Understanding Alpha School: Can We 10x Education?

Review of a review of the famous "school for Elon Musk's kids"

Can an AI tutor, used just 2 hours a day, with no human teachers, help a student score within the top 2% of competitive test scores? “Every student, anywhere”? Just check out both, the headline in the screenshot below, and the claim Reid Hoffman (founder of LinkedIn, and an overall smart and sensible guy) is making:

The answer is: of course not! The reality is more complicated, but once you dig past the hype, there are enough interesting insights about the future of education that we can all learn.

Here’s a supposedly impartial review of Alpha School by a parent. It is very long—if I open it as a PDF, it comes to 64 pages! But I would say it is a must-read for anyone interested in the future of education. And for the others, I’ve captured some of the highlights that I found most interesting.

Here are the claims Alpha School makes:

• Students learn for just 2 hours a day

• It is self-paced, and on average, their students go through the material at 2.6x the speed of students in other schools. That is, in one year, Alpha School students cover the same amount of material that requires 2.6 years in normal schools—even though the students are learning only 2 hours per day

• There are no teachers: students use AI tutors for their 2 hours of learning. There are “guides” who provide mentoring and guidance to the students, but all of the actual learning happens via the AI tutor

The first reaction of anyone sensible should be scepticism. And, if you search the internet for criticisms of Alpha school, you will find many pointing out things like:

• These are tall marketing claims which are probably false

• All the “results” they show are simply the results of selection effects: i.e. accept only smart students, and obviously your school’s results will be better than average schools¹.

• This will never scale

• All AI and no teachers sounds like a very bad idea

Here’s an outline of the Alpha school parent’s review:

Since last autumn I’ve collected the sort of on‑the‑ground detail that doesn’t surface in press releases, or is available anywhere online: long chats with founders, curriculum leads, “guides” (not teachers), Brazilian Zoom coaches, sceptical parents, ecstatic parents, and the kids who live inside the Alpha dashboard – including my own. I hope this seven-part review can help share what the program actually is and that this review is more open minded than the critics, but is something that would never get past an Alpha public relations gatekeeper:

1. Starting Point: My Assumptions: how my views on elite private schools, tutoring and acceleration shaped the experiment (and this essay). WHAT is the existing education environment.

2. A Short History of Alpha: from billionaire‑funded microschool to charter aspirations. HOW Alpha came to be.

3. How Alpha Works Part 1: Under the Hood: What does “2‑hour learning” actually look like – what is the product and the science behind the product? HOW is Alpha getting kids to learn faster (Spoiler: “Two hour learning AI learning” closer to three hours, with a 5:1 teacher:student ratio and zero “generative AI”).

4. How Alpha Works Part 2: Incentives & Motivation: The secret sauce that doesn’t get mentioned in the PR copy, but I have discovered is at least as important as the fancy technology. The “other HOW” that no one is talking about.

5. How Alpha is Measured: Effectiveness: The science says it should work, but how do you measure if it is working? How is the vaunted “2.6x” number calculated? WHAT data is Alpha using to make its claims and what does that data actually say?

6. Why this time might be different: Most promising educational initiatives fail to have impact when expanded beyond their initial studies. Bryan Caplan might argue this is because most education education is just signaling anyway (“The Case Against Education”). He also argues that most parental interventions have no impact (“Selfish Reasons to Have More Kids”) – He claims that how kids turn out is a combination of genetics and non-shared environment (randomness; nothing to do with parenting choices). How can we reconcile Caplan’s buttoned-up data with the idea that the “parenting choice” to educate your kids differently (like with Alpha) might result in different outcomes than would be expected from genetics alone? WHY could Alpha work?

7. What Comes Next? The Scaling Problem: The Alpha founders have a vision of completely re-inventing the way the world serves education. But even if Alpha works, it is up against a history of education programs that were never able to scale. It is also going to face resistance for being “weird”. WHAT comes next?

After twelve months I’m persuaded that Alpha is doing something remarkable—but that almost everyone, including Alpha’s own copywriting team, is describing it wrong:

• It isn’t genuine two‑hour learning: most kids start school at 8:30am, start working on the “two-hour platform” sometime between 9am-930am and are occupied with academics until noon-1230pm. They also blend in “surges” from time to time to squeeze in more hours on the platform.

• It isn’t AI in the way we have been thinking about it since the “Attention is all you need” paper. There is no “generative AI” powered by OpenAI, Gemini or Claude in the platform the kids use – it is closer to “turbocharged spreadsheet checklist with a spaced‑repetition algorithm”

• It definitely isn’t teacher‑free: Teachers have been rebranded “guides”, and while their workload is different than a traditional school, they are very important – and both the quantity and quality are much higher than traditional schools.

• The bundle matters: it’s not just the learning platform on its own. A big part of the product’s success is how the school has set up student incentives and the culture they have built to make everything work together

…Yet the core claim survives: Since they started in October my children have been marching through and mastering material roughly three times faster than their age‑matched peers (and their own speed prior to the program). I am NOT convinced that an Alpha-like program would work for every child, but I expect, for roughly 30-70% of children it could radically change how fast they learn, and dramatically change their lives and potential.

The rest of the 64-page review goes into detail for each of these 7 sections. But rest of this article has the parts that I find interesting

Two hours of learning are enough

First, I should point out, as the review does, that to get 2 hours of learning, the children have to be there from 8:30 to 12:30. So with the various overheads, 2 hours becomes 4 hours. But even that is far lower than what most other schools in the world are doing. Is this enough? I think so.

I studied at IIT Bombay. All our classes were from 8:30 am to 12:30 pm, Monday to Friday. On some of the days, we had labs from 2:30 pm to 5:30 pm, but those were never more than 2 days in the week. Other than that, we had some homework assignments and then some studying for exams—but we had lots of free time. Some students used it to do projects with professors or study a lot. Others used it to pursue other hobbies and interests. And some “wasted”² the time in cack sessions in the hostels.

Contrast this with the average engineering colleges in the rest of the country. I know many which keep students fully occupied from 9 am to 6 pm (or something like that). In addition, they have so many homework assignments and “submissions” that they are busy all the time. I think these students are clearly spending far more time on academics than we did at IITB. And yet, I don’t think they learnt more. In fact, I would argue they learn far less because most of the “submissions” are a stupid waste of time.

So, I think there is a lot of merit to the case that 2 hours a day of focused studies is enough³.

The “AI-Tutor” claim is hype

Alpha likes to talk about the “AI-tutor”. And the media loves to hype up that angle. But, the review has this to say:

A typical lesson involves watching a curated YouTube video followed by specifically selected problem sets within the third party tools. After every question the student is given feedback – either “Great Job!” (with the option to click and learn more) or “Incorrect” followed by explanations on why the question was wrong and a mini-remedial lesson to fill in the knowledge gap.

If a pattern of stumbles appears the system will automatically task the student to book a “coaching call” with a remote teacher (most of these teachers seem to be based in Brazil)

[…]

At the end of the day the students get a report on their achievements that day, as well as overall feedback (this is potentially where the “AI” comes in. The computers track both clicks and eye tracking. It can tell both what the student clicked, but also if they were paying attention or distracted, or how much effort they put into reading the feedback when they got questions wrong).

You can see that this is not really AI. It is just self-paced lessons with problem sets. And the problems are repeated until you demonstrate that you have mastered the subject. This is just deliberate practice and spaced repetition—things I’ve been talking about for years (both of those links are to FutureIQ videos).

How is this different from regular schools?

What is impressive about the Alpha program is that is seems to be built from the ground-up around three of the most powerful learning principles:

1. 1:1 learning: It’s not really 1:1, but in practice every kids is working on stuff at the edge of their ability, rather than the median of a group

2. Spaced Repetition: The system brings back topics on a regular basis “just before” the student forgets, so it is more likely to encode in long term memory

3. Mastery: In most education settings school students learn some percentage of the material and then move on. This can work for a while, but eventually students who missed something will struggle because they don’t have the tools to learn the next thing on the ladder. This is particularly noticeable in mathematics, but it can be a problem anywhere. With “mastery learning” the kids are not allowed to move on without mastering the subject. If there is something they miss, the system will bring it back again and again until they have mastered it. Liemandt in particular thinks this is a very big deal, and believes it is a significant driver of Alpha’s success (and the reason many kids are failing in traditional schools)

There is no Gen AI

There doesn’t appear to be any significant amount of Gen AI at Alpha school.

A common belief these days is that Gen AI-based one-on-one tutoring will change the world. I am a fan of Gen AI, and in fact I even teach a course on Gen AI for school teachers. But still, I would be surprised if just using Gen AI-based tutors could propel average students into the 2% nationally. The real world is messy, things get bottlenecked at surprising places, and it takes a long time—decades—for any new technology to start showing widespread impact.

So, it is not a surprise to me that the secret sauce at Alpha is not the AI.

Incentives Matter

My friend Ashish spends way too much time pointing out that incentives matter. “Show me the incentive and I'll show you the outcome,” said the very smart (and very successful) Charlie Munger. And the review has an entire section pointing out that Alpha school has an incentive structure for the students, which they don’t like to talk about, but which is crucial for the success of the school.

Alpha school has an in-house currency called Alpha Bucks. Students earn Alpha Bucks for doing the right things.

If the students complete their 2-hour learning “minimums” each day they earn about 10 GT Bucks. They get additional bonuses for every lesson they complete beyond their minimums. They also get a bonus if they finish their minimums within the scheduled time (vs going home and doing them later), additional bonuses if the entire class completes their minimums during the allotted time, and weekly bonuses for hitting longer term targets.

They only get credit if they both complete their lessons AND get 80% or higher on the problem sets within the lesson. If they get 79% they still move on (with the questions they missed coming back later for review), but they don’t get the GT bucks associated with the lesson (this stops gaming where the kids rush through the lessons just to get “bucks”)

A GT buck is worth 10-cents. So if they are really pushing a kid could be earning roughly $2 per day.

Once again, many of you are probably cringing. Because you probably believe that it is wrong to pay students to study. Because you think that they should have (or it is your job to make them develop) “intrinsic motivation”. Because you think that “extrinsic motivation” (i.e. money) will stop working as soon as the money stops, and of course, the money will stop at some point before the kids reach the real world.

The review points out that all these beliefs are wrong.

Anders Ericsson is a world-leading expert on expertise, and he studied motivation of various experts, for example top musicians.

Ericsson dug further to figure out where the motivation came from and he found it grew over three stages:

1. Parental and authority approval: Initially kids practice because they are given praise and attention from their parents when they do so, and are reprimanded when they don’t. He gives examples of mom saying “if you don’t practice an hour per day on piano I am going to stop paying for your music instructor”.

2. Peer approval: At some point the young musicians begin to care less about what their parents think, and more about their relative status among their peers. Part of this is that they can perform music for their classmates, which is very impressive, but a bigger motivation is that their skills are recognized by other young musicians – their true peers.

3. Self Actualization: Eventually the best musicians stop caring about their peers and start internalizing the desire to be great. They see themselves as musicians, and they do the hard, uncomfortable work of practicing because “that is what a great musician does”.

Ericsson found every musician followed the same path (and he repeated it with other adult experts and came to the same conclusion).

Does bribing kids work? Roland Fryer, a researcher on what works in incentivizing students, in this paper, points out:

providing direct monetary incentives to kids works to drive behavior if that behavior is easy for the kid to understand and execute on. When he paid kids $2 for each book they read, they read a lot more books (+40%). When he paid kids to show up to class and not be late, tardiness dropped 22% versus the control group. But when he tried targeting the end goal and paying students more for higher test scores he saw no effect. Tell a kid to read a book or show up on time and they know what they need to do to get the money. Tell them to get higher scores on tests and, while they have a rough idea how to do that (pay more attention in class, study longer and more efficiently), the actual things they need to do are not entirely clear and the inputs they put in (studying) are not directly tied to the outputs (test scores) – and the incentives have no impact.

I thought this point was important and insightful enough that I summarised it this way:

This tweet got added to my thread on karmanyevaadhikaaraste.

You’ll notice the parallels between this and the Gita’s timeless wisdom: karmanyevaadhikaaraste…

Surround Your Kids With the Right Kind of Kids

Don’t forget step 2 of the “incentives” discussion in the previous section. Parental approval (and bribing) only works when kids are small. Later, they want approval from their peers. At some point, the kids will start doing things that their friends find “cool”. Any parent of a teenager will tell you that after this point, the amount of influence you have on the kids will drop drastically.

So, one of the most important things you can do as a parent at this stage is to ensure that your kid is surrounded by the right kind of kids. You can’t and shouldn’t micromanage who your kids should be friends with—that will backfire, because teenagers have an evolutionary urge to rebel. But you have a little bit of control over this by choosing to stay in the right neighbourhood and choosing the right school.

This is one of the reasons why I always say that when choosing a college and branch after 12th std, pick the highest-ranked college you can get into and ignore the branch. One of the important benefits of this is that it is one way to surround yourself with smarter and more ambitious classmates. This works via step 2 of the incentives discussion to improve your outcomes in life. (This isn’t the only benefit. Go here for a full discussion.)

As Gen AI makes all the knowledge of the world accessible to anyone willing to pay ₹2000 per month, I believe the importance of book knowledge will reduce, the importance of credentials and degrees will reduce, and the value proposition of mediocre colleges will reduce. But I believe the value in going to a top college will remain high.

Bloom’s 2 Sigma or the Power of 1:1 Mentoring

Alpha school’s claim that there are 0 teachers is not true. As the review points out, the guide-to-student ratio is 5:1. Which is far better than the typical 20:1 that is found in even high-quality, expensive private schools. And these guides are paid much higher salaries than even the best teachers can get in other schools. So they are certainly a part of the secret sauce.

The best educational content from some of the best educators in the world is available for free (or very low cost) on the internet. And yet, that has not made any significant difference to educational outcomes around the world. Why?

The most important job of a teacher is to motivate students to learn. YouTube is a superlative medium to get explanations for concepts you’re interested in. But developing the interest in the first place is usually done by good teachers.

A famous (and infamous) result in the field of education is called Bloom’s 2 Sigma Problem. Benjamin Bloom (the guy who came up with Bloom’s Taxonomy) found that providing students with one-on-one tutoring improved their performance by two standard deviations. 2 SDs is a staggeringly huge number—in most research, an improvement of 0.8 SD is considered large.

There were various problems with Bloom’s research. Here’s an important criticism:

"... they weren’t obtained by tutoring alone. Instead, Anania and Burke mixed a potent cocktail of interventions that included tutoring; training and coaching in effective instructional practices; extra time; and frequent testing, feedback, and retesting”

Bloom himself concluded that one-to-one tutoring “too costly for most societies to bear on a large scale”.

You will notice that a lot of what Alpha school does appears to include the “potent cocktail of interventions” mentioned in the criticism above. As of now, even Alpha school remains “too costly for most societies to bear on a large scale”.

But, based on the review, it is possible to tentatively reach the conclusion that Alpha school might have managed to get a Bloom-style improvement but at a lower cost, by going from 1:1 tutoring to 5:1 tutoring with the help of technology.

So What…?

What should you take away from all this?

If you are in the field of education, you should read the whole review, even if you are not directly interested in Alpha school. It is well written and contains a lot of interesting information. My article covers only a small fraction of it. There’s much more there.

As AI disrupts society, there is going to be upheaval in our institutions and our culture and beliefs around education. There’s a high chance that what worked well until now—all the old heuristics and thumb rules that we and our parents used—will stop working and we’ll all need to figure out new ways to survive and thrive. For that, we’ll need to rethink everything from first principles.

And this article contains many of those first principles. Make sure you understand them well.

1. A lot of people argue that IITs in India are not really great colleges, and the only reason IIT students have such an outsized impact on the world is because they get the best students from the country. The claim is that any other college which got a similar selection of students would show similar outcomes.

2. These “waste of time” sessions have turned out to be quite valuable for me in the long term. I’ve talked about this in my video on learning by osmosis.

3. Even when preparing for the JEE Advanced, I studied just 3 hours a day (except in the last few months). But that claim should now be taken with a grain of salt: that was a different time; JEE wasn’t as well-known, the competition was much lower, the Kota-style coaching systems had not yet started. The world is different today, and it is quite possible that I wouldn’t have cleared the JEE Advanced today by doing the same thing.