Taming the Mindset - It's an Unlearning First Journey

The Urge Problem

There’s this constant urge we all feel - the urge to do something, to build a project, to apply for internships, to fill up our resumes with impressive-looking accomplishments. It makes everything feel chaotic and restless. I’ve been literally acknowledging this pattern and actively working on it, because I realized something important: many of the foundational things we’re learning come from listening to and learning from people who actually shaped this industry, who contributed significantly to where we are today.

This means we need to tune into that particular way of thinking and then try to adapt and unlearn a lot of things. When we come from industry, we might have very application-centered thoughts. We always want to add a couple of projects to our resumes. But I truly believe that graduate studies has to be a function of multiple things. It shouldn’t be just projects, but also working on mathematical theorems, reading original texts - that’s what I’ve been doing for some time.

The Wisdom of Slowing Down

We shouldn’t quickly give in to the urge of things. Over the years, I’ve developed a reading practice, and now I feel like I can more fully embrace that approach. My mind always tries to make me feel like “okay, we need to apply for internships, we need to do some projects,” but the thing is, in some subjects we simply cannot start with projects.

I had a wonderful conversation with Professor Elias, who’s teaching me causal inference. He gave me this incredible advice: until midterm, or about 70% of the course, you read and practice the mathematical part. That way, you get into the system. You understand what’s being shaped, how things are built up. That’s a quite good insight I can directly apply.

How can we build causal intelligence systems? How can we train causal intelligence systems? We may not be able to figure that out immediately, but if we have a very good unified view, we’ll be able to decode any problem statement based on our understanding.

Starting from the Source

I truly believe it should start from the original texts - deep reading, then trying to solve a couple of problems, then writing it as code. When things compound like this, we develop something substantial. I don’t want to run behind what people are sharing on social media. Sometimes there’s this urge when we see “Oh okay, this is a wonderful blog, let’s read it.” But reading that one blog out of context may not help. There has to be some continuity, something that complements what we’re already doing.

For me, the courses I’ve taken right now are interconnected. When I’m reading something about a probabilistic way of approaching data analysis and building models, it’s quite related to how causal inference works. It’s also related to how reinforcement learning utilizes the Markov process to model scenarios. At the end of the day, we need this foundational grounding so that new problems can be easily approached.

The Art of Slow Digestion

We need to really spend enough time going slowly through the proofs and all such things. I don’t want to rush thinking “okay, let’s go” because I’m quietly understanding my own vibe, my own pace. I’m able to interact and talk with PhD students in their language. I’m not in a position where I can’t understand what they’re teaching or discussing. I’m quite confident and motivated that I can get things quickly. But the thing is, I shouldn’t rush in processing that. I should slowly digest them.

This gives me something retrospectively - I’m able to fix certain things as I go along.

Building Rigor for the Future

My goal is to build the rigor that’s necessary for applied research or product innovation roles in companies. I’m just getting clarity now. People might think I already have clarity, but the thing is, I’m just getting things sorted.

Sometimes I get advice or feel motivated to build a benchmark-level problem or model. But I don’t want to do that just for the sake of doing it. I don’t want to take some random dataset and just use existing code - like sentence transformer code that’s already there. If I build a model, the effort is going to be spent on finding a new dataset, or just iterating and changing loss functions or improving hard negatives and positives.

What excites me are the conceptual elements of the problem. Why does model A perform better than model B? How are the architecture or loss functions contributing? Can I do some empirical analysis? Can I change this or that? Such questions give me more attachment to the problem statement, more relevance to how well I can understand it, rather than just following cookbooks.

The Daily Practice

I’m quite happy that I’m able to address some of the problems I’m facing and fix them while focusing on the work. Every day it’s like eight to ten hours - I’m able to read and work on something meaningful. Next, I need to slowly move toward research-focused problem statements. That’s where this fall should help me develop these habits and follow through on all these principles.

A Repository of Growth

I find that I need this kind of repository for myself - tracking where I was and how I’m progressing. People might wonder, “You’ve worked for so many years, don’t you already know this?” But the thing is, I’m in this state now. This is a journey, a continuous evolution. It’s not like we always have to be in a saturated state and stay there without any growth.

Life is like a Bayesian process. We start with some belief, some opinion, and over time, as we start getting more and more evidence and experience, we update our understanding. That’s the beauty of this journey - always learning, always evolving, always becoming better at understanding not just what we’re studying, but how we’re studying it.

---

This reflection captures where I am in my master’s journey - building the discipline to learn deeply rather than quickly, and finding joy in the process of genuine understanding.