At the fourth BEYOND International Science and Technology Innovation Expo, Zhang Wen, the founder of Bi Ren Technology, shared his views on AI large models. Zhang Wen said that the cost reduction of AI large models can be achieved from three aspects: chips, systems, and cluster capabilities. "At the chip level, training an AI large model used to take several months. Now, with the increase in chip computing power and bandwidth, it may be shortened to several weeks. In terms of system, we may integrate CPU, GPU, and DPU architectures in the future to improve system cluster capabilities. Now, clusters are getting better, and the efficiency of computing power has also greatly improved."