I don't think the author is referring to the C–H correspondence. Just the fact that

a) it can be actually helpful to check that some property holds up to one zillion, even though it's not a proof that it holds for all numbers; and

b) if a proof has a bug, a program checking the relevant property up to one zillion is not unlikely to produce a counterexample.