how to self learn anything
文章目录
self learning is for everyone. why did I say that before anything?
because many people feel self learning is not for everyone, its only for geniuses and people who can teach themselves.
and that's... a myth.
it's not some GOD gifted talent, it's just a loop that you need to do endlessly.
why you don't need another person to self learn things now
the reason many people felt self learning is not for them is they needed a teacher or mentor when they hit a wall.
but you no longer need another person. that problem is solved by a tool we now know as an LLM. large language model.
it can hand hold you, it can help you, it can guide you, it can assist you, it can help you clear your clutter and confusion.
so the point here is, self learning is for everyone.
this brings us to another question: how to self learn anything?
the self learning loop
-
ingest: decide a thing you want to explore and just randomly start ingesting information around it through videos, books, papers, etc.
-
build: apply it. this might feel like generic advice, but it actually made sense when I did it. as soon as I get "appliable" information, I just apply it. and build something cool using it or around it. so... start building.
-
hit a wall: this is non negotiable. you won't be able to build something around the thing you're exploring for the first time. you'll definitely hit a wall, if you don't hit one, that's definitely a problem.
-
fill the gaps: go to the roots of your problem, ask "what information do I lack?, what information can help me climb this wall and move ahead?" and...
-
go to the first step again. ingest.
and this works for anything, not just tech.
learning guitar: ingest a random video on chords. build by playing a song badly. hit a wall at the barre chord. fill the gap by learning the exact finger shape. play again.
learning a language: ingest videos, try talking to someone, hit a wall when you can't reply, fill the gap with the words you were missing. talk again.
you learn just like an AI model learns
AI model starts randomly, and the loss is very high at the beginning. the model gradually starts taking steps, not knowing which one is right until the loss tells it.
the wrong step is information too, the wrong step tells you which way NOT to go. and the right step decreases the loss and makes the model more accurate.
the model keeps nudging and making changes in it, it keeps learning. BUT the loss never goes to ZERO in real life. and if it does go to zero on the data it's learning from, we consider that BAD in machine learning, we call it over fitting, or memorization in simple words.
what do you understand from this?
YOU'LL NEVER LEARN ENOUGH. LOSS NEVER GOES TO ZERO.
the confusion you face in the beginning is just high loss. it's completely normal. it's NOT normal if you aren't confused, that means, you aren't pushing yourself hard enough.
the first days of “i understand nothing” are what the loop feels like from inside.
the 1:1 rule
"the purpose of knowledge is action, not knowledge." ~ Aristotle
you should never consume more than you build or apply.
keep a strict 1:1 (ideally 1:2) ratio while you consume. 1x consuming should mean at least 1x building something around it or applying it.
value practical knowledge more than just theoretical knowledge.
consumption with no application is tutorial hell, a model that only ever sees data and never computes a loss.
the resources and topics can be random, but the ratio of ingestion to application has to be very strict.
or else... you'll be in a delusion where you think you understand something, but deep down you are shallow.
AI's role in your learning (dont overuse it)
hit walls. dont run away from them. because that's where learning happens.
AI is your honest 3 AM friend.
you hit a wall, you go to AI and ask it to explain it to you according to your situation and your life ("explain it to me like I am a footballer").
and this is what almost no teacher can bring to you.
at your time, for free, in your own way, simplified. that's AI for you.
dont ask AI to hold your hand everywhere, that's not the way. the right way is to F*CK AROUND and come to AI when you are into a problem or when you lack information.
also, AI sometimes over simplifies things so much that the main details get lost. use it as leverage, not as your only learning tool. your only real learning tool is building.
the explain test
"if you can't explain it simply, you don't understand it well enough."
this has been the biggest weapon in my self learning journey so far.
as soon as you learn something, build something, try to explain it to others.
it can be your parents, friends, close friend, girlfriend, boyfriend, or your little brother, or maybe to strangers online.
I chose "strangers online", started writing articles, tweets, started recording myself explaining the things I learned.
how does this help?
it shows you the gaps that are still not filled by building, because explaining something requires extreme understanding of the thing.
this way you'll fill up the remaining gaps, and that's what we want. F*CK AROUND AND FILL AS MANY GAPS AS WE CAN (like the AI tries to reduce the loss as much as it can).
for eg. last week I wanted to decode and learn about the physics behind the new tesla roadster, so I dived into it, ingested a bit, and then started recording myself explaining it to the camera point by point. I had the bullet points, I just ingested and explained it in my own words.
right thing, wrong time
many times, you'll go to the right resource at the wrong time.
for eg. I went to CS229 (math heavy lecs) before I trained a neural net, and it was very boring for me, I would have quit ML long back, if I had not switched to CS231n (a more practical lec series), where I could actually build and learn things.
this is why I feel, roadmaps are the worst things. because not everyone learns the same way. some love doing the math before getting the feeling of building, some like to build and figure out the math along the way. math is just an example here.
a roadmap decides your path before you have any loss signal to steer by. and that's why I hate when people ask for roadmaps or when people give other beginners roadmaps.
what to do next?
no plan. I am not going to give you some checklist or roadmap. I just want you to think about the thing you really want to learn and explore.
and within 2 days (ideally right now), start by watching a random video around it, basic or complex doesn't matter, taking the first step does.
just start. ingest, build, hit a wall, fill the gaps. repeat. forever. that's it.
I am Aaryan. 19 - self taught ML guy who’s been actively building in the ML and AI space for 2 years now. I write from the perspective of a builder - breaking down what’s actually happening under the hood of the things everyone is talking about. You can find me at @aaryan_kakad.
