A lot of people ask me for advice on securing a job post-graduation and doing well in college. To avoid repeating myself, and because I think it’s useful, I want to share this unsolicited with all of you :)

I stand on the shoulders of giants and quote Andrej Karpathy — a Stanford PhD grad and co-founder of OpenAI:

Undergrads tend to have tunnel vision about their classes. They want to get good grades, etc. The crucial fact to realize is that noone will care about your grades, unless they are bad. For example, I always used to say that the smartest student will get 85% in all of his courses. This way, you end up with somewhere around 4.0 score, but you did not over-study, and you did not under-study.

Your time is a precious, limited resource. Get to a point where you don't screw up on a test and then switch your attention to much more important endeavors. What are they?

Getting actual, real-world experience, working on real code base, projects or problems outside of silly course exercises is extremely imporant. Professors/People who know you and can write you a good reference letter saying that you have initiative, passion and drive are extremely important. Are you thinking of applying to jobs? Get a summer internship. Are you thinking of pursuing graduate school? Get research experience!
…

That’s it. I wish I read Andrej’s post in its entirety in college — it is full of good advice. I wish I worried less about getting a B+ or an A-. Especially in the age of AI, grades and courses are fungible and what matters more is your reputation and skill as someone who does things.

Do the thing.

so much of work is ugly and so much of your ability to work is just your ability to eat shit and move on.

there’s a small sample of jobs reserved for the purists. the academics (full professors), the semi-retired CEOs (of successful companies), the film directors (with a final cut!). and even they have had to eat shit at some point in their lives: grinding through grant approvals or a successful exit is no noble work.

more than that: some positions’ whole point hinges on your ability to produce copious amounts of pretty low level menial work: junior investment bankers and consultants famously “move around icons in powerpoint” and have 100-hour workweeks. even in fields where we believe in a higher level of intellectual capacity as a requirement, the work is reduced down to an ugly floor: highly trained judges deal with self-representing criminals that shout in the courtroom, engineers move dots on a screen at 3am to deliver a design in time for a deadline, quantum physicists use scotch tape to get a good layer of graphene after yet another etching failure. is this just friction towards a goal or humiliation?

to restate my hypothesis succinctly: my guess is most careers are not limited by intelligence, taste, or even ambition but by how much sustained shit a person can take without quitting or becoming bitter and letting it poison their sense of self.

maybe this is why “success stories” feel dishonest. we like to tell tales about passion and brilliance because they’re pleasing, but the more accurate story is probably procedural and most value in the world is created not by eureka moments, but by people who can operate at (and even enjoy) the ugly floor long enough to, maybe, sometimes, touch the ceiling.

meta vp: “we now have MZ actively testing <redacted>, so please let’s be careful to maintain our quality bar”

zucc is testing my code👀

“There is, after all, always something wonderful and touchingly beautiful about a young man, for the first time released from the bonds of schooling, making his first ventures toward the infinite horizons of the mind. At this point he has not yet seen any of his illusions dissipated, or doubted either his own capacity for endless dedication or the boundlessness of the world of thought.”
— The Glass Bead Game by Herman Hesse

so many ugrad admissions decisions went out today and now as an alum i realize that if someone sent me back in time and let me choose any place… i’d do it all over again at brown.

this channel is a testament to a beautiful 4 years spent with my best friends. i wouldn’t give it up for anything. sorry for the sap but enjoy a picture of a very lanky 18 year old me for the first time on campus :)

been thinking a lot about horses lately

https://andyljones.com/posts/horses.html

it’s been a little scary to follow AI’s rise in ability to solve math problems. Navier Stokes especially was pretty shocking, and recently we got some more bangers like quasi-RH that i wont pretend to understand but that by all expert accounts is Fields Medal worthy.

but if we look beyond the upending of an entire academic field, i think it’s worth understanding the tools with which these and many others AI miracles have come about.

with math specifically, many of the solutions use Lean, a programming language that lets you write a proof and have a machine verify it's correct. fun fact: at Brown, i was the head TA for the class that taught Lean :)

we’ve also seen huge feats in large-scale software. this would include Bun’s rewrite from Zig to Rust (the famous 1M line PR), or the recently viral “porting PS5 to PC” project. in both cases, the agent has a ground truth to check against, like an existing test suite or the original behavior.

the common thread is that: each of these tasks allows an agent to self-verify its work with a guarantee. i’m quite late to this realization, but many thousands of dollars of token usage later, some of my most successful work has happened because of finding the right loop for the right task.

need to optimize performance? autoresearch! sandbox the workload, give the agent a benchmark to improve, and leave your desk. need to land code safely? build a strong CI pipeline and have the agents go ham. even if verifiability is a fuzzy concept, you can give the agent a goal and a way to self-evaluate. it's amazingly hands free.

and i think this is a great callback, because when i was learning Lean, the proof/solution itself was not always the hard part - it was mostly just being precise enough that a machine can check your work. that's the new regime with agents too: true productivity leverage isn't necessarily the prompt, but the check itself.