Author: Zhang Yifan Editor: Hard AI
Baidu, a tech giant that has emerged in the PC era, has faced challenges and competition in the mobile Internet era, and seems to have failed to fully seize the opportunities of the times.
As time enters the era of artificial intelligence (AI), Baidu hopes to take this opportunity to make up for the lost ground in the mobile Internet era and re-establish itself as a leader in the field of technology.
At Baidu's latest earnings conference that just ended, the company revealed the latest progress of the AI business:
• AI cloud service revenue increased by 12% year-on-year; • 11% of Baidu search content is generated by AI; • The number of API calls for large models has surged: from 50 million to 200 million in the past five months; • Greatly optimized AI model performance: The training efficiency is increased by 5.1 times, and the inference cost is reduced to 1 percent.
Robin Li, chairman and CEO of Baidu, said in the conference call that the company is shifting from an Internet-centric approach to an AI-first approach, and is promoting the use of Wenxin models to reconstruct the To C and To B businesses. At present, the Wenxin model processes about 250 billion tokens text per day on average, and the average daily call volume is 200 million.
Robin Li emphasized that search is most likely to become a "killer" app in the AI era, and Baidu Search's AI reconstruction is still in its early stages.
In addition, Baidu CFO Luo Rong said that in the coming quarters, Baidu will continue to support the high-quality growth of the AI business. The company believes that as a new era of generative AI unfolds in China, it will bring more opportunities for Baidu.
“All in AI” —— 百度的新机遇？
In order to achieve Robin Li's goal of "turning Internet-centric to AI-first", Baidu is investing heavily in the research and development and application of AI technology, especially in areas such as autonomous driving, deep learning, and natural language processing.
Baidu is trying to regain ground lost in the mobile Internet era.
• In the PC era, with its unique search engine technology, Baidu has become the leading search company in China; • In the era of mobile Internet, user traffic has shifted from PC to mobile, and although the company has a multi-faceted layout, it is still inferior to Tencent/Alibaba. • In the era of recommendation algorithms, it lags far behind the decentralized mode of byte information; • Entering the AI era, Baidu hopes to break through in this wave based on its deep accumulation in the AI field;
In the first quarter of this year, the revenue of the intelligent cloud business reached 4.7 billion yuan, a year-on-year increase of 12%, and the revenue from generative artificial intelligence and basic models accounted for 6.9% of the revenue of the intelligent cloud business.
AI cloud services, from chips to platforms
In 2016, Baidu CEO Robin Li proposed a cloud service strategy of "AI + cloud + big data".
Since then, Baidu Intelligent Cloud has built a full-stack AI infrastructure of "chip-framework-model-MaaS platform", which supports full-link AI services from data storage to model training, deployment, and operation.
It is worth mentioning that the Baidu PaddlePaddle platform has been compatible with more than 50 different chips, many of which are locally designed, and the developer community has grown to 13 million.
This AI infrastructure layout model also allows Baidu to better match the domestic AI cloud demand under the constraints of the external environment. Robin Li argues that this model allows Baidu to use less advanced chips for highly efficient model training and inference.
Robin Li said in the call that Baidu integrates GPUs from different vendors into a unified computing cluster to train large language models, "Our platform has demonstrated very high efficiency on GPU clusters, which are made up of hundreds or thousands of GPUs, which is an important breakthrough in the context of restrictions on imported GPUs." ”
1) Independent research and development of chips
The company has independently developed AI chips (Kunlun) and voice interaction chips (Honghu) to optimize AI technologies such as speech, natural language processing, and images, and support the company's deep learning framework.
Chip Latest Progress:
• AI chip: Iterated to Kunlun 2, the performance is 2-3 times higher than that of Kunlun 1, providing 128 TFLOPS of computing power and 512 GB/s memory bandwidth. • Voice interaction chip: iterated to Honghu 900, compared with the previous generation Honghu 818 chip, CPU performance increased by 200%, GPU increased by 160%, equipped on Huawei smart screen V5 Pro;
2）深度学习框架 —— Paddle Paddle
Paddle Paddle is Baidu's self-developed deep learning framework based on BERT, which integrates the core training and inference framework of deep learning, the basic model library, the end-to-end development kit, and rich tool components.
The framework serves companies in a wide range of industries, including energy, finance, industry, healthcare, and agriculture. For example, the "AI System for Pneumonia Screening and Pre-assessment Based on CT Images" developed by LinkingMed based on the PaddlePaddle platform has been put into use in the Affiliated Hospital of Hunan Chenzhou Xiangnan University.
Paddle Paddle adopts the form of open source, which is convenient for Baidu to better build its own AI ecosystem.
According to the information disclosed by the company:
• As of mid-April 2024, the PaddlePaddle developer community has 13 million members; • By the end of 2023, PaddlePaddle served 235,000 businesses, and the development community had created 860,000 models;
In addition, the data shows that Paddle Paddle is second only to the overseas first-tier platforms Pytorch and TensorFlow in terms of the number of Github users, the number of contributors, and the speed of technical iteration.
3) AI Large Model - Wenxin Series Model (Ernie)
Based on the powerful basic model (Paddle Paddle), Baidu has developed the Wenxin series model Ernie (interesting story: Wenxin Yiyan's English name is "Ernie", which is derived from the name of the character in the American children's show "Sesame Street", and Ernie's good friend in the show is Bert, an AI model released by Google in 2018).
At present, it is iterated to Wenxin 4.0, and the parameter scale may exceed 1 trillion, which is about 4-5 times that of Wenxin 3.5.
According to the company's disclosure, as of April this year, the Wenxin model has processed about 200 million API calls per day, far exceeding the about 50 million calls in December last year, indicating that Baidu's Wenxin model is being adopted by more and more people, and indicates that model inference has strong revenue potential in the future.
In addition, in order to make the model more cost-effective, the company continuously improves the efficiency of the model through its unique four-layer AI architecture and powerful end-to-end optimization capabilities. Compared to the March 23 version, the training efficiency of the Wenxin model has increased to 5.1 times, and the inference cost has been reduced to one percent.
4) MaaS (AI Development Cloud Platform)
To make it easier for developers to develop AI models, Baidu provides three sets of tools on the cloud platform:
• AppBuilder and ModelBuilder: for enterprise and individual developers to develop applications and build models; • AgentBuilder: You can easily create an AI Agent.
In terms of profitability, Baidu has focused its business on improving the profit margin of its AI intelligent cloud business since 2022, eliminating low-quality businesses, and at the same time committing to customizing standardized AI solutions for customers in different industries.
At the 24Q1 earnings conference, the company said that the growth of intelligent cloud revenue is mainly driven by generative AI and model training. Currently, most of the revenue comes from model training, but revenue from model inference is growing rapidly.
Looking ahead, Baidu sees model inference as one of the most important long-term opportunities and a key growth driver for AI cloud revenue in the future.
"AI+" business empowerment
AI technology is widely used in many fields such as search engines, information flow recommendations, and Baidu translation. In the conference call, Robin Li has more confidence in the future AI-empowered search pillar business, believing that search is the most likely killer application in the AI era: "AI search can enable users to do things that they couldn't do before, and Baidu has no plans to charge for this part for the time being." ”
Since the second quarter of last year, Baidu has used Wenxin Yiyan to rebuild Baidu search, and now more and more search results are composed of Wenxin Yiyan in different formats such as text, images, third-party links, etc.
In terms of model construction, Baidu used 3 lightweight models and 2 task-specific models, and used an expert hybrid model to assign tasks to large models, finding a better balance between performance and cost.
Judging from the data, AI is indeed improving the efficiency of Baidu's various businesses:
1) AI + Baidu search: As of 24Q1, 11% of Baidu search's content is generated by AI, and at the same time, AI search has driven the growth of advertising revenue; 2) AI + Baidu Wenku: Attracted 18% of new paying users by introducing generative AI features. These features include content summarization, creation, extension, and one-click conversion of ideas into Powerpoint. In 24Q1, the number of paying users of Baidu Wenku achieved double-digit year-on-year growth. 3) AI + Baidu Map: The "AI Wizard" function has been introduced, which intelligently calls many map functions and services on the basis of understanding user needs, and gives solutions quickly and accurately; 4) AI+Baidu network disk: Launched the intelligent assistant "Cloud Yiduo" based on the Wenxin large model, helping users to quickly search for files and videos, summarize knowledge, translate documents, and even create content in one sentence. As of 23Q3, the number of users has reached 20 million.
Terminal manufacturers cooperate with AI mobile phones/AI PCs
In addition to improving the efficiency of AI internally, Baidu also further expanded its external cooperation this quarter.
According to the company's disclosure, last quarter, the company has reached cooperation with Samsung China, Honor and other mobile phone manufacturers. During the quarter, the partnership expanded to include more leading smartphone manufacturers, including Oppo, VIVO and Xiaomi.
The business scope has also expanded from smartphones to PCs and electric vehicles, further expanding Baidu's AI ecosystem. Lenovo is using the Ernie API to power its AI assistant in its default browser, and NIO is starting to use the Ernie API to enhance the in-car experience.
In addition, the company has attracted many new customers, including Ctrip, Gaotu, Recruitment, Job Bang and the Singapore Tourism Board, to help them reimagine all their consumer-facing products to provide a better user experience through AI technology.
Outbound investment
In addition to independent research and development, Baidu has also continuously expanded its own AI ecosystem through foreign investment, covering three major links: chip-model-model application. But it's clear that, unlike Alibaba, Baidu's outbound investment is more skewed towards both ends of the spectrum.
This reason is not difficult to understand, because Baidu will rely on applications such as search to support revenue in the future and complete the closed loop of the business model.
The most representative proof comes from Robin Li's judgment on whether the large model is open source or closed source: "The open source of the large model is of little significance, and the closed source can get through the business model, and it can make money, and only by making money can it gather computing power and talents." ”
"Closed-source has an advantage in terms of cost, as long as it is the same ability, the inference cost of the closed-source model must be lower, and the response speed must be faster", which is also the thinking of a different business model from Alibaba Cloud.
Therefore, according to Baidu's thinking, the future road is relatively clear, but the competition in the AI era will obviously be more fierce than in the PC era.
⭐ Star Wall Street news, good content is not missed ⭐
This article does not constitute personal investment advice, does not represent the views of the platform, the market is risky, investment needs to be cautious, please make independent judgment and decision-making.