It has been about nine months now, and there is a quest that has stayed with me the whole time: as a student, what is the better thing I can do, and what is the extreme I can explore? Underneath that question sits another one — a search for what I’d call true scholarliness. It is the thing I feel I missed earlier. When I did my bachelor’s in India, the path was never very directional, and honestly I didn’t have much idea of what was possible. It was mostly about getting a mark. Now, with years of experience behind me and with all the information the modern era of AI has placed in front of us, I want to reflect on the right methodology, the right mindset, and how to stay genuinely optimistic as a student — how to actually embrace the learning experience instead of just surviving it.

This is mainly for students, and especially for those who share my segment: people doing a master’s abroad, people doing a master’s in the US, and anyone drawn to artificial intelligence, machine learning, data science, business analytics, and everything around them.

Let me start where the journey starts — with the decision itself — because every step along the way is a decision.

The good first decision, and the chaos that follows

Everyone reading this has already made one good decision: the decision to study, to learn something, to do a master’s. I’m not arguing from pure logic or hunting for evidence here. For someone in India who wants a real leap in their career, who wants to come to the US to learn from top professors, to go through the process properly, to get a shot at working at the best companies in the world — that first decision is genuinely a very good one. It is essential.

And then a paradox begins.

You take this excellent first decision, you apply, you get your admits — and then the chaos starts. Bad influences, misinformation, disinformation. For most people, the primary source of information turns out to be Reddit, followed by alumni, followed by YouTubers and social media influencers. Three channels, and the first one I simply don’t count. Reddit, when it comes to a master’s in the US, is mostly populated by pessimists. They rarely encourage anyone. They gaslight you — why are you even doing a master’s, it’s not worth it, it’s not a real career — and a lot of what gets written there comes from paranoia, or from envy, or from people who are just plainly demotivating.

Then comes the second trap: what looks good on a resume? Is university A better on the resume, or university B? Endless hierarchical and vertical comparisons — which department matters, which university is better, which city is better, is there an ROI, is it worth the money. All of that.

But the real conversation — the one that almost no one is having — should be about what is the best thing you can actually do. How can you do hard things? How can you learn as much as possible and thrive? Why would you want a world where it’s written on your forehead from day one that you’ll get a 100% guaranteed job, that you just have to exist and it will come to you? It doesn’t work like that. Go and check how many undergraduate CS students even from Stanford land a job versus those who don’t — there are always numbers, but there are also a thousand other stories behind them. Some don’t want a job at all; some want to start a company, some want to go on to do a PhD. The numbers never tell you the whole thing.

On the “cash cow”

Then there is that one word that haunts these decisions: cash. The cash cow.

I’d ask you to think about it differently. Consider how the paid-forward, entrepreneurial culture in the US actually works. Treat your master’s as an entrepreneurial, venturing thing — because that’s what it is. You come here, you pay for many things, you pay into an ecosystem, and that ecosystem genuinely has the potential to be life-changing, for you and even for your generation.

But here’s the question. I buy a shoe, an apparel, a phone, an internet subscription — and how much of what they actually offer do I ever use? The thing has so much to give, but are we really using it to the fullest? The same is true of a master’s program. The ecosystem offers an enormous amount. Most people never come close to drawing it down.

So don’t be governed by the phrase “cash cow.” Remember that a public university like the University of Michigan runs a master’s program around the 100k mark too, even though it isn’t sitting inside a major cosmopolitan city. There will always be ifs and buts, foreign exchange, loans, how much burden you can carry. If you can get a loan and you can see your way to a good outcome, often you should just go ahead. The deciding factor shouldn’t be the sticker. It should be this: do you like the college, the subjects, the professors? And — because seeing is believing — can you see yourself doing better there? Can you see yourself collaborating with people, doing research, doing hackathons, networking well, building ideas, running experiments? Will that ecosystem let you do that?

That, plus the quality of conversations, the curriculum, and what the city has to offer — that is what’s essential. The options are practically infinite. Whether we realize them is another matter, and often our own inherent bias gets in the way. With an open mind, you can explore an extraordinary amount.

The thing almost no one talks about

Here is what I’ve actually observed: there is very little conversation around the real curriculum and what one can genuinely do with it. Most of the worry goes into wanting someone to tell you this is a very good college, you made a great choice, this is better than the other one. People go looking to be pampered. It’s confirmation bias — we want some random person to validate that our university is good and our decision was right, even when that person may have nothing going for him in his own career.

Let me tell you a small story. Before I joined, I had around six or seven admits. I met a guy who had twelve, and he was flexing — bro, you made the wrong choice. But ten minutes in, if you actually looked at his resume, his projects, the courses he’d taken, what he had done over time — there was nothing. Literally nothing. He had gotten a sort of self-indulgent high out of I got into many universities and chose this one. And after getting in? Nothing. He never put in any effort once he was through the door.

This is the trap so many mindsets fall into. They stop iterating, stop experimenting, stop learning. They keep betting on the tag — I’m from a big college, a big program, I have the good label. The tag is not going to help you at any cost. I say this while doing a master’s at an Ivy League university myself. Saying “Ivy League,” “Big Ten,” “UC” only carries meaning when you actually embrace it and seek the real knowledge — not when you announce it on day one or in your first month. It genuinely pains me to watch people not explore, not do things, when the curriculum has so much to offer and the university has so many open doors — research, entrepreneurship, all of it.

Take data science specifically, since it’s close to me. I’ve watched people studying data science across many universities share one common limiting behavior: whatever program they get into, they never go deep. They never really ask what data science means to them. The curriculum offers a lot, the university offers a lot, but the serious work just doesn’t get done. How many original, individual ideas — beyond the assigned course projects — do people actually produce? Very few. How many are reading books? Reading papers? So this is the checklist a fresher, or anyone starting out, should keep running on themselves: Am I reading papers? Am I reading textbooks? Am I understanding things? Am I working on multiple things at once?

And then there’s the GPA problem. The system lets you craft your own curriculum — you choose electives, you register courses. But under the pressure of I need a better CGPA, people take the simplest courses they can find. They never push, never challenge themselves. Career-building, though, is precisely about how much you are willing to push yourself toward the hard things. Not everything — but how much.

You see it in the resumes too: very old projects, concepts not even explored at a fundamental level, one notebook passed off as one project, one thin repository, minimal exploration, minimal reading, minimal interaction. And I believe deeply in the lesson of complexity science here — the more you interact with the subject, the more it gives back. Interaction generates new opportunities, new ideas, new behavior. If I don’t interact with the subject, the projects, the ideas — I’ll just be standing there staring at something, or staring at somebody else’s work.

So we may need to be a little delusional about our careers. A little extravagantly ambitious. I want a genuinely unique experience out of this program. Test yourself. Run yourself like a startup. Explore. Do hard things. Fail at experiments — that’s completely fine. Sometimes you begin an experiment with a wonderful mindset and, over time, your happiness dips because the expected outcome isn’t arriving. But the learning is always assured.

The central idea: arcs of work

Here is the hypothesis I keep coming back to. We need to believe in an arc of work.

Take deep learning. Beyond building a transformer for simple language modeling — what else are we actually doing? Do we start from pretraining, then explore RL for LLMs, then move into post-training, then interpretability? Try to see the work as an arc. Keep two or three arcs of focus at a time. One arc might run the whole way from pretraining to agents inside LLMs. Another might be diffusion models and flows — generative modeling. Another could be reinforcement learning and robotics. For a lot of data science people, causal inference is one such clean arc: it starts at plain potential outcomes, at simple backdoor and front-door adjustment, and runs all the way up to causal representation learning, causal reinforcement learning, and causal abstractions in LLMs.

So picture it: one arc is causal inference, another is language modeling, another is generative modeling. Or — one arc is systems for AI, another is performance optimization in ML, where you learn to write CUDA kernels and design optimizations of all kinds. Or statistical inference and probabilistic modeling as an arc, where the destination is building projects on something as intricate as Gaussian processes — which open up an enormous amount, especially if you’re into stochastic modeling or finance. And then you ask: to reach there, what are the steps, the milestones, I can climb?

I’ll be transparent about my own. Over the last two semesters, one of my arcs has been energy-based models. I started simple and ended up working on JEPAs and JEPA-related projects. The other arc, causal inference, ran its course into a couple of projects on causal representation learning and causal state-space modeling. It’s still going. I won’t claim I’ve done a perfect job — but I have something like a real thesis forming, and I think it can help a lot of people.

I have about eight more months. So I’m recording this now, openly, as a hypothesis. By around January 2027 — roughly eight months out — you’ll be able to look back and judge whether I actually did it, whether I proved it or not. The hypothesis is simple: building arcs of work is going to help people enormously, and keeping two or three arcs is essential for building a strong career. Why am I confident? Because this is exactly what PhD students do. They don’t do scattered things. They work on arcs. That’s the whole point. Even I am following it.

So here is the structure as I see it. The curriculum gives you the range — the A-to-Z of a field, whether it’s machine learning, data science, or business analytics. On top of that range, you develop arcs — cumulative, compounding efforts. And inside each arc live the projects: you read papers, read books, write things, run experiments, share your learnings. Don’t do any of it for the sake of grades. When the motivation is grades rather than skin in the game or the sheer joy of doing it, the work never gets better. Keep interacting — through projects, through reading, through exploring, through collaborating. People do all of this for the sake of it and miss the real stuff inside it.

I know plenty of people who say they’ve “worked on agents.” But it’s one agent, maybe two — tutorial-level work, nothing that demanded an original idea from the learner, no further exploration into how you’d build self-improving agents, or agents trained with RL, or agents for a specific domain. The same hollowness shows up in statistical modeling, in learning systems, in generative learning. I’m not saying this from theory or from having read some book. It’s a culmination of what I’ve read and what I’ve lived first-hand — including all the things I missed. You can find every musing tracing how my thinking evolved. I’d rather share from a radically open mind, transparently, than yield myself to the collective narration where some random person says something and everyone repeats it. Beyond anyone’s opinion, I have my own first-hand experience — the sweating, the crying, the emotional attachment to certain subjects. Sometimes you don’t get the thing you cherished; sometimes you get unexpected results somewhere else entirely.

And nothing arrives through if I do this, then that will happen. You have to struggle. You have to actually do it. A plan in your head or a result in a paper will not carry you.

Decision engineering, and seeing yourself bigger

This is also why, especially for data scientists, I’ve been writing about decision engineering — because it gives you a holistic problem-solving stack. I genuinely empathize with people in data science here. So many see their entire opportunity as SQL, dashboarding, A/B testing, or some simple modeling. They can’t see themselves as big movers, as builders, as real engineers — as people who solve problems holistically. Data science of all things gives you that option. It is, as I’ve said before, a fluid discipline. It attaches itself to and moves hand in hand with any domain.

And the idea of data science — of these graduate programs and data science institutes — didn’t come from someone deciding to round up students and chase funding. There is real research behind it. You’ll find data science and decision science institutes, departments, professors building real work with real teams — at OpenAI, Microsoft, Nvidia, Apple. Under the umbrella of data science sit many arcs: causal inference, machine learning, deep learning, stochastic modeling, statistical modeling. Every one of them is a real career with real meaning and real opportunity — once you move beyond number-crunching, tutorial-level stuff, and bootcamp-level learning, and into genuine scholarliness, into the joy of building those arcs.

So push yourself. Come up with projects and workshop-level research. Publish it. Share it. Write it online. That gives you far more practice than endless résumé-crafting and shallow work ever will — because shallow work simply does not secure the better career. The whole thing starts with that one dilemma — what sounds good on a resume? — and then every action gets bent toward getting shortlisted instead of building the real thing.

When you actually build the real thing — which is what every one of these master’s programs is quietly aiming for — I am sure it becomes a great career and a great journey.

So don’t yield yourself to noisy opinions. Just believe in the hypothesis. Put your head down. Do things.

That’s it.