There appears to be a distinct lack of actual discussion of "predicting" in this article. It's actually fairly easy to determine the state of simple RNGs by observing their output... I was expecting the article to be about that, not just reiterating the (obvious) fact that using secret data for the seed is good. :)