The AI Revolution: The Road to Superintelligence

I'm not well-versed on the subject, but I see two problems (which I also voiced in the comments section of the article) with the claim that once an AI reaches human-level intelligence, it can improve its own intelligence in an exponential way:

1) Who says that each additional unit of intelligence doesn't need an exponential (or worse) amount of effort? This would mean that the increase in intelligence is very slow, and most importantly, linear or even sub-linear.

2) Maybe we can build an AI that has an IQ (relative to humans) of 100, and that's the best we can do. What if an IQ of 100 isn't enough to improve itself? Then there won't be any recursive self-improvement at all.

These two assumptions are simply taken as given, which I find a little disappointing in an otherwise very good article.