it’s kind of crazy that spaced repetition has completely revolutionized language learning and then not really changed the world in any other way at all. why are there no great scientists who are inhumanly good at remembering the corpus of their field through incremental reading? why are there no insanely good engineers who know every detail of their entire stack through spaced repetition?
oh yeah, that too. but my understanding is this is mostly a “makes you good at test taking” thing. does it make you exceptionally good at medical research or being a doctor?
H1: Anki Doctors misdiagnose more often, because they get stuck with thought patterns like “Ah, sniffly nose, sore throat, must be… Diseasitise”, instead of correctly thinking “Ah, sniffly nose, sore throat… might be Diseasitis, or the cold, or the flu, or...”
H2: Anki Doctors are worse because they over-fixate on interesting/novel diseases that are more salient in an Anki deck, instead of just putting some thought into what would be the most logical fit for the ailment. “Your big toe hurts? Must be super-cancer+”, instead of “Your big toe hurts… [thinks for a bit] did you kick it on something?”
H3: Anki Doctors are better, because they know a larger range of diseases, and can diagnose more broadly. While a normal doctor might exit med school, and start with a broad range of diagnoses they hand out, perhaps as they gain XP, their range shrinks, and they find their ‘go-to’ diagnoses. On the other hand, an Anki Doctor might exit med school, and keep their range broad, as long as they keep practicing their Anki deck.
a really nice thing with spaced repetition is that you haven’t replaced how your memory functions, just augmented it a bit. but it’s still fundamentally your same brain, one with the ability to ask questions like “am I zooming too much/not considering all options?” or “is what’s salient to me actually what’s happening?” doing spaced repetition doesn’t have any bearing here unless it gives you a false sense of confidence in what you know or doing it crowds out developing other skills.
I’d expect H3 to be true, ceteris paribus, but would also expect that having the ability to ask questions like the one above and just think better to be of greater importance, i.e. someone whose thinking habits are more developed and doesn’t use Anki would be a better researcher/doctor than an Anki savant whose thinking habits are underdeveloped.
For non-standard curricula, probably the overhead of making cards? I have memorized lots of details about AI stuff just from reading and thinking about it; I tried using Anki for this but found the overhead of creating and maintaining a card deck to not be worth it.
Imo, best way to make cards is to do it while you study.
Read textbook
Highlight things that are discrete packages of info
Once you’ve finished that section, go back and turn the highlights into Q:A pairs
Q: What is the capital of France?
A: Paris
OCR those written down Q:A’s and have Claude format them into proper Anki format, with a skill (Or, you could vibe code a program to do this. I was mostly adding cards before vibe coding was a thing)
Add the output .md file to anki
This was the best way I found to study head and neck anatomy. Worked especially well if I’d just come out of the cadaver lab, and was already in the right ‘vibe’
I did Anki for ~3 years, and have now—sadly—dropped it. I loved doing Anki, because it made me look/feel smart. But, I think I got more out of writing daily for a month, than doing Anki for a year.
Oh, no way. Hard disagree for me. But Anki is almost exactly the same thing every day, and I struggle to do tasks like that (Cleaning for example).
I found writing every day many times easier. Even though I hacked around with Anki to make it as addictive as I possibly could. It was always something I was sliding off. If I missed a few days, it’d take 30m-1h to catch up, and knowing that I had other work to do, that wasn’t so repetitive… it was hard to do Anki.
Also, I had cards that were like “Who is this k-pop person?” b/c I wanted fun cards in there to keep me motivated, and because I wanted to impress my gf by knowing k-pop people’s names. But that gave me an aversion to doing Anki in public—since people seeing me look at pictures of k-pop people on my phone felt embarrassing. But with writing, I was very willing to do it anywhere and everywhere.
Spaced repetition is good for training declarative (explicit) memory, but procedural (implicit) memory is the more important kind of memory—in science, engineering and life in general.
We can write software to train procedural memory. To learn to touch type for example, most people use an application for that purpose. But you cannot add cards to typing-training software that would make the software useful for learning something other than touch typing: you would need to write an entirely different application.
There isn’t software for training most of the procedural skills people try to acquire because of the sheer variety in the procedural skills people want to learn and because developing software is expensive and because procedural knowledge is acquired automatically in the process of doing things and because most people are intrinsically skilled at choosing things to do (e.g., things just beyond their current ability) that maximize their learning rate (without the aid of software).
Actually, let me qualify that last statement as applying only to the situation before about 2024: since then I wouldn’t claim it impossible for someone somewhere to have used AI to create software than can be used to train a large fraction of all of the skills people want to acquire.
so I’ve been using spaced repetition for ~1.5 y. now in my CS degree and life in general. it’s a great solution for lots of problems: no relearning what I understood in class two weeks ago; a cheap way to incrementally refine my understanding over time on topics I haven’t worked on directly recently (which is most of them); the rate of forgetting is slowed down, etc. but to use spaced repetition this way requires learning how to do so, it’s not like with language where you can download Anki and a premade deck and you’re good to go. there are also practical problems:
why are there no great scientists who are inhumanly good at remembering the corpus of their field through incremental reading?
my current prompt writing style is kind of similar to incremental reading, in the sense that I prioritize close cards written quickly over well-written Q/A cards. but even writing in this way is costly. for a 3h lecture I will write ~50-200 cards. even when writing as quickly as possible, almost automatically turning notes into cards, it takes ~30m. then tomorrow morning it takes ~30m to revise them + ~15m to revise the due cards. and that’s just one lecture. then there is the spread out cost of actually integrating that knowledge with other things I know, updating other understandings, etc. I don’t have any estimate for the time that takes up, but it’s definitely not free. so I doubt that it’s possible to actually remember the entirety of a field. there’s just too much information, and you need to chew it out to be able to make use of it.
why are there no insanely good engineers who know every detail of their entire stack through spaced repetition?
similar story to the one above. plus I find that little idiosyncratic details which don’t fit neatly into a wider picture are the most difficult to remember, thereby being the most time-consuming.
all that is to say that spaced repetition makes learning more efficient by solving some memory-related problems, but not so much more efficient that you can just download an entire field into your brain in a manner of months.
it’s kind of crazy that spaced repetition has completely revolutionized language learning and then not really changed the world in any other way at all. why are there no great scientists who are inhumanly good at remembering the corpus of their field through incremental reading? why are there no insanely good engineers who know every detail of their entire stack through spaced repetition?
Anki is super popular in med school, so I’ve heard
oh yeah, that too. but my understanding is this is mostly a “makes you good at test taking” thing. does it make you exceptionally good at medical research or being a doctor?
H1: Anki Doctors misdiagnose more often, because they get stuck with thought patterns like “Ah, sniffly nose, sore throat, must be… Diseasitise”, instead of correctly thinking “Ah, sniffly nose, sore throat… might be Diseasitis, or the cold, or the flu, or...”
H2: Anki Doctors are worse because they over-fixate on interesting/novel diseases that are more salient in an Anki deck, instead of just putting some thought into what would be the most logical fit for the ailment. “Your big toe hurts? Must be super-cancer+”, instead of “Your big toe hurts… [thinks for a bit] did you kick it on something?”
H3: Anki Doctors are better, because they know a larger range of diseases, and can diagnose more broadly. While a normal doctor might exit med school, and start with a broad range of diagnoses they hand out, perhaps as they gain XP, their range shrinks, and they find their ‘go-to’ diagnoses. On the other hand, an Anki Doctor might exit med school, and keep their range broad, as long as they keep practicing their Anki deck.
a really nice thing with spaced repetition is that you haven’t replaced how your memory functions, just augmented it a bit. but it’s still fundamentally your same brain, one with the ability to ask questions like “am I zooming too much/not considering all options?” or “is what’s salient to me actually what’s happening?” doing spaced repetition doesn’t have any bearing here unless it gives you a false sense of confidence in what you know or doing it crowds out developing other skills.
I’d expect H3 to be true, ceteris paribus, but would also expect that having the ability to ask questions like the one above and just think better to be of greater importance, i.e. someone whose thinking habits are more developed and doesn’t use Anki would be a better researcher/doctor than an Anki savant whose thinking habits are underdeveloped.
For non-standard curricula, probably the overhead of making cards? I have memorized lots of details about AI stuff just from reading and thinking about it; I tried using Anki for this but found the overhead of creating and maintaining a card deck to not be worth it.
Imo, best way to make cards is to do it while you study.
Read textbook
Highlight things that are discrete packages of info
Once you’ve finished that section, go back and turn the highlights into Q:A pairs
Q: What is the capital of France?
A: Paris
OCR those written down Q:A’s and have Claude format them into proper Anki format, with a skill (Or, you could vibe code a program to do this. I was mostly adding cards before vibe coding was a thing)
Add the output .md file to anki
This was the best way I found to study head and neck anatomy. Worked especially well if I’d just come out of the cadaver lab, and was already in the right ‘vibe’
I did Anki for ~3 years, and have now—sadly—dropped it. I loved doing Anki, because it made me look/feel smart. But, I think I got more out of writing daily for a month, than doing Anki for a year.
surely writing every day for a month costs a lot more willpower?
Oh, no way. Hard disagree for me. But Anki is almost exactly the same thing every day, and I struggle to do tasks like that (Cleaning for example).
I found writing every day many times easier. Even though I hacked around with Anki to make it as addictive as I possibly could. It was always something I was sliding off. If I missed a few days, it’d take 30m-1h to catch up, and knowing that I had other work to do, that wasn’t so repetitive… it was hard to do Anki.
Also, I had cards that were like “Who is this k-pop person?” b/c I wanted fun cards in there to keep me motivated, and because I wanted to impress my gf by knowing k-pop people’s names. But that gave me an aversion to doing Anki in public—since people seeing me look at pictures of k-pop people on my phone felt embarrassing. But with writing, I was very willing to do it anywhere and everywhere.
Spaced repetition is good for training declarative (explicit) memory, but procedural (implicit) memory is the more important kind of memory—in science, engineering and life in general.
We can write software to train procedural memory. To learn to touch type for example, most people use an application for that purpose. But you cannot add cards to typing-training software that would make the software useful for learning something other than touch typing: you would need to write an entirely different application.
There isn’t software for training most of the procedural skills people try to acquire because of the sheer variety in the procedural skills people want to learn and because developing software is expensive and because procedural knowledge is acquired automatically in the process of doing things and because most people are intrinsically skilled at choosing things to do (e.g., things just beyond their current ability) that maximize their learning rate (without the aid of software).
Actually, let me qualify that last statement as applying only to the situation before about 2024: since then I wouldn’t claim it impossible for someone somewhere to have used AI to create software than can be used to train a large fraction of all of the skills people want to acquire.
so I’ve been using spaced repetition for ~1.5 y. now in my CS degree and life in general. it’s a great solution for lots of problems: no relearning what I understood in class two weeks ago; a cheap way to incrementally refine my understanding over time on topics I haven’t worked on directly recently (which is most of them); the rate of forgetting is slowed down, etc. but to use spaced repetition this way requires learning how to do so, it’s not like with language where you can download Anki and a premade deck and you’re good to go. there are also practical problems:
my current prompt writing style is kind of similar to incremental reading, in the sense that I prioritize close cards written quickly over well-written Q/A cards. but even writing in this way is costly. for a 3h lecture I will write ~50-200 cards. even when writing as quickly as possible, almost automatically turning notes into cards, it takes ~30m. then tomorrow morning it takes ~30m to revise them + ~15m to revise the due cards. and that’s just one lecture. then there is the spread out cost of actually integrating that knowledge with other things I know, updating other understandings, etc. I don’t have any estimate for the time that takes up, but it’s definitely not free. so I doubt that it’s possible to actually remember the entirety of a field. there’s just too much information, and you need to chew it out to be able to make use of it.
similar story to the one above. plus I find that little idiosyncratic details which don’t fit neatly into a wider picture are the most difficult to remember, thereby being the most time-consuming.
all that is to say that spaced repetition makes learning more efficient by solving some memory-related problems, but not so much more efficient that you can just download an entire field into your brain in a manner of months.
How exactly did spaced repetition revolutionize language learning?
I had assumed most elite engineers are using spaced repetition?
I’m not elite but life would be much more difficult without it.
i find it’s much less useful for math/CS than languages.