AI models converge to similar internal structures

Johns Hopkins researchers analyzed 1,100+ AI models and found that, regardless of task or domain, neural networks converge to remarkably similar, low-dimensional internal structures.