I understand why it might seem that way, but it's really not. We've just put a premium on learning efficiency, and most math classes are highly inefficient - whether K-12 or university.

Just imagine how much more quickly a student could progress through a course if they were to work one-on-one with an expert tutor 5 days per week. Quite a bit more quickly in most cases. Well, our system effectively serves as an expert AI tutor and based on our calculations is on the order of 4 times more efficient than a traditional math class.

But I get the skepticism. I'd probably be skeptical myself.

Through trial and error and the application of some rarely employed, but highly-effective educational strategies like active learning, distributed practice, mixed review, mastery-based learning, interleaving, layering, etc., you end up with a pace of learning that's a little shocking. I realize it's hard to believe -extraordinary claims require extraordinary evidence and all that, but that's what got us started having our 8th-graders take the AP Calculus BC exam as as an outside, objective measure.

Opt-in educational programs of all sorts are usually heavily selection-biased toward students that are well-motivated and generally exceptionally easy to teach—good support from home, probably above-average SES, above-average IQ, et c.

Which doesn't mean such programs are useless, just that they're typically far less effective if you try to apply them to a representative slice of the overall student population.