Summary

Alpha School’s central wager is that AI can compress core academics to two hours while humans move up the stack into motivation and social development. Students complete personalized math, reading, language, and science work in 25-minute Pomodoro blocks, then spend afternoons on entrepreneurship, financial literacy, leadership, teamwork, and communication. MacKenzie Price’s pitch is that children can be “crushing their academics” by lunch without homework or a traditional lecturer.
The model’s reported outcomes are striking, but its evidence still carries a selection-bias discount. Price says Alpha classes rank in the 99th percentile across grades and subjects, except fifth-grade math at the 93rd, and claims students can move from the 10th to the 90th percentile within two years. She concedes selection bias at a private network commonly costing roughly $40,000 annually, while pointing to a $10,000 option and schools rolling out free as ways to broaden the data.
Alpha’s technology thesis is more specific than “give every child a chatbot.” Generative AI builds K–8 Common Core and high-school Advanced Placement lesson plans around each student’s gaps; a vision model measures accuracy, pace, explanation-reading, and guessing; and future lessons could overlay a student’s “knowledge graph” with an “interest graph.” The operational breakthrough is turning assessments from inert grades into instructions for what the learner should study next.
Motivation, not content delivery, is the bottleneck Alpha is trying to own. Price assigns only 10% of a strong learning experience to correct pace and level and 90% to motivation, making guides responsible for personal connection rather than explaining how to “carry the one.” Alpha Bucks let students earn, spend, save, invest, and donate; her defense of the extrinsic reward system is that only “5% of the population” is naturally motivated to learn for its own sake.
D. Graham Burnett argues that AI is already breaking the university’s assessment machinery, even if faculty remain defensive. Systems now emulate human analysis and expression well enough to make outputs indistinguishable from human work, undermining papers and the professoriate’s “police function.” His opportunity case is that universities can abandon students’ “karaoke dance” of producing inferior academic articles and return to shaping people “equal to the conditions of freedom.”
Burnett nevertheless considers institutional decline more likely than a humanities revival. As parents, students, and administrators optimize for tuition ROI and employment, he expects vital humanistic work to migrate into “thousands of new schools” financed outside the traditional university and perhaps offering neither accreditation nor diplomas. Princeton, Harvard, Yale, and Stanford may remain recognizable, but many other institutions must “get dynamic or die.”
Students who want to learn are already assembling adaptive-learning systems more responsive than many courses. One MIT student uses Gemini to identify prerequisite concepts, Perplexity to teach and quiz each exam topic until mastery, and a Gemini-written Google Apps Script to turn Notability backups into summaries and 10-question review quizzes. Casey Newton’s dividing line is motivation: for a student who genuinely wants mastery, AI may be “one of the best things that has ever happened to them.”
The same ubiquity creates a three-sided risk around cheating, inequality, and skill formation. A student admits using summaries to skip books, while Kevin Roose calls AI detectors an “absolute travesty” because false positives can punish original work; paid subscriptions also give wealthier students better answers and more queries. If computer-science degrees become four years of autocomplete, the deeper threat is not merely weak learning but reaching a point where AI no longer needs graduates to initiate the work.

Deep dive

1. Alpha School turns the school day inside out

Price rejects the “weird dystopian robot-led” caricature: Alpha is a full-time school where children arrive around 8:30 and begin with a “limitless launch,” described as “Tony Robbins for kids.” One morning’s impossible task had groups untangle strings around their hands by communicating—a “human noodle project” meant to prime collaboration.
Core academics occupy only two hours. Students rotate through roughly 25-minute Pomodoro blocks in math, reading, language, and science, with breaks between periods; by lunch, formal coursework is over and there is no homework waiting at night.
Afternoons belong to guide-led, project-based workshops in entrepreneurship, financial literacy, leadership, teamwork, communication, and socialization. Everyone reconvenes around 3:15 for “shout-outs,” then leaves time for family, friends, and sports—the broader product is a redesigned childhood schedule, not merely faster coursework.

2. Personalization—not a robot teacher—drives the two-hour claim

Price says Alpha uses neither a robot lecturer nor conventional chatbot instruction. Generative AI creates personalized lesson plans against K–8 Common Core standards and Advanced Placement curricula, first determining “what do they know, what don’t they know,” then assigning material that fills the holes.
A vision model measures engagement through accuracy, problem-solving speed, whether students read or watch explanations, and whether they are guessing. The system then coaches students on how to learn more effectively, making behavioral telemetry part of the personalization loop.
The next layer combines a learner’s “knowledge graph” with an “interest graph.” Price’s sharpest example is a reluctant reader placed, with soccer teammates, inside an Avengers-style adventure calibrated to that child’s exact Lexile level: personalization extends from difficulty into narrative motivation.
Her compression argument starts with classroom waste: 10 minutes settling everyone, one lecture boring students who already understand and losing those who do not, followed by a forced collective advance. Against that 50-minute class, Alpha offers 25 minutes at precisely the student’s level and pace.

3. Motivation becomes the human job and the real bottleneck

Roose’s pushback—worth keeping—is that self-guided software may favor already self-directed children. Price answers that a struggling fifth grader can work at a level that repairs second-, third-, or fourth-grade gaps, while the adult concentrates on getting that student to “lock in.”
Guides are deliberately discouraged from reteaching content. When a child asks for help, they ask whether the student read the explanation, searched the resource library, or found a useful video; the intended end state is a self-driven learner with the life skill of “learning how to learn.”
The human connection can be highly specific: a boy who loved birdwatching earned weekly time in the greenbelt with his guide by meeting academic goals. Guides study what children tell themselves under difficulty and introduce growth-mindset strategies around those individual motivators.
Newton challenges the tension between lifelong intrinsic motivation and Alpha Bucks. Price embraces “cold, hard cash”: children earn currency by completing daily goals, then practice spending, saving, investing, and donating. Her blunt defense is that only 5% are naturally motivated—schools cannot answer the remaining 95% with “tough on them.”

4. Alpha’s strongest claims still carry a selection-bias discount

Price “absolutely” concedes that private-school enrollment selects for parents willing to reject the default school bus, while Roose notes tuition of roughly $40,000 a year. Her counter is expanding price coverage: one school costs as little as $10,000, free schools are beginning to roll out, and future data should become “larger and more rigorous.”
Alpha does not screen transcripts or reject children for being behind, Price says; it looks for coachability and willingness to engage. Her strongest catch-up claim is moving a newly admitted student from the 10th percentile to the 90th within two years.
Price reports results at the 99th percentile “across the board”—all grades and subjects—with fifth-grade math the stated exception at the 93rd percentile. Those are her reported figures, while her own selection-bias concession leaves broader replication as the unresolved test.
Responding to edtech-fad skepticism, Price says technology is “never the answer in and of itself”: pace and level supply about 10% of great learning, motivation 90%. The shift from 2014-era apps to 2022 generative AI made assessments actionable—weak fractions can trigger repair before algebra—and lets afternoons build AI literacy for jobs she says do not even exist yet.

5. Burnett finds an intellectual counterpart, not an essay machine

Burnett’s baseline is categorical despite his warning that the technology moves too quickly for comfortable prediction: AI now emulates human thinking, analysis, and expression with enough sophistication to make its work indistinguishable from human output. Consequently, “a bunch of the ways that we’ve operated to educate students are off the table.”
In a course tracing attention from monasticism to the algorithmic attention economy, he asked students to converse with a chatbot about attention and edit the exchange into four pages. He had high hopes but no precise expectation for what this new assignment would expose.
After 30 years reading student papers, the results felt uncanny: he watched a generation encounter an “alien, familiar, ghost, god, monster child.” Students and machines seemed to test each other in a ring—“What have you got?”—while sensing that they would be partners for years.
One student found more than infinite patience. “I got to be inside my intelligence in a way that felt kind of new to me, because I wasn’t worried about the person I was talking to.” Freed from managing or intimidating a human interlocutor, she could “bring full force” and discover the shape of her own thinking.

6. AI breaks the university’s policing model before its deeper mission

Burnett’s faculty vibe check remains defensive: professors act as “the sheriffs,” worrying that papers are useless, readings will become NotebookLM podcasts, and blue-book exams may return. Humanities academics may not be politically conservative, he says, but their institutional “lag time on change” is substantial.
His own mission is not job training but “working to give shape to persons equal to the conditions of freedom”—people responsible to freedom, other people, and themselves. He finds AI exciting because it strips away policing and makes universities confront whether that deeper formation was ever their real purpose.
The disposable layer is students’ “karaoke dance” of imitating professors with miniature, inferior journal articles. If systems can iteratively generate “a shit ton of interesting articles about Hamlet,” the scarce resource is no longer production but readership: “Does anyone want to read them?”
Burnett’s honest forecast is darker: the more likely outcome is that institutions double down on tuition ROI and employability until the humanities end inside universities. Roose presses the price problem—self-discovery need not cost hundreds of thousands plus decades of debt—and Burnett expects the work he values to relocate elsewhere. His warning: “This is not a test. This is real.”

7. Higher education may unbundle as literacy turns oral

Burnett imagines “thousands of new schools” for exploring how to live, what to do, and how to understand history and tradition. His historical analogy is the rise of secular, vernacular schools after monastic training: AI may similarly break a university model at “a certain kind of endgame,” while Burnett says the humanities exist as fields precisely because they “don’t monetize.”
The urgent pedagogical shift is from textuality toward orality. Burnett says widespread long-form, immersive literacy is ending: people do not read as they did 10 years ago, two years ago, or 30 years ago, and even in elite circles the ability to surrender themselves to extended text is ending.
His response is preservation through transformation: assign short book passages, then dramatize, debate, perform, sing, memorize, cut up, or paste them on dorm-room walls. “People are not gonna get stupider. People are getting smarter,” but a tradition inscribed in books must now travel through different forms; podcasts themselves exemplify the oral turn.
Raised by a college president and a dean, Burnett still urges decent people to defend universities. Yet the American model’s exceptional 75- or 80-year run is “winding down”: elite “motherships” may barely change, while others must “get dynamic or die.” Casey Newton described his combined English-for-life and journalism-for-ROI degree and wondered whether that combination was an artifact of his era.

8. Students are already building their own adaptive-learning stack

Keith, a Princeton computer-science student, uses AI where professors hand-wave proof derivations or equation intuition. Weeks later, rather than reverse-engineering incomplete notes, he asks the model to connect “different moments of clarity” and fill gaps created when his attention or understanding lapsed.
Greta, an MIT junior studying AI and business analytics, uploads problem sets to Gemini but asks only for prerequisite concepts—for example, the relevant graph theory—rather than a solution path. The boundary preserves her chance to approach the actual problem herself.
For an optimization exam, Greta gave Perplexity the syllabus topics and lecture notes, directing it to explain and quiz each topic until it was sure she understood. She also had Gemini write a Google Apps Script that converts Notability backups in Google Drive into lecture summaries and 10-question quizzes, then returns the documents to Drive; “Gemini kind of did it all itself.”
Roose concludes that motivated students may “zoom ahead” and should sometimes teach faculty how to use these tools. Newton sharpens the condition: for someone who wants to learn, AI may be among the best inventions ever; if the desire is absent, the same system cannot supply it.

9. Ubiquity brings corner-cutting, false accusations, and a skills trap

Claire, a Fordham senior, admits the temptation plainly: she sometimes discusses books she never read after consulting ChatGPT summaries and is “not proud of it,” but the shortcut is “way too easy.” More troublingly, a professor suspected her original essay, and a friend was also accused of AI use without having used it.
Roose calls commercial AI-detection tools an “absolute travesty”: they produce large numbers of false positives, a model cannot reliably verify that it generated a passage, and authorship cannot be determined at scale. His answer to corner-cutting is redesigned assignments and evaluation, while noting that SparkNotes and CliffsNotes made avoidance possible long before AI.
Pia, an American applying to German master’s programs, wrote a 10-page essay in English from her own research, then used ChatGPT and her girlfriend to translate it. She calls the help a “godsend” but worries that “cheating’s cheating yourself”; Newton argues that language is a different system of thought, not merely a communication layer an AI earbud can replace.
Vikram, a Michigan CS student, says AI is now “like the internet or like a calculator”: Cursor’s Tab autocomplete is so good, clubs expect AI, and nonuse means falling behind. Roose adds that paid subscriptions give wealthier students better answers and more queries; Newton asks whether four years of autocomplete teaches computer science, while Roose sees a halfway point before systems code alone and shrink the jobs those degrees target.

Mercado Libre: E-commerce Empire - [Business Breakdowns, EP.227](Preview) Google and Apple Avert Disaster, More on Kpop Demon Hunters and Hollywood, Mail on Microsoft, Adobe and ASML