Nursing schools are all running at or over capacity because of the huge number of people wanting to be nurses. Whether they should become nurses is a moot point.

Writing a basic kernel is really not as hard as it sounds. The trouble comes if you want a more elaborate kernel, that supports a variety of hardware, and does this with a minimum of bugs, and is secure. I recommend writing a kernel, actually. It's pretty fun.

Take a look at the Android API sometime. With that thing, decent GUI design seems to be the path of least resistance.

Don't worry. We can liken Bash to a largemouth bass, and say that it doesn't like to be anthropomorphized (because it's a fish, not a human) and we'll have ended the loop. If you learn Haskell, you'll get this good at recursion.

CUPS? Why would you want to run the Common Unix Printing System on a phone?

Android is already a free software platform where users can write software without being locked out. Isn't that what you're looking for? Sure, it came from Google and is promoted by a consortium of telecom giants, rather than springing straight from the People, but as long as a cat catches mice, does it really matter what color it is?

Most of the beauty of the Android's design is looking good and being usable on a tiny screen. I doubt that most software written for people with big monitors is going to be pleasant on a cell phone screen.

It's valuable for sociologists, if nothing else. See how porn evolved over the years!

Are you sure? I found this paper about doing a fast Fourier transform with probabilistic CMOS hardware, with errors making it right through to the output -- but the errors were so small that, for many applications, they are irrelevant. And they were able to get by with 82% less power than a conventional approach with the same precision.

You're exactly right and should be modded up to 5. One of the applications that these researchers keep trotting out in their papers is neural networks. They figure they can make neural network hardware a few hundred times faster or more energy-efficient, as compared with a non-probabilistic CMOS implementation. They're also looking at Bayesian inference and classification problems, so that's nice too.

That's why these guys want to make systems-on-a-chip which integrate probabilistic circuitry with an ordinary ARM processor core. It can handle the general computing and the probabilistic coprocessor can handle a few specialized jobs. This won't replace your CPU, but it might make your graphics card faster and more efficient a few years from now.

This guy's soon-to-be-unveiled chip uses several different voltage levels. Higher voltages are used for more significant bits, and vice versa. There's your guarantee that the more significant bits will be less error-prone.

That's exactly what they're proposing: a system-on-a-chip with a conventional processor for most of the application and an on-die probabilistic coprocessor for the calculations that can be done in a probabilistic way. The only people talking about approximate accounting and other such inappropriate uses are Slashdot "smart"-asses.

The processors aren't entirely general purpose. They're stripped-down, simplified cores driving loads of SIMD hardware. There are instructions to control caching because a GPU does a lot of streaming data access. The core is pipelined and in-order, with 4-way SMT. It's very much optimized for throughput at the expense of everything else. That's a good thing for a GPU, of course.

It's a solid state hard drive that uses PCI Express instead of SATA. It looks pretty zippy. Decently large, too. Is there something else cool about their technology that I'm missing?