The N=2 Interview about Ph.D.s in Computer Science
August 2015 (perspective of an assistant professor)
In this one-hour interview, Eugene Wu and Keith Winstein, two new assistant professor of computer science, reflect on and give advice about the Ph.D. student experience.
In this hard-hitting, no-holds-barred interview, Keith Winstein and Eugene Wu, two new assistant professors of computer science, reflect on and give advice about the Ph.D. student experience. The format was inspired by Adam Marcus's N=1 Guide to Grad School, except that we've doubled the sample size to N=2!
Keith suggested to conduct this interview as a one-hour live editing session within a shared Google Doc, which ended up working super well. As the interviewer, I could switch back and forth between asking the two of them questions without much pause, and their responses often built upon one another's. After the interview, we each lightly edited the parts that we wrote. I then did some final polishing and “linearizing” to produce this article but tried to keep the original free-flowing spirit of the interview. Enjoy!
Philip Guo: Welcome to The Grind Google Docs Doccast. We're here with two new assistant professors of computer science who want to share some thoughts and advice on the Ph.D. experience. First, why don't each of you give a one-sentence blurb about yourself and your current position ...
Eugene Wu: I'm currently a first year assistant professor at Columbia University in Databases and Data Science. I work at the intersection of Databases and Visualization and Interaction, and studied at MIT under Samuel Madden and Michael Stonebraker from 2008 to 2014. Before graduate school, I did an extended year long internship at Google research, and after graduate school I hung out at Berkeley for half a year.
Keith Winstein: I'm an assistant professor at Stanford, in computer science. I was at MIT for 11 years (1999-2006 and 2011-2014), overlapping with Eugene and Phil (and Adam Marcus who wrote the N=1 guide). In the middle of my Ph.D., I took a break to work as a newspaper reporter and then helped my friends for a year at a startup (Ksplice Inc.).
PG: I want to preface this interview by saying that this is an N=2 slice of Ph.D. life views, and that these guys are on the most accomplished end of Ph.D. students at a top Computer Science (CS) department. So let's see what they have to say.
PG: OK first question, starting generic ... if you went back in time to Year One of your Ph.D., knowing all that you know now, and could do something differently or approach something differently, what would it be?
Wu: It would be to separate and value both life and research. On the life side, I went directly into graduate school from undergrad, and spent several years figuring out what it is I wanted in life. It was a weird time in life because the self-imposed pressure to do “good work” coincides with a quarter life “crisis”, so it helps to acknowledge this and set aside time for your life. Spend time meeting new people, making and hanging out with friends, talking about non-technical things, and going for walks.
On the research side, it was scary because I had followed a relatively well defined track from kindergarten through undergrad, and now had intellectual freedom. The freedom comes at the cost of a huge amount of ambiguity because so many choices are available and it's easy to want everything and end up doing nothing (the Paradox of Choice). I spent a lot of time trying to find golden problems, reading papers, getting scared and discarding them. Instead, I suggest finding projects that could use a hand, and provide free labor for as many existing projects as possible (though probably not concurrently). It's a good way to bootstrap.
KW: I think for me, going straight through from undergrad to the M.Eng. and then into the Ph.D. at MIT turned out to be a mistake. The problem with staying in one place is that you don't really have to prove yourself to people again – you can just keep all your bad habits. I got obsessed with trying to solve the relay channel (a deep and long-outstanding problem in information theory), which I was totally unqualified to work on but seemed, to me, like the only problem in the universe worth solving. Because I figured, like, why be in grad school if not to solve the most important problem?
On the side, I had gotten involved in the student newspaper while an undergrad, and did a summer internship at The New York Sun (a now-defunct right-wing newspaper), and then another summer internship at The Wall Street Journal in 2005. So when they had a full-time job open up, it seemed like a good escape hatch to go work as a reporter for a while, gain some maturity, and figure out what I really wanted to do. I wish I had figured that out a bit earlier instead of feeling guilty for a few years not making progress on hard problems in Shannon theory (which, btw, remain unsolved).
PG: Eugene – go flow with those thoughts, and I can insert in follow-up questions as you and Keith write in parallel ... no need to wait on him.
Wu: When I first started in grad school, I spent a huge amount of time reading papers and feeling depressed – it's easy to convince yourself that every problem has been solved. It's easy to feel this way because 1) you don't know much about anything yet, so every problem seems new and every problem seems solved and 2) research papers are often written in a way that sounds like they solved every problem under the sun and then some. The reality is, most problems have not been solved, and there are more than the number of atoms in the universe (back of the envelope estimate). Basically, don't be paralyzed by what others have done – even the wheel has been reinvented and improved for millennia!
Given this, also don't discard ideas too easily. A professor once told me (paraphrasing): ideas are like seedlings – they need room and time to breathe, and with enough nurturing they'll blossom into trees.
PG: Yep yep, that's super fundamental. You never see how the sausage is made behind those papers, or what dead-ends they encountered, or what subtleties there were. Papers are like people's “Facebook posts” of well-curated life (research) experiences – the final products.
Wu: That's true! From the outside it always seems like there was a deep and fundamental insight that sparked an entire successful project. That's so rare. What actually happens, in my lab, my advisor's lab, and every other successful group I've encountered, is that they pick a simple problem and just start building, and prodding and encounter misstep after misstep. Eventually, they find an issue that existing techniques just can't solve – and they solve it. Then you wrap a story around it.
PG: Yep, we get the benefit of somewhat-fast iteration cycles since we can code. It's harder to do this when, say, you need to train monkeys in a lab and sweep up their poop for months before they can do an experimental task. Life sciences are a harder grind in that respect. Can't iterate as fast.
PG: OK now you guys made me feel inadequate for not doing much at the beginning of grad school. <flashbacks style=“cry: true;”/>
Keith – what were the original tidbits of advice that you wanted to share, which inspired us to get together for this interview? Maybe you can write those down and we can direct the conversations and questions around them.
KW: One thing I found very helpful (and probably picked up from my reporter days) was this idea that you never want to have an important conversation, the first time, with the person where it's important. When I was trying to come back to grad school, I met with a few candidate advisors to see if they would take me on. And I had learned that you never go to your first-choice person at the beginning! First I went to another professor, and I pitched her the idea I had been thinking about working on. And she (constructively) found a lot of problems with it that I hadn't really considered. Then the next week, I had an appointment with the professor who became my advisor (Hari Balakrishnan), and I pitched him the same idea. By that point, when he asked, “Ok, and what about X?” I could say, “I've thought about that a bit, and my response is...” like I had really thought deeply about it. So I think that helped!
I continued to use this technique all through grad school. Your time with your advisor can be sort of precious, so you want to make sure you make the most of the meeting. So when I had a new idea, I would always first pitch it to other people (friends in grad school, roommates, other professors who weren't my advisor) and see what kinds of issues they brought up. So when I went to my advisor with a new idea, I knew I was probably in a relatively good place to make the most of a one-hour meeting and keep up with the conversation.
PG: Wow that's really smart. The Winstein Technique. It's like all the good answers are already in your cache by the time the important conversations come up with key people.
Wu: I like that point. One thing I found effective is to never walk in with an idea and start selling it to my advisor. I always built a demo or prototype of what the end product could enable, or could look like, and start there. Basically sell him on the vision (kind of like a VC). The reality is that advisors are busy people, and would love to shoot the shit and talk about ideas all day, but don't have the time. So it's best to try to be as efficient with your time as possible.
PG: Yeah the implicit burden is always on the junior person to prove their idea's worth, since senior people have already seen half-baked ideas thousands of times so they might be a bit jaded or skeptical of seeing new things. Yep, and advisors are also extremely bandwidth-limited. You'll both see this in the coming year as you hear more student pitches :)
PG: Also, this approach is similar to giving the same talk multiple times and fielding similar questions, so that your answers end up coming out smooth and thoughtful after a while. The first time I give a talk, I always stumble when answering questions.
KW: I think that's a good point. The people who look like they are a natural at giving talks – they're probably not. They just practice a ton, and you're probably not seeing their first time and maybe not even their tenth time. I am very lazy about these things, but when I was on the job market, the faculty on the 9th floor at CSAIL basically forced me to give like four or five practice talks before my first interview, and they asked hard questions. I also signed myself up to give my job talk several times at an industry conference the week before I had my first interview. And that forced me to prepare! I had arranged to have Stanford as my very last job talk of the interview season, because it was where I wanted to go the most, and by the time I gave my job talk at Stanford, I had given it 15 or 16 times in the prior six weeks. And still, when I started the talk, I was super nervous! Literally, I am not making this up, my knees were knocking in front of everybody. (That hadn't happened to me since I auditioned for an a cappella group at MIT as an undergrad...) But fortunately I was practiced enough that the words could still come out of my mouth. Anyway, the point is, people who seem like naturals at giving talks probably aren't – they just practice a lot, with and without audiences.
PG: Keith, do you watch your own talks on video? Or listen to audio of it? I do that sometimes but should do it more.
KW: I do. I think it helps. The other thing I do for a conference, if it's a 20-minute talk, is I just do it 8 or 10 times in a row in the hotel room the night before, over and over, to see if it flows. That gives you a chance to adjust the placement of things on the slides if you're stumbling over something. The other advice I got is to absolutely get the $60 Logitech slide advancer that has the timer and vibrator built-in – it will give you a five-minute warning, etc. If you practice with it, you can see the timer and which slide you're on and you'll know whether you need to speed up or slow down a little to keep on schedule.
PG: Oh yeah, I like the rapid-fire rehearsals for short talks. But it's so hard to practice a 45-minute talk repeatedly. Oh wow, I didn't know there was a vibrating reminder thing ... that's super helpful since you don't need to glance at your watch like George Bush during the Clinton debates in 1992 and look unengaged.
Wu: There are some people I know that completely disappear from a conference until the day of their talk in order to practice practice and practice. It makes sense and is worth putting in the time, since that is the only 20 minute span of time where everyone (in the audience) is focused on you and your research.
PG: YES, talks are immensely powerful since you're on stage. And people come up to you afterward to ask questions and banter about your talk topic. I've always had serendipitous things happen after good talks that I give, often interacting with unexpected people whom I didn't know would be there. A talk focuses attention on your work without you needing to schmooze or appear like a salesperson in the conference venue hallway or whatever.
Wu: A wise man once told me that at some level, you need to be a little selfish in graduate school. The metric is how well you picked and executed and shared an idea which is very hard. And it's very easy to help other people's work – it's intellectually easy and feels less risky. Ultimately, you end up without anything to call completely your own, and he's seen that with many very smart students that have a lot of potential.
This was a bit surprising to hear because almost all of my good work has been the result of collaborations, and other colleagues have similarly extolled the virtues of collaboration. The way I reconcile this, is that you want to do both. A good model is to collaborate far and wide, with as many smart people as possible. You'll learn a lot, make many friends, and naturally position yourself in the community. Try to make sure many of them are busier than you – because they are busy you'll naturally have to take control of the grunt work and will end up owning the core pieces. At the same time your collaborators could contribute in meaningful ways (e.g., provide valuable direction, feedback, writing) while focusing on their primary directions.
PG: Word, collaborations helped me early in grad school in terms of learning the ropes and building confidence, and getting some papers out so my CV didn't look empty. I wasn't first-author on those papers, but I didn't need to be anyways since I was young.
PG: Regarding collaboration, though, what about if you're a mid-stage or senior student and working 50/50 split with someone else? That can be challenging in terms of credit for your dissertation, no? And then on papers, on the job market, etc.? I've always liked the 80/20 split or something similar where one person is clearly senior or has ownership on their project.
Wu: If it's 50/50, you'll need to have a sober conversation – there's no getting around that. On that note, it's rarely a win-lose situation. If a system was really 50/50, then what you built is just the ladder for picking the fruit. Take large systems projects such as PostgreSQL, or Telegraph (in the database community), it's always a team of students laboring for a few years on a giant system, and once it's built then everyone scatters off to reap the rewards of the interesting problems that the system entails. Keep in mind Sayre's Law: “In any dispute the intensity of feeling is inversely proportional to the value of the issues at stake.”
PG: True, true, good computer systems projects aren't solo affairs. I always think of research projects in terms of “how many grad students 'big' is this project?” (this is relevant for planning grants as faculty). In some CS sub-fields, it might be 5-ish (systems), in others it might be 1 or 2 (theory, HCI), in others (giant autonomous car robot droids) it might be 10. So how to split credit and effort depends on sub-field.
KW: In the right light, I think “be a little selfish” might be good advice for grad school in general. Often the hardest part of grad school is knowing what you want so you can ask for it. It seems common that you might be unhappy, or your friends in grad school might be unhappy, and they don't even know why. And sometimes it turns out to be something simple, like, I want a new laptop, or I want a one-hour-a-week meeting with my advisor that is on the calendar and we do it every week, or I want to go to the X conference, or I want to spend a summer collaborating with Professor Y at the University of Z. And usually if you figure out what you want, if you ask your advisor for it (at the kind of rarified top-20 programs we're talking about at least), the answer will often be yes. But it's that figuring out what you want part that's hard.
Wu: On KW's note – the hardest part of starting was to understand myself well enough to know what I needed. It's super easy to feel frustrated for a huge number of reasons – feeling like other people aren't contributing enough, like your advisor is not helping you enough, etc. Unfortunately, (or fortunately), no one owes anyone anything and it's up to you to understand what's missing that you want, clearly articulate it, and look for a resolution. For example, if a meeting was frustrating because your advisor didn't seem excited about your work, maybe that's because validation is important to you. When you realize that, you can seek it out elsewhere.
I highly recommend taking non CS courses. I especially like business school courses – think about it: the job of a business person is to form, lead and work with teams, so they're the best to study! I particularly like negotiation courses because the curriculum is to learn about what you want (your interests) and what the other party wants (their interests), and how to make everyone happy. Basically how to look for win-win situations.
I also recommend seeing a therapist. I think it's often taken in a negative light because it's perceived that you have “some problem” and thus need to see a therapist. That's ridonculous! If you wanted to improve your running form, you would see a sports therapist. If you want to better understand how you think and approach the world, why wouldn't you see a mind therapist?
KW: The other part of “be selfish” might be that it's ok to be a little annoying. Advisors are often interrupt-driven and they might let emails fall off the stack. If you're frustrated that your advisor won't get back to you, you don't necessarily have to seethe about it. You can also camp outside their office while they're in a meeting with somebody, and as soon as it's over and the person leaves, just walk in and grab them. This is ok – your education is the priority and your advisor is there to help you, even if they are a bit frazzled.
Wu: A corollary to KW's point is that most people are on your side, and if not they are not against you. It's perfectly normal to feel scared that someone will destroy your work, or ridicule you, because you know the flaws better than anyone else. Just keep in mind that people are just too busy to spend the time on this because it takes away time from what they're focused on: building new ideas.
PG: Yes, we're also lucky that we're in a creative production-driven research field, and not a narrowing discovery-driven field where, say, ten research groups around the world are competing to scoop one another to discover X fact about Y physical process while running similar experiments to converge on one “truth.” For us, everyone is building up separate by related systems and getting inspired by one another's complementary ideas.
KW: I tried to get my students to read “The Ph.D. Grind” and they don't want to do it. They just say it seems really long and depressing. But I think it's helpful to read other people's experiences and see how bad it can get – and that the final outcome is obviously not bad in the end, even if there are slumps in the middle, which are very common. Almost everybody goes through a sort of mid-grad-school crisis. I left for four years in the middle of my Ph.D. I don't think it's a bad idea to understand that this is a common part of the experience. (Recognizing that “The Ph.D. Grind” represents a portrait of what, I think, roughly a 30th-percentile-happiness Ph.D. experience looks like. Not sure if Phil would agree with that...)
PG: Maybe a 30th percentile at Stanford ... but people at non-top-ranked schools often have it worse due to scarcity issues – less funding, less surrounding resources, etc. My situation was also kind of self-imposed since I was on five-year fellowship and decided to go rogue rather than sticking to more established paths. I could've aligned with a more traditional research group and maybe had a smoother time.
KW: That's quite fair. Obviously there's a lot of benefit to being at a top-20 program.
Wu: I've noticed that graduate students that came back from e.g., the Army or “real jobs” tend to do really well in graduate school. The PhD takes way too much grinding and motivation to do on a whim. The small sample of older graduate students I know (Jonathan Perry, Keith W, Tyson Condie, Peter Alvaro) all seem similar in that they've decided to put their drive and focus towards doing good research. A lot of other life questions that I had (should I do this? would industry be better? Maybe I should go farm) don't seem to distract them. Basically, they seemed mature.
KW: I think this is an important point. You want to get a Ph.D. as an act of agency. When I got laid off from The Wall Street Journal, I thought of going back right away, but “I decided to come back to grad school because I lost my job” is not a good answer to the question of why you're getting a Ph.D. I first went to the startup instead because I wanted my decision to do the Ph.D. to be something I did on purpose.
PG: OK we have 10 minutes left ... lots of good material so far. Any other final words?
Wu: I said earlier that most people are not assholes. That also means that you should try not to be an asshole yourself – try to see the positive in any piece of work. Every paper you read, write down at least two things that were interesting about it, even if it was not a top-tier paper. If you can't appreciate the interesting aspects of other people's work, you won't be able to pick up on the promising directions in your own research career (thanks Dr. James Cowling for this idea).
KW: Right, I think “Yes, and...” (Wu: everyone go read Getting to Yes) is always the way to respond to a colleague's nascent work. Research is kind of like improv comedy in that way. Your job is not to poke holes in your grad-school colleagues' work; it's to help find and fill the holes. The program committee is the shared adversary; you all want to armor your work enough to satisfy their objections. :-)
Wu: Marketing: There is somewhat of a distaste in the academic community about “marketing” one's work. I think there's a fine line, and the approach I take is to be as truthful and clear as possible about your work. That goes in two ways: it doesn't mean you should oversell and say your work is better or more extensive than it is – that is certainly distasteful. But if you obscure, or hide or don't accentuate the goodness of your work, it's also taking something away from the world and is untruthful in the opposite direction. Ultimately, there's a reason you worked on a problem and it was hopefully exciting. It's your responsibility nay, your duty, to convey that feeling to the rest of the world!
PG: Yep, positioning yourself on the same side of the (metaphorical) table as your colleagues is key when having discussions rather than sitting across from them in an adversarial stance. I'd even extend this further to real physical tables ... I try not to sit across from someone if I can avoid it. Being at a 90-degree angle, sitting on the same side of the table, or even taking a walk, makes it feel much more like a constructive conversation.
PG: Three more minutes left ... so let's do rapid-fire mode. How about each of you just hits up some bullet points of tidbits of grad school or life advice wisdom. Bullet point style. Lightning round. Jeopardy music.
PG: Wow you guys have gotten me all nostalgic for old times :)
KW: Thanks for doing this!
Wu: Thanks! That was a lot of slobber.
PG: I expect my paycheck in the mail.