"In the era of AI, 'super capable' applications are more important than 'super applications' that only focus on DAUs (active daily users)." On July 4th, at the 2024 World Artificial Intelligence Conference and High Level Conference on Artificial Intelligence Global Governance Industry Development Main Forum, major' leaders' in the AI industry reached a consensus on the commercialization path of the AI industry: identifying application scenarios and avoiding traffic thinking.
"In fact, applications are not far away from us." Robin Lee, chairman and CEO of Baidu, said that applications based on basic models have begun to gradually penetrate into all walks of life and various fields. In the past two months, the daily adjustment volume of Wenxin Big Model has skyrocketed from 200 million to 500 million. "In just two months, such a significant change in the number of calls indicates that it represents the real needs behind it. Someone is using it and someone has truly benefited from the big model, gaining value."
Taking the express delivery industry as an example, he utilized the ability of large models to process orders. The express delivery company achieved the goal of "sending express delivery with one picture and one sentence", eliminating the need for other cumbersome processes and shortening the time from over 3 minutes to 19 seconds. "Moreover, over 90% of after-sales problems are also solved by large models, resulting in significant efficiency improvements."
In the development direction of AI applications, Robin Lee is most optimistic about agents. He believes that with the growing strength of the basic model, the development and application are becoming more and more simple. Among them, the simplest is the agent. As long as the workflow is clearly described with "human words", coupled with a proprietary knowledge base, a valuable agent can be made, which is "simpler than making a web page on the Internet".
Ant Group Chairman and CEO Jing Xiandong also believes that professional intelligent agents can solve the key problems of general large models in rigorous industrial applications. The industry generally believes that the implementation of the general large model in a rigorous industry faces three "capability shortcomings": a relative lack of domain knowledge, difficulty in making complex decisions, and dialogue and interaction not equal effective collaboration. Jing Xiandong introduced that the future intelligent user experience will not rely solely on a large model, but requires deep collaboration across the entire industry, and requires the participation of many professional intelligent agents to perform their respective roles.
However, the commercialization of large models is not a smooth road. The "traffic trap" is a detour that many domestic large models have taken.
Robin Lee said that over the past year, the industry has been in the midst of a very lively "roll model", and there was a grand occasion of the 100 model war, which caused a great waste of computing power. "We need to avoid falling into the 'super application trap' and believe that a 1 billion DAU app is necessary for success. This is the logical thinking of the mobile era."
At the same time, the commercialization of large models on the enterprise side also faces many barriers. For example, Fu Sheng, Chairman and CEO of Cheetah Mobile, said that large enterprises have discovered many problems when using artificial intelligence, such as difficulties in feeding internal data, interface debugging, and internal processes.
Wang Jian, an academician of the CAE Member and the founder of Alibaba Cloud, also felt the same way. "People are the factors that are easy to ignore in AI.". He believes that due to institutional and other issues, it is difficult for large companies to embrace AI for everyone. Large companies believe that AI is a revolution of tools, while small companies believe that AI is a tool of revolution. If large companies also have the awareness of "revolutionary tools," then AI will have a revolutionary impact.
"Whether it's the chat applications brought by ChatGPT or the cultural and educational videos brought by Sora, although they are impressive enough, they have not yet reached the super moment because they have not truly entered a vertical application in an industry, causing widespread changes." Xu Li, Chairman and CEO of Shangtang Technology, believes that artificial intelligence must break through the three core points of intelligence, experience, and controllability to usher in the "iPhone moment".
He explained that in order for large models to generate the intelligence of deep thinking, they cannot do without data from advanced thinking, and the more applied the scenario, the more high-quality data can be generated. Secondly, real-time interaction performance brings a smooth experience. Only by making good use of computer resources in both the end and cloud can it become a fully natural interaction mode. "The last and most important point is controllability. Whether it's text, images, or videos, if the big model doesn't have content and boundaries, it can't achieve AI controllability and sustainable development."