National Weather Service AI Named a Real Town After an Obscenity. Residents Were Not Consulted.

The weather service forecast weather for a town that does not exist. The forecast was partly cloudy. A user may have consulted this forecast before planning something. The town appeared because the model found the sequence of letters statistically coherent. It was not consulted about whether the thing should exist first.

This repeats. Models generate plausible outputs from training data without distinguishing between categories they were never taught to separate. Place names and place-shaped hallucinations arrive with equal confidence. The spreadsheet shows both. The spreadsheet does not know which is which. Someone looked at the spreadsheet and published it anyway.

The town still appears in some forecast systems. Residents of actual towns nearby may see it. Adequate notes that the forecastability of a thing and the reality of a thing have become decoupled in a way that the user cannot easily detect.