applying to grad school p2

my reflections on this cycle of applications

Hey all,

reporting back in the midst of grad school applications, once again for PhDs in AI, and so far this time has been much more succesful (fingers crossed). Compared to last time, I've had interviews with potential advisors, so that is miles better than where we were two years ago.

I am still not the most traditional candidate, I did not have a corpus of published work from undergrad, or go on to do AI research as a career, but I leveraged the position I was in to make the most out of it.

The lessons I learned from last application cycle I think have proven themselves to have helped, so I want to share them in case they might help others.

academic connections

I traditionally hate networking, and thus left undergrad with a pretty small and distant set of connections on whom I could ask for advice. After last cycle's applications, I was desperate for feedback, on some sense of what I could have done better. I did not get that from the advisors I applied to, but that forced me to reach out to PhDs at my workplace in different fields, friends of friends doing programs, and even professors who I only found out about online and were extremely helpful in providing me actionable feedback. A lot of their advice I already read about online, but they each helped develop my ability to talk the talk - communicate what I wanted to study and why I wanted a PhD in a way that resonated with academics. It was like a translation activity, I knew why I wanted to pursue academia, but they helped me develop the vocabulary to translate it into language that would resonate with academics.

research

Everyone will tell you research is the most important thing, and although I didn't conduct rigorous AB-testing this time around, having a paper seemed to help. I was fortunate enough to be able to contribute to a reserach paper through my work, by pinging AI researchers at my company and mentioning I could help out outside of work hours. I found a project that was getting started from the ground up that was a joint effort between Capital One folks and UMD folks (the project became PersonaLedger), and got my foot in the door. By the time I applied, we had submitted PersonaLedger to ICLR 2026 but it is still under review, but having a recent, concrete experience to anchor to really helped. The project also gave me way more confidence in my ability to do AI research, being able to contribute to a piece of it without a PhD, and it also gave me tons of material to talk about in my SOP, while also helping me figure out my reserach interests.

statement of purpose

I started my statement of purpose from scratch, and it was much easier to write this time around. One thing that really helped was my research interests were much more focused, instead of saying "I want to make machines that think like homans" I phrased three research questions:

• How can AI systems self-improve?
• How can researchers ensure self-improving systems remain safe and don't develop biases?
• How can insights from human reasoning-such as hierarchical planning, memory, or counterfactual reasoning-inform new approaches? These questions are the ones that are actually interesting to me, and still very broad (to not exclude researchers who I am actually interested in working with), but they're a little more confined, organized, and use slightly more specific vocabulary which helped a lot. The content of my SOP was all about how my academic and professional experience has honed my interest in these questions in particular, and my desire to pursue these questions through academia. It was less personal and philosophical than my previous SOP, which helped a lot.

letters of rec

I had some great people who I really trust write me letters of reccomendations. Having 2-3 years to really think about pursuing academia gave me more time to build more recent and relevant connections who could vouch for my work and passion.

programs/advisors

I cast a wider net! I applied to 11 schools, focusing on younger, newer labs who would be hungry to grow with more students. I talked to above mentioned academic connections to reccomend labs, I used ChatGPT to propose professors, I utilized CSRankings and went through school faculties to build my list. This is one of the most important, time-consuming, and deliberate parts of the process, so its important to be very mindful of this. In the end, I am really happy with the advisors I chose to apply to, but in retrospect there are a few that I probably would not join their lab. When you are applying, there is an appetite to cast a wide net to maximize your chances of getting in, but it comes at a cost of time, energy, and money.

Last thoughts

Time is a river, and this time applying felt much different than my last time. Now that I am actually seriously entertaining the thought of PhD study, it is a much harder decision to weigh. In 2023 I probably would have sold my soul to be accepted into any program, to get me foot in the door. Nowadays, I am more fulfilled in my current career, I have the opportunities to to do AI research, I have a few more responsiblities and it is a little scarier to take a leap into academia. I still feel a desire to pursue AI research in academia, but my choice is a harder one to make so I have to think about it a lot more and be deliberate with what I choose, which is hard but a good thing. My most overused quote is "anxiety is the dizziness of freedom" by Kierkegaard. I feel that a lot in my day-to-day life, but it is exciting. There are many good doors ahead of me, and I need to trust myself, and walk through one with faith in myself and no looking back.