AI made weather forecasting fast and cheap, but a third of the world still gets no warning
alex_verem · x · 2026-09-02
A recap of key discussions from July's AI for Good Global Summit in Geneva: WMO head Celeste Saulo noted that as a PhD student, simulating a single day of weather took a week — now AI models produce a 3-4 day forecast in minutes, running on a personal device instead of a supercomputer.
Around 90% of disasters worldwide are weather-related, so faster forecasts let rescue teams act days ahead and farmers move livestock, with prevention costing a fraction of rebuilding. But UN disaster risk chief Kamal Kishore laid out where the chain breaks:
- Only 128 countries (65% of the world) run a multi-hazard early warning system
- Cell broadcast, the gold standard for alerting every phone in an area, is live in fewer than 40 countries
- Forecasts are only as good as the observations feeding them; weather radar has always been a rich-country technology, and satellites are only starting to fill the gap, with Global South data still thin
Kishore's target is 100% warning coverage: the forecast is solved — the message still has to reach the person standing in the path of the flood.
More from AGI Musings
- Midjourney founder David Holz: rent 8x GB300 racks for a weekend to attack 3000 open math conjectures — ethanniser · 2026-09-02
- Sam Altman: Declaring AGI is becoming meaningless — ChrisGPT · 2026-09-02
- Compute dominance equates to the largest army of computer-use agents — djcows · 2026-09-02
- Fable 5.1 First Impressions: High Pricing, Less Optimization, Shift to Complex Tasks — Aizkmusic · 2026-09-02
- Dwarkesh argues incident plausibly supports open source AI — connoraxiotes · 2026-09-02
- Opinion: Are AI Labs Responsible Enough to Manage the Risks of Machine Intelligence? — joshua_saxe · 2026-09-02