Li Yanhong’s internal speech: Computing power can be bought, but innovation ability cannot be bought

The big language model is setting off a new industrial revolution.

Geoffrey Hinton, known as the "Godfather of Deep Learning", began to warn of the risks of AI, but also believed that it was comparable in scale to the Industrial Revolution, electricity, and even the invention of the wheel.

A global AI competition has already started, and the most anticipated ones in China are big manufacturers, who have huge and necessary computing power and data resources, but the most important thing is "innovation".

Robin Li, founder, chairman and CEO of Baidu, gave an internal speech today. He said, "Computing power cannot guarantee that we can lead in general artificial intelligence technology. Computing power can be bought, but the ability to innovate cannot be bought." The ones that come here need to be built by themselves.”

He called on employees to cultivate AI-native application thinking and reconstruct every existing product and business with new ideas. I believe that after reading it, you will have a deeper understanding of Baidu's innovation in the AI ​​era.

The following is the full text of the speech:

Dear Baidu students,

On such a special day, I am very happy to see you all. First of all, I would like to congratulate every student in our award-winning team. You are the pride of Baidu and mine! I would also like to thank the family members of Baidu who were present today, and thank you for your silent dedication and support behind the scenes!

Today, 6 years ago, also here, we officially released Baidu's new mission, using technology to make the complex world simpler, and officially sounded the clarion call for Baidu's transformation into an artificial intelligence company. But what I want to say is that our dedication to innovation is consistent, and our investment in technology is long-term and has never wavered .

Ten years ago, that is, in 2013, we realized that the development of artificial intelligence technology will make unmanned driving a reality one day. Since then, we have started to invest heavily in autonomous driving technology for ten years. Today, Carrot Kuaipao has far exceeded 2 million orders, making it the world's largest driverless operator. Also in that year, we participated in the bidding of the team of Geoffrey Hinton and his two doctoral students Alex and Ilya. I saw from the media that Baidu did not understand business negotiations and missed the opportunity to make ChatGPT first. In fact, it was not the case. , Google paid more than 40 million US dollars more than we did at the time, didn't they also fail to make it? Ilya Sutskever turned around and left Google to participate in the creation of OpenAI.

Eight years ago, that is, in 2015, we realized that artificial intelligence was gradually becoming a core general-purpose technology and basic technology, which would inevitably lead to changes and leapfrog development in many fields. That year, I spoke at a conference in the Great Hall of the People at the National Two Sessions, called the "China Brain Project". At that time, artificial intelligence had not yet come out of the circle, and AlphaGo would not come out until 2016. My suggestion at the time was to develop and build the world's largest artificial intelligence platform at the national level. This platform can be opened to government agencies, enterprises, institutions, research institutes, etc., so that everyone can carry out corresponding technological innovations on this platform. In recent years, the surge of intelligent computing centers in various places has more or less reflected this idea. At the same time, at Baidu, we also started to build our own full-stack artificial intelligence technology.

Four years ago, we first launched Wenxin 1.0, a large language model, with over 100 million parameters. At that time, 100 million parameters were the definition of a large model. And we continue to polish our large-scale model technology through core application search and Xiaodu.

Two years ago, we realized that artificial intelligence technology had undergone a directional change, from discriminative artificial intelligence to generative artificial intelligence. Since then, AIGC has become a direction of our focus, and resources are gradually tilted in this direction.

On March 16 this year, Baidu was the first to release a large-scale model product Wenxinyiyan, which is benchmarked against ChatGPT, among major global manufacturers, empowering thousands of industries and allowing all products to be reconstructed accordingly. The reason why we were able to make it ahead of big companies such as Google, Meta, and Amazon is because we have achieved the world's only full-stack layout and layer-by-layer leadership in the four levels of chips, frameworks, models, and applications. We are conducting joint optimization. At that time, most friends were still lying on the three-tier architecture of chips, operating systems and applications of the previous era. Fortunately, today, the concept of four-tier architecture has begun to be recognized by the industry. End-to-end optimization allows us to enjoy big data, big models, and big computing power like a fish in water, and we can have both sides. Recently, I saw a data that said that the global computing power scale, the United States accounted for 34%, China accounted for 33%, the computing power is about the same. I want to say that computing power does not guarantee that we can take the lead in general artificial intelligence technology. Computing power can be bought, but the ability to innovate cannot be bought and needs to be built by ourselves.

After more than a month of internal testing, Wenxin Yiyan has completed four major technical upgrades. The cost of large-scale model inference has been reduced to one-tenth of the original, or the inference performance has been improved by nearly 10 times. When others are just beginning to think about how to train, we have already gone a long way in reasoning.

Of course, the tech race has only just begun. In the true AI era, new applications will be developed based on large models, just like the emergence of various Android and iOS-based applications more than ten years ago. Therefore, I have said many times that the big model is a Game Changer, which will completely change the rules of the cloud computing game. In the past, cloud computing mainly sold computing power based on speed and storage. Today, customers purchase cloud services based on the quality of the framework and model, not just the computing power. In the future, when enterprises develop applications based on large models, they will find that Baidu Smart Cloud is the most convenient, fastest, and low-cost.

All the past is prologue. With the growing maturity and involution of the mobile Internet, the competition in today's artificial intelligence field is more intense than ever, and the challenges we face are also unprecedented. In an interview some time ago, I said: The gap between us and ChatGPT is about two months, but to be honest, this is a bit out of context. I then said: This is not the point, the point is how much we need to use the gap between these two months It will take a long time to catch up, maybe soon, maybe never.

It depends on our own efforts, on our ability to seize opportunities, and on our ability to execute. Whether our strategy, talents, organizational mechanism and culture can adapt to the new situation, and whether we can continue to innovate in the new environment, is a matter of life and death.

In the future , we must consciously cultivate the way of thinking and concepts of AI native applications, which requires us to use new concepts to reconstruct each of our current products and businesses. It's not easy, it's hard, because among the 40,000 of us in Baidu, none of us are natives of the AI ​​era. We have all experienced the mobile era, and many of us have also experienced the PC era. Our way of thinking In a sense, it was solidified. I remember when I used the browser for the first time, I felt that the efficiency of that thing was very low. The problem can be solved in a second through the command line. Why do I have to wait for the pictures to be loaded one by one? Only then can I start to operate, but then I realized that it is much easier for ordinary people to click the mouse than to let him learn touch typing;

I also remember when I first started using Apple mobile phones, I felt that the screen was so small, why would users use such a thing to surf the Internet, but when I saw that a few years old children quickly learned to use the touch screen to surf the Internet, I felt Determined to spend two years, Baidu's mobile search traffic exceeded PC search traffic. With the advent of generative AI this time, our first impression is that a problem can be solved by typing one or two keywords and clicking a mouse. Why do we need to use a long paragraph to describe our needs in order to get what we want. But we need to be aware that how many functions developed by engineers are hidden under layers of menus, and how many inspirations that cannot be obtained through hard thinking can be inspired by some simple prompt words. You should have the courage to change your way of thinking. Of course, changing the way of thinking will be painful, and we will experience various setbacks and sufferings. We need to use perseverance to overcome this inertia and look at this new world with the perspective of discovery and exploration.

In today's world, no great product is made by one person, it needs a dynamic organization. It is so difficult to change a person, but it is even more difficult to change an organization. This requires us to think together and work together. It requires all departments to fully communicate, open discussions, work together, support and help each other, and continuously launch the most competitive products and technologies in the market.

This is a contest of comprehensive capabilities, which requires us to unify our thinking, think positively, be brave in innovation, seize this opportunity, devote ourselves wholeheartedly, and forge ahead.

Finally, let us keep in mind the mission of using technology to make the complex world simpler, live up to our youth, go through the clouds and fog, and pursue the expedition of light!

Today is also the May 4th Youth Day. I wish you all a happy holiday and stay young forever!

#Welcome to pay attention to Aifaner's official WeChat public account: Aifaner (WeChat ID: ifanr), more exciting content will be presented to you as soon as possible.

Ai Faner | Original Link · View Comments · Sina Weibo