Ask PhD students what matters most in a PhD and you’ll hear all kinds of answers: the school, the topic, the advisor’s character… But almost every answer has one thing in common: the advisor.

Before my PhD — three years of research as an undergrad, plus a gap year after that — I had worked with more than a dozen professors. I met every type and learned plenty of lessons the hard way; I’d seen nearly every kind of advisor, character-wise, that people run into when choosing a PhD advisor. There was the kind who knew nothing and was short-sighted, yet was an expert at exploiting people with every trick in the book. There was the kind who wanted me to target ICLR, three months away, when I had barely started on a topic. There was the kind who deliberately cut my pay. And there was also the kind whose goal was to train students, who looked out for their students at every turn, and who gave me plenty of freedom and respect. So, having seen what a good advisor looks like before my PhD, I wasn’t going to lower my standards on an advisor’s character for anything else (like the research topic). In fact, in every collaboration since, I have never again lowered my standards on “character” for anything. In my three years as a PhD student, very few people or situations have required me to swallow my pride and go against my own nature. I’ve crossed paths with the wrong people now and then, but I’ve almost never met anyone truly bad — anyone who left me with real psychological scars. (For reference, look up the Keyu Tian “poisoning” incident.) We shouldn’t view people in academia through rose-tinted glasses. If you start a PhD wearing them, you may go through a very painful stretch of taking them off.

When it comes to advisors, Cornell CS has a pretty good culture. I rarely heard of faculty with a terrible reputation, or ones who felt awful to work with — so much so that I seemed to have almost too many options to try. So for the first year and a half of my PhD, I bounced back and forth between three advisors: all three of them were great, and I was in an exploring phase, so I tried a lot of things.

Throughout my first year, I never quite figured out what I wanted, so I was doing theory with Sarah on one side and NLP with Claire, my current advisor, on the other.

Sarah was the first person who really taught me how to write a paper.

I’d had a few papers before my PhD, but no one had ever taught me the way she did. Back then I barely used GPT, so every sentence in my papers — including the clumsy, convoluted ones — was typed out by me, word by word.

Sarah would leave lots and lots of comments: what she thought the problem was, how she thought it should be changed, and a more detailed revision I could use as a reference. This part is really my personal preference. If she had only listed a pile of issues without telling me how to fix them, I might not have known what to change; if she had just rewritten things herself — even though that would have saved her a lot of time — I might not have understood why she chose to change them that way. Except for the appendix, she read every corner of the paper carefully, from start to finish, and revised it many times over.

That’s also one of the reasons I didn’t want to leave the theory community back then: your collaborators and advisors don’t just put their names on your paper — they genuinely polish the work with you, even if no one outside a handful of people in your subfield will ever care about it. They’re willing to spend time training a student. The days I spent revising papers with Sarah are among the few happy memories of my PhD. Later, during rebuttals, we discussed every single one together: I’d write a draft, and she’d comment and revise. We did have small conflicts now and then, but in the days before a deadline, writing the paper together, I’d once again feel that Sarah was the best advisor in the world.

In the beginning, I honestly didn’t know how to work with Claire. I didn’t yet have the ability to do research on my own. What I needed was an advisor who would guide me step by step. I was also extremely insecure about my research — I often felt that what I was doing was garbage — and I needed to draw energy from someone else (my advisor, for example). I needed my advisor to care about my research more than I did. (My advisor before my PhD, Preslav, was exactly that kind of person.) On top of that, I still leaned toward continuing with theory, but the kind of theory Sarah did was quite different from what I’d worked on as an undergrad; I wanted to do theory closer to RL/ML. I was trying hard to find a direction of moderate difficulty (i.e., one where I could actually follow the proofs), and anyone who did theory back in my day might agree: RL theory was a pretty easy place to crank out papers. So I went to Wen. He was already gradually moving from RL theory to LLMs, and he told me plainly that he was no longer interested in theory. I can’t remember what I did after that, but somehow it ended with him and Sarah co-advising me on an RL theory project, and I formally said goodbye to Claire.

Wen agreed to co-advise on one condition: we would all meet together and work on a joint project. But — maybe I was being too sensitive — once I’d forced the two of them together, the vibe in our meetings felt a little strange. They both seemed to have plenty of ideas and things they could do. If they hadn’t cared about each other’s feelings, each could have steered the project toward their own research direction, but out of respect for each other, neither did. Eventually, the project was bound to lean toward one person’s interests, leaving the other without any real sense of ownership. I had never been co-advised before, let alone in an arrangement I had forced together myself. I quickly realized the problem, so, awkwardly, I went back to Claire and asked whether she would take me back in — co-advising me with Wen. (Yes: less than two weeks later, I went crawling back to Claire with my tail between my legs.)

And so began the Wen-and-Claire co-advising chapter. Later on, because of some things that happened, my mental state became extremely bad and I wanted to drop out. Claire supported every decision I made. During that period I probably came to almost every meeting with a “new decision” — that I wanted to quit, and later, that I wanted to try to graduate in three years — and every time I brought her one of these “surprises,” she never once brushed it off as something I’d said in an unstable moment. Claire walked with me through the hardest stretch of my second year. Later still, my research skills improved, and I gradually became able to do research independently. I no longer needed a hands-on advisor to feed me ideas or tell me what to do at every step. At that point, having Claire as my sole advisor seemed to fit better than co-advising (there were other reasons too, of course), so I told Wen I no longer needed to be co-advised, and Wen agreed. A year later, in that great wave of assistant professors leaving academia, Wen left for industry. Half a year after that, he suddenly came to mind one day, so I set up a meeting with him just to catch up — and heard Wen speak Chinese for the first time. Somehow his English seemed better than his Chinese. He seemed very happy in industry.

Over these years, I’ve heard plenty of horror stories from other schools: departments with several faculty members notorious for their bad character; departments where the Director of Graduate Studies (DGS), instead of protecting students, goes out of their way to find fault with them; and places where PhD students compete fiercely with one another to “grab” advisors.

But Cornell CS does well on all of these fronts. Most of the faculty are good people and very supportive, and so is the DGS.

More than half of the people in our department have switched advisors, or gone through a period without one. Whenever that happens, everyone goes to the DGS, and the DGS always helps figure something out. As someone independent of the advisors, the DGS always takes the student’s side and gives students as much support as possible.

So in my two years at Cornell, I hardly heard of anyone quitting. In other departments, though, and at other schools of the same caliber, I’ve heard of plenty of students being driven out by their department or advisor over these years.

Admittedly, not many people in the department have tried this many advisor combinations in such a short time, and I’m very grateful to my advisors for their patience and understanding. My story with Claire deserves a post of its own. I actually met her by chance at the end of 2022, at EMNLP, on the bus to the conference banquet. I wasn’t doing NLP yet and didn’t know who she was; she just looked familiar, so I started chatting with her. Luckily, Claire was fairly low-key: she wasn’t surrounded by a crowd, and apart from her former students, almost no one stopped her to talk. As an outsider who had finally found someone to talk to, I stuck with her the whole way, until she left the banquet.

Back then, I had no idea that two months later I’d get an offer from Cornell — or that this woman with short blonde hair and a kind face, as calm as a summer sky and as warm as spring sunshine, would end up becoming my PhD advisor, walking with me through the whole long, painful PhD grind and watching me grow into who I am now.

I know that many advisors are nothing like mine (Claire, Sarah, and Wen). Some are hypocrites. Some take putting other people down as their only joy in life. Some will do anything to get what they want. Some show one face to their peers and to people more senior than them, and quite another to their students…

Even with a great advisor, a PhD hurts in all sorts of ways. Many people reach a point where they want to quit, and then, because of certain people or certain things, they hang on and see it through to the end. I think writing these things down might help some people see what a good advisor looks like, so that they stop treating some advisors’ behavior as normal. As independent individuals, it isn’t our place to change how other people behave, but at the very least, we can choose who we spend our time with.