AlphaEvolve and Gemini autonomously derive disaster prediction theory

A new paper describes autonomous algorithm discovery using AlphaEvolve and Gemini 3.8 Flash on data from 175,363 Netatmo IoT stations. The generated model detected all 21 historical disasters with 0.0% false alarms and was framed as a unified geophysical theory.

• AlphaEvolve wrote and refined Python algorithms without manual threshold tuning
• Model achieved 100% detection across 21 historical disasters with 0.0% false alarms
• Data came from 175,363 Netatmo IoT weather stations
• Transparent white-box code enabled a unified theory for rainstorms, quakes and eruptions