WAIC2022 ｜ Shangtang Technology Xu Li: AI landing must have "certain regulations", standardization is expected to drive the outbreak of the AI industry

21st Century Business Herald reporter Zhang Sainan Shanghai report

The ancients built different buildings and followed some standardization principles. So, is there a common standard in the large -scale landing of AI technology at the moment?

On September 2nd, the 2022 World Artificial Intelligence Conference "Great Aiwai, Shangtang Technology Yuanchuang Rong" was held at the Shanghai West Bank Art Center.论坛上，商汤科技董事长兼CEO徐立以古代著名传统建筑为例，探讨人工智能落地如何实现标准化，并发布商汤科技的四大创新人工智能平台体系，覆盖AI云、车路协同、 Games and future hospitals and other fields.

Xu Li pointed out that through standardized and more efficient and easy -to -use platform services, it will build AI "creating French" for industrial innovation, and drives the large -scale outbreak of the artificial intelligence industry.

Xu Li takes the famous traditional buildings such as Drunk Waigon, Foguang Temple, and Teng Wangge as an example. It elaborates between today's promotion of AI large -scale landing and building a clever building with the ancients. "".

"These norms and consensus are continuously accumulated with people's experience, and finally gradually formed by summarizing summary, model deduction, simulation simulation, etc." Xu Li said in his speech. Similarly, there are "certain rules" in the landing of AI industrialization.

First of all, artificial intelligence landing elements from the bottom layer's computing power, data, algorithm model to the upper layer have formed a "standardized structure"; followed by the "standardized model". Artificial intelligence and complicated scene application, such as assisting driving, urban governance, surgical planning, game intelligence, etc., behind the artificial intelligence algorithm model, each type of model can be customized, just like the Tang Dynasty architecture The tenon -and -mortise structure.

After the standardization model, it also needs to cooperate with division of labor, that is, standardization of production. At present, artificial intelligence from the underlying computing power infrastructure, data labeling, pre -training large models, and industry models to upper applications all form process standardization. For example, automatic labeling with a large model can increase the label efficiency by 600 times.

Xu Li believes that "no standards and industries", standardization will bring a large -scale outbreak of AI industrialization. By constructing a strong infrastructure, production of high -quality and standardized models and reuse to achieve "cost reduction"; at the same time, modular management of the entire process of model production is to achieve "efficiency". It is worth mentioning that standardization is not restricted, it will bring the industry's prosperity and flowers.

"We understand the track, power socket, Wanwei Internet Internet agreement, etc. These industries have ushered in explosive growth after forming a common paradigm and consensus. In short, standardization will allow the AI ​​algorithm , Rapid promotion, promote the scale of AI applications, and create unlimited possibilities. "Xu Li said.

The 21st Century Economic Herald reporter noticed that at this World Artificial Intelligence Conference, the standardization of artificial intelligence has become a hot topic of discussion among all parties.

Huo Kun, the chairman of Huawei's rotation, also said in the opening ceremony of the 2022 World Artificial Intelligence Conference that the future computing power network will become a new infrastructure of the national economic development like today's power grids, communication networks, and high -speed rail. Relatively uniform standards, the relative unity and compatibility of the standards such as hardware application interface, node interconnection, data sharing and other levels of computing power, so that different computing power centers can sing a song together to achieve efficient operation of the entire network. "

"The current data generated by various computing centers does not match the format and algorithm. It cannot be directly called each other. It can only be used in the local area and brings great difficulties to unified operations. It is recommended that all parties work together The standards, the hardware application interface, node interconnection, data sharing and other levels of computing power are compatible with standards. Different computing power centers can sing a song in order to form computing power, data, and ecology. Efficient operation. "Hu Houkun said.

- END -