Fractal patterns reveal fundamental limits to AI predictability

University of Sydney researchers identify a mathematical reason AI systems can learn divergent behaviors from nearly identical training conditions.