Review for submission 25

The paper addresses the question of generating pseudorandom data in a provably secure way. Essentially, despite well theoretical understanding of pseudorandomness, practical random generators are not analyzed in a provably
secure way, and usually based on some heuristic argument. In contrast, the
point of this paper is to (1) give a clean and formal definition
what it means to have secure (pseudo)random generation which withstands a
variety of powerful adversarial attacks (such as learning some of the outpus of the generator, resetting its state, affecting the fresh randomness used to
refresh the state, etc.; (2) give a simple and elegant scheme which once and for all should settle the general paradigm of how to construct such generators.