Translated data: The team led by You Yang at the National University of Singapore recently introduced the p-diff diffusion model, which can generate neural network parameters at 44 times the speed. This model combines autoencoder design to learn parameter distributions and produce high-quality model parameters. Its accuracy is close to or even surpasses that of manually trained models, with excellent generalization capabilities. LeCun has expressed his approval, considering it a significant breakthrough in the field of AI. The p-diff model accelerates the efficiency of neural network training and offers new insights for the development of AI technology.