2025-12-25
the education system is a scam.
not because the knowledge is bad, but because the delivery mechanism is dogshit. you sit in a lecture, someone talks at you for an hour, you forget 90% within a week. that’s not learning, that’s theater.
the entire system is optimized for the wrong thing. universities optimize for credentials. duolingo optimizes for engagement metrics. anki optimizes for card count. none of them optimize for actually making you smarter.
the 2-sigma problem
bloom showed in 1984 that 1-on-1 tutoring produces 2 standard deviations of improvement. read that again. that’s taking an average student and making them better than 98% of peers. not through magic, not through genetics - just through having a tutor who pays attention to you.
we’ve known this for 40 years and done nothing. why? “it doesn’t scale.” you can’t give every student a personal tutor. the economics don’t work.
except now they do.
llms scale. infinitely patient, available 24/7, remembers everything. bloom’s problem was that tutoring doesn’t scale. llms solve this. every user gets a tutor that adapts to them specifically. the 2-sigma solution isn’t a fantasy anymore - it’s an engineering problem.
invisible infrastructure
here’s where everyone else gets it wrong: they make the system visible. “time to review!” “you have 50 cards due!” “start your practice session!”
fuck that. learning shouldn’t feel like work.
the user never thinks about “saving” knowledge or “scheduling” reviews. they just talk. the system silently extracts what you understand, builds a knowledge graph, tracks decay, weaves review into conversation. you don’t even notice it happening.
the magic is invisible. the best infrastructure disappears.
the graph is grown, not designed
no predefined curriculum. no topic trees designed by some committee. the knowledge graph emerges from what you actually explore.
most platforms hand you a roadmap and say “follow this.” that’s not how learning works. your path through a subject is unique to you - your background, your interests, what clicks and what doesn’t.
your map is unique to your path through the territory. it shouldn’t tell you what to learn next. it should make sure you actually understand what you’re learning now.
review is not a mode
anki users know the drill. open the app, grind through cards, close the app, forget it exists until tomorrow. review is a separate activity. a chore.
no “practice sessions.” no anki-style grinding. review is woven into the texture of every conversation. the tutor brings things up when relevant, challenges your understanding, connects new ideas to old ones.
like a study partner who actually remembers everything you’ve ever discussed. “hey, this relates to that thing from last week - how do you think they connect?”
that’s how real learning works. not isolated flashcards. connections.
generative, not recognition
duolingo fails because selecting from 4 options is nothing like producing language. anki fails because recognizing a fact is nothing like applying knowledge.
recognition is easy. generation is hard. that’s why tests feel different from flashcards.
the llm generates novel problems that require synthesis, not pattern matching. it doesn’t ask “what’s the formula for kinetic energy?” it asks “a car is moving at 30 m/s and weighs 1500kg. it needs to stop in 50 meters. what’s the minimum braking force required?”
you can’t bullshit your way through that. you either understand it or you don’t.
the endgame
optimize for actual learning. for building genuine understanding that persists. for solving bloom’s 2-sigma problem at scale.
if something actually makes people smarter, makes them retain knowledge, makes them solve novel problems - the credentials become worthless. the degree becomes a cope for people who didn’t actually learn.
you are responsible for your own education. the tools just got a lot better.
i’m building something to solve this. if you want to try it out, reach out.