This might be actually much faster for the following reasons:

• HW acceleration
• Neural networks are probabilistic and self-organizing, thus errors in the underlying HW are acceptable (much to the contrary to classical computing). It is much easier to build chips if they need not be 100% error-free.
• Once we understand the brain we might be able to wire the logic much more efficiently then it is done in real brains.

Pandora is certainly not a scam, they already have our money but they are still working. It might be a vaporware but I don't think it will be since it is a Linux device fairly compatible with Beagleboard and I am sure there will be enough enthusiast who will port new software to this device.

I am really tired of this. Some blogger/journalist shares with us his ingenious thoughts what is Linux doing wrong and how it can get better. The crowd disagrees, discussion under the blog/article turns to flameware and nothing is accomplished.

The blogger/journalist is angry that the crowd didn't agreed and developers didn't "fixed" the "obvious" failures he has pointed out. So he writes another blog/article how "these Linux guys" do not like critics.

Dear blogger/journalist, everyone "has great ideas" and everyone "knows how to fix the world". But at the end of the day it is the doers who change the world while nobody cares about blogs.

So next time please, either spend money for some commercial distro and write your suggestions/criticism to them or keep using your free of charge Ubuntu (at least it seems according to TFA that this is what you were using) and help the community to improve the product, there are plenty of tasks even for non-programmers. But do not keep saying "they" should improve that, "they" should listen to my criticism...

Sorry, we can't. We just have an chip that has the same number number of neurons. But the thing is that we suspect that the real neurons are much more complex than the classical Hebbian model. We also have no clue how to organize the neurons to do anything non-trivial and make use such a hign number of neurons.

Nevertheless, this is a great achievement. And it is good to know that once the "software" problems are solved we will have a good hardware too.

This scheme seems to me very similar to audio/video watermarking, combined with on-line check how many copies with the same watermark are played at one time. The on-line check part seems to me very DRM-ish, but the watermarking part seems to be quite reasonable.