This is an essay told from an interview with Govind Gnanakumar, co-founder of Automorphic. The following has been edited for length and clarity.
I spent my senior year of high school thinking about reading computer science, philosophy, and linguistics in college.
This last year of high school was a pretty exciting time where I had endless leeway to explore my interest in computer science. I spent many of them playing, building and learning in my free time. In fact, I was used to moving very quickly.
I ended up enrolling in Georgia Tech with plans to major in computer science. When I actually got there, though, I was pretty surprised by the slow pace of learning.
Don’t get me wrong – I still think school is a fantastic way of structuring your life, and that it’s valuable.
But when you get used to moving fast, you may start to feel a little suffocated at school. You start to wonder if you’re getting enough price for it. Computer science, in particular, is anything that can be done. learned anywhere. You don’t need to act as a cloistered nun in academia to examine computer science. You can read a manual and build things yourself.
So, I didn’t feel like I was getting much value out of class. I stopped going to classes as often. It all sort of reinforced itself.
At the same time, my friends and I were also building a bunch of things outside of class that were much more interesting.
I ended up applying for winter entry to startup accelerator Y Combinator along with two other Georgia Tech interns: Mahesh Natamai, my randomly assigned roommate, and Maaher Gandhi, another student I met at a concession stand on campus. Other wise people at Georgia Tech, however, not many of them are entrepreneurs, so it’s possible that we met.
In our first request, we proposed a task to index your knowledge in your private applications. It’s a compelling task, but we didn’t articulate it well enough and it wasn’t accepted.
But our moment was successful.
The startup we are building now is called Automorphic and our goal is to help developers iterate and use traditional language models cost-effectively and efficiently.
Currently, other people use massive models like GPT-4 that involve billions of parameters. However, in the future, we believe that other people will need to run more task-specific models, albeit particularly smaller ones.
There are some obstacles. The first is that there is a lot of dark magic wisdom involved in how to exercise and refine the patterns. The average JavaScript developer sometimes doesn’t notice what’s happening on the cutting edge. So we continue to exist by reading the studies and presenting them in an available form. The current challenge here is that there is regularly a long feedback cycle to refine and exercise the models. Therefore, our goal is to reduce this delay from a few weeks or months to a few days.
If I’m looking to get things moving faster and faster as a freshman at Georgia Tech, I’d say I’m in the most productive position possible right now.
Once we got accepted to Y Combinator in May, I had a discussion with my dad. He understood that I was passionate about what I was building with my other cofounders.
He even said “if I were you I’d go build your own startup.”
And honestly, Y Combinator is a good substitute for a degree. In a sense, academia is a focal point for talent, and one of the richest places for talent you can find is Y Combinator.
So, I told him I was going to stop by to take a leave of absence from Georgia Tech. However, I have no plans to return, so that means I’m officially a dropout.
Gonna