I thought this paper was interesting and found some of the toy examples in the beginning particularly amusing. However, as others have mentioned, I wish the paper had given a bit more real world motivation. Even the real world tasks that were covered (byte and word prediction) seem somewhat useless (though I could be wrong). I would have preferred to see motivating examples that actually needed to be solved.

I also thought the paper could have done a much better job explaining notation. Although most of the predictive methods are fairly straightforward, I found myself confused sometimes because I couldn't figure out exactly what n or some other variable stood for.