For a very specific subset of problems, which I believe now are the ones amenable to "simulated annealing" solutions so for example, finding local optimisation points in a complex problem which can be expressed as a surface plot in 3D optimising for height (or depth)

But not (for example) breaking RSA via Shor's algorithm.

It also appears to be a shotgun: you get "the awesome" for a brief window and then the magic pixie dust disappears and its incoherent and indistinguishable from noise for a bit again.