Home About us

Baidu AI Trek

Qingcheng Finance 2024/06/27 15:24

Qingcheng Finance, Baidu AI Trek


Editor丨Six sons

A few days ago, OpenAI announced the termination of the large model API service in China, in response to which Baidu Intelligent Cloud released the "Large Model Inclusive Plan": Wenxin's flagship large model is free for newly registered enterprise users for the first time, and supports users to switch to Baidu Intelligent Cloud Qianfan large model platform at 0 cost. In most people's impressions, Baidu is the earliest and largest AI company in China, with mature technical reserves and continuous commercialization of application cases. Qingcheng Finance found in the comparison of Baidu AI and competing products that although Baidu AI has achieved remarkable results, there are hidden dangers - domestic counterparts are catching up in technology, and they have not opened up a new situation in the application link. Perhaps for Baidu, which is "All in AI", AI is not only an opportunity, but also the beginning of danger.


"Invest 100 billion, forward-looking bet on AI"

AI is regarded as the fourth industrial revolution, and under the wave of artificial intelligence, there is Open AI abroad, who are the AI companies that can handle it in China? The answer is Baidu. In 2015, Baidu launched the world's first large-scale neural machine translation system, more than a year before Google; In 2016, Baidu officially announced Baidu Brain, and in the following year, it completed the upgrade of Baidu Brain 2.0 with a complete technical system, opening more than 60 AI capabilities to the outside world; In 2022, four months after the release of GPT-3.5, Baidu released a new generation of large language model Wenxin Yiyan, and after continuous upgrades and iterations, the number of users of Wenxin Yiyan has exceeded 200 million, the average daily API call volume has also exceeded 200 million, and the number of customers served is 85,000; in September 2023, Baidu released the first large model in the quantum domain and two AI native applications; in June 2024, Baidu released three large model reconstruction applications, including the "Keyue" intelligent customer service platform and "Zhenzhi" The knowledge management platform and the "Xiling" digital human platform focus on the internal knowledge management, customer service and marketing (digital human) links of the enterprise; When other major Internet companies in China are still building platform ecosystems, Baidu is fully committed to the research and development and application of AI technology, so Baidu is regarded as the leader of domestic AI, and Robin Li is the "lone brave" in the domestic AI industry. How much does Baidu invest in AI? In 2022, Robin Li revealed to the outside world that since 2012, Baidu has invested a total of 100 billion yuan in R&D, accounting for more than 15% of the annual R&D. What is smashed out of real money is not only the slogan of "All in AI", but also the real "record". There are about 4,000 merchants using Baidu Merchant Smart Twin, covering the healthcare, education, tourism, legal, B2B and automotive industries. In terms of technology, the invocation volume of Wenxin large model alone exceeds that of more than 200 other large models, and the inference cost of Wenxin Yiyan has been reduced to 1% of the original, and the weekly training efficiency of the model exceeds 98%; In terms of commercialization, Baidu AI has begun to achieve revenue, 18% of new paying users of Baidu Wenku are attracted by AI functions, the total service volume of Radish Express exceeds 5 million, and the proportion of fully unmanned orders in Wuhan has reached 45%. *Source networkQingcheng Finance, Baidu AI Trek

AI is Baidu's future and has a strategic role. In the 2023 earnings call, Robin Li said: "In 2023, we have made significant progress in iterating the Wenxin model and Wenxin Yiyan, reconstructing products and services, and commercialization. At the same time, Baidu's core business remains resilient and healthy. Going forward, we will continue to invest firmly in generative AI and foundational models to lay the foundation for creating new growth engines. In Robin Li's eyes, AI is Baidu's development direction, and AI is Baidu's second round of growth curve. But in the view of Qingcheng Finance: Baidu is the hardest working in the field of AI, but Baidu is also the most critical.


"From leading to parallel, Baidu AI lacks bright spots".

"Although Baidu has a high turnover rate and is known as the Whampoa Military Academy in China, the strongest technology is Baidu", earlier, a media teacher told Qingcheng Finance that Baidu is taking the technical route, but the commercialization ability is lacking. But now, with the development of AI technology, Tencent, ByteDance, iFLYTEK and other manufacturers have entered the field of artificial intelligence, betting on the development of AI large models, and Baidu's technical advantages accumulated in the previous 12 years have begun to be quickly erased. Previously, qubits had been sorted out for nearly ten years of essay topics in the national paper A and national volume I, and nine large models were asked to predict the 2024 college entrance examination questions, in order to test the multimodal ability of the large models. Wenxin Yiyan gives four directions: cultural inheritance and innovation, youth growth and responsibility, science and technology and society, and social hotspots and issues, but the overall is relatively general and no specific prediction topics are given. The answers given by Wenxin Yiyan are not fundamentally different from those of the other eight models, and compared with the four models of Tongyi Qianwen, Doubao, Tiangong and Tencent Yuanbao, which give specific questions, Wenxin Yiyan is far behind the above four in terms of understanding of the questions and the accuracy of the answers. This kind of question that wants a clear answer, but gives you a vague answer, makes many users throw themselves into other large models. According to data from market research agency Sensor Tower, as of April this year, on Apple's iOS platform, Byte's "Doubao" was downloaded nearly 9 million times in a year, while Wenxin Yiyan was downloaded 8 million times. In addition to being caught up by domestic large models in technology, in terms of commercialization, compared with foreign large models, Baidu AI is in a bottleneck period. On Baidu's official website, Baidu said that its AI can be applied in many scenarios such as education and training, medical health, transportation, and finance, but Qingcheng Finance found that in these scenarios, Baidu's AI application plays more of an auxiliary role and does not bring new productivity to scene customers。 Taking medical health with the highest requirements for data as an example, Baidu's official website introduces that there are currently 11 cases of medical health, including 4 cases of digital management of personal health, 2 cases of medical waste management, 1 case of auxiliary diagnosis, and 4 cases of medical process management. From its case, Baidu AI is more biased towards medical digital management in the medical field, such as its project with Jianke "search for drugs by pictures and build a smart health service platform", consumers can find the drugs they need through pictures, and the function of searching for goods with pictures has long been available on JD.com and Ali e-commerce platforms, and they have not classified it into AI business

On the other hand, OpenAI reached a partnership with Formation Bio, an American biotechnology company, on May 21OpenAI will provide advanced AI capabilities and expertise to help pharmaceutical companies accelerate the drug development lifecycle. Some data show that the use of AI technology can increase the time for drug research and development, which takes 12-18 years from concept to clinical in traditional drug research and development, while AI research and development only takes 5 years or less to achieve. BCG data shows that the success rate of AI-developed drug molecules in phase I clinical trials is as high as 80%-90%, compared with the historical average of about 50%. Baidu AI is currently in a relatively embarrassing situation, technically overtaken by domestic large model manufacturers, and inferior to foreign large model companies in commercial applications, Baidu AI's future prospects have to be worried.


「AI is here, Baidu crisis".

In addition to being caught up by domestic peers in technology and inferior to overseas competitors in commercial applications, Baidu's biggest problem in AI business is its own internal - AI is overly serving the advertising business, while the advertising business of traditional search engines is declining. According to Baidu's financial report in 2023, Baidu's core business revenue will be 103.465 billion yuan in 2023, a year-on-year increase of 8.5%, of which Baidu's annual online marketing revenue will be 75.1 billion yuan, accounting for about 73%, and non-online marketing revenue will be 28.4 billion yuan, accounting for about 27%. Advertising and marketing revenue still contributes the vast majority of revenue, and the reason for Baidu's growth mentioned in Baidu's official statement is that with the help of AI, Baidu's advertising business, cloud business and other revenues have increased. In other words, Baidu AI has not become an independent revenue-generating business, but more to provide performance improvements for traditional businesses such as advertising. Qingcheng Finance also found that when Baidu introduced the merchant agent, it mentioned a milk powder brand, and with the help of Baidu's AI merchant agent, it conducted multiple rounds of dialogue with consumers before making purchase decisions, providing users with more accurate and personalized services. The introduction of this "milk powder brand" proves that Baidu's AI products mainly serve customers' advertising and marketing. *Source: Baidu 

But AI-powered advertising has helped Baidu grow its revenue, but this growth is seen more like a short-term act to the outside world. "I don't need Baidu search now, I either go to Douyin or Toutiao to search for things, and there is more information there," said a media teacher, adding that in the post-mobile Internet era, Baidu search is not a necessary option, nor is it the only option. According to Gartner's insight report, it is predicted that traditional search engine traffic will decline by 25% in 2026, and search marketing will lose market share. In Qingcheng Finance's view, if Baidu wants to achieve growth in the AI era and open up a second growth curve in addition to advertising, AI should not be an auxiliary tool, but an independent business tool. In 2023, Robin Li's most common word at the Baidu World Conference is "refactoring", using AI large model technology to reconstruct businesses, reconstruct products, realize the vision of de-advertising, and use AI to open up a second growth curve. However, advertising is still Baidu's main source of revenue, and Robin Li's AI business, which is regarded as a strategic position, does not support Baidu's future growth, but AI is more of a supporting role and occupies a subordinate position. From Tianyan, it was found that Robin Li accounted for 99.5% of Baidu's shares and was the actual controller of Baidu. It can be said that the direction of Baidu's big ship sailing is completely determined by Robin Li alone.

Then, the era of AI has come, from AI pharmaceutical, education, to unmanned driving, perhaps with the gradual maturity of technology, it will not be long before AI will achieve a technological explosion like the mobile Internet. As the leader of AI technology in the past, Baidu's use of AI should not just be a patch, but an independent AI-led transformation of traditional businesses, and only in this way can it win the future. Let's see, how the handsome Robin Li will be the captain of the big ship Baidu.

This article is from Xinzhi self-media and does not represent the views and positions of Business Xinzhi.If there is any suspicion of infringement, please contact the administrator of the Business News Platform.Contact: system@shangyexinzhi.com