WorldParticle unifies physics simulation with single AI model

Researchers from MIT, UBC, and other institutions have developed WorldParticle, a unified AI model capable of simulating multiple types of physical phenomena. The system uses a neural network to predict particle movements over time, eliminating the need for hand-coded equations specific to each material or force. Demonstrated on six different physical regimes including granular materials and proteins, it outperforms traditional numerical solvers in accuracy and speed, particularly showing a 100x acceleration in protein folding simulations. The work aims to create a general-purpose simulation foundation model for scientific discovery.