Techub News reports that researchers from PhAI Labs, The Chinese University of Hong Kong, Fudan University, Stanford University, the University of Oxford, and Princeton University have released JEPA-Anything, a domain-agnostic framework for building world models. The framework uses a method called Orthogonal Predictor Factorization (OPF) to extend the Joint Embedding Predictive Architecture (JEPA), applying the same learning paradigm across seven very different domains: vision, biology, clinical trajectories, control, molecular dynamics, physical fields, and weather. The research team validated the framework across multiple benchmarks. On tasks including single-cell data analysis, clinical event prediction, molecular dynamics simulation, and physical-field and weather forecasting, JEPA-Anything matched or outperformed existing specialized models. For example, in 100-step molecular dynamics simulations, it achieved the lowest prediction errors for substances including water, quartz, acetaminophen, and benzene. (MarkTechPost)