Garrett Baker comments on Supervising strong learners by amplifying weak experts

[Garrett Baker]

In the article it says,

Supervised learning required tens of millions of examples in order to learn these algorithms. This would be a prohibitive cost if the examples were provided by an external expert. In contrast, Iterated Amplification required tens of thousands of examples in order to learn the much simpler decompositions.

But this seems to disagree with the graphs shown above. Are the number of questions here different from examples?