Baidu is never short of traffic and topics, and this time is different.
Surrounded by the wave of AI large models, Baidu, a "veteran" of the Internet era, has returned to the table. This time, it is not only because of the large model, but also because of the prosperous AI industry ecology that Baidu has cultivated for a long time.
Baidu, which has always adhered to long-termism, is only 24 years old, but some data shows that the AI-driven company already has the ambition of a "century-old store".
At a recent Baidu internal awards event, Baidu's founder, chairman and CEO, Robin Li, reviewed the achievements made in the field of AI large models in the past year: Wenxin Yiyan was first launched on March 16 last year, and after continuous iteration of versions 3.5 and 4.0, the basic capabilities of large models have always been at the forefront of China's large models. He said that since the reconstruction of all product lines with large models in May last year, "today, 10% of the big search traffic has been generated through Wenxin Yiyan's model; Today, there are 2.5 million users every day, using the AI capabilities of Wenku. ”
The reason why it can take the lead in large models is inseparable from Baidu's insistence on high-pressure and marathon R&D investment, and its investment in AI has exceeded 100 billion yuan in more than ten years.
Robin Li once set a "hard target" for Baidu: 15% of its revenue should be invested in research and development every year.
However, judging from the actual investment, Baidu's R&D investment in recent years has already exceeded this proportion. For example, the R&D investment in 2018 was 15.8 billion yuan, accounting for 15.4%, the proportion of R&D in 2020 reached 18.22%, and the figure will exceed 18% in 2023.
Insisting on the R&D investment of the marathon is a manifestation of long-termism, and cultivating relevant talents for the society is also a manifestation of long-termism, which is actually the underlying logic to ensure long-term development.
It is reported that Baidu's internal R&D personnel account for more than 60%, and it has also completed the goal of cultivating 5 million AI talents for the whole society in 5 years ahead of schedule.
Baidu is not stingy in R&D investment, which may be closely related to the personal style of its founder. As we all know, technology is Robin Li's belief, but also Baidu's belief, Baidu can come to this day because of the belief that technology can change the world.
In March 2021, when Baidu was listed on the Hong Kong stock market, Robin Li also wrote a shareholder letter, which contained an impressive sentence: "If we choose a technology strategy, we must resist the temptation of short-term opportunities and firmly face the challenges of long-term investment." ”
Obviously, based on the research and judgment of the development trend of the Internet, AI and other technologies, Robin Li has clearly realized that the artificial intelligence selected by Baidu, as well as various vertical tracks such as intelligent driving and cloud based on artificial intelligence, is where the future lies.
Although due to the relationship between the technological breakthrough cycle and market maturity, some tracks have not yet reached the stage of real explosion and large-scale realization, but on the road to technology investment, Baidu has put its hands into the mud.
As Robin Li said, Baidu "can survive loneliness and hide the sea of stars", although people always overestimate the short-term value of technology and underestimate the long-term value of technology. "But Baidu believes in technology and aims for the long-term value of technology."
AI, stark labels
From the traditional Internet, to the mobile Internet, to the explosion of generative AI, Baidu, which has a history of more than 20 years of development, has never been absent from every wave.
In 2010, Baidu officially laid out artificial intelligence and established the Natural Language Processing Department (NLP), which was the beginning of Baidu's in-depth understanding of the field of artificial intelligence. In 2013, Baidu established a deep learning research institute to recruit a large number of AI talents, and began a solid bet.
One after another, Baidu has borne lush AI fruits. In March 2019, Wenxin Model 1.0 was launched, and it has been continuously iteratively upgraded and innovated in technology, and now, Wenxin Model has evolved to version 4.0, which is not inferior to GPT-4 in terms of comprehensive level, and has become a leading and widely used AI basic model in China.
Based on the leading basic model, Baidu has also carried out a full-stack layout in the multi-layer technology stack of AI chips, frameworks, models, and applications, and each layer has representative self-developed technologies and products.
Thanks to its long-term and continuous investment in AI technology, Baidu has made breakthroughs in the field of artificial intelligence and large model technology, and has also achieved outstanding results in related patent layouts. According to the "Baidu Artificial Intelligence Innovation and Patent White Paper 2024", by the end of 2023, Baidu has applied for 19,308 patents and authorized 9,260 patents in all fields of artificial intelligence, ranking first in China for six consecutive years.
In 2023, among the world's major manufacturers, Baidu will take the lead in releasing Wenxin Yiyan, a large-scale model product that benchmarks ChatGPT, and open it to the whole society, with the number of users exceeding 200 million so far.
Based on leading R&D experience and deep patent accumulation, Baidu realizes that the competition for large models is entering the next stage. Starting in 2023, Baidu will shift part of its focus to AI-native application development. As Robin Li said, large language models do not directly create value by themselves, and AI applications developed based on large models can meet real market demand.
At the 2024 AI Developer Conference, Baidu has conveyed the logic of using AI thinking to create AI native applications, not only reconstructing Baidu library, search, network disk, map and other products, but also preparing three "out-of-the-box" tools for developers, including agent development tool AgentBuilder, AI native application development tool AppBuilder, and model customization tool ModelBuilder of various sizes.
Behind these innovative technologies and products are the support of a new generation of artificial intelligence patents with large models as the core, as well as Baidu's long-term experience in exploring AI.
It can be seen that large models have become an important starting point for unlocking new quality productivity, and AI native applications are being integrated into thousands of industries. As an explorer and pioneer of artificial intelligence, Baidu promotes the innovation of AI native applications and injects new ideas into the high-quality development of the industry.
Questioning with Everest
In the long-term growth chart, the trend is often not always steady, and it is actually full of misunderstandings and doubts before the return is obtained.
However, once the time comes, the speed of the payback, is lucrative and staggering.
This is the case of Baidu's determined climb of "self-driving Mount Everest".
Since 2013, Baidu has begun to deploy autonomous driving. In 2017, Baidu opened its first autonomous driving platform in China, Apollo. In the history of its operation, Apollo has also been questioned, but Baidu has persevered in it.
根据全球领先的法律和商业分析商“律商联讯LexisNexis”最新发布的《驶向未来：自动驾驶创新洞察》（Driving Toward Tomorrow:A Deep Dive Into Autonomous Vehicle Innovation）报告，百度自动驾驶在专利资产指数、专利包规模等多项排名上位列全球榜首，展现了创新技术与商业化的成功结合。
Two years ago, Robin Li said at an internal executive meeting that Baidu has been doing unmanned driving for six or seven years, but it has not yet been fully commercialized, "Then we will do it for another six or seven years." ”
In fact, it doesn't take more "six or seven years", and now, the commercialization path of autonomous driving has gradually become clear.
According to public information, Baidu Apollo L4 autonomous driving safety operation test mileage has exceeded 100 million kilometers, and the number of autonomous driving licenses has exceeded 1,000, becoming a "water seller" in the era of intelligent vehicles, and many car companies have joined Apollo's circle of friends.
The industry expects that by 2026, high-end urban intelligent driving will enter a period of volume. For Baidu, which has been deeply involved in autonomous driving for many years, this will also be a critical period for good cash.
Radish Express, which focuses on self-driving car sharing business, has achieved large-scale commercialization, providing autonomous driving travel services in more than 10 cities across the country, and is also the first company to launch unmanned in-car travel services in Beijing, Wuhan, Shenzhen and Chongqing.
As of January 2, 2024, the total service volume of Radish Express exceeded 5 million, and the number of service orders increased by 49% year-on-year to 839,000 in the fourth quarter, and the proportion of driverless orders in Wuhan reached 45%. In March 2024, Baidu Radish Express announced the expansion of autonomous driving travel service hours to 7*24 hours in some areas of Wuhan. This major service upgrade means that users can not only "take unmanned cars to visit Wuhan", but also enjoy the autonomous driving travel service of Radish Run at night.
Another piece of good news is that the 14th Five-Year Plan mentions that it is necessary to achieve advanced autonomous driving above L3. Some analysts also said that by 2025, the overall scale of high-level assisted driving will reach 400 billion yuan.
With the continuous evolution of intelligent technology, urban pilot assisted driving is moving from the "initial market" to the "mass market", and Baidu, which has been deeply involved in intelligent driving for more than 10 years, has more advantages in industrial integration to a certain extent, and the precipitation of time is also more valuable as a probe.
Mount Everest represents the sea of stars, which is a long-term plan, but before the arrival of the sea of stars, the large and small results in the process are also the best feedback for long-termism.
Professor Chen Chunhua of the Peking University Development Institute divides victory into three types: opportunists can only win temporary victories, pragmatists can win phased victories, and long-termists can win sustained victories. There are all kinds of signs that Baidu wants to do the third.