Recently, the Standing Committee of the Hebei Provincial People's Congress invited Shen Dou, Executive Vice President of Baidu Group and President of Baidu Smart Cloud Business Group, to give a special lecture on "Large Model: A New Engine for Industrial Intelligence". Shen Dou said: "The emergence of large models has redefined human-computer interaction. In the future, people only need to use natural language to put forward requirements, and large models can understand the requirements, regenerate content, upgrade tools, and finally provide services. Academia, The industry believes that the fourth technological revolution has arrived."
Shen Dou, Executive Vice President of Baidu Group and President of Baidu Smart Cloud Business Group
Different from the previous three scientific and technological revolutions, this technological revolution represented by artificial intelligence, especially the emergence of large models, has given birth to a new force, which is evolving from a role of assisting brain power to a role of replacing brain power. This is a higher level of technological productivity that will revolutionize all industries.
But it is not easy to train a good large model, and the process is very complicated. Different chefs use the same ingredients and the same kitchen utensils, and the taste of the dishes they make is different. The same is true for large model training. Using the same data and the same parameter scale, the effects of large models trained by different companies may be different.
"So we predict that in the future, a few major manufacturers may make basic large-scale models, and then various industries and enterprises can develop large-scale industrial models, large-scale enterprise models and even large-scale models of local government affairs on this basis." Shen Dou said . Since the internal test of Wenxin Yiyan on March 16, 150,000 companies have applied for the test within a week, and each company has obvious scenario needs. The large-scale model services provided by Baidu Smart Cloud for enterprises include Wenxin Qianfan large-scale model platform, Baidu AI large base, etc.
In January of this year, the Hebei Provincial Government issued the "Action Plan for Accelerating the Construction of Digital Hebei", proposing that by 2027, the digital economy will account for more than 42% of GDP. As of the end of March this year, the total number of cloud companies in Hebei Province exceeded 80,000. The cloud access rate of enterprise industrial equipment reached 19.7%, ranking first in the country for two consecutive years. Hebei is listed by the Ministry of Industry and Information Technology as the only pilot province in the country for industrial Internet platform application evaluation.
Shen Dou said, "Hebei's digital economy has a very solid foundation. The engine of the digital economy must be artificial intelligence in the future. Therefore, we must vigorously develop the digital economy and pay more attention to the implementation of artificial intelligence in all walks of life." World-renowned investment institutions Goldman Sachs and Ark The fund predicts that the large-scale model may increase global GDP by 7% per year and increase the productivity of knowledge workers by more than 4 times.
Regarding how to make good use of artificial intelligence and large-scale models and improve the digital economy, Shen Dou believes that the first thing is to do a good job in the construction of new infrastructure, including intelligent computing centers, data labeling bases, artificial intelligence chips, etc.
The general construction period of a smart computing center is about one to two years. In view of the current trend of rapid development of large-scale models, Shen Dou suggested that local governments make arrangements as soon as possible. At present, Baidu is participating in the construction of intelligent computing centers in many cities across the country, such as Fuzhou, Tianjin, Jinan, Shenyang and other places.
In the era of large models, a large amount of labeled data is needed. In addition to labeling data, the data labeling base can also continuously train AI trainers, model fine tuners, and command engineers to help local enterprises get through the last mile of large models . Shen Dou believes that data labeling bases can also drive local employment and create economic value. Baidu Smart Cloud participated in the construction of 12 labeling bases across the country, all of which were jointly built with the government, accumulatively contributing more than 11,000 local jobs, indirectly driving 50,000 jobs, cultivating tens of thousands of AI data talents, and attracting more than 200 high-quality enterprises to incubate Home.
In terms of artificial intelligence chips , Shen Dou said that whether it is cost-effective or safe and controllable, it is necessary to vigorously support domestic independent chips. At present, Baidu's self-developed Kunlun chip has been widely deployed and used in the business of customers such as finance, industry, and education. The third generation of Kunlun core is expected to be mass-produced in 2024.
In the past few years, Baidu has been deeply involved in the construction of digital Hebei, building a data center in Baoding, and building risk control industrial bases in Handan and Tangshan. Baidu's subsidiaries and branches have a total of nearly 3,000 employees in Hebei. Baidu also introduced a number of corporate customers in the ecosystem to Hebei, directly driving nearly 30,000 jobs.
Shen Dou said that with the improvement of the capabilities of large models, multi-modal large models that integrate text, voice and vision will soon mature. This will further promote the deep integration of artificial intelligence and the real economy, and further enhance the level of intelligence and productivity in the fields of steel, automobiles, industrial manufacturing, cultural tourism, and autonomous driving.
Click "Read the original text" to cooperate and consult immediately