Hmm, I would think it will be the reverse...AI will be useful in diagnosing mid and long range patterns, but the physics-based models will do better in the short term, as they are analyzing loads of fresh data....and there is less time for "compounded errors" to mess things up. Some of the early statistical comparisons are indicating that AI is most useful beyond four days. But I agree, it is another tool in the box and not a panacea.