Li Yanhong once again mentioned "heavy traffic" to build AI innovation power

"What is generative? For example, draw me a picture of a busy traffic. This kind of thing, in the past, people didn't think it was something that artificial intelligence should do, but now it can be done." On May 18, Baidu CEO Robin Li was in Tianjin He said in a speech at the Seventh World Intelligence Conference held.

The popularity of generative artificial intelligence has once again aroused widespread global attention on the development of artificial intelligence, and made people realize the importance of large models for the development of artificial intelligence. Li Yanhong specifically mentioned the large language model product Wenxin in his speech Vincent figured out the case of "heavy traffic". It is understood that on March 16, Baidu was the first to release Wenxinyiyan, a large-scale model product that is benchmarked against ChatGPT among major global manufacturers. After Wenxinyiyan started the internal test, it was accused that the one-click picture generation function could not understand Chinese well, resulting in the generated picture words not expressing the meaning. For example, when inputting "heavy traffic", superimposed pictures of "car + dragon" elements appeared, and some voices even questioned Wen Xin's words of "covering" foreign AI.

Why did Li Yanhong mention "heavy traffic" again at this time?

In fact, due to the impact of public datasets and galleries, there were still many inaccuracies in Wenxin Yiyan's generation of pictures at the beginning, but in just a few days, based on many feedbacks from users, the large model learned the correct understanding and drawing The method can well draw the bustling and bustling street scene with crowds of vehicles and horses, and this just reflects the characteristics of generative AI based on the ability of large language model: based on feedback and fast iteration.

"Wen Sheng Tu" is Wen Xin Yi Yan's very special and world premiere ability. It comes from ERNIE-ViLG, Wen Xin's cross-modal large model developed by Baidu itself. It is worth mentioning that ChatGPT-4 still does not have the ability to generate pictures from text so far.

As the leading artificial intelligence enterprise with the earliest investment, the strongest technology and the most complete layout in China, Baidu is not only the builder of AI infrastructure, but also the leader and promoter of AI technology and application innovation. Since the establishment of the deep learning laboratory in 2013, Baidu has carried out in-depth and comprehensive layout in the four layers of artificial intelligence chip, deep learning framework, large model and application layer. , framework layer, model layer and application layer have a full-stack layout of the company.

Wen Xin's words reflect Baidu's leading strength in the field of artificial intelligence technology. The reason why Li Yanhong mentioned "heavy traffic" at the World Intelligence Conference is precisely because of his technical confidence in Wen Xin's words, and he effectively responded to the question of "covering the shell". Wenxin Yiyan completed 4 major technical upgrades after more than a month of internal testing, and the inference performance of large models has increased by nearly 10 times.

Wen Xin's words also reflect Robin Li's forward-looking layout of AI. As early as four years ago, Baidu launched Wenxin 1.0, a large language model with over 100 million parameters, for the first time, and continued to polish the large model technology through core application search, Xiaodu, etc. "We are looking forward to the future, based on the large model, there will be good application scenarios for various applications in all walks of life, and the efficiency can be greatly improved." Li Yanhong emphasized that AI will not replace human work, stop innovating, follow The various unpredictable risks brought about by the inertia is the biggest threat to human beings.