My 2025 Resolution: Beyond the Roadmaps, Beyond the Timelines

Last Monday was a national holiday in Japan due to 成人の日, so I had 1 extra day off to do some much-needed self-reflection.

This has been a massive week for me in terms of rearranging my thoughts and strengthening my belief in what I want to do and achieve by the end of 2025. Like always, I did some research and firstly built a resolution list. Nothing unusual.

Something slightly different that I did this time — unlike how I did it for last year's resolution — was to actually reach out to people who have learned the things I want to learn and get their constructive criticisms so that I would ensure I avoided doing things with diminishing returns or things that wouldn't matter as much as some other aspects. A full-time job takes a lot of toll on me, and while I still can dedicate 3-4 hours per day to studying and learning outside of work, this in my opinion is still quite a limited amount of time per day and I wanted to ensure I was wasting as little of it as possible.

Back to the resolution talk. As you can probably see when you check it out, it's pretty much like a detailed roadmap. I've always followed roadmaps religiously ever since I started seriously learning things during my 4th-year at university, and they haven't gone wrong for me. Something was different here, and it was likely because of what I wanted from myself: efficiency in learning.

One of my achievements in 2024 was being able to conquer discipline and consistency towards learning. I've always been a passionate learner and I have had, for a long time, a strong confidence in myself — confidence in my strengths and my weaknesses alike. Last year was truly the year I was able to prove to myself that I, in fact, also bear the discipline in consistency & putting in effort; almost every day since mid-August, I dedicated time to getting things done in the morning pre-work & also late in the evening after work. Learning became a second nature to me, so much so that I wouldn't even think about it. I just did it. On the other hand, after I had firm control over my habit, I started getting the feeling of whether those 3 hours a day would be enough. As this one saying by an economist goes: "There is never enough of anything to fully satisfy all those who want it". Used to hear it a lot during my university days. Edit while proofreading: you know what, maybe this quote kind of means something else (lol). Anyway, you probably get what I'm trying to say here, so I won't bother deleting this.

Which meant that my next stage here was efficiency: getting more done in less time. And that's pretty much what the people whom I had asked for feedback indicated as well. The "roadmap" I had indicated a more bottom-up approach, which is fine in normal cases but I received clear-cut examples from people who mentioned that they managed to achieve paces that were multiple levels higher by just breaking things down from the top based on concepts rather than conventional roadmaps with assignments and deadlines. Elliot Arledge, one of the people whom I'd asked for feedback, took a portion out of my roadmap and shared something that genuinely made me wonder — and I'm saying this unironically — why I wasn't already doing it:

"Doing practice problems by hand works for some, but there's a high chance you will go 10x faster building cool projects. You're allowed to use AI tools to speed this up. No it doesn't count as cheating."

This was an interesting outlook for me. I've tended to first try to get the theory down (let's say I learn the definition, theory, etc. like the ML course by Andrew Ng), followed by a mini project or just playing around with code to get my hands dirty. And as I feel confident enough, I move on to some other topic. So, it's pretty much been an approach that builds up mini concepts one by one. Since I've been thinking of ways to up my pace of learning anyway, I've decided, after a week of self-reflection, that I want to try it this way and see how it goes for me.

Words are cheap, however, so I need to first prove to myself that I am, indeed, doing it properly over the next few months.

So this means my 2025 resolution list is basically already obsolete. But that's quite exciting.

Which means? I am going to concurrently juggle Python, Math, and traditional ML first; I need to finish the Math for ML course on Math Academy, get myself more comfortable with Python + traditional ML, and then hop on to building project pipelines.

This all pretty much connects with the advice that I'd received from Karpathy as well:

"go go go and count the hours and iterate. Pick up projects, complete them, publish them, write about them, repeat gl"

2024 was my year of discipline and preliminaries. 2025 is the year where there will be multiple echelons of elevations in my knowledge fabric. Watch this space.