17y/o @AlphaSchool Building the next generation of Ed-tech that will reach 1 billion kids.
Joined August 2023
- Tweets758
- Following202
- Followers6.1K
- Likes777
As an @Alphaschool student for the past 7 years, I’d love to clear up some misconceptions:
“Alpha kids spend hours on screens every day.”
Alpha School: ~2 hours a day learning on screens.
Average public school: ~98 minutes a day on screens.
The numbers are close, so screen time isn’t really a fair criticism of Alpha specifically.
“Replacing teachers with AI is what no parent wants.”
Alpha doesn’t use AI chatbots to teach. In fact, they are banned during academic time!!!
The lessons are designed by learning scientists, and AI mostly works in the background.
Teachers become guides, which gives them more time to build relationships with students and coach each one individually.
Arguably, Alpha is more human than other schools.
TIL Texas Governor-elect @GinaHinojosaTX isn't just against school choice, she’s strongly against @Alphaschool, saying "If we have a generation of kids being taught by AI instead of teachers, we will raise a generation of sociopaths instead of productive citizens."
Source for the public school minute claim: lightspeedsystems.com/news/l…
2
6
1,246
Woke up to this.
Every morning, my @bot team picks one study in my field,
breaks it down to first principles,
tests what would falsify it,
checks it against new research,
and turns it into a video.
github.com/AustinAWay/daily-…
The most valuable thing @Alphaschool taught me is acting with high agency and high urgency.
Sophomore year, I got into a long email war with one of Alpha's academic leads.
Advanced Placement courses, the college-level classes you can take in high school, took us 80 to 120 hours each. I was sure we could do it in 40.
Weeks of back and forth ended with: "You're a kid. I have a Harvard degree."
My guide (@em_findley)'s response: "If you really believe you're right, stop arguing and go prove it."
Two weeks later, a small team and I launched our own AP courses. In about four weeks, students went from the 35th percentile to the 85th, with under 40 hours of work.
Today, I do that academic lead's job, without the Harvard degree (yet). With high agency and high urgency, I didn't need it.
What should school look like if AI can help students learn dramatically faster?
Alpha School principal Joe Liemandt joins a16z's Erik Torenberg and Gagan Biyani of Horowitz Andreessen Academy (HAA) to discuss Alpha's model, where an AI learning engine compresses core academics into two hours a day and students spend the rest on teamwork, entrepreneurship, and other real-world skills.
They explore why high standards make kids happier, how HAA is using motivation and real-world building to redesign college, what parents need to believe before choosing a new kind of school, why Joe expects AI to drive an explosion in entrepreneurship, and why Gagan thinks students should pick what they care about over whichever field seems safest from AI.
0:40 How Alpha reinvented the school day
4:47 Making Alpha believable at scale
6:13 Why kids want to do hard things
8:30 Fifth graders vs. the Wharton MBA test
10:48 Keeping kids in the 80 to 85% zone
13:08 Gagan on redesigning college
15:00 Why Stanford and MIT kids are terrified
19:30 Why college lectures are dying
22:45 Why Ivy League parents want a new path
25:00 The 1910 tractor moment for education
27:02 Why families should take the risk
34:35 Joe's dad on his Stanford dropout
42:10 What to study when AI is everywhere
48:33 Founders School's $1M guarantee
51:53 A 14-year-old cold calls Jensen
YouTube: youtu.be/T4u2FoIm3pw
@jliemandt @gaganbiyani @eriktorenberg
I've been building a team of @bot
They take research papers and turn them into easy-to-understand videos.
1
28
1,456
A new study found that ~26% of teens reported feeling happier after using social media.
The teens self-reported their happiness levels, and the results could be attributed to complete randomness.
$10 budget each.
Hakiu 5.5 vs GPT-6 Luna.
Prompt: "Show off your motion graphics skills and build an educational video from this [script]"
17
1,016
Jev grades essays as well as Opus 5.5, for a tenth of the cost.
I used both models to grade 36 AP (Advanced Placement) essays from College Board.
On average, Jev's score was 0.5 points from the AP graders' score. For Opus, it was 0.6 points.
In this test, Jev compared each essay to essays that had AP scores. For each pair, Jev answered yes or no to the question "Does this essay deserve the higher score?"
Jev grades an essay in approximately 0.3 seconds. Opus takes approximately 15.
Jev costs approximately $0.002 per essay. Opus costs approximately $0.02 per essay.
ElevenLabs v4 is invaluable for the education industry.
Gone are the days when you couldn't effectively teach science and English with just voice models.
Introducing Eleven v4 and Eleven v4 Turbo, our fastest and most emotive voice models yet.
Ranked #1 by Artificial Analysis.
"AI tutoring encompasses lesson planning, text-based conversational tutoring, and practice problem generation." ... "Human tutors taught one-to-one in live video calls."
Could the AI results be summed up to students simply answering more practice questions?
Today we establish that AI is as good as expert human tutors for immediate GRE learning gains (p=.015, n=2,383 students, study conducted July-Sep 2026).
One hour with an expert GRE tutor: $75.
One equivalent hour with an AI tutor: 7 cents .
918x cheaper, equivalent learning (p=.044, n = 140).
In 5 of 7 academic topics, the top AI tutor beat the expert human tutors on average.
How we measured it: pre-test -> 1 hour with (AI or human or no) tutor -> post-test.
There are no LLM judges. Learning gains are post-pre. Everything is real students.
Why we did this:
> 1 trillion USD is being spent per year on machines making machines better. StudentBench shows us how to shift these resources so machines also make humans better.
Why I did this:
I grew up in rural Kentucky. My dad, grandad, and great grandad were all mailmen. Getting into a good college changed my life. Now we have done a rigorous study to help AI labs evaluate LLMs and help show how bright AI can be for students of all incomes.
I believe in good science that leads to the democratization of opportunity.
The research paper behind StudentBench: arxiv.org/abs/2609.28470
Data, code, details: in this thread
If you're wondering what I'm selling, sorry to disappoint: we released all study data for free on Hugging Face, and made studentbench.org free to use. Enjoy.
lots more in the video / thread!!
Jev has accelerated the timeline for personalized education by years.
The biggest bottleneck in building personalized education with AI is validating the quality of the material it generates.
For the courses I’m building, generating lessons is only ~5% of the content cost.
The other ~95% is validation and improvement.
One of the biggest costs is checking whether a lesson gives a student too much information to process at once, overloading their working memory.
Running that analysis across all of our content with frontier models gets incredibly expensive.
At our current scale, even if Astra used half as many tokens as Jev, it would cost us ~$250,000.
With Jev: ~$2,000.
I built the tool we’re using for the working-memory check and made it available here:
github.com/AustinAWay/Workin…
I tried 17,000 times to get AI to write at a first-grade reading level.
It failed every time.
So I fine-tuned a Qwen model that scores near-perfectly on reading-level benchmarks.
huggingface.co/AustinWay/qwe…
“Have AI tutor you” is useless advice.
I tested 255 AI models, reviewing over 20,000 chats across 10 subjects.
AI’s ability to write at the requested grade level varies significantly by subject.
Despite the variation, the models never reliably hit the target reading level.
x.com/Austin_Way/status/2098…
AI tutors have no place in the classroom.
They repeatedly fail to deliver on their promises (as I show below).
But there's still hope for AI-powered education...
💯
One of the most valuable things you can do is test with students. You learn things you'd never think of.
I've seen so many edtech products w/ content two grades above the student's reading level.
It's so hard to realize that's even a problem until you watch students use your product. (One of many examples.)
Genuinely one of my favorite edtech teams right now!
AI is bad at writing educational lesson content.
3 problems:
1. Reading level is inconsistent.
2. AI ignores working memory limits.
3. AI avoids Direct Instruction.
You can tell AI to do all of these things in a prompt, but it still won't do them reliably.
My team and I have been working on a pipeline that checks and fixes all three:
"Just use ChatGPT to learn it" is useless advice.
AI doesn't know what to teach you first.
In a new study, AI tutors had to teach simulated students an entire subject. GPT-5-mini got them to full mastery just 3% of the time.
(btw: picking the next topic at random averaged 28% mastery).
So we mapped all 454 concepts in AP Human Geography and how they fit together.
If used in AI, this likely would result in high mastery.
At 17, I pitched my school's founder to let me build the AP social science edtech. (AP = Advanced Placement.)
That was 2 months ago. Today the practice test results are in.
AP Human Geography (13 students): avg 5.00
AP US History (9 students): avg 5.00
AP World History (9 students): avg 5.00
A 5 is a perfect score. The national average on these exams is a 2.91. Only 14% of students ever score a 5.
In just 6 weeks of using the program, students' practice test scores jumped:
Student 1: 1 → 5
Student 2: 2 → 5
Student 3: 2 → 5
Every student got there through advanced, personalized edtech. ( @AlphaSchoolATX is the future of education )
A student with a textbook will outperform a student with ChatGPT.
Hear me out.
A team at UPenn took about 1,000 high school students and split them into two groups.
One group got ChatGPT to help them with their math practice. The other group got nothing but their textbook and notes.
The ChatGPT group scored 48% higher on homework than the textbook group. By every metric, it looked like the question was already answered.
Then the researchers took ChatGPT away and gave everyone the same exam.
The ChatGPT group scored 17% worse than the students who never had it.
That is insane.
None of this means AI can't work in education. It means a raw chatbot is not a learning tool.
(Peer-reviewed, published in PNAS, 2025.)
