China Mobile Communications Group Co., Ltd. (China Mobile) on April 2 announced that its Jiutian Natural Language Interaction Large Model, capable to provide generative artificial intelligence (AI) services, have been filed with the Cyberspace Administration of China.
It is the first large model developed by a central state-owned enterprise to be filed in both the “Record Filing for Generative AI Services” and the “Domestic Deep Synthesis Service Algorithm Filing”.
Centered on the originally-systematized AI technology and based on the profound technological accumulation formed by long-term research on intelligent systems, China Mobile realized independent innovation of core technologies throughout the entire process of large model data construction, pre-training, fine-tuning, and inference.
The Jiutian model has outstanding performance in security and reliability, and can be domestically developed. So far, various versions with parameters of 9 billion, 13.9 billion, 57 billion, and hundreds of billions have been developed, which can be flexibly adopted in different scenarios such as cloud, edge, and the terminal side.
Based on the Jiutian model, China Mobile created a co-construction platform for backbone industries, accelerated construction of more than 20 large models for industries in the fields of networks, customer service, households, government affairs, tourism, healthcare, industry, comprehensive governance, and enterprise communication.
China Mobile started independent research and development of AI technology and products in 2013, serving as a leading force in national scientific and technological innovation.
It developed landmark products such as the Jiutian large multimodal language model and AI platforms, and established a professional AI team consisting of talents at the national level and high-level doctoral experts.
In the era of general intelligence, the company is expanding its business from “5G plus” to “AI plus” and plans to build a new-type intelligent foundation based on large computing power, large models, big data, and big platforms.
At present, the company has built a multimodal large model with intelligent computing power exceeding 10EFLOPS and parameters of 200 billion. The large model aggregation platform is to open soon.
