Beyond Rote Memorization: Why I Love Flashcards for Deep Learning

Two really good modern books on how we actual learn are.

1. The ABCs of How We Learn by Daniel L. Schwartz

2. How Learning Happens 2nd by Paul A. Kirschner & Carl Hendrick

I can really recommend them. Now back to flashcards and active recall. They are the secret sauce. People in the thread are saying that understanding is important, while I agree you should understand things its not THAT important.

Important is that you remember not only *facts* what something is. But also how to do something. If you stare at a blank page not being able to do something like writing a a simple typedef struct in C. But knowing what a typedef and struct is, you are missing procedural knowledge.

If you can only programming with an IDE giving you cues like autocompletion you will never form the memory for the actual symbol. Only when you active recall something it will be rememberd until it becomes automatic freeing your working memory from the load and reducing cognitive load overall.

I strongly believe that an AI rephrasing your backside and giving some additional knowledge from your own notes is helpful. You should still write your own notes and flashcards, an AI can help here too but just dont copy and paste them.

Learning can be so much, creating mental models and checking their validity, creating anecdotes, connecting new knowledge and ideas to already learned ones. Have fun!

There are many people that start using anki or flashcards only to stop doing them after a month.

Another interesting […]

I do a really lightweight version of flash cards. Everytime I'm learning a new tool or tech, I grab oversized notecards (my favorite are 8x5" dot-grid cards). I put a label at the top, and create bullet points of each item i want to remember. I then review. No individual cards for each item or anything. Just all the things grouped on one card as bullet points.

For example, I'll have a `sqlite` card, and put all the commands and everything on it, as I learn them. I'll use it as a cheatsheet, but then also a few minutes of mindful review. This for the toolings that I want to know well enough to not get slowed down googling the commands. I do this for a lot of CLI tools, but also things I need to remember about the business of my company and working across group, etc....

Eventually the five or six working cards I have, get put on a pile and new ones come in.....

> Most of my cards are handwritten by me and accommodate my brain.

I'm guessing from all the talk about Anki that people are missing this part. I think these frictionless, digital replacements for note cards are solving the wrong problem. You're never going to spend more time on a given flashcard, and better absorb the information on it, than the first time, when you're sitting there making the thing. Making this step as brief as possible is a bad idea.

My sense is the same as the authors that LLMs + an Anki-like -> mediocrity. But I feel like there ought to be such potential there! Even just things as simple as rewording the same question a bunch of equivalent ways to avoid recognizing the sentence structure...

Downside for AI potential as a whole, framed broadly: We don't seem to be good enough yet at identifying what friction is functional and what we should strive to automate/eliminate.

> Some people also use LLMs to generate flashcards. And of course, the result will be those impersonal, mediocre cards.

> I won’t say LLMs are useless for this. But from my trials, I get about 1 card that’s useful to me out of 10, and even that 1 card still needs rewriting.

I don't know the specifics of how the author tried to do so, but from what I've seen the majority of attempts are, let me drop a chapter of a textbook and say "make flashcards." If that is what we are talking about, then yes, LLMs are useless.

In my mind, though, this is sort of like looking at the very first GitHub Copilot LLM autocomplete from a couple of years ago and concluding, yeah it's nice for one-liners, but it cannot write an app.

If you create a framework around your card-creation AI so that it can use tools, and verify its work to ensure common card-creation pitfalls don't happen, you can get pretty high-quality cards. In my experience, you go from a 10~20% hit rate to a ~90% hit rate, which in my mind is good enough. I got to ~75% quality just from a two extra LLM calls that would assess a potential card against a standard set of rules (adapted from [0]). There are huge Pareto gains to be had here.

I've generated thousdands of cards over the last few months this way. I let the AI add it directly to Anki via AnkiConnect. Then, if when I go to review I find a card that my AI created and I don't like it, I just delete it.

Removing the limitation of card creation is really quite compelling, and I […]

> From this perspective, fields that require deep understanding, like math, require memory just as fields with a breadth of shallow knowledge do, though in different ways.

I'm interested in understanding how others use Anki for conceptual subjects like pure math or physics. I believe many fundamental rules in Spaced Repetition (e.g. like keeping cards concise) are thrown out the window for conceptual subjects.

The best advice I've ever read about flash cards is if you are dreading to review, because you are forgetting your cards or they are too complicated, you are writing cards wrong. Learning is supposed to be fun! Also, Common Core should ship Anki decks. I seriously think so many problems with education stem from students not realizing that memorizing is actually very easy with FSRS, and thus struggling and hating learning.