I'm excited for ML's effect on weather forecasting, even as a general skeptic in the AI/LLM space. I use Windy to get my forecasts and have taken to comparing the outputs of the various models, and it's interesting to me that some of the models consistently get things wrong for my region - ECM for example consistently over-predicts precipitation when you push it past a few days. I'm not exactly sure what this looks like on the backend (I think it's still a physics based model?), but I think it's pretty likely that a model trained on weather data directly would not make that type of error.
