This Wired article claims that a knowledgeable scientists thinks that the "off-target mutations are already a solved problem."

Even if the Crispr protein itself doesn't cause mutations you likely will have to doublicate DNA a few time via PCR which produces additional errors.

According to Wikipedia:

This means that a human genome accumulates around 64 new mutations per generation because each full generation involves a number of cell divisions to generate gametes

I think we are very far off from reaching the exactly the same level of mutations or a lower level. The difficult question will be what level of mutations is acceptable.

Within a decade we should know a lot about the genetic basis of intelligence. Wouldn't it then probably be easy to create embryos that give birth to extremely smart people, far smarter than have ever existed?

If gene A raises IQ and gene B also raises IQ than that doesn't mean that both genes together will raise IQ even more. The might cancel each other out. A few people will grow to be extremely smart but I don't think that will be the case for every embryo in the project.