On July 4th, 2024, at the World Artificial Intelligence Conference and the High-Level Meeting on Global Governance of Artificial Intelligence held in Shanghai, hundreds of representatives from the academic and industrial sectors engaged in in-depth discussions on the development direction and application of AI. Experts attending the conference generally believe that the focus of AI development has shifted from theoretical research to practical application, with how to create real value from AI technology across various industries becoming a key concern.
Image Source Note: Image generated by AI, provided by image licensing service Midjourney
The Core Issue of AI Application Implementation
Li Yanhong, the founder of Baidu, pointed out that the AI era should not fall into the "super app trap," but should focus on "super capable" applications that can bring benefits to industries. He particularly favors the application direction of intelligent agents, believing that search will become the largest entry point for intelligent agents distribution.
Wang Xianzong, the chairman of Ant Group, believes that the implementation of general large models in rigorous industries faces three major challenges: lack of domain knowledge, difficult complex decision-making, and dialogue interaction not equaling effective collaboration. He proposes solving these challenges through deep connections of professional intelligent agents, predicting that AI will bring a generation of service upgrades similar to the internet.
Xu Lili, the CEO of SenseTime, emphasized that application is the key to pushing AI into the "super moment." He points out that to promote the wide application of AI, it is necessary to break through three aspects: high-quality data, smooth interaction, and controllability.
The Development Direction of Large Models
Zhang Peng, the CEO of Zhipu AI, believes that the core breakthrough of large models lies in multi-modal capabilities, which will make AI closer to the way humans solve problems in the real world. Yan Junjie, the founder of MiniMax, emphasizes that improving model accuracy is the key to application implementation, aiming to reduce error rates from the current 30%-40% to single digits.
About open-source models, Li Yanhong said that open-source models are valuable in specific scenarios such as academic research but are not suitable for most application scenarios. In a competitive business environment, closed-source models have more advantages.
AI Security and Ethical Issues
Zhou Bowen, the director of the Shanghai AI Laboratory, pointed out that the investment in AI security is far behind that in AI performance, with only 1% of resources invested in alignment or security considerations.
Yao Qizhi, the winner of the Turing Award, believes that AI risks mainly come from three aspects: the expansion of network risks, the potential subversion of social structures, and the existence of risks. He emphasizes the need to seek a balance between controlling AI and not destroying its potential.
Industrial Transformation and Opportunities
Zhang Pingan, the CEO of Huawei Cloud, emphasized that AI innovation cannot be separated from the innovation of computing infrastructure, especially the release of edge hardware AI computing requirements to the cloud.
Michael Yang, the chairman of Qualcomm China, predicts that transferring 20% of generative AI workloads to the edge can save 16 billion US dollars in computing resource costs by 2028. He believes that the close integration of the edge and the cloud will drive the scaled expansion of generative AI.
About the opportunities brought by AI, Wang Jian, the founder of Alibaba Cloud, said that while large enterprises may have more advantages in AI development, this does not mean leniency. New large companies will inevitably emerge, and some existing large companies may be reborn through AI.
This conference reflects that the AI industry is shifting from theoretical research to practical application, and how to effectively implement it has become a focus of attention. At the same time, issues such as security and ethics are also receiving attention, and the industry is striving to seek a balance between AI development and risk control.