Ferran Alet

Excited to join a live Q&A tonight with Bryan Norcross to talk about our experimental cyclone model!

Great question! We'll follow up with you+other CIRA folks+NHC with more details as part of our research collaboration. I'll say we have some understanding of the causes. We'll also probably push FNv3.1 in 1-2 weeks, which will give some improvements (but not full fixes) in the short lead times.

Early in the Atlantic season, but FNV3 is doing pretty well so far🎉

Seeing our model analysed and used by experts forecasters like James has been one of the most rewarding parts of the project!

Finally, we're releasing deepmind.google.com/science/weat..., a preview of our AI models. Huge kudos to @tom-andersson.bsky.social who played a key role improving Weather Lab with expert trusted testers since last Fall.

Here's a current cyclone in China, with our model (blue) being more confident.

According to internal evals, reproduced by the Cooperative Institute for Research in the Atmosphere (thanks to @franklinjamesl.bsky.social!), our experimental model excels at both tracks and intensities. We're looking forward to running real-time evals with CIRA & other worldwide partners.

At Google DeepMind, we're announcing a partnership with @nhc-atlantic.extwitter.link for expert forecasters to access our newest experimental model. We hope to enhance their forecasts this cyclone season! Looking forward to working with experts like @pppapin.bsky.social & @ericblake12.bsky.social!

In my case, the conversion happened with:
1) Kenji Kawaguchi's (my labmate at the time) "Deep Learning without Poor Local Minima" 2016 paper and
2) empirically finding gradient descent highly effective for a very non-smooth problem (modular meta-learning) in 2018.

YouTube video by Welch Labs The Misconception that Almost Stopped AI

Beautiful visualizations of optimization in high dimensions & why local minima became less of a concern as networks scaled: www.youtube.com/watch?v=NrO2...
It was interesting to see this realization start in 2014-15, and propagate around the community as theoretical & empirical evidence mounted.

This is an amazing resource to learn about LLMs and how to make them scale in practice!
I read the internal version of it a month ago and it's very insightful, you can feel they put a lot of passion into writing it.

WeatherNext WeatherNext is our most advanced weather forecasting AI technology.

Our GraphCast and GenCast got new names (WeatherNext Graph & WeatherNext Gen) and their real-time predictions are now available on EarthEngine & GraphQuery. Great to see our research advancements being deployed, looking forward to seeing what people do with them!

deepmind.google/technologies...

I'll be presenting GenCast, recently published in Nature, tomorrow Tuesday 8h45AM at #AMS2025 in room 339.

GenCast is a diffusion model that outperforms ENS, the top operational ensemble forecast, giving skillful probabilistic forecasts up to 15 days ahead.

ams.confex.com/ams/105ANNUA...

I'm attending #AMS2025. If you're interested in AI for Weather prediction(particularly GraphCast/GenCast) or in tropical cyclones, send me a DM!

Oops, I'm new to Bluesky :) I've now enabled DMs and sent you one

The Tackling Climate Change with ML #NeurIPS2024 workshop will be livestreamed for free!

If you're attending in person and want to chat about AI for weather prediction or Science&Sustainability more generally, send me a DM!

Hi Blanka! I'm a research scientist at Google DeepMind working on AI for weather; would love to be part of this community!

Probabilistic weather forecasting with machine learning - Nature GenCast, a probabilistic weather model using artificial intelligence for weather forecasting, has greater skill and speed than the top operational medium-range weather forecast in the world and provid...

Paper: www.nature.com/articles/s41...
Joint work with Ilan Price, Alvaro Sanchez-Gonzalez, @tom-andersson.bsky.social, Andrew El-Kadi, Dominic Masters, Timo Ewalds, Jacklynn Stott, @shakirm.bsky.social, Peter Battaglia, Remi Lam & Matthew Willson

Weather is fundamentally chaotic and with finite sensors and computation, forecasts need to be probabilistic past a few days. In GenCast, we used diffusion models to generate an ensemble of realistic future scenarios, showing skill up to 2 weeks into the future!

Excited to share our new Nature paper on probabilistic weather forecasting today @neuripsconf.bsky.social! I'll be presenting at the DeepMind booth 1pm Vancouver time.

I'll be at NeurIPS the whole week. If you're interested in AI for Science and sustainability, feel free to reach out for a chat!