When I was applying to graduate school, I sought out posts like this one to understand what I should put in my statement of purpose, and later, after I received acceptances, how to select a university and advisor. I’ll put the same caveat on this post that many professors have put before me—these are my personal preferences, and I know they are shared by some, but certainly not all of my colleagues.
The advice in this post is targeted at junior researchers in 2026 interested in AI who are deciding between attempting to do a PhD with me (or one of my colleagues) or instead taking a job in industry.
These are the three key traits I believe distinguish successful PhD students.
Possession of longterm goals
Doing a PhD requires you to believe that the skills and credentials you will slowly acquire over 5+ years of the degree will still be valuable at the end of the degree. These days, there are some contingents of AI researchers and practitioners who are consumed with the shortterm. They believe that the world in 5 years will look so incredibly different from the world today that there is no point in setting personal goals that span more than a year or two. If this is your attitude, it’s my opinion that you are ill-suited to a PhD; you’d be better off looking for a job in industry that shares your short-term priorities.
This is largely because the design goal of a doctorate in philosophy is to certify that after completion, the student is equipped to join the Academy as a peer. Of course, the skills one acquires along the way may be useful in other realms, and many (indeed, most) students don’t end up becoming professors, but that aim is what doctoral training is optimized for, which means that many of the intermediate goalposts are structured around long time scales (e.g. we force you to take classes outside of your narrow interest because we expect you will shift your research focus sometime over your career). Even if your long-term goals do not include staying in academia, having long-term goals is important because they motivate students to maintain morale through the inevitable adversities that come with the territory. Students with long-term goals are better able to persevere through hard classes, being scooped, paper rejections, and the like.
To be clear, I don’t even especially care what your long term goals are. A few people enter AI PhDs because they are passionately interested about some weird, esoteric topic they can’t possibly hope to convince profit-seeking companies to work on. A lot more people enter PhDs because they see value in having the degree; perhaps they want the legitimacy the title of “PhD” bestows when conversing with policymakers and other decisionmakers, perhaps they want to become a professor so they can teach and train the next generation of students; perhaps they are simply pursuing an “American-dream” path to citizenship and prosperity.
Some level of intolerance or obliviousness to fads and popular wisdom
It’s easier than ever to be inundated with other people’s opinions. I admit there’s some irony in me saying this as I add to the maelstrom my own opinions. Lately, I’ve been noticing a convergence toward a singular hivemind when it comes to academic advice. The collective voice of social media will try and convince you there is a correct way to write a statement of purpose, a correct way to approach a professor for research opportunities, a correct research topic to select, a correct way to optimize your life choices around a career goal. Students’ reliance on AI for writing and advice is compounding this homogenization, with AI leading students toward undistinguished, look-a-like research plans and prose.
While it’s almost always a good idea to solicit advice from others’ who have achieved your goals before you, it’s also useful to not take this advice too seriously. Not only does this have the benefit of making you less stressed, it’s also likely to lead to a stronger PhD application and more interesting, impactful research in the long run. Most importantly, just because some strategy worked for some student doesn’t mean that strategy will work for you, or that all other strategies are inherently worse.
As this pertains to PhD admissions, it’s important to remember that faculty are reading anywhere from 50 to 100+ statements of purpose. In the ~5 minutes I spend on each statement, I am looking for evidence that you have the creativity to pitch interesting research questions and the technical skills to develop good solutions. If your statement starts with the same opening paragraph referencing the same “grand challenge” as a half a dozen other statements, it will almost inevitably fall through the cracks. If your statement is a bit weird—perhaps you start with a quote from someone you admire, include a small personal anecdote, or mention taking classes in a niche topic—your statement has a much greater chance of standing out from the pack.
And this brings me to my last piece of advice.
Academic interests outside of AI (or better yet, computer science)
There are many important topics that need to be tackled for AI to become a useful and trustworthy tool for humanity.
I’ve found that students who have exposure to how people think and work outside of the field of AI, and better yet, outside of STEM entirely, end up doing more interesting and impactful AI research.
While there are many PhD students who write highly impactful dissertations on making AI work better for AI researchers/practitioners, I am much more excited about making AI work better for people outside the field.
This requires the patience and openmindedness to learn not just the jargon and core concepts of these other fields, but how people in these fields think and what they value.
Lately, it seems like an increasing number of AI people subscribe to the notion that once we build AI that solves building AI (what some call recursive self-improvement), all other traditionally human forms of inquiry and creativity will be made irrelevant as superhuman AI adapt to perform all these acts for us.
I’ve found that the more exposure students have had to disciplines far adrift from AI and CS, especially ones where “goodness” is not easily quantified or verified, the less likely they are to hold this viewpoint.
Personally, I am more keen to find ways to make AI a useful tool for humans rather than a replacement for them, and I hope to recruit students who hold a similar sentiment.
Returning back to PhD admissions: if you are a senior undergrad who honed in on NLP/LLM research in sophomore year or before and has already published multiple first-author papers in top ML conferences, I am sure there is a prospective advisor who will be impressed by these early successes, but that advisor is not me.
If you are a senior who double majored in CS plus something like history, journalism, archaeology, or astronomy, and who only decided they’d like to pursue advanced studies in AI in the last few months to year, please apply to CMU, and put down my name as a prospective advisor–I’d be excited to look through your application.
A last reminder about cold emails
Lately, I’ve been getting many emails a week from prospective PhD students asking about joining my research group. These emails are a waste of your time; I don’t have the bandwidth to sort through for the human-written ones, let alone read closely enough to retain any memory of who says what. The best way to get my attention is to apply for my department’s PhD program and put down my name as one of the advisors you’d like to work with.
I do my best to look through all PhD applications which pass initial departmental screening and list me as a desired advisor.