Weather prediction is shifting toward data-driven AI models, which have been shown to beat leading numerical weather prediction systems on average while using far less computation. But average performance can hide weaknesses on rare, high-impact events. A new arXiv preprint (2610.22077) takes up that question, evaluating probabilistic AI models on tropical cyclones.

The abstract confirms the motivation: AI models are now competitive or better on typical forecasts, yet it remains unclear whether that success carries over to extremes. The paper's title and framing indicate a focus on tropical cyclones, but the released abstract stops short of reporting results. Readers will need the full preprint to learn whether the models hold up where it matters most.

This digest is based on a single source, so there are no independent findings to compare or contrast.