As someone in grad school for related topics, this stuff will undoubtedly surpass physical weather models (climate change is a different story). As I chart my own research direction, I wonder what kinds of problems meteorology will still be working on in a few decades? AI models like this will render many physical models obsolete. And with the leaps made recently by cube satellites and other remote sensing tech, direct observations may soon no longer be sparse. So what does that leave? Understanding the physical principles? Yes, and doing so will augment these advances. But what else? And what kind of applied problems could the physical principles solve? At some point, AI weather forecasts may cross the threshold of being "good enough" for 99% of use cases. I don't doubt that there is much more important science to be done (I recall someone famously claiming science was "done" in the 1800s), but admittedly it takes some creative thought to imagine. The field will have to evolve. I don't know if it is possible to know how it will evolve before it does. Just something I wonder about.<p>Also, I certainly don't mean to detract from this achievement . This is great work and it is truly awesome to see these kinds of leaps. There is plenty of vanity in science and the urge for self-preservation, but such advances in weather forecasting will save so many lives and do a lot of good in the world. I wholeheartedly cheer on the scientists behind this work and am excited to see what is around the corner.