I had glanced over that paragraph, and indeed, it's wrong. One, the stream only changes if the seed is dependent on the system time. Two, if you modify the result with another pseudo-random number, then you have a function whose randomness if as only as good as its inputs.