366 ‒ Transforming education with AI and an individualized, mastery-based education model

KEYPOINTS

• 01

  US K-12 spending is huge and results keep slipping

  Peter notes the US spends about $1 trillion on K-12, roughly a seventh of global education spending, with poor returns — paralleling healthcare. Joe Liemandt cites the national NWEA MAP norms: the average eighth grader in 2025 knew less than in 2020, who knew less than in 2015, so it isn't just COVID. Only the 99th percentile improved. AP exams were re-normed easier because college students now know less.

• 02

  Straight-A transcripts hide years of gaps

  Placement tests at Alpha show incoming A students range from one year ahead to three years behind, and B students from three to seven years behind — a seventh grader missing many fourth-grade questions. Joe tells parents paying $40,000 for private school that the grades were misleading, but stresses the gaps are quick to fix.

• 03

  Fact fluency frees working memory

  Drawing on cognitive load theory, Joe says memorized facts don't occupy working-memory slots, so skipping multiplication tables in favor of calculators handicaps higher math. 'Careless errors' often signal overloaded working memory from a missing foundation rather than a lack of grit.

• 04

  Knowledge is hierarchical, so time-based promotion compounds holes

  Physics is calculus with word problems, chemistry is algebra with word problems, algebra is advanced fraction manipulation, and fraction scores predict algebra scores. Passing kids on with 70–80% mastery creates holes that surface in algebra or chemistry. His sports analogy: a point guard losing the ball 20% of the time works on dribbling, not dunking.

• 05

  With mastery and AI tutors, a grade level takes 20–30 hours

  Instead of ~180 days of class plus homework, Alpha students finish a grade level of a subject to mastery in about 20–30 hours (around 22 each for sixth- and seventh-grade math). Bloom's 'two sigma' finding — individual tutoring to mastery lifts average students to roughly today's top 10% — underpins Joe's claim that, excluding the bottom 5%, 95% of eighth graders could reach top-10% performance in math.

• 06

  Motivation is 90% of the problem; 'time back' is the motivator

  Edtech handles placing kids at the right difficulty; it fails at engagement. Pitching '2x learning' fell flat with parents — fewer than 10% prioritize academics — but '2-hour learning' that frees the afternoon for workshops, sports or projects motivates every kid. Lecture classrooms retain maybe 5%. Their AI watches screens and coaches against scrolling and guessing; a 'waste meter' tracks unproductive time.

• 07

  Keep students in the zone of proximal development

  Questions should be answered correctly about 80–85% of the time — 99% is too easy, 50–60% causes disengagement — with lessons completed first-time-right 90–95% of the time. A Stanford lecture conveys one to five facts or concepts per hour; their engines reach 20–40, aiming for 100 with generative AI.

• 08

  Mastery turns ability into a decision

  Joe argues good grades in a time-based system track IQ plus conscientiousness; mastery removes the IQ ceiling for K–8 concepts, and extrinsic motivators compensate for low conscientiousness. Once kids see their remaining gap in hours ('you're 60 hours behind'), reaching the top becomes a choice of effort. He says the 'I'm not a math girl' story has no basis in learning science.

• 09

  Paying kids is kindling, not the fire

  Alpha uses extrinsic rewards — time back above all, but also money, such as $1,000 for middle schoolers who reach the top 1%. Critics worry it undermines intrinsic motivation; Joe says those studies cover already-motivated kids, while his students gain belief in their capability. Peter's framing: kindling you don't need forever.

• 10

  AI is education's light microscope

  Peter compares the light microscope enabling germ theory to AI enabling 40 years of learning science to finally work at scale: precise teaching plus closed-loop measurement of what each child knows. Joe expects this year's math curriculum to teach more in 20% less time based on last year's data. He's adamant that schools, adults and classmates remain; the teacher's role changes.