LLM-Style AI Predicts Material Motion

⊲ Notebookcheck ⊳ MIT has developed WorldParticle, a single AI model that simulates the physical behaviour of many materials. Rather than relying on separate equations and custom code for each substance, it breaks objects into tiny 3D particles, then predicts each particle’s next state of motion, much as a chatbot predicts the next word. The model covers granular materials such as sand, liquids, deformable solids and proteins, with protein simulations running about 100 times faster than conventional methods. The work is described as a step towards foundation models for physics.
◂ Foundation models for physics would shift trust from validated equations to learned particle behaviour, raising hard questions for verification in safety-critical training and engineering. ▸
⊲ Featured Image – MIT ⊳
⊲ www.simulation.news ⊳