Knowledge of, but as a statistics geek, I see a lot of people (like this guy) throw around the term "hidden Markov model" when they don't know what "hidden" means in context. In this context, an ordinary Markov model would be more appropriate.

A hidden Markov model denotes the existence of some dynamic parameter that the computer cannot directly observe, such as the user's current mood. Such a model would take into account transitions between "happy" and "angry" in addition to transitions between "Firefox" and "TextMate". These additional hidden variables are very useful in a wide variety of circumstances, but in this situation I suspect it would cause a combinatorial explosion of state space with little measurable benefit to the user.

Aza Raskin is also a developer: http://code.google.com/p/pythonicjavascript/