What is usually missing from these articles is even a simplified error analysis.

We can all count the floating-point operations, and today caches sometimes influence matrix algorithms more than raw flop counts. The problem gets more interesting when the matrix becomes ill conditioned and different algorithms will produce very different results.

Yes! Please.

I'd be interested, but probably if you went into less common subjects. "Don't invert that matrix" is probably the first thing I, or anyone else, learned in numerical analysis.

Side note: My professor also mentioned it at least 3 times in each lecture... so it is important.

In the same vein as "Statistics Hacks" I'm sure those same fans will be fond of "Numerical Hacks". :)

I did numerical analysis too...I don't think most CS departments cover it though, so chances are a lot of people would be interested in it.