Should I do a postdoc?

By Niloofar Mireshghallah
With the help of Sasha Rush

What you won't learn here: How to apply to postdocs or how to prepare for and do well in interviews. Basically, this post is not about how to land a postdoc.

What you will learn: What types of postdocs there are out there, and advice on how to decide if a postdoc is right for you. You will also learn about my experience, take it with a grain of salt!

Target audience: Anyone who is confused about what to do after getting a PhD, specifically in CS and more specifically, in AI/ML/NLP.

What you need to know about me: I finished my PhD in ~5yrs (2018-2023) at UC San Diego, I started my PhD in computer architecture/privacy and switched to NLP at the end of my second year (I was in small labs), and I have been a postdoc at UW NLP for almost 1.5 yrs now, where I am co-advised by Yejin Choi and Yulia Tsvetkov, so I'm in relatively big labs now.

My personal biases and point of view that you should take into account when reading this: I have personally always wanted to do academia and become faculty, and I have limited industry experience (I have done 4 internships at different teams in Microsoft Research and collaborated with different industry labs, but I have never held a 9-5 job or been a full-time employee). However, I am not biased in the sense that I would think everyone should become an academic and it's the one true path. In fact, I usually encourage people to do industry. I think if you are not super sure about academia, you probably shouldn't do it. It takes a lot of patience, dedication and passion, and the pay gap (between academia and industry) is increasing rapidly, so unless you absolutely love what you are doing, it's not worth it.

What is a postdoc? What types of postdocs are out there?

When people ask me what a postdoc is or what it's like being a postdoc, I often start my response with: it's like having a situationship, as a job. It's temporary, and there really is no hard set agenda or commitment to it. You are usually hired as a researcher, and the duration is 1-2 years (some places extend to a third). It's whatever you want it to be, depending on the type of postdoc you are doing:

  1. Academic postdocs: Doing a postdoc at an academic institution, you usually apply by reaching out to the lab PI, however, there are fellowship programs like CRA or Simons Institute, where you are not directly assigned to a PI.
  2. Industry postdocs: Doing a postdoc in an industry research lab. These are rather limited and more scoped. As far as I know Amazon, Meta and Microsoft Research have such opportunities.

Usually in academic postdocs you get more freedom (in terms of where you work from, when you work and what you work on) and don't really have much responsibility, however you are paid almost half of what an industry postdoc would be paid (this could be higher or lower, but on average half is a good guess, as most academic postdocs range 50k-85k based on location, and industry postdocs are 120k-165k). Industry postdocs tend to be relatively loosely defined as well, however there is a higher level agenda and goals you need to adhere to.

Another difference is access: to students, collaborators and diversity in people and fields you get exposed to. I think the highlight of my postdoc was all the amazing people I got to meet, and the broad range of collaborations I had: being at UW NLP, apart from the UW NLP folks, I had access to folks at AI2, and also the security and policy groups at UW, from whom I learned so much and also ended up collaborating with! I personally value community, collaborations and people a lot, especially since I do interdisciplinary work so it really helps to be around a wide range of people (granted, you get access to interns and great collaborators in industry too, though more limited). You also get to benefit from access to courses and seminars, and would get opportunities to teach and lecture, which are all very helpful if you are considering academia.

Apart from the categorization above, in practice, I think there is another way you can categorize postdocs which is more operational:

  1. Hands-on, low-level tackle a specific problem, get shit done postdoc → smaller academic labs and industry
  2. Hands-off, high-level, mentoring/collaborative manage projects postdoc → larger academic labs

Where the first type is more common in smaller academic labs, or in industry postdocs, and the second is more possible and common in larger academic labs. The first is suitable for you if there are still specific problems you want to work on yourself, or if you want to learn the nitty gritty details of a field you didn't really work on in your PhD, but you are interested in and want to roll up your sleeves and get your hands dirty.

The second one, however, is more suitable for you if you already kind of know you want academia, and have an idea of what your research agenda is, and want to get experience advising students and learning softer skills. You want to make sure you can context-switch between projects fast, have days filled with meetings, and have to work with and mentor students at different stages with different backgrounds. You can also ask your advisors to involve you in grant writing and fund-raising, which is another aspect of a faculty job, if you want to see whether academia is for you or not. I would classify my postdoc as the second and I personally really enjoyed this.

What postdoc is right for me?

There are so many things to consider before you choose what postdoc you want to do, or if you even want to do a postdoc. Here is a list of questions to ask yourself, before you decide:

  1. Personal/life constraints: Are you trying to start and support a family soon? Do you have people who financially depend on you? Are you constrained to certain locations/have a two body problem?
  2. Are you leaning towards industry more, or academia?
  3. You are towards the end of your PhD, you want to do academia, but are wondering: should I stay in my phd institution for a year or two longer, or should I go do a postdoc somewhere?

Have no fear: Don't be afraid to make a decision, and remember things are not this rigid in real life, you can always mix and match too, you can be co-advised, and be part of both a big and a small lab! You can keep collaborating with industry! There is more than one 'correct' way to get what you want/need!

A note to international folks, considering the US: if you end up deciding to do a postdoc within the US, and need to change status from f-1, watch out for the J visa!! A lot of folks switch to J visas for their postdocs, however they are not aware of the home-residency law, which would require you to leave the US for 1-2 years after you finish your postdoc! Consult your lawyer before the status change!

One additional note to consider, however this is extremely context-dependent: if there is a school you want to end up in as faculty after your postdoc, and you think you have a good shot there, don't go there for a postdoc! In academia, you are basically burning options when you get affiliated with a school, this is more the case with where you do a PhD, but also holds true with postdocs in many cases. However, this doesn't mean you can never become faculty at your former institution, it just might take a few years. And the main reason is, in the eyes of the department, you are redundant with respect to your advisor. So, you do not bring much added value. Again, this doesn't always hold, but it's something to take into account when considering your next step!

More on my personal experience

How I decided to do a postdoc

To be honest my postdoc was a happy accident, and I feel extremely grateful that it happened! I ran into Yulia at a hallway during NAACL 2022, and I went over to introduce myself to her but she already knew of me and my work! I was very surprised, and then she said she is interested in privacy and was wondering if I wanted to do a postdoc. At that point, I didn't have work authorization in the US and technically could not defend, I was also in a weird place mentally so I didn't really have an answer. A few months later I got my work authorization, and I really felt that UCSD has run its course with me and there isn't much more for me to learn there. So I talked to my advisor and we decided I would defend and do a postdoc! I then reached out to Yulia, asked if she still wants a postdoc and she said Yejin is interested in co-advising, so it happened!

I didn't explore other options, mainly because I felt this was a great opportunity for me, to be advised by two great female mentors, in diverse labs with wide reach. I also LOVE the pacific northwest.

Acknowledgments

I want to express my deepest gratitude to all the amazing people I've met and worked with during my PhD and postdoc journey. With special thanks to my incredible Postdoc and PhD adivsors and mentors: Yulia Tsvetkov, Yejin Choi, Tadayoshi Kohno,Taylor Berg-Kirkpatrick, and Reza Shokri