A new training technique streamlines the production of fast, trustworthy, and adaptable AI models for climate prediction, known as climate emulators. The technique, developed at MIT, boosts the performance and adaptability of a climate emulator while using only a fraction of the data typically required.