Generative AI enters the "second half" of spelling and application: Yuan Foyu

Baidu Group Vice President Yuan Foyu

At the 25th Peking University Guanghua New Year Forum held on January 7, Baidu Group Vice President Yuan Foyu emphasized the importance of technology The value lies in the application. He pointed out that large models themselves have little value. What is truly valuable is the prosperous industrial applications built on large models. Only by applying technology to actual scenarios can the value of technology be maximized.

According to her point of view, generative AI has entered the "second half" of application innovation and is currently at the most critical stage. Generative AI brings new opportunities, mainly by giving full play to the four core capabilities of AI intelligence: understanding, generation, logic and memory, and using this to create a new user experience and improve production efficiency. This is Yuan Foyu’s point of view.

In 2023, generative AI and large model technology have attracted great attention from the global market, and are called the "first half" of industry transformation by some.

Yuan Foyu pointed out that artificial intelligence has experienced many cycles of rise and trough. However, although there are still some challenges in applications and scenarios, such as limited application scope and scattered scenarios, artificial intelligence still has huge development potential.

“However, the characteristic of this wave is the universality of intelligent applications. In other words, once you master a set of excellent basic technologies, you have the opportunity to quickly develop products in various industries. Valuable applications. The real value of technology lies in its application. Those native AI applications that are not based on huge models, these huge models themselves have no value." Yuan Foyu pointed out.

She believes that repeatedly developing multiple large-scale models with no practical applications is a waste of social resources. Once we have usable large-scale models built by many enterprises, the next focus should be on application drivers. We should promote the advancement of basic model technology based on the needs of practical applications.

In addition, in practical applications, the gap between basic large models gradually increases. With the continuous development of tool chains, platform ecology and applications, the capability differences between large models have become increasingly obvious.

Yuan Foyu judged that the development of the application layer with the greatest opportunities has just begun, and there is huge room for innovation. Moreover, because the underlying technology is becoming more and more perfect, innovation is breaking out faster and faster. In the AI-native era, the time for popular applications to be born will be further accelerated.

"2024 is a year that everyone is looking forward to. We believe that many valuable and star applications will continue to explode. Now is the window period for application innovation, and we must actively promote the implementation of generative AI in all walks of life. ." Yuan Foyu said.

In the face of generative AI, what are the most innovative opportunities that should be seized? Yuan Foyu believes that problems that could not or could not be solved in the past should be solved based on the four core capabilities that were not available in the past but are now possessed.

It is understood that based on the concept that technology is valuable only when used, Baidu Intelligent Cloud launched the Qianfan large model platform in March 2023 to help enterprises use generative AI more simply and economically.

In specific practice, Yuan Foyu divided the enterprise's related application methods into 5 categories and 3 levels.

In the application layer with common needs, Qianfan App Store lists many AI native applications and provides an APP builder development workbench, which integrates common patterns, tools, processes and paradigms of large-model application development, and supports applications with the lowest threshold. Developed to meet the individual needs of enterprises.

At the model layer, Qianfan launched the Model builder tool chain to support proprietary model development, model fine-tuning, etc., which can complete the development of proprietary models in one stop and support its final application.

In terms of AI computing power requirements, Baidu’s Baige AI heterogeneous computing platform has an intelligent computing infrastructure with Wanka cluster capabilities and supports the training of Baidu Wenxin series large models and the intelligence of many customers. Calculate demand.

Yuan Foyu said that the actual implementation of many professional scenes is the model of large models within small models. This model meets the needs of performance, efficiency and cost at the same time, that is, the industry-specific model is faster, smaller, lower cost, and more usable; the large model has stronger general capabilities, is smarter, and can be enhanced and protected.

"An obvious change. After experiencing the excitement of 2023, the current focus of the industry is on practical applications, and ultimately returns to the essence of value: who is more efficient and who can create more certain business value , who will win, not just the so-called general ability itself. This is a long-term round of major changes that will reset a lot of existing experience. Only by acting faster can we truly seize the opportunity." Yuan Foyu emphasize.

The above is the detailed content of Generative AI enters the "second half" of spelling and application: Yuan Foyu. For more information, please follow other related articles on the PHP Chinese website!