Classic articles, especially the Gaffer one.
After reading deWitter, I am wondering: Is it really fair to say that the "constant game speed" solutions help with input response time on slower machines? The article certainly seems to suggest that, stating it as a flaw of the other approaches. Furthermore, this idea is reinforced by the notion that your game logic will be independent from your rendering. However, what you are really doing is giving your physics simulation time to catch up if the rendering is running slow. The catching up is done in a rapid sequence of fixed-length incremental steps. If I am not mistaken, the main purpose of this is to make sure that your simulation is well behaved. However, this should not really affect input response time, right? In a sense, the simulation is running in virtual time, whereas user input happens in real time.
Either way, it is a good article on an interesting topic. The intent of my post is to verify that my understanding is correct and to help others that may be confused.