Learning is a Passive Skill

Learning is a passive skill. I was having a discussion with someone on what they need to learn to become an AI Engineer and we talked about system prompts, skills, chat templates, and many things, and we finally reached the conclusion that learning is a passive skill while reading, writing, memorizing, and all that are active skills. If you are a gamer or someone who reads novels or fiction, you would know that active skills have a cooldown but passive skills don't (iykyk).

Many people ask me about:

• How do I learn about harness engineering?
• How do I learn about LLMs?
• How do I use agents better?
• How do I become an AI engineer or something?

To be honest I never had any kind of answer to give because even I don't know. Today, on a walk, the same person from that discussion was telling me they had set some goals to become an AI engineer, or just to learn about agents. The high-level plan was to read about the topics, try them out, and write blogs on it, which sounded like a good plan in general. This reminded me of something from college.

I was doing a project for the first time at the semester level and it was related to machine learning, steganography to be specific. I was discussing it with the professor who was mentoring us and I said, "This is the idea of the project that I'm planning to do this semester." I told him I would go learn some basics first, because I had no idea about data science, Python libraries, OpenCV, or ML in general, and that I would get back to him next week.

While I don't remember the exact answer he gave, I remember the gist of it. It was that this is perhaps the most inefficient way to do it, because computer vision is so vast that it will take me ample time, and if I try to learn about ML it will take years. Even if I want to learn only the most minimal things and I ask someone to help me decide the topics, I would probably learn a lot of unnecessary things. It's not that they go to waste, but they definitely won't help me today, this semester, for this goal. And he suggested to me to just read a research paper, the exact reference for what I am trying to do. In this case it was steganography. Read it with Google open on the side. For every word you don't know or every line you don't understand, just search it and read as much as you can understand about it. Not too deep, just the minimal thing you need to know, and keep doing this until you feel like you can try to copy what they did.

And I did that. To my surprise, just within two weeks I had good enough depth that I proposed a small optimization, to drop ASCII and use a random charset, which surprised him so much that he asked if I had come up with the idea myself. Looking back at it, I didn't put any active effort into learning. I was curious, so I was reading about it and searching, but I never had any goal or plan about doing xyz daily, and perhaps most important, I was not miserable while doing all this.

And this stuck with me even now. I'm doing many things today that I never set as a goal and never read about first. I know Go Language in general. I have used it but I never read a Go book at all. I never read a system design book either, but I'm very confident in both of them.

What I did was know the syntax, start doing the thing I was trying to do, and read only the next thing I needed to do. That's it, and somehow it compounded and it worked. Go and system design came off the back of things I was already doing anyway. But I had some, I guess, obsession with Rust. I was just trying to learn Rust. It's a fascinating language. I tried to read the Rust book, this and that. And Rust is the one I stalled on, which I think is the whole point: it's the only one I tried to learn directly.

For some reason I'm very bad at reading books and I never was able to complete the Rust book, or any other book. For a long time I thought that was a discipline problem. I don't think it is anymore. Reading is an active skill and active skills have a cooldown. I was hitting it. I already had too many things to do, or rather too many interesting things to try out, so I was never able to follow up with it, and whatever I would learn by reading, sitting for an hour, I would forget within a week if I hadn't used it, and this had happened so many times with me lol. I realize now, nothing was wrong with me. I was just casting the wrong thing.

Summaries were the same trap for me by the way, just from the other side. I learnt this the hard way: don't try to summarize a topic or a blog or a paper you're new to. Summaries are for things you already know and are confident you know most of. A summary gives you every term and none of the trouble, and the trouble is where the learning was.

I told them all of this on the walk. Then they asked, "Okay, this is good, but what if I want to become an AI engineer, or some designation which requires some qualifications?" I reflected that many people I know are overqualified for their position. I've realized that even in the resume and the job description, you just put things in for the sake of putting them in.

So it became, "Okay, but what do I need to learn? I won't try to actively learn anything, but how do I become qualified enough to be an AI engineer?" We reached a point that instead of trying to gather all the knowledge that designation needs, it might be much easier to just try doing what someone would do in that role. If it was an AI engineer or something, then that would be customizing your harness, or contributing to a harness at the minimum I would say.

Then they asked the one I was worst equipped to answer: "Okay, but what should I set my goal for, say, one month, two months, or three months?" I've always been bad at setting goals. I read parts of some books and one thing that stuck with me was that setting a proper goal is perhaps the most important thing one can do, because that will affect many aspects of your life, especially mentally. What they said was to set something big and specific. Big and specific never really worked for me on its own. What worked was a filter. I'm selfish with my effort. Ever since college, whatever the project was, I would always avoid doing anything for anyone else. I hated the idea of doing some kind of project for something I don't use or won't use in future. Library management system, for example. The most common project anyone would find. I hated that one most, cause I don't read books, nor do I work in a library, and I'm not a saint to build and donate software, and come on, no one needs a Library Management system.

Even for the hardware projects, I decided them on that basis. If I do it and if it works, then at the minimum I should be the user of it, or it should be remotely helpful to me. If it doesn't fit that criterion, that's something I shouldn't do. And I think that filter is doing more work than it looks like it is, because a passive runs the whole time you're in the game, and you only keep showing up for the things you actually want. This made me pick very difficult and insane projects (cause I had zero skill and expertise to complete those projects). Most of them I never finished, partially completed would be the correct word. I feel like I have learnt most of what I know, perhaps 80%, from college to now, by just this.

And yeah, the discussion ended and they set their goals. I realized I forget this from time to time, so I wanted to note this down as a reference to myself in the future so I don't repeat the mistakes I do lol. Perhaps this would be an interesting story for someone else as well, because I think everyone has kind of similar questions now, considering you have AI to teach you everything in the world possible, and every content creator out there is trying to guide you. We probably will always be in surplus of knowledge around us and people trying to teach us. But all of that is an active skill being sold to you. A course, a roadmap, a book, a curriculum. Nobody can sell you the passive.

Let me know if someone found this useful. That's it. Bye Bye