There's also the issue of Linux support, not just to developers, but even the end users. As an end user, I've repeatedly been ignored by Nvidia when submitting detailed bug reports about their drivers, both to the support forum and by e-mailing email@example.com. Oh, and even on Twitter. Not even a "thanks, but we're not fixing this". Stone-walled. I won't be making the same mistake next time I buy hardware.
Why is it so wrong to use the proprietary drivers?
Here's an example. I bought a laptop with an Nvidia Quadro FX 3800M specifically for triple-head support (via docking station) and I can't even get dual-head to work properly due to an infinite loop somewhere in the binary drivers when mode-switching. I've done most of the investigation work, even running X through gdb and they're not interested in helping - not even some basic debug symbols.
I've been completely ignored by Nvidia via both of their official support channels: (1) the nvnews.net forum and (2) their firstname.lastname@example.org email address touted in their driver README. Even on Twitter.
What options do I have now with a binary driver except to run Windows (where triple-head works fine) or change hardware? I can't imagine why I would ever again support a company that can't even acknowledge my pain with their product, let alone help me to fix the problem myself.
On the contrary, Jeff Hawkins argues that memory and intelligence are very closely linked, in this TED talk. He says that intelligence is prediction, i.e. pattern recognition which needs knowledge to predict. Extending his insights, I guess we could look at the typical knowledge vs. intelligence debate as nurture (learned knowledge) and nature (innate knowledge through evolution, i.e. what we traditionally think of as intelligence).
Those memory-based qualifications aren't useless. They don't provide as detailed knowledge as a degree, so the resulting prediction system isn't as intricate. Yet I argue that they still produce intelligent behaviour, but, of course, relative to more advanced education / training.
I would be curious what the article thinks is so fundamentally wrong with the sockets paradigm.
TFA doesn't say there's anything wrong with the sockets "paradigm". It basically says that the API has performance issues with two use cases (low latency and high bandwidth) and doesn't support multi-homing very well.
Wrt multi-homing, it mostly talks about applications which I'd call mobile IP. Also, it curiously doesn't mention the ability to bind to all addresses as the current (crude) solution to multi-homing.
An opt-out system (or presumed consent) was debated a lot here in the UK in 2008. Due to objections from various sources, our government is trying a "major publicity campaign" to boost the numbers of donors. If that's not successful, we may end up with the system anyway.