The 4th Optics Valley Artificial Intelligence Conference was held in Wuhan, drawing academicians and experts to discuss frontier AI technologies. Academician Qian Depei proposed that "deep convergence of intelligent computing and supercomputing will be the key path to cracking the compute bottleneck in the large-model era" — noting that a tenfold increase in model parameters drives roughly a 68-fold rise in compute demand, and that existing systems still face constraints in energy consumption, reliability and programmability.
Qian advocated a "super-intelligent convergence" concept in which AI computing and supercomputing are not substitutes but mutually supportive, deeply integrated partners — using an integrated compute network as the entry point, building a supercomputing internet and exploring new models and mechanisms that unify the interests of multiple parties.
Yu Dengyun, chief designer of the fourth phase of China's lunar exploration program, said the country has completed nine lunar-related missions, and that under AI empowerment, lunar robots are being upgraded toward embodied-intelligent robots capable of handling the "high-risk, high-complexity, high-uncertainty" nature of space scenarios. Academicians Wang Yaonan and Chen Xuedong noted that embodied intelligence — a key vehicle for deep AI-economy integration — is at the critical juncture from technology R&D to industrial application. More than 60 experts from Tsinghua, Zhejiang University, the CAS Institute of Automation and other institutions presented reports on industrial large models, brain-computer interfaces and medical AI.
Source: China Science Daily / ScienceNet
