In the beginning of the Year of the Dragon, various companies have successively released their annual financial reports for 23 years. The innovative 2023, which is full of various industries, is also unfolding before our eyes. The AI large model field that we are most concerned about is especially so.
Throughout 2023, from the birth of domestic large language models to the emergence of AI native applications, and then to the exploration of the industrial large model path. It is still too early to say that China's large models have provided answers, but it can be confirmed that 2023 is a year of great development for China's large models, like the main note in a musical score, setting the tone for this piece.
On February 28, Baidu released its fourth quarter and full-year financial report for 2023, with total revenue reaching 134.598 billion yuan, and Baidu's attributable net profit (non-GAAP) reaching 28.7 billion yuan, a year-on-year growth rate of 39%; fourth quarter revenue was 34.951 billion yuan, and Baidu's attributable net profit (non-GAAP) was 7.755 billion yuan, a significant increase of 44%.
Baidu's financial report far exceeded market expectations, marking the beginning of the symphony for China's AI large models.
Thunder and Spring Rain: The Commercial Potential of Large Language Models
The first note is the commercial potential of domestic large language models.
As a large language model that has received a large amount of human and computing power input since its birth, the industry has been waiting for an answer - whether large language models have commercial sustainability?
Baidu's 2023 annual financial report provides this answer. The financial data shows that in December 2023, Baidu App's monthly active users reached 667 million, and the online marketing revenue for 2023 was 75.1 billion yuan, with 19.2 billion yuan in online marketing revenue in the fourth quarter.
Baidu's specific business success comes from rapid action at the level of large language models.
In March 2023, Baidu first launched "Wenxin Yiyuan," which is benchmarked against ChatGPT, among global tech giants; in August, Wenxin Yiyuan was the first to open to the public; and in the October iteration of Wenxin's large model 4.0 version, its comprehensive level has already been comparable to GPT-4.
The birth of Wenxin Yiyuan is like a thunderstorm in spring, followed by the spring rain that promotes the growth of all things.
Just four months after its official public opening, Wenxin Yiyuan's user base exceeded one billion. For users, the large language model is no longer the terrifying creature rumored to come and replace us, but a reachable and interactive assistant.
This undoubtedly confirms the commercial success of large language models in China, marking the first super traffic entrance born in the AI era after the Internet era.
Similarly, on the technical foundation built by Wenxin's large model, Baidu has also carried out a large amount of business restructuring internally. Search, network disk, document library, smart speakers, and various businesses have spawned new possibilities due to the more powerful capabilities of Wenxin's large model. Undoubtedly, the Wenxin large model has become Baidu's new "internal combustion engine."
The assistance provided by the Wenxin large model to Baidu's own business proves the smooth and feasible commercial path of large language models. The entire large model industry and Baidu are basking in the spring rain, and can move forward more firmly in 2024.
Turning Flywheels: Developers Sing in Harmony
The second note is the harmony of large model developers.
The reason why we can hear the main theme of China's AI large models in Baidu's financial report is that Baidu's developer ecosystem is pulsating along with Baidu's technological development.
In the past year, in addition to continuously advancing its self-developed technological capabilities, Baidu has also been working to lower the technical application threshold for developers and provide more handy deep learning and large model development tools.
The process of large model development kits has been upgraded in pre-training, fine-tuning, compression, and inference, and the generation capability of large models has also been enhanced.
After launching the "Star River Plan" for AI talent development, Baidu provided integrated services for AI native application development and communication for developers in the PaddlePaddle Star River community. In the community, developers can use the PaddlePaddle industrial-level model library, Baidu Brain AI capabilities, Wenxin Yiyuan, and other tools. At the same time, PaddlePaddle also provides developers with a multi-tool intelligent orchestration development mode, allowing even those with zero code foundation to develop multi-modal full-featured customized AI applications.
At the same time, under the layered optimization of Baidu's four-tier AI architecture, the inference cost of Wenxin's large model has shown a significant downward trend. Wenxin's large model has become the most powerful and efficient basic model, with a model training efficiency of over 98% and a training GPU effective bandwidth of 95%. Meanwhile, Baidu also provides small language models to help customers achieve the best performance through the MoE mode.
The lower threshold for development and development costs is attracting more and more developers to PaddlePaddle and Wenxin's large model, making various industries realize that AI applications based on large models can be used for their benefit.
The financial report shows that as of December 2023, PaddlePaddle has gathered 10.7 million developers, serving 235,000 enterprises and institutions, and has created 860,000 models based on PaddlePaddle. At the same time, there are over 4,000 innovative applications based on Wenxin Yiyuan in the PaddlePaddle Star River community, covering many scenarios such as marketing, education and training, and intelligent office.
This is a true portrayal of China's AI large models in 2023. A single technology company cannot make a difference alone, but Baidu is able to drive millions of developers to sing in harmony.
The rapidly growing number of AI native applications and developers means that Wenxin's large model is not only easy to use and effective, but also can support countless innovations to solve industrial problems.
In the era of large models, we are galloping on the wilderness, with rich and soft soil under our feet, which is the perfect time to plant the seeds of industry.
China's AI Choice: Galloping on the Wilderness, Laughing as Heroes
The third note is China's AI choice.
From the overall financial data, Baidu is the first company in China to achieve commercial success with large models and is expected to lead large model developers towards commercial success.
The melody played by Baidu, to some extent, also responds to recent confusion about the large model Sora in the video of Wensheng. From ChatGPT to Sora, from a global perspective, it seems that there are enough advanced large model applications born every year, but large models are not a single-track race, and there is no single correct solution.
Rather than competing with their choices, China's AI large models are better off focusing on hard work.
This is also what Baidu is doing.
In addition to AI native applications developed by developers based on Wenxin's large model, Baidu itself is also creating more industrial possibilities through Wenxin's large model.
On the terminal side, Baidu has successively cooperated with Honor MagicOS and Samsung Galaxy S24 to integrate Wenxin's large model into the terminal. On the Galaxy, it not only achieves functions such as call translation and intelligent summaries, but also creates the "circle and search" function, using multi-modal technology to allow users to search through gestures in videos, images, and text.
In Internet applications, Baidu has reached a cooperation with Autohome to empower its business with AI native applications based on Wenxin Yiyuan's large model capabilities, helping Autohome improve efficiency in content intelligent generation, multi-scenario dialogue generation, and gold medal sales consultants, among many other scenarios.
In the physical industry, Baidu has conducted strategic cooperation with Changan Automobile to explore R&D, manufacturing, supply chain, marketing, and service in multiple areas based on Wenxin's large model.
Bowing down to plant a seed, standing up to feel the vastness of the world.
Based on the cooperative exploration of Wenxin's large model, Baidu has opened up new profit space and found its own "confidence" for China's AI.
We have a technology foundation that is no less inferior, a huge developer ecosystem, and a vast industrial soil full of enthusiasm for new technologies. The pursuit of technology is always uncertain, but as long as China's AI is galloping with all its might in its industrial wilderness, there will always be the confidence to laugh as heroes.
The development of 2023 is not the answer to China's AI large models. Similarly, this wonderful financial report for Baidu is not so much an answer sheet as it is a new beginning, with the industrial exploration and commercial validation of AI large models already showing results. After the starting notes, there is a new and moving chapter about China's industrial AI.
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
