Physics-based and AI weather models usually lose reliability beyond two weeks. Microsoft researchers say they have found a cost-effective machine-learning method for more accurate forecasts two to six weeks ahead, which could help farmers and emergency responders.
A post on the Hugging Face blog presents HiRO-ACEv2, a model for simulating extreme weather at 3 km resolution at a fraction of the usual cost.
Google Research presents Deep Review, a tool for analyzing technical documents.
Google Research highlights ten ways its work aims to amplify human ingenuity, as part of its Research@ 2026 event.
Google Research describes geospatial AI agents built on its Earth AI work and aimed at major planetary challenges.
A new machine-learning approach set records for weather forecasts two to six weeks ahead, MIT News reports. That window is especially important for emergency planning.