Jing Kun, DuerOS has become the largest conversational AI operating platform in the Chinese market, with activations of DuerOS devices exceeding 90 million and monthly active devices exceeding 25 million.
DuerOS covers almost all kinds of scenes such as TV, mobile phone, speaker, children’s equipment and automobile. In each field, DuerOS has reached cooperation with domestic leading manufacturers, such as Huawei P20, P20 Pro, Vivo X21, NEX, OPPO FindX and other popular mobile phones. There are also DuerOS.
With the launch of DuerOS, the core capabilities of DuerOS are being upgraded as the 3.0 era approaches.
DuerOS 3.0 also has a module design update, the new module is the size of a business card:
Baidu vice President Shen Dou announced the official launch of intelligent small program, Shen Dou said: Baidu intelligent small program fully access baidu Brain 3.0, developers can call a few lines of code, at the same time, intelligent small program will be fully open source in December this year, in the future, intelligent small program can not only run on Baidu apps, but also bilibili, 58.com and other external partner apps.
Intelligent applets AI capability framework
Last year, Robin Li got a ticket for driving his car on the fifth Ring Road. A year later, Baidu’s driverless car was officially mass-produced. We witnessed the development of Baidu’S AI and were even more amazed by its speed of evolution. Lu qi has laid out an ambitious AI blueprint for Baidu, and Li is gradually turning it into reality.
Baidu AI’s entry into the 3.0 era has brought more than simple changes to users and developers. Just as Robin Li said: AI for Everyone, only by crossing the intelligence gap can artificial intelligence truly enter people’s lives. For Chinese AI, it is equally important to develop calmly and avoid bubbles in the process of rapid technological development.