"Every PhD Is Different"
Five Maps for Finding Your PhD
“Every PhD is different.”
I heard this phrase a handful of times when I started my PhD.
Like most platitudes, I immediately dismissed it. It seemed like a way to make people who were having a crappy PhD not give up. “Maybe success is just around the corner! Keep trying! Remember famous person X? They failed miserably at the beginning of their PhD, and then got it together with stunning discovery Y. You could too!”
But unlike the hundred other aphorisms thrown at you when you start the journey of becoming Dr. Last Name, this one stuck with me for some reason. It has come to mind several times in the last nearly six years.
I think I’ve been trying to figure out what it really means.
Which, of course, is hopeless. There is no single interpretation to “every PhD is different,” no more than there’s one way to understand other vital but vague prescriptions like “be nice” or eat healthy.”
The value, I think, lies in discovering the meaning for yourself. The quote is like scaffolding, upon which you can construct your own advice.
In other words, I now think of “every PhD is different” as a prompt.
Finding Your PhD
If we turn “every PhD is different” into a prompt, we might ask, “what is your PhD?” or, “how can you find your PhD?”
At least in the PhD program I’m in, classes are seen as an entertaining distraction—kind of a departmental insurance policy against accidentally graduating a complete buffoon. So this question then amounts to, “how can you do your best research?” This is great progress already because we’re getting more concrete.
So, what does “your best research” look like? How can you discover it, and hopefully find your PhD as well along the way?
I’d like to present a handful of conceptual maps you can use to think about yourself as a researcher. These won’t be exhaustive, but they’ll give you the flavor of some concrete ways I looked for my own PhD.
I will sketch five areas:
- your advisor
- types of research
- comparing yourself to others
- balancing improving with being effective
These are geared towards PhD students, but I think if you abstract a little bit, they apply more broadly.
I have found balancing collaborators on a project both surprisingly difficult and surprisingly important.
If I was working entirely alone on a project, several problems arose. The most obvious were strategic issues. I would get tunnel vision into the way I’d framed a problem, what aspects of a project were actually important, or—most dangerously—whether the whole line of inquiry I was pursuing was even interesting at all.
Less clear were mundane problems that could sprout into big issues given enough time. The prime culprit here is dwindling motivation. A PhD can already be so lonely, you don’t need to make it worse by toiling by yourself. One pattern that I learned to be vigilant for is delaying or cancelling project meetings because I hadn’t gotten enough done. This causes more pressure to deliver even more results the next meeting, which had a funny way of sandbagging the whole process even further.
Having another collaborator can be an immense help. In fact, I think given the right intellectual and personality chemistry, a two-person research team can be monstrously effective and fun. I would look out for this, and embrace it if you find one.
But even for two people for whom this might work, lots of little factors—time, interests, expertise—influence its fate. If it’s almost working, you might find some of the downsides of going solo totally alleviated, but some amplified. If two people have tunnel vision about a mediocre idea, a positive feedback loop can self-sustain the inquiry way past its reasonable bedtime. Talking to others—and listening carefully—seems helpful in these situations, but of course there’s no silver bullet.01
On the other side of the spectrum, I found it paralyzing to have too many collaborators. At best, it feels like working in public, where I fret over each thing I share. This yields higher quality artifacts when I present preliminary findings, but at the cost of slowing everything down. You also run a greater risk of bikeshedding, which is when everyone spends a great deal of time arguing about a trivial detail. At this point, I just abandon slack channels that have too many collaborators in them if I’m the one driving the project.
For my own work, a balance of one to three engaged collaborators seems like the sweet spot. But the truth is, even this depends on the phase of a project. Here’s a deeper dive for how I like to have collaborations on my own projects:
Early: One to three collaborators. Like a tiny trusted committee, they help sculpt the research direction into something fruitful. I often still prefer individual meetings. One highly involved collaborator (of the three) stays up-to-date on all of the ideas and decisions. They’re like a copilot.
Mid: One collaborator. It might be the same copilot. They help hammer out the details. Some of my best working times in grad school are at this stage. Most of the scaffolding was in place, and in order to nail down what we were doing, a close collaborator and I would regularly spend four-hour sessions at a whiteboard figuring it all out, sometimes for several days in a row.
The Grind: Solo. At this point, there’s a big chunk of work, you’re author no. 1, and you need to get it done. You’re still sharing progress and getting feedback from all collaborators, but maybe only weekly or monthly.
Late: Everyone. Suddenly, everything needs to be done at once. Stuff like model ablations, statistical analysis, human evaluations, making figures, and writing. The work is branched and you now balance two main roles: contributor and coordinator. Fresh off the grind, this stage can feel wonderful, because suddenly so much is getting done by so many people who all give a damn.
All of this is just me. Now that I’m cranky and old (for a PhD student), and because my advisor is flexible, I can just work this way—I don’t feel like I need permission. But as an early grad student, nobody described to me how collaborations might work, and so I didn’t know, for example, that you’re allowed to have contributions ebb and flow like this. Mileage may vary, and your lab culture certainly shapes all this.
Summary: Discover what shape and structure of collaboration styles work best for you. Be open to this being messier than you might have expected.
The PhD student/advisor relationship is fascinating. As a PhD student, but not an advisor, I have only a narrow window into this. But since I’m not a professor, I actually have time to drivel on in a blog, so you get to hear my perspective.
Your advisor is likely going through career changes of their own during your PhD. So rather than you each learning to work with a single person, you and your advisor will both learn to work with the people you are each continuously becoming. As an example, when I started my PhD, I had never done any work in my new field, and my advisor had just a few students. We would sometimes meet for multiple hours multiple times a week, and she would teach me the details of algorithms or linguistic concepts. Now, she runs two enormous labs—one at our university, and one at a company—and her other obligations (giving invited talks, being on committees, raising funds, and so on) have grown accordingly. I’m grateful I got to have a period of close collaboration early on in my PhD, and, fortunately, I no longer need quite so much hand holding. But it’s vital I don’t still expect the same kind of interactions now as six years ago.
While the adage, “learn how to work with your advisor” captures this, I don’t think it sets the expectation quite right. It sounds like you do that once and you’re done. Rather, I think learning how to work with your advisor—and anyone else—is a process that you never finish. Another benefit to conceiving of it this way is that you allow others to change when they want to.
So, how do you do this? Try to figure out which things your advisor is best at helping you with. Here’s an incomplete list: assessing a project plan, resolving technical questions, finding related work, introducing you to someone, planning funding, improving your writing, or helping with a personal difficulty. It’s helpful to consider other variables operating on their end: their current available time, interests, expertise, and stress level. Some of these change often, some slowly, and some never.
I think the best way of going about this is to make mistakes. Try a lot of stuff out. If you try to classify your advisor too early, you will miss out on potential fruits of your relationship. There have been countless times when my advisor was more technical, creative, empathetic, or strategic in a meeting than I had expected. Probe the dimensions of your working relationship. You will almost certainly get more out of it.
Summary: Keep learning how to work with your advisor as you both change. Ask for help in many areas.
3. Type of Research
I suspect in most research areas, there are different unofficial “genres” of research that people do. This is usually kind of implicit, but you get a feel for it. Here are a handful of the ones that appear in my field:
technical: “we tried a new neural network architecture,” or, “we used a lot of data”
non-technical: “we did some human studies or evaluation”
math-y: “here’s some weird distributions and maybe some proofs”
opinionated (often requires famous coauthors): “I believe something”
dataset-y: “here’s some new problem nobody cares about yet, I hope someone will now”
Sometimes people self-select for what kinds of research they can do based on their background. For example, if someone is non-technical and resists becoming familiar with technical things, they’ll be stuck doing non-technical things their whole PhD. Or, if someone is math-y and finds that quality important, they’ll resist working on problems they don’t find well-enough defined. And so on. I think it’s worth being aware of this early on so you can decide what styles of research you might want to try before you unconsciously pigeonhole yourself.
Summary: Be vigilant to discover which genres of research are out there in your field, and which you naturally gravitate towards. You get to decide how to balance becoming an expert in one style, with the risk of underdeveloping another style and having a blind spot.
4. Comparing Yourself to Others
This double-edged sword gets a bad rap.
Even just reading this section header, it feels like my mind has been trained to scream, “don’t do it,” and, “impostor syndrome.” It’s difficult for us to even conceive of a culture and world view where we don’t take every opportunity to measure the self, judge its value, and wrap it all up into a big painful package of misery. It’s tough to see actually trying to compare oneself to someone else as anything but stirring up the hornet’s nest of your own mind.
Which… fair enough. It might simply be too risky.
But let me offer one idea: watching someone else can completely change your idea of what is possible.
One early memory I have of this was in one of my weekly piano lessons. I think I was in high school. My teacher was dutifully tolerating another painful attempt at the piece I’d been working on for three weeks, waiting to give the universal advice I never seemed able to internalize (“try playing it slower”). I don’t remember how it came up, but for the first time, I asked something about how long it could take me to learn this song. What he said blew my mind.
“Probably a couple days.”
What. WHAT?! This was, like, ten times faster than I had ever imagined. Did I believe it? My mind racing through the myriad practice tips and techniques he’d taught me over the last five years, most of which I didn’t apply faithfully or patiently (or with enough sleep), I realized, yes, that completely makes sense.02
In the PhD, I had a few similar eye-opening experiences. Here’s one: some time in my third year, I learned about someone implementing a machine learning model in a single day.
One day. ONE DAY?! This was, like, ten times times faster than I had ever imagined. Did I believe it? My mind racing through all of the myriad ways I spent my time during the day, and the countless engineering hurdles that I never considered working around, I realized, yes, that completely makes sense.
You can immediately recognize the trap here: learning something like this, and then ladling unreasonable expectations onto yourself—which, of course, you won’t meet at first.
But the benefit is opening your eyes to the full playing field. Seeing completely different ways of thinking about what you’re doing.
For me, this made me rethink my whole process. Yes, I understood things better if I implemented them myself. Yes, I would like to balance research with my classwork, and maybe leave work just a teensy bit early to go see that movie. Yes, there was no real time limit to finish this project, and I could take the opportunity to read a few auxiliary papers related to the subject. But this kind of death by a thousand cuts lull can creep in and slow everything down, where suddenly what could take one day takes ten or thirty. The whole ineffable but magical feeling of momentum and energy dries up. And then this pace gets normalized. Out of the many twists and pivots that a project usually takes before it reveals the golden interesting nugget inside it, you may only have time for the first couple. Then the deadline comes around, you write up something okay but not great, and you fight the reviewing cycle for months and months…
In other words, while there’s value in trying to balance all aspects of work and all aspects of life every day, there’s also value in focusing on one thing and making a lot of progress.03
These two examples were both about speed. But I’ve found the same thing with engineering tricks, routines, presentation quality—you name it.
I think that in other jobs, when you’re working closely with more experienced folks, you naturally get exposed to more of this. If you’re a designer in a design shop, an apprentice carpenter, or an engineer at a software company, you’ll probably enjoy hands-on exposure to others’ work. In my PhD, most people code alone, and it’s pretty rare to discuss your working process or speed. So, getting these insights is more difficult, but no less instructive.
Summary: Consider getting a close look at someone else’s workflow. It may be eye opening.
5. Improvement vs. Effectiveness
Stepping back a level, there’s an aspect of balance to all of the axes so far. As you discover what your perceived weaknesses are, you usually have a choice: do I work on this issue, or do I find ways around it such that I can work more effectively? In AI we traditionally call this “explore vs. exploit,” but let’s go with the more boring “improvement vs. effectiveness.”
Let’s look at the idea of “every PhD is different,” through the lens of “improvement vs. effectiveness,” and peer backwards into the conditions that people bring to the PhD. Why is every PhD different?
People have different sets of strengths and weaknesses that are revealed during a PhD. In other words, a PhD reveals only a subset of your (huge set of) strengths and weaknesses as a human being. I would venture that this is why PhD work is easy for some people—it just fits their current strengths and avoids their weaknesses. This probably sounds too trivial to be worthwhile, but I find it great at cutting unhelpful comparisons (like impostor syndrome) when they arise.
People have different rates of discovering their own strengths and weaknesses. Depending on the whole research cocktail—let’s say: field, advisor, research problem, collaborators, environment—someone may simply hit the ground running, or hit the ground and stay there (i.e., faceplant). I would venture this is why people will stall or accelerate when switching up an ingredient in their mixture, such as finishing their first project, or changing advisors. Suddenly, they’ve realized issues they must address, or that suddenly evaporated.
People have different inclinations to address vs. work around their weaknesses. In other words, people have different default preferences for where they land on the “improvement vs. effectiveness” spectrum. (I think it’s common to admire people who fall elsewhere on the scale.)
Here are some examples of this playing out in my own PhD:
Improvement, since abandoned: “ML Bootcamp.” When I started my PhD, we all took a machine learning (ML) class that was statistics-heavy. After we finished the class, I felt like most of the foundational material was still beyond me. As a freshly motivated first-year PhD student, I organized an ML “bootcamp,” where we read chapters from a statistics textbook and met weekly to do problems and discuss. In hindsight, I don’t think I’ve needed to know basically anything I learned from that time.04 But it was an itch I felt like I had to scratch. And to my past self’s credit, had I not tortured myself computing a zillion expected values, I might have always had a more crippling case of math envy, worried those skills were locking away juicy research contributions. So, I’m both glad I did it, and glad I gave it up by the time it outlived its effectiveness.
Improvement, have kept it up: Figuring out time management. Managing your time when you have tons of open hours and only long term goals is a fiendishly difficult task. I have spent literally years working on this. I think I have a long term life goal to learn to live a loosely structured, self-directed working life. In this sense, the PhD has been a great training ground. I have not figured it out yet, but six years in the trenches has helped.
Ignored, now exploiting: How I work best with collaborators. I ignored figuring this one out for a long time. Now that I realize how important it is, I’m purely attempting to figure out how I can collaborate best (exploit) rather than take on any kind of self-improvement journey.
Summary: Because the PhD is so long, and such a formative time of your career, you have many opportunities to balance improving yourself with doing work effectively. Almost no one talks about this, and most of your pressure will be to work effectively. Be aware of where you tend to balance on this scale.
You must find your own PhD, because you come to the PhD with your own unique set of skills, troubles, and inclinations. While your conditions may be similar to those around you, your own situation is unique. Like so many things, you’ll find yourself balancing others’ advice with listening to your own instincts. Some axes to be aware of are:
- your collaboration process,
- the breadth and development of your advisor relationship,
- your styles of research,
- how much you understand others’ processes,
- and how you balance improving with doing.
As always seems to happen, this whole conceptual framework is only now becoming clear in hindsight. It’s so easy to get sucked into the trap of believing you can both maximally improve and optimally perform in all aspects of work, rather than viewing your time as finite and the balance explicit. I hope that some sliver of these considerations finds a helpful home in one of your brains.
Most of all, if you’re in your PhD, I wish you luck finding it.
Thanks to Ari Holtzman, Julian Michael, and Maarten Sap for reading drafts of this.
A thriving advisor/advisee collaboration can avoid this pitfall, since the advisor probably has a finely tuned research barometer, but such relationships are rare given how busy advisors are these days. ↩︎
As an aside, I think the delivery here was crucial. My teacher didn’t shout at me, or say something like, “my other students could have this done in two days.” In fact, he didn’t even say it until I asked. This was probably the least psychologically scarring way of having my mind opened to it. The flip side is, since it was so many years before it occurred to me to ask this question, I had plenty of time to ingrain low expectations into myself. ↩︎
A great topic I’m not covering at all is, “how many projects should you work on at once?” Fortunately, the answer is simple: try out a few numbers (make sure to try “just one”) and see what you like. ↩︎
Of course, my advisor dutifully mentioned this at the time, and I dutifully ignored her. ↩︎