Bad article. The seed of a cryptographic generator isn't unknown to the programmer. That's not what makes cryptographic generators different. What makes them different is that it's computationally infeasible to derive the seed from the output, or even distinguish the output from random, e.g. tell whether a certain sequence came from this generator.
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. :)