Zhou Hongyi: 360 builds a 'large model factory' for cities to help implement AI industrialization

"The performance of '360 Intelligent Brain' in some scenarios is close to the international advanced level", on September 17, jointly sponsored by the Nanjing Municipal People's Government and China Academy of Information and Communications Technology, Nanjing Economic and Technological Development Zone Management Committee, China The 2023 Nanjing Artificial Intelligence Industry Development Conference co-organized by the Artificial Intelligence Industry Development Alliance will be held in Nanjing. Zhou Hongyi, founder of 360 (601360.SH, hereinafter referred to as "360") Group, delivered a keynote speech titled "Vertical Large Models Leading the New Industrial Revolution".

Zhou Hongyi said that large models are not a trend or a bubble. As a new generation of productivity tools, they will lead a new industrial revolution, become the standard configuration of digital systems, and empower the digital development of industries. However, general large models cannot be used directly in enterprise-level scenarios, and there are seven major problems, including lack of business depth, failure to "understand" the enterprise, and data security risks. In this regard, he proposed that the future development trend of large models is "verticalization." In terms of timing, Zhou Hongyi believes that you should not wait for the large model to be perfect before you start working on it. You can start working after you think about the scene clearly.

He said that with the increasing improvement of the open source ecosystem, large models themselves will no longer become a barrier, and will even become "cabbage price" in the future. The real challenge is how to carry out in-depth customization of vertical large models. He divided the customization of large models into four steps: the first step is business analysis and scenario selection, the second step is data collection and cleaning, the third step is training enterprise-specific large models, and the fourth step is developing enterprise scenarios. application. He also emphasized that vertical large models must also be trained on the basis of general large models that have been proven by the market and have sufficient capabilities. Only in this way can we achieve twice the result with half the effort.

Currently, China has entered the “Battle of 100 Models”. Zhou Hongyi believes that enterprises must be cautious when choosing basic large models, and can evaluate them from the six standard dimensions of "generality, security, controllability, usability, ease of use, and scalability". According to reports, 360's basic large model is supported by a large visual model, a large language model, and a multi-modal large model, forming an "integrated training and promotion tool and platform" that can provide enterprise-level users with one-stop large models. Development Kit.

Zhou Hongyi revealed that 360 is developing enterprise-level AI solutions and has built a large model training framework to process the data of vertical large models and perform the complex process of pre-training and fine-tuning. This framework has been streamlined and standardized, and has successfully supported the training, deployment and delivery of nearly 20 industry vertical models. He took the application of "360 Intelligent Brain" in urban governance scenarios as an example. Through the empowerment of large models, the area's incident discovery, acceptance, allocation, tracking and case closing have achieved significant efficiency improvements, fundamentally Changed the way of safe operation and management of cities

Facing the digital construction in various places, Zhou Hongyi said that 360 can do two things for the development of the urban AI industry. One is to create city-level exclusive large-scale models to help build smart cities; the other is to create scenarios around large-scale models to build The "large model factory" that supports the training, fine-tuning and deployment of vertical large models is also the R&D and training center for large models. 360 is responsible for providing basic large models and training platforms, and developing and cultivating a new profession "artificial intelligence trainer". While driving high-quality employment, it also makes large models smarter and more suitable for enterprise-level scenarios.

In his speech, Zhou Hongyi said that building enterprise-level large models should be done step by step. First, let the large model play the role of "co-pilot" and maintain isolation from the existing system. When the time is right, the Agent model can be developed, so that the large model can not only serve as The "brain" recognizes human intentions, thinks and decomposes tasks intelligently, and can also grow "hands and feet" to automatically use tools, call various APIs, perform tasks, solve problems, and achieve target results, becoming a general-purpose intelligent agent. system. At the same time, he also emphasized that before solving the safety problem of large models and putting large models in cages, this is only the best wish of human beings, and people still need to stay in the decision-making loop.

Source: Financial Industry Information

The above is the detailed content of Zhou Hongyi: 360 builds a 'large model factory' for cities to help implement AI industrialization. For more information, please follow other related articles on the PHP Chinese website!