Can AI make your town's weather forecast better?

A trained AI model can produce a ten-day forecast in minutes on one chip, which could make local forecasting cheaper. It still needs good observations and effective warning systems.

A September 22 report from the Rockefeller Foundation and University of Chicago says trained AI weather models can produce ten-day forecasts in minutes on a single chip—a fraction of the computing required by traditional physics-based models. That could make detailed local forecasts much cheaper and more widely available. But affordable computing does not replace good local observations from weather stations and radar, or warning systems that actually reach people.

Faster and cheaper forecasting is meaningful progress; it is not magic, and the forecast is only as useful as the data and public-safety response around it.