Source: Phys
Researchers at the University of Sydney have discovered that certain fractal patterns in data create hard mathematical boundaries beyond which AI systems cannot improve, regardless of scale or training. This is a structural constraint embedded in how predictability itself works. The industry's assumption that bigger models solve harder problems will eventually hit a ceiling tied to underlying data geometry. The finding reframes AI capability debates from "how much data/compute do we need" to "what problems are mathematically unpredictable," forcing a reckoning with which real-world applications actually have tractable patterns versus which ones don't.