This is a follow-up of my earlier article on lessons learned in my first two years of PhD. This one is more tailored to later, more experienced PHD students or researchers, and includes things I have relefected on deeply during the last two years in my PhD.

Two years ago, I published a Michgian AI Blog article to help students just starting out in the PhD program navigate their degree. I am very grateful that, to this day, people are still telling me how much they learned from that blog post. However, the article primarily targeted early-career PhD students, focusing on how to begin research and develop good habits. It paid far less attention to the second phase of the PhD—when you have already dipped your toes into research but are still uncertain how to make a meaningful impact.

Unfortunately, many students sail through the second half of their PhDs without refining their research skills enough to set themselves up for the research career they aim for. Even worse, some students do so without having a chance to work on problems they really care about.

As my perspective has evolved much since the first article, and I’ve gained some wisdom as I approach the end of my own PhD journey, it made sense for me to write this follow-up blog post. This post is a bonus set of eight lessons on how to make the most out of the second half of your PhD.

1. Protect your curiosity at all costs

If you’re two to three years into the PhD, you’ve likely already published a paper or two, and you may start to feel like you now “get it.” You may feel that you know much about your topic and that “nothing is new under the sun.” This could be a sign that you are losing your sense of curiosity. One thing I learned from some of the amazing researchers I worked with is to always have fun with your research. Someone who worked closely with Nobelist Geoff Hinton told me that Geoff would show child-like excitement every time he got a new research idea. I’d argue that is the type of love that every researcher should aspire to. As Brain Tracy once said, “If you turn work into play, you’ll never have to work a day in your life.” Curiosity has that power to turn work into play. That also means you adopt a curious language. Instead of saying, “I have to understand why this is not working,” say, “I’m curious why this is not working.” Simply starting a sentence with “I’m curious…” can dramatically transform your perspective on the problem at hand.

Maybe you’ll object: “I’ve never actually been curious in my life. Curiosity just isn’t my thing!” Well, If you never had it, build it. Curiosity is a muscle and the more you use it, the stronger it becomes. I believe curiosity is an all-or-nothing trait. You are either curious about everything, or you’re not curious at all. You can’t just be curious about why your loss curve is plateauing early during training, but not about why the moon looks bigger when it’s in the horizon than when it’s in the sky. This is a muscle you should train all the time, especially outside the lab. So remember to ABC: Always Be Curious.

You will notice that in different parts of this article, I will invite you to ask questions about yourself and your research. Questions are fantastic. They are the “reps” with which you build that curiosity muscle.

2. It’s time to work on important problems

Richard Hamming has a great lecture titled “You and Your Research,” which is also transcribed here and which I highly recommend reading. Richard suggests that one should always be asking, “What are the important problems in my field?” and “If what I am doing is not important, and if I don’t think it is going to lead to something important, why am I working on it?”

Working on super ambitious problems may not always be feasible, especially during early stages of a PhD. However, if you are in the middle part of your PhD and if you have already published a paper or two, you should feel more secure and less inclined to pursue low-hanging fruits. This might be a perfect timing to start tackling more ambitious problems.

You need to be careful about balancing ambition and feasibility, and so what I would suggest is that you ask yourself, “What is one problem or area I could not work on two years ago, but I can work on right now?” This is an extremely powerful question because it will push you to capitalize on the experience you’ve accumulated in the first half of your PhD. You do not want to start from scratch—you want to build on the skills and the knowledge you’ve already accumulated, and move from there toward a more ambitious problem. If you start from scratch or switch completely to a new area, then you’ll need one to two more years to build expertise in that new area. Some of my quite impactful work such as ThinkPRM was achieved when I built on knowledge and skills I accumulated from earlier projects on reasoning.

3. Develop a vision

Academia places high value on published papers, and this in turn leads to us thinking about our research in terms of individual and isolated papers. Yes, it’s great to have papers around different domains because it shows breadth, but from my experience, depth is just as important, if not more. You cultivate depth through a vision, and you develop a vision by thinking not just about what’s relevant today but what will be relevant 2-3 years from now, and weave your individual papers around that vision. You think in terms of research goals, not in terms of ideas. This is something I wrote about in the first article as well, so revisit that if you need to.

Ask yourself questions like What class of problems do I want my research to address?, what research directions/techniques/methods I believe are most promising towards solving these problems? These are just example questions, and I’m sure you can come up with better ones, but if you do this right, your research will have an arc—a storyline, not a few scattered papers that don’t connect well. Not only will this develop you into a strong researcher, it will also help tremendously during job search—especially academia.

4. Don’t mistake busyness for progress

It is much easier to be busy than it is to be productive. Mid-Ph.D., you are often swarmed with tasks—many of which feel urgent, but only a few of which are essential to your success. Between weekly meetings, paper reviews, internship applications, emails, and administrative duties, it is easy to reach the end of a semester feeling exhausted yet realizing that little meaningful progress was made on the one thing that actually matters: your research.

The danger is that busyness feels productive. It gives the illusion of “doing something.” To that note, Paul Graham wrote “The most dangerous way to lose time is not to spend it having fun, but to spend it doing fake work.” Research rarely moves forward in short bursts between meetings. It requires long, uninterrupted stretches of deep thinking, experimentation, and reflection. Without protecting this time, weeks can pass with nothing substantial to show for them.

One practical habit that helped me a lot was learning to deliberately schedule deep work. Something I picked up from Matthias Galle—my internship host at Cohere AI—is to block focused research time directly on your calendar. Time blocking is powerful because it treats deep work as a commitment rather than something you squeeze in when everything else is done. Make it a rule to protect this time. Progress in research is not about doing more; it is about consistently doing the right things deeply.

5. Learn how to speak

Early in the Ph.D., your audience is usually a single person: your advisor. Most of your communication effort goes into weekly meetings—presenting progress, pitching ideas, and calibrating direction. This is necessary, but it comes with a cost. Your advisor already knows the problem space, the literature, and many of the unstated assumptions behind what you say. Over time, you begin to overfit your communication style to a single expert listener.

But being able to explain your work to someone with a good background is very different from explaining it to an audience encountering it for the first time. Without realizing it, the first 1–2 years of your PhD may turn you into someone fluent in expert-to-expert communication but incapable of conveying complex ideas to broader audiences.

The fix is simple—though uncomfortable and might take some time: talk about your work to people who don’t know it. Put yourself in situations where you are forced to explain your research from first principles. The good news is that you do not have to wait to be invited. Many of the talks I gave during my Ph.D. were self-initiated. This is a form of tactical stress: deliberately placing yourself in mildly stressful situations in order to grow. Other than formal talks, department events, reading groups, or lightning talks—such as the AI Tea in our department—are excellent low-stakes opportunities to practice. At the same time, study great speakers. Watching a strong TED talk can teach you more about structure, pacing, and clarity than many technical presentations ever will.

It is especially unfortunate to see a strong researcher whose ideas do not land simply because they are poorly communicated. The curse of knowledge is usually to blame: Once we understand something deeply, it becomes difficult to remember what it felt like not to understand it. The most effective way I’ve found to counteract this is to constantly adopt the listener’s perspective. When preparing for a talk, I ask myself: If I knew nothing about this topic, how would I want this to begin? What is the best way to structure my slide, so the information flow is logical? This question alone forces me to strip away unnecessary complexity and rebuild the explanation from the ground up. Watching this lecture on speaking from Patrick Winston is probably one of the highest-return investments I’ve made in improving my presentation skills.

Writing is just as important; scientists communicate through writing more than any other means. Writing is thinking. By putting us face-to-face with the messiness of our own thoughts until we turn it into something that others can understand. Write long blog posts, short blog posts, well-thought tweets, good emails, whatever you can write. It is fine to use AI for writing, but only for System 1 tasks of writing to AI e.g., polishing, fixing typos, etc., but never let AI think for you.

6. Build independence

Early in the Ph.D., it is natural to rely heavily on your advisor for direction, ideas, and validation. This is not only expected, but often necessary. However, as you progress, you want to gradually move toward a different goal: being able to carry a project from start to finish on your own. True independence is about developing the ability to reason, lead, and push a project forward without needing constant external guidance.

I believe it is perfectly okay to be opinionated. I do not quite understand why being opinionated is usually treated as a negative trait in academia. If you look at many influential researchers, they tend to have strong, sometimes controversial opinions about where the field should go—even when those opinions conflict with the consensus. Yann LeCun has spoken many times about how large language models are a dead end. Richard Sutton argues that artificial general intelligence will only emerge through learning by exploration. Andrej Karpathy has expressed skepticism about reinforcement learning (RL), describing it as inefficient and tedious at a time when everyone is excited about RL. Whether you agree with these views or not is beside the point. What matters is that these researchers have clear, well-reasoned takes and are willing to argue for them.

Forming your own opinions—and being able to defend them—while still remaining open to changing your mind when presented with strong counterevidence, is one of the clearest signs of intellectual independence. Note that I’m not talking about being stubborn; I’m talking about having a coherent internal model of the world that you are constantly refining, but willing and able to argue for when needed.

7. Do not take yourself too seriously

I learned later in my Ph.D. that life becomes much easier when you stop taking yourself—or even your research—too seriously. This does not mean being careless or uninvested. You can care deeply about your work while still leaving room for playfulness and experimentation, where failure doesn’t reflect badly on you in any shape or form. You become okay with the fact that your ideas may not work, your research may not land as you had hoped, or when your coffee turns out bad. Jensen Huang recently said in an interview that the biggest obstacle to success is having high expectations. Not taking yourself too seriously lowers expectations, stops constant comparisons, and regulates overly high standards that can harm your resilience and mental health.

I write this from experience, as this is something I still struggle with from time to time. I sometimes demand so much of myself that I leave little room for mistakes or imperfections. What I have found to help is to shift my mindset to think of myself as a student of life. If I am a student, then it is perfectly fine to make mistakes. I have found this mindset especially helpful whenever I catch myself obsessing over a less-than-perfect answer I gave in an interview or beating myself up over a paper rejection.

It also helps to not identify with your own research. You are not what you produce. What you produce is a separate entity with its own destiny. Whatever happens to it does not reflect well or badly on you. You do your best to produce something of as high quality as possible, but then once it’s finished, it is no longer up to you what happens to it.

8. Appreciate every second of the PhD life

There are few environments as conducive to growth and learning as a PhD program. It is a rare period in life where you are given the time and intellectual freedom to learn continuously, and reshape (how you see) the world. I know I will always look back on my time here with strong nostalgia, remembering—more than the papers and successes—the day-to-day things and the person I was becoming along the way.

Unluckily, most work environments are not like this. Outside academia, and perhaps a handful of research-driven roles, work is often constrained by deliverables and business objectives. Industry research, except in a handful of labs, has little tolerance for playing with ideas. Even a post-doc is not the same, as post-docs tend to be shorter, and are typically tainted with the anxiety of securing an academic job. The PhD offers something unusually generous: the freedom to work on problems you genuinely care about, to spend weeks tinkering with an idea, and to grow intellectually without constant pressure to justify every hour.

Time, however, moves faster than you expect. I am writing this less than one week away from my PhD defense, and it feels surreal how recently this journey began. What once felt far away is now a few days away. If there is one lesson here, it is this: be present. Appreciate the flexibility, the struggle, the late nights, the conversations, and even the uncertainty. This is a uniquely transformative chapter of your life, and once it passes, it is unlikely to exist in quite the same way ever again.