Steve Bellovin on Nostr: Google's new AI-based weather forecast system ...

Google's new AI-based weather forecast system (https://arstechnica.com/science/2024/12/googles-deepmind-tackles-weather-forecasting-with-great-performance/) is amusing because it's in some sense a return to how forecasting used to be done. Today, forecasts are created by large physics-based models, requiring many hours of supercomputer time. Earlier, though, they were compiled by looking at historical weather maps, seeing what happened, and matching that against current observations. This approach was pioneered in the 19th century by Gen. A.J. Myer, who was known as "Old Probabilities" for this (and who also created the US Army Signal Corps during the Civil War). See https://media.defense.gov/2021/Jul/13/2002761507/-1/-1/0/MASKED_DISPATCHES.PDF, and nprofile1qy2hwumn8ghj7un9d3shjtnddaehgu3wwp6kyqpqeak06lshwmktyeqm9vs929spfhlrxs4taq4gtpr2g2w7v7zvz6nq64x704 (nprofile…x704)'s "Turing Cathedral" for the transition to physics-based computational models.
An AI-based system is once again looking at historical data to predict the future—but today, we call it training the model…