Local minima basically don't exist in high dimensional spaces, so in practice the tools work for a wide variety of problems.
Definitely not true. Not even close.
Not much of a response. Care to expand here? This is a well-understood principle in the existing body of research on convergence in high dimensional spaces.
This has been very lucrative for computers so far.