Learned emulators accelerated climate and materials simulations, complementing classic HPC.
Surrogate models trained on physics-simulator outputs cut climate-scenario and materials-design costs by orders of magnitude.
Validation against conservation laws and observational data is critical: a fast but biased emulator can mislead policy.
AI for science institutionalized in national labs and big tech with physical-impact metrics, not chat alone.
Around September 4, 2026, DeepMind unveiled WeatherNext 3 with hourly forecasts on a 5 km grid (vs 25 km in WeatherNext 2), live satellite data versus ~6 h NWP lag, 100 m wind for turbines, and up to ~50% better precipitation; it powers Search, Gemini, Maps, and Earth Engine.
Autonomous-engineer products plan issues, edit repos, and open pull requests with variable human oversight.
Districts and edtech deployed LLM tutors; studies showed gains when pedagogy is designed—not generic chat alone.