will say that that sequence of distributions is pseudorandom if for
all probabilistic polynomial time adversaries A.
There's some negligible function epsilon such that if we look at the difference in
probabilities, between the probability that A(x) outputs one when x is sampled
according to D(n) and the probability that A(x) outputs one when x is sampled
according to the uniform distribution over strings of length p(n), then
the absolute value of the difference of those probabilities is at most epsilon n.
So intuitively, for any statistical test A, running in pro probabilistic polynomial
time, the distinguishing gap, right, the probability with which that statistical
test can distinguish whether it's given a string sampled according to D(n), or, or
whether he's given a string sampled according to the uniform distribution, or
the strings of the appropriate length.
It's at most epsilon(M).