Recently, a research team led by Li Guoqi and Xu Bo from the Institute of Automation at the Chinese Academy of Sciences, in collaboration with relevant institutions, has successfully developed a brain-inspired spiking neural network model named 'Shunxi 1.0' (SpikingBrain-1.0). The model is based on the team's unique 'endogenous complexity' theory and has undergone complete training and inference processes on a domestically produced GPU platform. This achievement significantly boosts the efficiency and speed of large-scale models in handling lengthy texts or data sequences.

When compared to mainstream Transformer models, 'Shunxi 1.0' achieves comparable performance in various language understanding and reasoning tasks using only approximately 2% of the data volume. Moreover, by capitalizing on the event-driven nature of spiking neurons during the inference phase, the model demonstrates substantial improvements in both efficiency and speed. The research team has made the model openly accessible, along with providing a testing website and releasing technical reports in both Chinese and English for public reference.

The large-scale model unveiled this time presents a novel technical pathway, featuring a non-Transformer architecture, for the advancement of next-generation artificial intelligence. It is poised to drive progress in the theory of low-power neuromorphic computing and chip design.