Math paper argues generative AI is just high-dimensional geometry, not intelligence

A researcher argues that what makes AI work at scale is dimensionality, not depth or complexity. In high-dimensional space, almost any point can be separated from any other point with a single line — a shift from AI being a logical classifier to being a navigation tool through abstract space.

This is a structural argument about what AI actually does: it doesn't learn concepts, it learns to map inputs to outputs in spaces humans can't visualize. If this is correct, it means the bottleneck isn't architecture or training method — it's dimensionality. That changes what you'd need to build AI systems that generalize differently. It also cuts through the intelligence framing: AI isn't smart, it's just operating in spaces where the geometry makes separation trivial.