At the beginning of 2023, there was a lot of doubts about "Can China have large models?" In the sound, Wen Xinyiyan's sudden appearance brought answers to all walks of life. Since then, whether it is open to the public, version upgrades, or technology iterations, Wenxin Large Models have always been at the forefront of China's large models. The entire industry maintains an industrial pattern of "Wenxin is the best, Internet companies and large model unicorns are strong".

As the time comes to the end of 2024, we have noticed that some media have put forward the view that the inflection point of large models has been reached, believing that the future will belong to the field of large models that started late but has commercialization advantages such as capital and publicity. Late comers. However, has the pattern of large models in China really changed?

On November 30, China Internet Network Information Center released the "Generative Artificial Intelligence Application Development Report (2024)" (referred to as "CNNIC") at the Fifth China Internet Basic Resources Conference. According to the report , Baidu's Wenxin large model has the highest usage rate among netizens, achieving an overall lead in the number of users.

This information will make us think about such questions: In the ongoing large model competition, what do users rely on to make choices? What is the competition going on in the noisy and bustling large model arena?

Judging from many clues, Wenxin still maintains the pattern of China's big model with one super and multiple powers unchanged.

Behind this change, there is also a logical core: the background of China’s big model is by no means a short-term zero-sum game, but about truly treating AI as a long-term undertaking and doing deep research. , do it broadly and do it solidly.

For AI and even all technological fields, core technology They are all the basis for the future. China's large models cannot be willing to be followers of technology. Only by mastering core self-developed technologies, achieving a systematic innovation flywheel, and always taking the initiative in AI technology can enterprises have a future on the large model track.

The core element of China’s large-scale model pattern remains unchanged, and the core reason why Wenxin is able to distance itself from other large-scale models is that it has chosen to embrace the core technology self-research route and has achieved success.

For example, "slow thinking" capability is the hottest large model technology in 2024. In September this year, OpenAI released the o1 large model, which can handle relatively complex reasoning tasks, and has since quickly become a hot spot in the global AI industry.

However, nearly a year before o1 was released, Baidu has predictedProposed the "slow thinking" ability, proposed a system 1 dominated by the immediacy of content generation, paired with a technical implementation route of system 2 with the ability to understand, plan, reflect and evolve. Therefore, this year we can already see the implementation of multi-agent collaboration technology in the Wenxin large model system. China's large models have thus once again been at the top of the world's AI wave.

For another example, after the rise of large models, the most controversial issue is the large model illusion. In order to overcome this problem, Baidu has developed a "comprehension-retrieval-generation" collaborative optimization retrieval enhancement technology, which can effectively improve the model application effect and eliminate the trouble of large model illusion.

From knowledge enhancement, retrieval enhancement to multi-agent, Wenxin Large Model has gradually built a complete large model core technology system and an innovation mechanism for sustainable development. According to data from IDC, Sullivan, China Software Evaluation Center and other institutions, Wenxin Large Model exceeds the average performance of leading international large models, ranking first in China and the first echelon internationally.

Being able to control the sharpness of the continuous evolution of large models in one's own hands is what keeps Wen Xin ahead in the large model competition.

Has the industrial structure changed in the development of China's large models? Has it reached an inflection point? The answer must be decided by consumers.

We often say that whether a store is good or not depends on the publicity, consumers will vote with their feet. So Wenxin’s success on the C-side comes from the voting completed by users with their fingertips.

According to the "Generative Artificial Intelligence Application Development Report (2024)", Wen Xinyiyan ranks first in the usage rate of typical generative artificial intelligence products among netizens, far exceeding similar competitors such as ChatGPT and Gemini. Taste. This survey is very authoritative. It conducted a survey with a total sample size of 30,000 domestic Internet users, and the survey time was the first half of 2024, which is the time when competition in the domestic large model market is most intense.

It can be seen that the Wenxin large model always maintains the lead in faults at the C end.

The widely popular Wenxin large model has also been continuously Empower Baidu's business system and activate new growth in the consumer market. For example, Baidu Wenku has been restructured into a "one-stop AI content acquisition and creation platform" based on the Wenxin large model, and has released hundreds of multi-modal AI projects such as intelligent PPT, intelligent writing, mind mapping, research reports, and photo-writing. ability. As of October 2024, the number of Baidu Wenku AI users has exceeded 230 million, and the cumulative number of AI functions used has exceeded 2.8 billion.

Technology leadership and better experience still play a core role in the C-side competition of large models.

When talking about AI and large models, we must realize one thing: the core difference of China’s AI lies in the maturity and prosperity of the industrial intelligence market . This is the core opportunity for China’s AI industry and an irreplaceable competitive factor in the large-scale model stage.

We all know that large-scale models will bring a series of values ​​to the industry, but industrial intelligence projects have higher technical requirements, and greater demands for service capabilities and industry understanding capabilities. This has resulted in many large model manufacturers preferring to focus on large model product models that are highly similar to those of European and American manufacturers, and relatively despise or even ignore the industrial intelligence needs of the B-side.

One of the differences of Wenxin’s big model is that it always regards empowering intelligence in thousands of industries as the core of the big model’s development goals. Starting from this, Wenxin has completed in-depth exploration in a large number of key industries, and can bring large model capabilities into the core of the production system, bringing effective value realization.

For example, intelligent agents are widely considered to be the future of AI The mainstream form of application. For the industry, the combination of intelligent agents and industrial intelligence is still in the exploratory stage, but Baidu has made intelligent agents a new fulcrum for industrial intelligence. For example, in the field of government services, Zhongwei Huitong uses grassroots government intelligence to help villagers answer various problems in life. Its service scope has covered more than 6,000 villages, and about 30,000 grassroots government workers have used large models to serve more than 10 million residents, achieving rapid and accurate feedback processing of residents' needs. In the field of corporate operations, BYD and other companies have applied intelligence to sales and after-sales fields, and the sales lead conversion rate has increased by 119%.

Wenxin’s capabilities have penetrated into thousands of industries, forming a large-scale and replicable industrial structure. In the industrial field, Sinochem Information cooperated with Baidu to create the "Hua Xiaoyi" knowledge assistant based on Qianfan AppBuilder, which can quickly retrieve and answer professional knowledge such as molecular properties and molecular synthesis routes based on natural language questions, replacing traditional manual queries. This method has improved the search efficiency of specific molecules by more than 5 times.

In the field of transportation, Guiyang Municipal Public Security and Transportation Bureau jointly Baidu Intelligent Cloud has selected 31 important traffic nodes in the core area of ​​Guanshan Lake to implement intelligent information control, and has achieved an accuracy of more than 80% in predicting traffic flow distribution within 5 minutes. On this basis, with the combination of multiple measures, average vehicle delays on weekdays in the region have dropped by 15% on average, the all-day congestion index has dropped from 1.7 to 1.58, and the morning peak congestion time has been shortened by 30 minutes, significantly improving citizens’ transportation experience. .

Autonomous driving is not only a key industry for the development of new productivity, but also a naturalLarge model floor-to-ceiling stage. Luobo Kuaipao's sixth-generation vehicle is equipped with the Apollo ADFM, a large autonomous driving model. Combined with the multi-modal fusion of point cloud and vision, it can more accurately detect and understand obstacles in complex environments, thereby promoting the expansion of autonomous driving services across regions. Scenarios such as traffic, cross-river traffic, and airport expressway traffic.

The deep application of these Wenxin models in the industry is the choice of Wenxin models in thousands of industries. These choices are the foundation of Wenxin and the foundation of China's large model business.

After the explosion of AI, more about large models Discussions of business competition also arise, which is of course a good thing. But what must be noted is that if we are accustomed to attribute the answers to all technological problems to commercial competition, the complex AI track will become flat and full of distortions.

AI is a technological undertaking that is in the exploratory stage, has a high technical threshold, and is changing rapidly. The future is not determined by pure commercialization. Recently, some media believe that there are only two companies that have shown the determination to fight to the end on China's large-scale model battlefield: one has unlimited ammunition, and the other has unlimited firing rights.

But if you analyze this logic carefully, you will find that the so-called The "ammunition" and "right to fire" are ultimately capital, not technology. In fact, Baidu, which has the most core technologies, is still the undoubted core player with absolute strength in the large model battlefield. In the foreseeable future, Wenxin, which maintains its core technology leadership and can continue to iterate, will still maintain China's industrial structure of one large model, one super, and multiple strengths.

Of course, from a media perspective, we do not object to changes in the pattern of the large model industry, and even look forward to the arrival of an inflection point, which will bring more stories worth telling and discussing. However, the basis for judging the occurrence of an inflection point is It should come from user choice and respect and recognition of the value of technology.

In the future, the factors supporting changes in China’s large-scale model landscape should not be the quantity purchased or the amount spent, but should be the evolution of core technology and changes in user choice.