Actually, it is more general than that. Any polynomial time algorithm that takes as input a sequence and which has access to a unbounded source of uniform rand

Actually, it is more general than that. Any polynomial time algorithm that takes as input a sequence and which has access to a unbounded source of uniform random numbers (i.e. a "true" random number generator) should output a 1 when given a uniform random sequence with "nearly" the same probability as when given the output of a PRNG ("nearly" here means within a negligible amount e.g. the algorithm might just try to guess the PRNG seed). In other words, Pr[A(prng(seed, n))] - Pr[A(uniform(n))] < 1/negl(n) Where for all polynomial functions p(n), there exists an N such that for all n > N, 1/p(n) > 1/negl(n); uniform(n) is an n element long uniform random sequence, and prng(seed, n) is the first n elements of the output of the PRNG for a (uniform) random seed. This is a common definition in cryptography and a variant is mentioned further down in the paper in Section 2.