This $40M AI Company Is Using AI Tutors to Teach 2 Hours/Day | #233
Peter Diamandis and co-hosts Dave Blundin and Salim Ismail interview McKenzie Price and Joe Liemandt, the co-founders behind Alpha School, a K-12 private school model that compresses academics into two hours a day using AI tutors and 40-year-old mastery-learning science, freeing the rest of the day for life-skills workshops. They lay out grim baseline stats on US education (falling reading/math/science proficiency, collapsing trust in college, exploding tuition), then explain Alpha's five pillars: kids must love school, learn 10x faster via AI-personalized instruction, spend the afternoon on life skills, have teachers reinvented as motivational 'guides' rather than content deliverers, and be surrounded by strong culture and peers. They detail the AI tech stack (the Timeback platform, vision models that monitor and coach focus, a closed feedback loop that iterates curriculum), the current ~$10,000/student/year AI cost that they expect to fall sharply, and the business/political obstacles to scaling — parental skepticism, teacher-supply economics, and near-total rejection by public charter-school authorities. The episode closes on optimism that AI will give today's kids "superpowers" even as it disrupts adult jobs, and on the founders' next moves: new school formats (Texas Sports Academy, gifted, Montessori, wilderness) rolling out in 2026, and a push toward a large-scale randomized controlled trial to prove the model works.
State of US K-12 and college educationEconomy▶ 3:44
Diamandis presents data showing US high schoolers at historic lows in reading, math, and science proficiency, collapsing public confidence in college, exploding tuition, and college graduates now facing the longest unemployment spells of any group.
Alpha School delivers a full K-12 academic program in 2 hours/day using AI-personalized tutoring, with students scoring far above national averages (SAT averages of 1410 school-wide and 1535 for seniors vs. a 1024 national average).
The five pillars of a '10x better' schoolOther▶ 11:04
Joe Liemandt lays out the first-principles redesign of school: kids must love school, learn 10x faster, spend afternoons on life skills, have teachers reinvented as motivational 'guides,' and be embedded in strong character/culture/peer groups.
AI tutors vs. chatbots and the Timeback platformAI▶ 15:36
Alpha distinguishes AI tutors (which generate personalized, mastery-based lesson plans using learning science) from chatbots like ChatGPT, which they say are used to cheat; their in-house Timeback platform plus apps like Math Academy power the academic block.
Because AI now handles content delivery, Alpha hires 'guides' not for subject expertise but for their ability to motivate and connect with kids; guides come from teaching, coaching, and corporate backgrounds and start at six-figure salaries.
Mastery-based learning vs. time-based gradingOther▶ 47:04
In Alpha's mastery-based system every student can reach the top percentile because they don't advance until they've mastered material, reframing academic success as a function of effort rather than fixed IQ or 'grinder' traits.
AI-driven difficulty tuning and cognitive loadAI▶ 50:33
Alpha's AI targets an 80-85% correct-answer accuracy zone per lesson (the optimal challenge level per cognitive load theory and game design); accuracy below ~66% causes disengagement.
Vision-model screen monitoring and the closed-loop data cycleAI▶ 54:58
A vision model watches each student's screen in real time to coach focused-learning behavior (an 'XP' system where 1 XP = 1 minute of focused learning), currently costing roughly $10,000/student/year in AI compute; Alpha's learning-science team also runs an iterative closed feedback loop that updates curricula based on measured outcomes.
Screen time, chatbot cheating, and parental pushbackAI▶ 51:44
Addressing parental concern about screen time, Liemandt argues 'good screen time' (10x faster mastery learning) is worth the trade-off, while unmanaged chat access is disabled during academics because ~90% of students used it to cheat.
Scaling challenges: cost, quality, and motivation without owning the school dayEconomy▶ 1:17:53
The founders identify their core scaling bottlenecks as driving down AI cost, maintaining quality (love of school, academics, life skills) across new locations, and motivating a billion kids whose school day Alpha doesn't control — prompting a planned motivation-plus-learning product bundle with game designers and influencers.
Public-sector resistance and the case for large-scale proofGeopolitics▶ 1:14:51
Alpha applied for charters (public funding) in 10 states and was rejected everywhere in-person, receiving only one virtual charter approval (Arizona); the founders argue education needs a 'pharmaceutical-grade,' million-student randomized controlled trial to overcome institutional skepticism.
AI's diverging impact on kids vs. adultsAI▶ 1:24:29
The founders argue this is 'the best time in history to be a five-year-old' because AI will give children lifelong superpowers, even as it disrupts adult livelihoods (e.g., Uber/taxi drivers displaced by robo-taxis).
Predictions made
openDave Blundin: Fully interactive, AI-driven online tutoring — where the AI can take on any chosen voice or personality to maximize engagement, unlike today's fixed-personality human teachers — will become viable and widely available.
openJoe Liemandt: AI will become good enough to handle the earliest stage of bootstrapping young children (around age 4) into reading, a task current AI tutors still can't do.
“We think we need another 18 months before AIs can do that.”
Your call:
openJoe Liemandt: The roughly $10,000-per-student annual AI compute cost for Alpha's vision-model screen monitoring will fall by about a factor of 10, moving toward on-device processing.
“That'll come down a factor of 10... We're going to get it down to on device.”
Your call:
openJoe Liemandt: Several new Timeback-platform-based school formats beyond Alpha itself (Texas Sports Academy, a gifted/talented school, a Montessori school, a wilderness school) will roll out broadly.
“Next year in 2026, you're going to see all these other models beyond just Alpha that roll out based on this time back software.”
Your call:
openJoe Liemandt: A packaged 'motivation plus learning' product, built with some of the world's best video game designers and top influencers, will be released to help motivate students Alpha doesn't directly control.
“I got 20 years, so we're going to build 10,000 schools, you know, and that's going to be great and I love it.”
Your call:
Numbers that matter
Only 35% of US 12th graders are at or above reading proficiency, down from 40% in 1992Baseline US education crisis data presented by Diamandis
Only 22% of seniors are math proficient and only 31% are science proficientSame NAEP-style proficiency data on US high schoolers
Share of Americans who see college as important fell from 75% (2010) to 35% todayCollapsing public confidence in the value of a college degree
US college tuition has grown 893% since 1983Cited alongside falling perceived value of college and college closures
Alpha School's whole-high-school average SAT is 1410 vs. a US national average of 1024Comparing Alpha's academic outcomes to national norms
Alpha School seniors' average SAT score is 1535Corrected/clarified figure from Joe Liemandt on senior-class performance
90%+ of Alpha students report loving school; 40-60% say they'd rather attend school than go on vacationSelf-reported weekly survey data cited as proof of the 'love of school' pillar
The US spends over $1 trillion a year on K-12 education, with some cities spending over $40,000 per studentFraming how much money already flows into education relative to outcomes
Alpha/Timeback has invested over $100 million building its AI learning platformCost of building the Timeback engine and related apps over ~3 years
80,000 people applied for Alpha's guide/teaching rolesVolume of applicants used to find highly motivating non-traditional educators
Kindergarten student-to-guide ratio at Alpha is 5-6 to 1Higher staffing ratio for the youngest students because AI can't yet bootstrap reading
AI-generated lessons target 80-85% accuracy; disengagement occurs below ~66% accuracyThe optimal-challenge zone from cognitive load theory and game design applied to lesson generation
About 90% of students given unrestricted ChatGPT access in school used it to cheatReason Alpha disables chat functionality during the academic block
Vision-model screen monitoring currently costs roughly $10,000 per student per year in AI tokensDescribed as the most expensive part of Alpha's AI stack, expected to fall sharply
Average one-to-one teacher-student time in a traditional school is 22 seconds per dayContrasted with Alpha's weekly 30-minute one-on-one guide meetings
Worth digging into
🕳️ The Timeback platform's actual AI architecture
The entire 10x-learning claim rests on a proprietary engine (personalized lesson generation plus real-time vision-model coaching) that is described only qualitatively on the podcast and costs ~$10,000/student/year to run.
🕳️ Bloom's 'Two Sigma Problem' and 40 years of mastery-learning research
Liemandt grounds Alpha's entire approach in a 40-year-old, allegedly under-implemented body of learning science; verifying the strength and consensus of that literature would test the core premise.
🕳️ Independent verification of Alpha School's SAT and proficiency claims
All test-score statistics (1410/1535 SAT averages, near-100% mastery rates) are self-reported by the school itself, and Alpha is a self-selected, tuition-paying private-school population, raising obvious selection-bias questions.
🕳️ Alpha's charter-school rejections across 10 states
A 10-for-10 rejection rate from public charter authorities, described by Liemandt as school boards saying 'I was put on this earth to stop people like you,' points to a concrete regulatory/political conflict worth digging into.
🕳️ Vision-model screen monitoring and student data/privacy
Streaming every student's screen to a vision model for real-time behavioral coaching is a novel and potentially sensitive use of AI on minors that wasn't discussed from a privacy or data-governance angle.
🕳️ The proposed million-student, 'pharmaceutical-grade' RCT for education
Liemandt explicitly frames the lack of rigorous, large-scale randomized controlled trials as education's biggest credibility gap and floats running one — a concrete, trackable commitment.