Huawei Zhou Bin: There is no problem with Ascend AI's computing power carrying GPT-4

Huawei Ascend Computing Business CTO Zhou Bin

“The Shengteng AI basic software and hardware platform has incubated and adapted more than 30 mainstream large models. More than half of China’s native large models are based on the Shengteng AI basic software and hardware platform. From the perspective of underlying software and hardware technology, Shengteng AI It has been verified on a large scale, and there is no problem in carrying the computing power requirements of ChatGPT or GPT-4." On May 26, at the 2023 Zhongguancun Forum, Zhou Bin, CTO of Huawei's Ascend Computing Business, made the above statement in an interview with The Paper and other media.

The large-scale model that has been "flying" all the way has become the focus of attention in the domestic technology circle. Perhaps affected by the increase in computing power of large models, cloud vendors have recently made collective price cuts for the first time. Cloud vendors including Alibaba, Tencent, JD.com, etc. have all announced large-scale price cuts. In this regard, Zhou Bin said frankly that as the development speed of large models continues to increase, the demand for various peripheral industries such as data, computing power, network communications, and infrastructure is also increasing. Competition is a manifestation of the normal development of the market economy, "This is a This is a good thing, because behind it is the increasing demand for computing power, and this demand is valuable."

As a manufacturer of underlying computing power, Huawei faces dual challenges from both international and domestic sources. When talking about the competition with foreign manufacturers, Zhou Bin said frankly that different manufacturers have their own advantages in terms of functions, and they need to consider comprehensive conditions such as cost, basic software and hardware facilities, etc. Each manufacturer has its own advantages. “There are also those who choose friends. There are many people choosing us, and currently everyone is in a relatively healthy market."

Zhou Bin revealed to The Paper that Huawei’s computing product lines have been deployed and used on a large scale in a large number of mainstream manufacturers, and there has been good growth in recent orders. “Huawei improves cloud computing density and operation through large-scale distributed parallelism. Efficiency, overall product quality is very good, and we are very confident about the future."

Regarding the future development trend of large models, Zhou Bin analyzed that the first characteristic of large models is "big". It is predicted that they will rapidly grow from the existing hundreds of billions to the trillions level, and will consume extremely large-scale computing power. At the same time, large models will become the core of various business systems in the future, which also puts forward higher security requirements for the underlying computing base. Without a powerful computing base, large models have no roots.

How to ensure data security? Zhou Bin said that since large models need to absorb a large amount of data when upgrading, it is necessary to intervene at the source of the data to prevent large models from abusing wrong data or obtaining user privacy. To ensure the security and reliability of the artificial intelligence environment, a stable basic operating environment must be maintained at the technical level and effective protection and encryption measures must be provided.

The above is the detailed content of Huawei Zhou Bin: There is no problem with Ascend AI's computing power carrying GPT-4. For more information, please follow other related articles on the PHP Chinese website!