What Alpha School got right, and what is still unproven
Alpha School built its school day around AI tutoring. Some of its ideas are sound. Its results have not been independently tested.
Alpha School is a private school that runs its academics on AI tutoring software. It deserves a fair reading: some of what it does is worth copying, and some of its claims are not yet established.
How it works
On its own description, Alpha students spend about two hours a day on AI-driven academics, and the rest of the school day in life-skills workshops (the program, FAQ). Adults in the room are called guides rather than teachers; the school says their job is motivation, mentoring and coaching, while software handles routine instruction. Students are expected to reach 90 percent mastery of a concept before moving on. Tuition has been reported at 45,000 to 75,000 dollars a year, depending on campus. [primary source for the tuition range]
What it got right
Mastery before moving on. A child who scores 75 percent on fractions and moves on carries a gap into everything that builds on fractions. Requiring mastery is an old idea, and Alpha applies it consistently.
Separating practice from everything else. Concentrated, individual practice in the morning leaves time for work that needs other people. Whatever one thinks of the specific workshops, the idea that software should take less of the day, not more, is sound.
Adults where adults matter. Putting adults in charge of motivation and coaching, rather than lecturing, is a reasonable use of scarce adult time.
What is still unproven
There is no independent study of its results yet.Alpha reports strong results on standardized tests its students take, such as MAP Growth. No outside researcher has yet studied those results, so for now they are the school's own account.
Critics cite selection effects. Families who choose, and can pay for, a school like this differ from the general population in ways that affect results by themselves. In one Hacker News discussion, a commenter argued that the article under discussion mentioned family income only once, when it should have been central. In another thread, the point was made that gains may come from the number and quality of adults more than from the software. These are arguments, not findings. Only a careful outside study can settle them.
What we take from it
ayyayo is not a school and does not teach academic subjects. But two lessons carry over. First, mastery: an ayyayo lesson only completes when a child passes a check and explains the rule in their own words. Second, restraint: we make no claim we cannot measure. We are measuring ayyayo in pilots and will publish what we find, including what doesn't work.