MIT Finds Scientific AIs Converging on a Universal Representation of Matter

A viral MIT paper shows that AI models trained on different scientific domains are independently converging on similar internal representations — suggesting emergent universality in how neural networks understand physics.