How can artificial general intelligence empower thousands of industries? The industry is hotly discussing the development path of large models

Central Broadcasting Network Beijing, July 22 (Reporter Huang Anjin) With the emergence of general artificial intelligence large models represented by ChatGPT, large models have gradually become a new engine to drive huge changes in social productivity and production efficiency. At the "Development Path of Big Model" forum of the 2023 (22nd) China Internet Conference held recently, participating industry experts discussed hot issues in the field of artificial intelligence big models such as "big model era, general big model, industry big model and mutual collaborative development path".

It is necessary to accelerate the establishment of an open source ecosystem for artificial intelligence

In view of the definition and development status of general artificial intelligence, Huang Tiejun, president of Beijing KLCII Artificial Intelligence Research Institute, pointed out that "general artificial intelligence" often corresponds to GAI (General Artificial Intelligence) and AGI (Artificial General Intelligence) two concepts, GAI refers to (with) general artificial intelligence, AGI is technically called artificial general intelligence. There is a fundamental difference in the meaning of the two, general artificial intelligence emphasizes versatility, while AGI emphasizes an artificial intelligence that comprehensively surpasses humans. "When we talk about the big model today, we actually refer to the transition stage from general artificial intelligence to truly superhuman artificial intelligence."

"A big language model is to train all human languages, mainly written language, into a neural network of hundreds of billions and trillions, so as to have the ability to integrate them." Huang Tiejun said that unlike the certainty of traditional IT systems, real intelligence has now emerged, "in a sense, this model has its own cognition of the world model space behind it."

In November 2022, as soon as ChatGPT was launched, it set off a boom in the development of large models, and more than 80 large models have been publicly released in China so far.

In view of the current status of R&D and application of general large models, Huang Tiejun stressed that an open source ecology should be established. "From the perspective of the evolution of the technical system, the Intel Alliance has dominated the PC era in technology, and the two ecosystems of iOS and Android in the mobile Internet era have dominated the ecology of mobile phones so far. In the era of intelligence, such an ecology will also appear and is emerging. ”

Huang Tiejun pointed out that at present, there is a problem in China that has changed from a "big refining model" to a "big refining model", resulting in a serious waste of resources. "The degree of intelligence and intelligence of a large model, and whether it can even replace researchers to discover new laws of physics, is what we pursue." He believes that for China, many enterprises and institutions can train big models, and the key is that everyone should jointly iterate innovation through open source and move forward.

The large-scale model landing industry needs to solve the problems of security and trustworthiness

As early as February 13 this year, the Beijing Municipal Bureau of Economy and Information Technology officially released the "2022 Beijing Artificial Intelligence Industry Development White Paper", proposing to comprehensively consolidate the foundation for the development of the artificial intelligence industry. Support leading enterprises to build large models that benchmark against ChatGPT, and strive to build an application ecology of open source frameworks and general large models.

The Politburo meeting of the CPC Central Committee held on April 28 proposed to "attach importance to the development of general artificial intelligence, create an innovation ecology, and attach importance to risk prevention", pointing out the direction for the development of the general artificial intelligence industry.

How to make general artificial intelligence empower the development of all walks of life? What are the remaining challenges at this stage? What are the future development paths?

Li Wei, General Manager of Big Data and AI at HUAWEI CLOUD, pointed out that unlike the traditional AI era, where there are many branches of AI technology and many small models in the industry, the Transformer architecture brings the industry an opportunity to use pre-training and super large-scale data to make large models have high generalization capabilities, massive data, and powerful fitting capabilities, thereby breaking through the limitation of accuracy, and one large model can cover multiple scenarios.

"In the early days, a large number of discrete small models such as vision, speech, semantics, and generation capabilities gradually converged into large models. At present, the application of To C basic capabilities has been extended to To B general scenario capabilities", Li Wei also said that there are three challenges in the general large model landing industry, one is that the large model has strong versatility, but the professional ability is relatively weak; Second, although there is a lot of knowledge about large models, the skills are insufficient; Data is one of the core assets of enterprises, and large models can effectively store and mine knowledge, but training and using large models needs to ensure enterprise data security compliance.

Feng Junlan, chief scientist of China Mobile Group, also pointed out that the theoretical core technologies related to large models are constantly making breakthroughs, the level of intelligence has been greatly improved, and controllability, security and credibility are urgent to be tackled. "The idea is to rely on the current ubiquitous computing resources and network resources to flexibly configure, schedule and deploy AI capabilities in an open environment to meet increasingly complex requirements."

"Every industry is a very complex system in general, and how to really integrate (data, knowledge and various systems) in such a complex system, there are still great challenges for the current big model." Feng Junlan believes that large and small model collaboration and human-machine collaboration will be the path of future industry development.

In view of how large models can enter thousands of industries, Wei Kai, deputy director of the Institute of Cloud Computing and Big Data of the Chinese Academy of Information and Communications Technology, said that one path for the development of large models in the future is from the basic model to the industry model, and the second is the development of model as a service. "In the future, it may be that large models drive small models, and models as a service may be nested in each other" "One model calls another model to stitch these models together, connect them, and connect them together to complete a complex scenario-specific task."

Lin Yonghua, vice president and chief engineer of Beijing KLCII Artificial Intelligence Research Institute, said that the general big model is an important foundation for the big model in the industry. First of all, the general large model is the basis of the knowledge system, and it cannot be forced that the general large model contains the expertise of all industries. Secondly, the general large model or basic model is an important capability framework for downstream industry models. At the same time, the development of the general model itself needs to be paired or iterated with the industry model.

For more exciting information, please download the "Central Radio Network" client in the application market. Welcome to provide news clues, 24-hour reporting hotline [phone]; Consumers can also complain online through the "Woodpecker Consumer Complaint Platform" of the Central Broadcasting Network. Copyright statement: The copyright of this article belongs to the Central Radio Network, and shall not be reproduced without authorization. Please contact: [email protected], we will be held accountable for acts that do not respect the original.