Reflections on My Time at Primer
From April to June 2022, I worked at Primer to create a math program that teaches math as play.

For those of you who know me, you know my focus is on improving education – for example relating to the work I did with my previous project the Feynman Mafia.
I thought working at Primer was a great opportunity because we shared a similar philosophy of taking kids seriously.
And Primer is exactly the type of company I would want to start myself, except it already has the backing of many reputable investors like Founders Fund and Khosla Ventures.
More specifically about how I joined Primer. Earlier in the year, a friend of mine told me about the Primer Fellowship, an opportunity to work with Primer to kickstart their dream school.
I applied, and talked with Ryan and Maksim. However I didn’t hear back right away.
A month later, I recorded one of my video updates that I send out every quarter, and decided to add Ryan and Maksim to the mailing list. A few days later, I heard back from the team that I was being invited to move forward in the process!
My Experiments at Primer
Over the next 3 months I taught kids on a daily basis, refining and experimenting with my interactions to figure out what did and did not work. I now have a much deeper understanding of the problem. This is what I learned.
By default, kids don’t really want to do math. Most kids I interacted with started off neutral about it. My goal was to engage them and show them how to like it.
What does it mean to play with math? Most people think it means embedding math into a game. For example Baugarten is trying to do exactly this.
I disagree. I think math itself is an intrinsically enjoyable activity to do. The key is to focus on what gives pleasure to the mathematician: the prolonged contemplation of a tantalizing mystery. This is what I mean by play.
Note that the focus is on reaching an affective state, not a cognitive achievement. This is very important. Most people think that math is about understanding complicated concepts, but this is inaccurate.
So what did I try? The first thing I did is “Do Things That Don’t Scale”, and directly engage kids and contemplate math mysteries together. Kids reacted very positively and enjoyed it.
Over the next months, I tried different ways of removing myself from the interaction, to see kids would still enjoy playing without constantly being engaged.
I tried having the kids read directly from “You Are A Mathematician”. I tried having kids solve problems directly from sheets. I tried a game where they create problems for me. All of these would scale, but didn’t work.
The failure point was common: they sometimes gave up and got distracted. Fundamentally the problem is that kids needed to be engaged while they contemplated problems, which makes any approach difficult to scale.
What does engagement actually mean? In practice, it means that whenever a student gets stuck, I ask them to explain what they’re thinking, and give them questions and suggestions that will lead them towards solving the problem themselves.
Can this be automated? I don’t know. But that is the next direction I’m excited to explore.
Are there any other directions I can go in?
I can continue to personally engage students. This would be successful, but it wouldn’t be scalable.
I can focus on students that already have a predisposition to math, and accelerate them by showing them how to play with math.
I can create a curriculum that marginally improves on what already exists.
These are all great ideas, but they don’t really interest me because they’re all being explored.
What’s Next?
I’m still very interested in figuring out the cargo cult education problem Paul Graham describes in his essay “The Lesson To Unlearn”.
The project I’m working on now, Digital Socrates, is looking at creating better tests that actually test if you deeply understand a concept, using large language models.