On the 23rd, Nature reported an artificial intelligence (AI) model. The model is called "NeuralGCM" and combines fluid dynamics with neural networks to accurately predict weather and simulate climate. The model surpasses some existing models and is expected to save a lot of computing power compared to traditional models. General Circulation Models (GCMs) can represent the physical processes of the atmosphere, oceans, and land, and are the basis for weather and climate prediction. Reducing the uncertainty of long-term forecasts and estimating extreme weather events are key to climate prediction. Machine learning models have always been considered an alternative means of weather prediction, with advantages in saving computational costs, but their performance in long-term forecasting is often inferior to general circulation models. In view of this, the Google Research team in the United States has designed the "NeuralGCM" model, which combines machine learning and physical methods and can perform short - to medium-term weather forecasts as well as decades long climate simulations. The accuracy of this model for 1-15 day forecasts is comparable to the prediction results of the European Centre for Medium Range Weather Forecasts (ECMWF), one of the best traditional physical weather models. For forecasts up to 10 days in advance, the accuracy of "NeuralGCM" is comparable to existing machine learning techniques, and sometimes even better. The climate simulation accuracy of "NeuralGCM" is comparable to the best machine learning and physics methods. When the team added sea level temperature to the 40 year climate forecast of NeuralGCM, they found that the model's results were consistent with the global warming trend found from ECMWF data. The new model also surpasses existing climate models in predicting tornadoes and their trajectories. The team concluded that these results collectively indicate that machine learning is a feasible means of improving general circulation models. (New Society)