beginner's mind
one of the qualities i've admired in people i've come across on twitter is their ability to be a beginner. there are different ways to learn something, but what i feel that the most important skill is being able to focus on small parts of the whole and think about them deeply enough to realise you don't understand them at all. then you can start from there and fill in your gaps.
in the information age, it seems difficult to start from "scratch" because no one wants to reinvent the wheel. what we often prefer to do is reverse-engineer our learning: take something up and try to update our knowledge as we move from the top to the bottom.
there might be cases where you need to learn topics which are connected to what you are doing but necessarily does not backtrack to it. these are special cases but such things come with experience once you begin to know good-enough about what you are doing.
does this approach work all the time? i'm not sure. but the very idea of understanding things takes a lot of courage. asking questions is easy, but finding their answers is hard. it does take effort.
if i want to understand diffusion models, i can ask chatgpt about the building blocks of current state-of-the-art diffusion models. i can pick a block and ask it to explain it further. all just a prompt away. but there's a downside: i will forget about it the next day.
this might sound counterintuitive, but it's what i've felt about learning from chatgpt over time. you can't truly learn anything if you don't understand its history, where it's coming from, and where it might go. this helps us better appreciate what we are learning and, therefore, helps us remember it for longer duration.
moreover, a mathematical approach (ik a lot of you don't like it) gives you a good feel of what is going on internally. it help you visualise the ideas better. maths actually helps us get our hands dirty and make us play on the level field rather than looking from outside. i think coding does the same when we want to tinker with what we have learned and test our understanding and update our own prediction models.
i'm no expert on learning, though.i've realized lately that simply looking at something on a board and writing it down doesn't help much either. what really helps me is writing things in my own words; it helps me understand the gaps in my own concepts and well as understand how stupid i am.
(yes, feynman, you are right.)