MIT Creates Multimodal AI for Materials

MIT researchers developed a multimodal AI model that combines 3D atomic structures with natural language descriptions to identify better material designs than text-only models. The system, detailed in a new paper in Nature Communications, helps scientists find crystals for applications like solid-state lithium-ion battery electrolytes by learning geometric representations from experimental research papers and computational databases.