Researchers from BIFOLD and Google DeepMind have developed MD-ET, a transformer-based molecular dynamics model that omits traditional physics constraints like energy conservation and equivariance.