Tag
Microsoft Research's newsletter highlights Aurora 1.5 for open weather forecasting, Flint for AI-driven visualization, FlowDAgger for robot adaptation, and other advances in AI security and conversational memory.
This paper presents NIVA, a multimodal foundation model trained on Earth system simulations to learn coupled atmosphere-ocean dynamics for subseasonal-to-seasonal prediction. Initial validation shows the model captures key modes of climate variability by accurately predicting major climate indices.
TerraBench is a new benchmark for evaluating AI agents' ability to reason over heterogeneous Earth-system data, including gridded data, satellite imagery, and simulator outputs. It reveals significant limitations in current frontier models, with top performers achieving only 59.2% tool-use score on average.