>but not the bits as determined by his
>5-bit encoding.

right :-)

Well, it's been a few years since I've had information theory lessons, so I looked it up on wikipedia:
http://en.wikipedia.org/wiki/Information#Measuring_information_entropy

There is no easily quotable part of this explanation, so I'll try to do something else instead. Imagine that you write a letter and store it as a text file. This letter contains a certain amount of information - namely, the text you've written. It's stored in ASCII code, probably. Now, imagine that you zip this file. Now it's not stored in ASCII anymore, it's stored as an ASCII file inside a ZIP file, which is smaller. However, this smaller file contains exactly the same information as the text file you started out with. Less bits, same information.

I agree with your second point. I should have listed learnability along with the "other factors."

Second, his "Efficiency lets you know when you can stop looking for a better design" meme is wrong. It's true if you're designing for robots. If you're designing for humans, it's wrong, because humans learn, and even "worse," they constantly forget and thus are forced to re-learn. The most efficient interface is only the most efficient interface for a user who uses the interface correctly. But the whole issue of interface design is that users *never use the interface correctly* and that we thus need to design interfaces which work for people who are "stupid" from the POV of the programmer.

His article is interesting, but in my opinion completely useless unless you're designing an interface to be used by robots.