Big models need to be used in order to come to life, "Cao Peng, Chairman of JD Group's Technical Committee and President of JD Cloud Business Unit, said in a recent interview with Shanghai Securities News. General big models can be built on computing power, but good enterprise big models must be used by business to promote the integration of data and reality. The key is to find the breakthrough point that understands both technology and industry. Cao Peng stated that JD's supply chain is a super incubator for big model applications, and JD's goal for big model applications is to create value for the industry through simple, open, and secure big model application practices, combined with enterprise business.
According to Cao Peng, JD's big model strategy will be divided into two lines. One is to solve the "last mile" landing problem of combining big models with industries, including trying around JD's own scenarios such as retail, logistics, health, and finance; The other line is to build JD's own large model, with the aim of serving the industry and building the foundation, core, and underlying capabilities of its own large model. The ultimate application of the big model still needs to realize industrial value in order to be effectively implemented. Otherwise, no matter how well the basic model is done, it is just empty talk
The past year has been a year of sustained enthusiasm for big models, and we have experienced a thousand model war, "Cao Peng said frankly. According to statistics from the China Academy of Information and Communications Technology, there are currently over a thousand basic models worldwide, with China accounting for 35% of the world's large model population. But at the same time, we also see that on the one hand, although the basic models are constantly improving, whether in the United States or China, there has not yet been a true super application in the C-end. On the contrary, in many B-end scenarios, related applications have gradually spread out.
On the enterprise application side, whether it is the architecture and quantity of models, or the diversity and fragmentation of scenarios, a more open platform is needed to support multiple models, while meeting the complex application scenarios and security needs of enterprises. "Cao Peng admitted that this open demand coincides with JD Cloud's open concept that has always been adhered to.
In his view, general models are built on computing power, while enterprise models are built on business operations. JD Supply Chain is a super incubator for big model applications. Data shows that more and more big model applications are growing on JD's supply chain, and are fully deployed in over 100 AI scenarios within JD. More than 350000 JD owned delivery personnel, over 230000 merchants, over 50000 deputy director level or above doctors, over 20000 procurement and sales operations personnel, and over 10000 R&D personnel have received support from big model applications.
Currently, there is no single model that can solve all application scenarios for enterprises. In the future, in differentiated application scenarios, closed source models, open source models, enterprise exclusive models, and business models will coexist and coexist to meet the actual business needs of customers. JD Cloud always adheres to an open route to serve enterprises and accelerate the application of large models, "Cao Peng said frankly.
Recently, JD.com officially released a new generation of one-stop AI agent development platform - Yanxi Intelligent Agent Platform. This platform has integrated dozens of large models, allowing users with or without programming background to quickly build various intelligent agents based on AI models at low cost, from solving simple Q&A to handling complex logical business problems. Nowadays, there are over 3300 intelligent agents active within JD.com, and the platform has accumulated more than 100 industry solution templates to support the rapid implementation of industry applications. In order to enable more enterprises to enjoy the convenient services of intelligent agents, JD.com has decoupled the underlying capabilities of developing intelligent agents and officially released the Yanxi Intelligent Agent Platform
According to Cao Peng, through the Yanxi intelligent agent platform, the exploration and implementation of AI scenarios has been shortened from 3 months to 1 week. Previously, it required the work of a team of more than ten scientists, but now one person can complete it. The development of new applications has changed from spending several months on just one application to now being able to produce thousands of applications per month.
When interviewed by reporters, Cao Peng emphasized openness the most. Based on an open Agent ecosystem, large model ecosystem, and cloud native ecosystem, JD's large model does not bind, but instead leaves the choice to the customer. He admitted that in the process of upgrading the industry and promoting the integration of data and reality through big models, it is important to promote better understanding of technology, especially in traditional industries, and to encourage technology companies to understand the industry better and find breakthroughs in technology implementation. In this regard, industry leaders and chain owners need to take responsibility. "As a new entity enterprise and platform company, JD.com can play an important role in both the consumer chain and the supply chain. Our goal is to become a link between big models and applications, and through simple, open, and secure big model application practices, we can combine with more fields of enterprise business to create value for the industry