Would you consider switching to an easier / more horsepower-dependent rather than effort-dependent major instead of dropping out? You are clearly bright and I suspect that with some good course selection you could get a b+/a- average in mathematics or economics while committing only 15 or so hours a week, which would free up a fair bit of bandwidth compared to CS or engineering. If you think GT's CS education is out-of-date for the AI era, an easier degree (poli sci?) may even be fundamentally more useful for your ambitions.
You are very different from me (and my college experience was some time ago, and at a liberal arts school) and I suspect you will do fine no matter what you choose. But if AI transformation timelines are longer than you currently suspect: having a respectable degree, being able to enter the standard recruiting pipelines as a junior / senior, and so on, are a pretty good insurance policy. Plus - for me at least - college was super fun, and I often wish I could go back; I've sometimes been tempted to do an MBA solely so I can get a little way back to that experience...
Now while agency / drive / ambition levels are harder to judge from a text sample than intelligence, you seem to be pretty aggressive. If you truly think you are truly exceptionally agentic - 4 or more standard deviations above average, well into the 99.99th percentile - then, probably you should just go full Thiel and try to make something of yourself in the real world.
I am absolutely biased by being: 1) less agentic / aggressive than you, 2) a big college enjoyer, 3) not a particularly strong believer in short timelines for AI transformation.
One year ago, I was preparing for my first year of undergrad. Today, I’m considering dropping out. What changed?
Before writing this post, I attribute my decision to variety of (unordered) reasons:
This is a pretty important decision. I’m not super confident in my motivations, so I’m writing this post to hash out exactly why (or why not) dropping out is the right decision to make. I’m posting it because 1) getting feedback from others is the fastest way to test my ideas, and 2) this could be helpful for someone in the same position as me.
Thanks to Zephy Roe, Ishan Khire, and Naren Manikandan for comments, and my sister, Anish Kallu, and Meru Gopalan for relevant discussion.
My perspective on the different levels of dropping out
I’m somewhat against viewing university as a binary choice, i.e. either I enroll or drop out. A more intuitive perspective is measuring my engagement with school as a continuous scale, with dropping out representing 0% engagement and my maximum academic capacity as 100%.
When I think about my academic engagement over the past academic year, my first semester fluctuated around the average student range (65%), starting above and then dropping below. My second semester was noticeably lower (~40%), as I began to prioritize organizing and engaging with AI safety over my coursework. This upcoming semester, I’m much closer to “taking 1 class” (15%), and slightly considering dropping out completely.
Why does this matter? When I compare dropping out to staying in school, I can’t assume my engagement with school will be at 100%. Rather, it's much more likely that staying in school would only mean a 10% level of engagement with my classes, which is qualitatively much different than maximal engagement.*
What am I dropping out for?
When I speak about dropping out, it's unclear what my concrete plan is, apart from working in a generalist capacity in the near future. I’m still working out exactly what shape I want my career to take, but for all of the generalist-adjacent roles I’m considering, dropping out is instrumentally valuable.
Some of my current ideas and plans look like:
Defining the different factors / assumptions at play
I briefly touched on this at the start, but I want to clearly detail what key factors/questions are influencing my decision.
Arguments for dropping out
Here are my current arguments for dropping out, derived from the factors I listed above. The biggest drivers pushing me to drop out are {Short Timelines, Quality of Education, Relative Competence, Bandwidth}.
At any given time, I’ll likely have multiple projects/responsibilities I’m actively leading or contributing to. Throwing school into the mix heavily constrains my time, and adds unnecessary complications (e.g. going to required sections, studying for exams, completing psets) that limit my capacity for high-quality work.
Additionally, the AI safety community largely convenes around Berkeley and DC, meaning being in Atlanta could be a major downside. Thus, dropping out maximizes the time and resources I have to work on AI safety and minimizes external distractions. This first reason feeds into most of my other rationales for dropping out of school.
I would probably get a better CS education by deferring my degree a few years into the future, with advances in education helping me learn more (this is a weak guess, not a strong prediction).
Arguments against dropping out
Now some strong reasons for staying enrolled in school. The biggest drivers pushing me to stay enrolled are {Maturity, Relative Competence, Current Responsibilities}.
This goes beyond just AI safety knowledge too. I’ve only completed my first year of undergrad, without taking highly challenging courses. There’s an argument that AI safety is a really hard problem, and that I haven’t developed intellectually enough to meaningfully contribute in the long-term.[3] College is a great environment for learning the fundamentals and cultivating a deep understanding of anything, and it could be valuable to spend time studying the core questions of the field, rather than rushing into short-term work that may or may not lead to long-term impact.
My broader conflict between upskilling and taking action
I often feel the need to act quickly, do things, be agentic, etc. in order to be impactful - timelines are short, and we don’t have the time to sit around and think about what to do. Rather, we need to act ASAP!!!
On the other hand, I feel the need to upskill - actually understand the arguments, motivations, threat models, and make sure I get what I’m actually working towards. I suspect I have a fuzzy sense of this, but I don’t know if it is too fuzzy or if I am even a good judge of this.
I recognize that these two aren’t mutually exclusive, but there's probably a tradeoff at play where I can’t invest fully into one without losing a major part of the other. In some ways, I think my internal debate on whether or not to drop out of school is a manifestation of this same conflict.
Uncertainty
I’m really uncertain about a lot of the writing in this post, and would really like to hear your thoughts. This is a really important decision for me to make, and I don’t want to take it lightly. Post-writing, I still feel confused about what to do, but I think I have much more clarity on what factors are really at play here. My current plan is to come up with a plan by 2027 - by the end of this semester, I’m confident I’ll know what decision to take.
This is my current belief. I've currently found a lot more personal growth from working on AI safety compared to my actual curriculum. However, the curriculum is question is primarily gen-eds and intro cs courses (DSA, comp org), so take this with a grain of salt.
My belief in short timelines is pretty unprincipled and I want to think long and hard about this before making my decision.
A comment from a friend: "it seems like the hidden point here is that for the most demanding ai safety work rn, deep technical experience isn't necessary and work such as fieldbuilding gives more value.
i feel like this general push towards nontechnical work for the mission of AI safety doesn't necessarily mean you have to give up your own journey of learning CS and going deep into technical material. however, this depends on how much conviction you have towards short timelines and the reality of TAI"