If H1, the human-equivalent generation of machines, takes one year to create H2, which doubles H1's speed and computing power, then H2 would take only 0.5 years to create H3, which would take 0.25 years to create H4, and so forth. The math series sum up to two or that the whole big bang would only take twice the amount of time for one doubling cycle (assuming that the delay caused by physical limits is negligible). IF this actually holds, then the progress when H1 is created will be way faster than the simple exponential illustrated in the article.
My first point is really just that the exponential is not hyperbolic. It's tempting to see an exponential graphs and think the value is rising against some invisible wall, like it is going to infinity. But of course it is not.
My second point was between the linear-exponential shooting skywards or the log-exponential rising incrementally which is "more real"? In the case of computer capacity I claim the incremental one is real, it reflects how we experience the increases.
None of this really speaks to whether the Singularity will happen or not. I'm just saying if it happens it won't be because of some upward surge in computing capacity just prior to The Singularity. Because no such surge is predicted.
Kurzweil does not say much about software. He feels we will get everything we need by reverse engineering real brains. I think this is possible, but it's certainly not obviously true, and it's not a simple extrapolation. We could be sitting there in 2040 with gobs of computer power, and detailed maps of the brain, and just not understand how it all works. Look at the Genome today.
I saw Kurzweil speak in 2008. I thought he was compelling but certainly leaned heavily on hype about the exponential which I don't think is entirely justified: http://www.kmeme.com/2010/07/singularity-is-always-steep.htm...