Am I right to assume one could also sample from a Gaussian distribution for the method to work? Of course, the probability e of sampling between the two real numbers would be very small, but it would be nonzero.
So ... yes, choose from a Gaussian, but then you have to tell me exactly how.
That's tricky.