Optimization in high-dimensional spaces stalls at saddle points, not local maxima

roon, a user on X, says that in high-dimensional space hills are rare, so optimization usually ends at saddle points rather than local maxima.

The post gives no supporting detail, data, or context for the claim.