While I am happy that this worked out for you, I would caution others against trying to get into PhD programs on a last-minute basis after receiving the GRFP. I had a NSF GRFP fellowship (in "chemistry of life processes" not computer science) and this was my experience:
I was in a rather unusual but broadly similar situation, and ended up needing to enroll in a PhD program to avoid losing my GRFP while I did a 1-year Master's program in England through the Churchill scholarship. None of the PhD programs I wanted would let me do this, so what I ended up having to do is formally enroll in the PhD program at my undergrad school, and then transfer. I made the deadline by about 3 days and was very stressed out.
This is good advice. It is very helpful to 1) have a home school that will let you stick around as a backup and 2) have a sufficiently general research topic and flexibility to apply to make many schools be good fits.
Beautiful. Now if only someone would top you and get to work as an ML researcher at a top lab without a PhD. Bonus points if they have no degree at all. The last person I recall doing something like that was Freeman Dyson.
Does the industry route count? Many people in private labs (Google, OpenAI, etc.) got there the private route without school. That was actually going to be my backup plan. In fact, one of the first authors on GPT-3 is now at Columbia working on her PhD somewhere down the hall. She's in year 3, so do the math... Now I know the true meaning of imposter syndrome.
There are industry places that will, at least as stated, take you seriously with no PhD as long as you have some publications (many job postings don't require a PhD or say "or equivalent research experience"), and it's unusual but not unheard of for people do this.
The thing is, a PhD program is a reliable way to build a research track record. And you don't see too many PhD dropouts who want to be scientists because if you've got a research track record, the extra cost of just finishing your dissertation and graduating is pretty low.
My unclear interpretation of the rules was that when you apply for GRFP you must state an "intended school". It felt disingenuous to state a school I didn't apply to, ergo you should apply to a school. But I have no reason to believe it is verified in any way shape or form.
Fellowship is All You Need: Getting into Grad School by Hacking the NSF GRFP
Last year I started having delusions of grad school. By delusions I mean that on paper, in the numeric and text fields that actually make it into the application, I looked utterly unqualified to do machine learning research, let alone at a top school. I had no computer science publications, no masters degree, and a 3.3 GPA, in materials engineering. I was a self-taught washed up startup founder six years out of school with no advisor and three months work experience in my proposed field of NLP. Only two people had ever seen me write code; one was my current boss and the other was my ex-cofounder, who also happened to be my brother, so I couldn’t even scrape together a single letter of recommendation attesting to my programming talents. I did take a programming class once, CSE142: Intro to Computer Programming I, where I got a C. But next month I’ll be starting my PhD studies at Columbia advised by Zhou Yu and Luis Gravano. They’re really really good at what they do and there’s no school I’d rather be at. This is the story of how I leveraged the National Science Foundation Graduate Research Fellowship Program (NSF GRFP) to get my unqualified ass into grad school. The technical guide follows.
Table of Contents
The Hack
In June of 2021 I forgot to register my +1 at a wedding and got placed next to a cool dude named Logan who told me just to shoot my shot on PhD applications even though I didn’t feel qualified. After all, they’re kind of a crap shoot anyway. Lacking any proper mentorship I took this advice at face value and spent the rest of the summer reading papers and the autumn drinking cappuccinos and banging out applications at Amy’s Merkato and Fort St. George after work. At the end of the year I fired off my freshly minted applications and swapped caffeine for alcohol while the responses rolled in.
I was rejected from every school. Don’t get me wrong, I knew my application was weak, but that doesn’t make it suck less. All seven schools turned me down without even waitlisting me. There’s not much more to say on this except that plan A was kaput.
Plan B, however, was still cooking. Promptness and agility are not virtues commonly ascribed to the Federal Government, so despite grad school fellowship deadlines (GRFP and NDSEG) closing before those of colleges, results are released afterward by several months. The GRFP is a three year fellowship for first or second year PhD students in STEM fields. Unlike most grants, the GRFP funds a person rather than a project. This fact carries several implications, the first of which being that you can sort of win a GRFP without a school or an advisor, in in mid-April I did just that.
The GRFP is considered rather prestigious because it’s a bit difficult to get (~16% acceptance rate) and because it’s dripping with the world’s best lubricant: money. It’s not a master key to the ivory tower but it is a big shiny ⭐ NSF Sticker of Approval ⭐ that can get a busy prof to reply to a cold email four months past the official application deadline. The reason I had a shot at the GRFP but not at the traditional application process was due to the second implication of the GRFP’s person > project philosophy: doing cool stuff matters a lot, and I did a lot of cool stuff instead of going to class. I invented some polio eradication tech deployed to six countries, founded three companies, and led a community safety initiative that maybe saved a few lives and definitely put out some literal fires. If you’re clever about framing then you can make all of this relevant to why the government should give you money to teach computers how to learn.
Even though you can win a GRFP without being at a school, you still must declare a school to accept the fellowship. Furthermore, there is no possibility to defer and they only give you three weeks. Apparently the NSF never considered the eventuality that someone could be talented enough to get a GFRP but not talented enough to get into any school at all. Cue cold emailing.
I emailed every NLP professor at a top 25 American school working on anything remotely related to my proposal, 45 in total. I also owe a massive favor to a mentor and old friend from high school marching band who had the credentials to cold email department heads explaining my unique situation and to my little sister Mia who I bribed with boba to help me manually scrape those email addresses. Within 48 hours I went from summary rejection to nine interviews scheduled, four them being the very next day. I also received a few responses chastising me for trying to circumvent the official application process, stating “that’s not how it works”. Au contraire.
Doing back-to-back zoom interviews for two days is a curious experience. In a way the dynamic is flipped versus traditional visit days. Everyone knows I’ll be getting in somewhere, so advisors are competing with each other to attract me, and there are no other candidates against which I can be benchmarked. After I explained my situation and research proposal they would usually give me their take on the ideas and ways it would fit with their expertise and existing projects, almost like a pitch. One prof in her first year seemed downright nervous and even fumbled softball questions like “what is your lab culture like?” Although nobody could guarantee me admission without submitting a formal application I got the impression from everyone that rejection was unlikely. Two weeks and many emails later I accepted an offer from Columbia, exactly ten years after my undergraduate application was rejected.
What follows is a meticulous breakdown of why the GRFP is the way it is and how to exploit that.
Technical Guide
Abstract
The National Science Foundation Graduate Research Fellowship Proposal is a three year fellowship open to US citizens who have not yet reached year 3 of PhD. The GRFP carries enough weight to near single-handedly get otherwise unqualified candidates admitted to certain top schools. Therefore with its relatively high acceptance rate (~16% vs <2%), a single GRFP application may have higher expected value both in terms of percent chance admission to a choice school and marginal increase in percent chance admission per hour of effort as compared to conventional application strategies like optimizing for quality or quantity of school applications. In order to maximize the chance of GRFP success, non-traditional candidates should lean into their unique life experiences in order to differentiate themselves from other applicants. Here I present a case study on a successful GRFP submission and a theoretical basis for its success grounded in the stated objectives of the GRFP. In general, the GRFP-First algorithm is:
Background
The Normal Route
Grad school application strategy has been discussed extensively elsewhere. In brief, the most successful candidates have as much as possible of the following, roughly in order:
If you don’t have these, your best option seems to be slowly building up connections to labs through informal channels (cold emailing, volunteering) over a long period of time.
Unfortunately, none of the above can be remedied on a short (< 1 year) timescale, or even at all for those trying to break into a new field and lacking a time machine. Last minute options are restricted to either writing a small number of very targeted and well researched applications (I tried this and failed) or sending out as many applications as possible.
Unfortunately, top computer science schools these days have an acceptance rate under 2%. This means a random candidate applying to all ten top-ten schools would have less than a 1−(1−0.02)10=18.3% chance of admission anywhere.
However the math is actually much worse for median candidates (as opposed to uniformly sampled random candidates). A median candidate will actually, by definition, have a 0.0% of being in the top 2% of any given department applicant pool. And on paper, otherwise qualified non-traditional applicants can appear well below average. Clearly for this kind of candidate must use an alternative strategy to avoid mediocrity at all costs.
GRFP is All You Need
Getting a GRFP is like being turned away from seven different nightclubs for how you’re dressed then turning up two hours later with four hot women in tow and being let in no cover.[e] Nothing about you has changed but somehow the old rules no longer apply. This is how the GRFP works, except the bouncer is your department’s admissions committee and somehow you have a 16% chance of finding four attractive ladies who want to go to the club with you.[f]
My point is the GRFP has a 16% acceptance rate and top schools have a 2% acceptance rate and the GRFP is stacked in your favor if you happened to spend your weekends building polio surveillance equipment or whatever instead of studying.[g] Most people consider GRFP supplemental, but I argue that the expected value of a strong application is so high that it can actually be ones primary strategy for grad school admissions. If relying on the GRFP, one can reduce school applications from typically 8-15 (from what I’ve heard) down to as low as 1, since the number of school applications is independent of GRFP success. This way the applicant can spend more time on the GRFP and waste less time on school applications that will be sent straight to the shredder.
What Does the NSF Want?
To successfully practice medicine, one must understand the human body. To win a GRFP one must understand the NSF and why the NSF has different priorities than potential advisors and university admissions boards. Fortunately the NSF makes its mission quite clear:
Advisors, on the other hand, need to publish or GTFO. So while profs are looking for candidates who will produce maximum h-index for minimum risk and supervision (read: conventionally gifted candidates), the NSF is looking for candidates who will make America kick ass at science. The NSF also funds about 1000x as many candidates (2750/year) as a single advisor, so they can place numerous high-risk, high-reward bets without unacceptably high risk. The official solicitation never mentions how many papers awardees publish, but it does mention the “numerous individuals… honored as Nobel laureates”. Much like Y Combinator, the NSF is in the business of farming unicorns.
The NSF is quite clear on this. They’re not looking for past success, they’re looking for potential.
and later
Sure, high impact papers and a good GPA suggest potential for significant research achievements. But so does grit. So does getting your hands dirty. So does applying for project funding, getting rejected, then building the company and the products from scratch on your kitchen table anyway.[d]
Methods
How to Win a GRFP
Take Risks
Okay, let’s say I’ve convinced you to go all in on the GRFP. How should you actually write the application? The main approach is to not be mediocre because humdrum applications have 0% chance of winning. Instead, take gambles. Be dramatic. Stand up for something. Go big or go home. If you commit hard to whatever it is you believe in, you increase the chance of falling flat on your face, but also the chance of striking a chord with the reviewer, except the expected value of being the worst candidate and the expected value of being a median candidate are both approximately 0 because neither will win. Therefore it is far wiser to roll the dice, say, committing an entire paragraph to how the joy and beauty of propagating Ocimum basilicum on your windowsill inspires you to research nanostructured catalysts for organic transformations or whatever than it is to write colorless platitudes like everyone else.
The Hero’s Journey
This strategy works best when your entire proposal follows a coherent narrative. Try to think of your proposal like a story. Not in the chronological sense (that’s boring!) but in terms of the Joseph Campbell’s Hero’s Journey. Storytelling is perhaps the oldest of all human art forms and carries tremendous ethos. By retelling the academic journey that led you to apply for the GRFP in the framework of the Hero’s Journey, you can guarantee that your story follows the arc of self-discovery present in legends and myths across time and space. Your narrative will represent you as a meaningful and authentic human who has grown over time into the young but mature researcher you are today. Try to address most of the Hero’s Journey, even just in a single sentence, roughly in order. The idea is to explain how you went from a naive freshman who doesn’t know their purpose in life to young but wise researcher who through gallant struggle learned the ways of science, now whole-heartedly prepared for the next great adventure, the PhD.
The Hero’s Journey is composed of twelve steps:
Not all elements are strictly necessary or need be explicit. Life can be stranger than fiction, so trust your gut, but if you’re stuck then return to the standard formula here.
Editing
I cannot stress this enough: get as many people to edit your application as possible. Don’t be afraid to cold email grad student researchers on papers that influenced your proposal. Chat them up a bit first asking, then see if they have time to give your proposal a read over.
Timeline
Once you submit GRFP, immediately begin trimming it down to 2 pages to submit to schools as your statement of purpose. Then you get a few months to chill and wait for results. If you get into a school, congratulations, you’re in! If you get the GRFP, double congratulations, you have funding! Either way, this guide ends here. But if you manage to win the GRFP but get rejected from schools as I did, here’s the playbook.
What to Do Once You’ve Won
Have you seen the Shining? Cuz this is where things get hacky.
First, the NSF will ask you if you want to accept the GRFP. Accept it. Then they will ask you what school you’re going to. This is a problem. You don’t have one, and you can’t defer the GRFP unless you’re in the hospital or in the military and I don’t recommend either of those. You must find a school and you have until the end of April before the NSF yoinks your hard earned fellowship. I hope you practiced cold emailing during the Editing section.
First, email every single potential advisor in the United States. I emailed 45. More is better; some profs are surprisingly openminded about what constitutes relation to their expertise. One professor at Northwestern pitched me on some pretty obscure social science applications. Fear not rejection; it will only only hold you back.
Make your email clear and to the point, but sell yourself pretty hard. Here’s my actual email. Feel free to plagiarize.
The calendly link is important. It massively reduces the activation energy to scheduling a meeting with you.
After that, compose a similar letter to department heads at every school you might be interested in. If you have a mentor with any sort of credentials, have them send the letter on your behalf. All in all I received positive responses from 8 professors.
Interviews should start rolling in. Don’t screw them up and profs will probably invite you to submit formal applications. It this point it should be hard to screw up. Congratulations, you’ve made it.
Results & Conclusion
It’s messy, but sometimes there’s only one play for the hand you’re dealt. If you lack the academic pedigree to get into grad school the normal way, this might be your best bet.
Besides being your golden ticket in though, the GRFP provides some notable perks.
Firstly, you basically get to choose your school from the list of those that responded to you. Even if you don’t love these options, since the funding is attached to you and not your PI, you can transfer schools much more easily than others.
The funding also gives you the freedom and security to pursue your own interest and work on a longer time scale than if you were on one of your PI’s projects.
Finally, the GRFP application is largely a superset of the typical school statement of purpose, so finishing by the early deadline gives you a distant head start on those.
Footnotes
[b] Assuming applicant aptitude is uniformly or normally distributed and therefore mean and median can be used interchangeably for large n
[c] https://www.nsf.gov/pubs/2022/nsf22614/nsf22614.htm
[d] Steve Jobs’ parents garage would have been a heck of an upgrade
[e] This is a thought experiment mixed with a metaphor, not my lifestyle. I have no idea how clubs work.
[f] Is there some sort of Nash’s law where all illustrations of expected value eventually devolve into male dating strategy?
[g] True story
Appendix 1: Useful Links
https://uh.edu/~lsong5/documents/A sample proposal with comment.pdf
US NSF - : Broader Impacts Review Criterion -- Dear Colleague Letter nsf07046 https://mitcommlab.mit.edu/broad/commkit/nsf-research-proposal/#:~:text= NSF GRFP Research Proposal 1 Criteria,is part of an application that... More https://www.katenuss.com/advice/us_fellowships/
https://oge.mit.edu/fellowships/fellowship-tips/tips-for-specific-fellowships/
https://www.alexhunterlang.com/nsf-fellowship