share_log

对话阿里云CTO周靖人:“炼出”一个大模型应用,最快需要多久? | 科创100人

Talk to Alibaba Cloud CTO Zhou Jingren: How soon will it take to “create” a large model application? | 100 science innovators

新浪科技 ·  Nov 2, 2023 21:20

Text丨Sina Technology Zhou Wenmeng

At the 2023 Yunqi Conference, Alibaba Cloud announced that it will upgrade its cloud computing system to create the most open cloud in the AI era.

Alibaba Cloud CTO Zhou Jingren said in an interview with Sina Technology's “100 Science Innovators” that models have become a very important production element in business system development and cloud computing, and everyone is redeveloping and designing their own development systems, even business systems, around models.

So how long does it really take to develop a large model application?

“You can't just focus on models”

As the core technology of this wave of AI, the quality of the basic model for large models has largely determined the prospects for AI industrialization. Training big models is a systematic engineering that includes complex technologies such as computing power platforms, networks, storage, big data, AI frameworks, and AI models. Only powerful cloud computing systems can train high-quality big models.

“In 2009, Alibaba Cloud proposed the concept of 'a data center is a computer'. Today, in the AI era, such a technology system is even more necessary.” Zhou Jingren pointed out, “As a supercomputer, cloud computing can efficiently connect heterogeneous computing resources, break through the bottleneck of a single performance chip, and collaborate to complete large-scale intelligent computing tasks.”

“Although we are focusing on models today, models are only one part of our business system development today. I hope to not only focus on models, including the data, storage, and network capabilities under the model, which are also necessary for technological innovation today.” In his view, “Today, big models have become a very important element in business system development and cloud computing, and cannot be missing.”

Furthermore, Zhou Jingren revealed, “Currently, half of the largest model companies in China run on Alibaba Cloud. A large number of leading companies and institutions, such as Baichuan Intelligence, Intelligent Spectrum AI, 010,000, Kunlun World Wide, Vivo, and Fudan University, are training big models on Alibaba Cloud.”

As the industrialization of AI gradually deepens, demand for large-scale intelligent computing power will surely explode. Zhou Jingren said that at present, Alibaba Cloud has built 89 cloud computing data centers in 30 regions around the world, providing more than 3,000 edge computing nodes and sufficient computing nodes, further exploiting the advantages of low latency and high elasticity of cloud computing.

“The model will be an important production element”

At the end of 2022, Zhou Jingren proposed the MaaS (Model as a Service model as a service) concept for the first time. In his view, “model as a service” represents both an infrastructure upgrade and a series of development paradigm shifts around models today. In his introduction to Sina Technology, he said, “In fact, many of today's development models are more model-centered, and models have become an important production element.”

“Some of our developers create the model itself. After having the abstraction of model development, it can be used very well by other business system developers, while business system developers are more users, who use and configure business scenarios around the model. In fact, for the most part, the model is a “handshake” handover between these two groups of people, or many technical experts like this.” He emphasized, “Everyone today revolves around models to redesign their own development systems, or even design their own business systems.”

At a time when all industries want to use big models to change production and service methods, the high threshold for using big models has kept most people out of the technological wave. Whether it's customizing a large exclusive model or building an innovative application based on the big model, there are high talent, technology, and capital requirements.

At the Yunqi Conference, Zhou Jingren released a one-stop large-scale model application development platform, Alibaba Cloud Bailian. The platform integrates mainstream high-quality large-scale models at home and abroad, provides services such as model selection, fine-tuning training, security kits, model deployment, and full-link application development tools, simplifying complex tasks such as low-level computing power deployment, model pre-training, and tool development for users. Developers can develop a large model application within 5 minutes, and “refine” an enterprise-specific model in a few hours. Developers can focus more on application innovation. At the same time, in order to make it easier to integrate and implement big models in thousands of industries, Alibaba Cloud has launched 8 major industry models based on common sense.

“Today, Alibaba Cloud's position is to serve these developers in the intelligent era who look forward to using digital innovation or artificial intelligence to innovate in technology. These companies are all our customers, and we hope to effectively support them with a complete product matrix. This is our current position.” Zhou Jingren said. While launching big models developed on its own, such as the Thousand Questions of Common Sense, Alibaba Cloud also hopes to use its technical capabilities to serve more big model companies and “build the most open cloud in the AI era.”

Disclaimer: This content is for informational and educational purposes only and does not constitute a recommendation or endorsement of any specific investment or investment strategy. Read more
    Write a comment