We built our WeatherNext forecasting models to help meteorologists and scientists more accurately predict weather — including potentially devastating events, like cyclones. To learn more about how our prediction technology works and how far we’ve come with WeatherNext, we sat down with Ferran Alet, a research scientist at Google DeepMind.

Learn more:

I appreciate the technology and the continued efforts. However, I wish a few more resources would be geared toward improving your own Pixel Weather app. The UI and other features are great—it’s just nowhere close to being consistently accurate.

AI making weather prediction faster and more accurate — that's a game-changer. 🌦️

Hopefully it will save many lives! That also will change a lot in a forecaster work as the scientist has to learn how to use it literary so that would be interesting to see too

AI predicting extreme weather more accurately could be a huge win for early warnings and disaster preparedness. 🌍🔥

decade of progress in one gen is huge

Hopefully WeatherNext can finally figure out if I actually need an umbrella today instead of just guessing like my local news station does every single morning.

Forecast quality becomes operational value only when uncertainty is calibrated and delivered early enough to change decisions. The useful benchmark is not one headline score, but lead-time reliability across regions and rare events.

Weather forecasting is one of those areas where AI can have an enormous real-world impact. Better predictions aren’t just about convenience—they can mean earlier warnings, better preparation, and potentially lives saved when extreme weather hits.

The learned model still eats the physics models' output as training data, so it's less a replacement than a very fast interpolation of forty years of ECMWF reanalysis. Which raises the interesting question of what happens to forecast skill if everyone stops funding the expensive simulations that generate the training set.

the value isn't just predicting the storm. it's giving people more scenarios to plan around before it arrives

Trump bootlickers

One of AI’s most valuable uses could be warning people before disasters strike. If WeatherNext can improve both accuracy and speed, its real-world impact could be enormous.

What is the name of the dino?

🩵🩵🩵

Great 👍🏾, one step ahead for humanity

Go fuck yourself, cowards.

ai predicting cyclones in real time is actually crazy

Using Ai to solve problems, while also getting rid of the data sources it bases things on is dumb.

👀🍿🧲

Google great

Forecasting doesn't need a lab budget to be real. My own version runs off FRED, Yahoo Finance, CoinGecko and TradingEconomics, all pulled into one SQLite database I built myself, no team, just the discipline of feeding it daily.

and built more flash models

Cyclone prediction is the use case where a few extra hours of lead time genuinely saves lives, way more consequential than most AI forecasting headlines.

What is the name of the dino?

@google is promoting pedos suportes by pedo @potus

You're great @Google and what this is available all public and free?

AI 주식을 싫어하지만 저는 구글을 좋아합니다

Disaster predictions for weather, geology, and oceans should be done together, just like with the mudslides in Nepal and Tibet. If we could predict them, thousands of lives would be saved.