I was anti-memorization until I went back to graduate school for mathematics. I had forgotten (or never learned) a lot of things needed to pass qualifying exams. At some point I ran across the spaced repetition idea (maybe from the Wired SuperMemo article [0]) and I gave it a try. I ended up using it to memorize large portions of baby Rudin and Munkres' Topology, as well as some algebra and a bunch of qualifying exam questions.
The qualifying exams were difficult until I reached some "critical mass" of knowledge. Then I could regurgitate proofs and even attack novel problems easily.
There's an analogy here somewhere to the "leetcode" style of software engineer interview. On one hand qualifying exams and leetcode questions are a stupid gatekeeping mechanism, but on the other hand the best researchers/engineers I know have a huge number of facts and examples memorized and ready at their fingertips. I didn't think I needed to do so, but perhaps there is something to suffering through the rote memorization phase to make what comes next that much easier.
That Supermemo article in Wired hooked me. I still remember where I was when I read it and the feeling of reading it. I still use Supermemo daily, and it's one of the programs keeping on Windows. Anki (and every other SRS program I've tried) just doesn't compare as soon as you move beyond a list of flashcards.
That said, having used Supermemo for over a decade at this point, the hardest thing about SRS is deciding what's actually worth reviewing for a long period of time. I delete (really remove from repetitions) cards from my collection almost on a daily basis.
There's a lot of stuff that seems really important that I just didn't care about after even three months.
Supermemo's incremental reading basically lets you schedule chunks of text or images (alleged video too) like a flashcard from Anki. So, instead of bookmarking articles and never reading them, I can put them into Supermemo and know I'll eventually review it.
It basically counts as a separate type of flashcard, but all your reviews are mixed by default. So on a typical day, I'll have maybe 20 flashcards to review, and then another 10-20 articles.
Supermemo saves where you last were reading, so when I get bored of an article, I just hit next and go to the next one. Eventually, you'll process an article down to individual flashcards like you'd put in Anki, or remove it from your review process altogether. Also, you can just leave the entire article in there if you like rereading it.
What's your use case? In my case, obsidian + anki seems to do very well (I can keep a decent amount of context in my cards) for math/CS topics.
Can you also give an example how you benefit from your workflow, and say a bit more why Anki wouldn't be sufficient? Wouldn't keeping a track of articles and making cards in Anki achieve the same purpose, or is there something else? Of course the implementation matters, and having a system for paper-reading and revising by itself is a nice feature.
According to the wired article, the key to remembering is reviewing the information the moment you’re about to forget. It makes it sound like the scheduled review times are designed to maximize the effect.
Never tried it but all the comments made me curious.
Yeah I agree. Anki has a failure mode for me that I eventually accumulate > 5000 flash cards and a review session can take an hour or more. Knowing what to review is really difficult. I'll give Supermemo a shot, I've always heard of it as the gold standard but never tried it.
> Yeah I agree. Anki has a failure mode for me that I eventually accumulate >5000 flash cards and a review session can take an hour or more.
You don't really need to keep up with Anki's review sessions, even a partial session is very effective when you have lots and lots of cards. The tradeoff is that you might not actually maintain 100% recall of all items, but you'll recall most of them, and the scheduling algorithm will space out the items that you do recall even further.
I'm familiar with that problem, and overcame it by really working at making good cards. Today, I have 6000 cards in my collection, and daily reviews take 15-20 minutes; usually 100-120 cards. (I have been using Anki for a lost time. My collection is almost 10 years old.)
Granted, writing good cards is easy to say and really hard to do. Some specific pointers:
- cards should take less than 10 seconds to answer
- most questions should be 7 words or so
- if your question has multiple clauses, split those up into separate cards
- cloze deletions are great, consider multiple clozes per card
- around 10% of your cards should have images
- good images help a lot, mediocre images make it worse -- when in doubt, don't
- Answer this Question cards should have an opposite Question for this Answer like Jeopardy
In my experience, you really only start to see these problems at the tens of thousands of flashcards level. I have ~70,000 mature flashcards and also have the growing backlog problem, even if I don't add new cards for weeks. My collection is 8 years old.
I agree that making good flashcards helps alot. Another big thing is to make sure the knowledge graph is connected, eg there are no 'orphaned' individual cards / groups of cards. Those tend to suffer seriously from decay for me.
I'm working on a program that solves this problem: we don't have mandatory daily reviews and we also have a Spaced Repetition Algorithm for reviews. Also it's much more flexible than Anki & Supermemo. So you don't have to stress over as to complete the daily grind.
Yo, I've been using SuperMemo for 16 years every single day (Did my cards this morning :), and you're spot on! I would love to chit chat about SM sometime, I've not met many people that use SRS for so long! Send me an e-mail at thesupermemoguy (at) gmail (dot) com
No it's primarily an awful Delphi for Windows desktop application that is extremely fragile.
But once you know how to "hold it correctly" it will be your companion for life.
I run it in a virtual machine with all its legacy dependencies like IE. Forget about running it on Wine, it barely works on any version of Windows (lol)
What's https://www.supermemo.com/ then? I just signed up to give it a go (requires CC #, but free for a month). Still haven't found good spaced repetition vocab learning site that works well for me.
Ah OK, though I'm still curious how there can also be an apparently unrelated online offering at supermemo.com. FWIW, from the very brief experience I've had so far I'm not super impressed, but I'll try sticking with it for at least the month of the free trial.
they are working together. supermemo.com is based on the same algorithm but is designed to be more mainstream (no advanced stuff like incremental reading & hard to make your own cards).You can buy high quality language courses there(no need to make all the cards). I have both since supermemo.com has an app for the phone.
If actually using it has this huge friction, how do you manage to use it every day? I always have a hard time creating habits out things that are hard to "start" doing every time.
I remember the exact point in my life when I realized how much difference rote memorization makes.
I had the opportunity to write some MC68000 assembly code, at a time which was not too long after having written a complete emulator for it in C. Writing the emulator required me becoming familiar with every single instruction in all of its nuances, and exactly what they all do.
So, having that behind me, I sat down to write this code and it was like wow ... I could just spew the code without having to look anything up. It was so easy!
> I remember the exact point in my life when I realized how much difference rote memorization makes.
I suspect this might have something to do with neurological development. At least in my experience. Something about memory and recollection "clicked" (not a skill but more a capacity) at a really adult age for me.
Almost similar to being taught calculus after not knowing how most physics formula are derived and then looking back in confusion how you struggled with a straightforward thing.
I'd be really interested to see some research into integrating spaced repetition into our actual education system. Almost everything I see about it is adults learning, I wonder how much we could speed up primary school education considering so much of the "base" stuff needed to advance is rote memorization anyway
seems like countries should be investing money in this, potentially trillions in unlocked economic potential by improving and speeding up education. I think I read something like 20-30% of medical school students use SRS, yet only a fraction of the general population uses it. Insane to me that we have a tool like this and almost nothing is being done to improve adoption.
>see some research...spaced repetition into our actual education system.<
I have followed the research in this area closely for several years. Most of the (excellent) foundational research was done by university psychology professors who used college students as subjects; it's much harder for them to use primary school students, hence far less research with young learners. Implementing these techniques in elementary schools is more challenging then many would guess. Two good books: "Make it Stick" Brown, Roediger, McDaniel; "Powerful Teaching" Agarwal & Bain. PM me if you'd like more info. The second book describes research with middle school students. Pooja Agarwal also hosts the informative site www.retrievalpractice.org which informs K-12 teachers and has lots of free downloads. I am involved in a startup that is beginning (pilot teachers in September) to implement spaced repetition for elementary school math.
I'm interested in this, but you don't have any contact info in your profile. HN itself doesn't have a messaging feature, so you have to leave an email address or something there. If you'd rather just message me my address is in my profile.
No research here, but I use the great Fresh Cards app (by allenu here on HN) with my daughter's homeschooling, and it works great. We've used it for phonics and all sorts of basic math facts, and I use it for scheduling spelling word review as well. We're about to start teaching science and I anticipate using it a lot there as well.
Hundreds of millions of the world's kids were remote learning for up to a year and a half. If there was any opportunity to develop real tools to help with remote learning, it was then, but we ended up with nothing, and remote learning still sucks.
If anybody is going to integrate memory techniques into a curriculum, it'll probably be some Silicon Valley charter.
Given that so much of modern education happens in large-group settings, it seems hard to apply SRS principles to it. A group of students will be a lot farther from 100% recall of all items than any individual student might be, and the items that require review will correspondingly differ.
Spaced repetition could replace time currently wasted on homework. As a kid I would have much preferred to sit in front of Anki instead of being laden with hours of homework, and it probably would have done me more good over the long term.
> but on the other hand the best researchers/engineers I know have a huge number of facts and examples memorized and ready at their fingertips
The US is staunchly anti-memorization. It wasn't until I did a year abroad in a French university where the power of memorization was so obvious. One of the math students I spent time with told me it's the French tradition to "carry everything around in your head". They were studying symplectic geometry in their sophomore year of their bachelor's!
I don't understand where this coming from. In my experience learning and teaching math in the US, I've found it's almost _entirely_ about memorization. Calculus students are usually able to regurgitate algorithms for differentiation/integration, but can't answer the most basic questions about what they're doing really means.
This is just an aside. I'm not against memorization. I think memorization is the first step necessary for deeper internalized understanding since it lets you take the basic stuff and put it on autopilot so you can think about the rest that isn't as basic. That said my main complaint about US teaching of math is that that second step seems rarely to be taken and instead focus entirely on the memorization.
I guess our experiences are both anecdotal, but I'm having trouble wrapping my head around your idea that the US is anti-memorization.
This might be the case for k-12 and maybe lower level courses, but this was _definitely_ not the case for when I was taking group theory, real analysis, etc.
With that said, someone else in the comments mentioned that while it might seem there's an emphasis "regurgitation" it's mainly because going beyond that and really digging into the why of things is vastly out of scope for the class.
In grad school it seems obvious what you want to retain, after that it can become a bit trickier to make the choice. General question for everyone using this approach: how do you go about selecting topics to put into your SR schedule?
I mean there are so many things that would be good to remember: math problems and leetcode are actually continuously relevant if you're in employment and ever thinking about moving jobs, software engineering tools that are important for daily work (those are changing so fast in some cases that I'm wondering again how much is worth committing to long-term memory), maybe there are some academic fields where you want to stay up-to-date on the research, all sorts of life hacks in the household / cooking / DIY space, literally everything about your family life, maybe some strategies in hobbies (just recently got into speedcubing, which is a lot about algorithm memorization),...
Maybe the problem is just that I have too much of a "finite memory" mindset and we can easily remember all of those, lol.
The problem is quickly solved, and it is important not to fall into " paralysis by analysis." Sometimes, while writing a card, we tell ourselves that it is not so important to remember that information and discard it, other times we discard it when during the review process we sense or realize that we are not interested in remembering that information .
Assuming that there is some sort of logic behind the decisions made, I recommend using the same strategy that Napoleon is said to have used in his campaigns: first we start, then we will see.
Yeah, the whole thing is probably easy to overoptimize. Any recommendations wrt to the software? My preference for these sort of things are maximally compatible/portable solutions based on simplistic file formats (I keep a lot of personal tracking data in csv, eg). Here I'm thinking something along the lines of local markdown files for the cards, a database for the sampling info and a web app that does the sampling.
I use Anki in a "straightforward" way. In the sense that I write the card using Anki's interface without using excel or markdown files.
For example, if I want to memorize a word I found while reading, I highlight it in blue in Apple's books, it goes to readwise, then I take it from readwise by copying and pasting from Apple's dictionary the translation or definition.
I am for the simplest things possible, the point is to learn/memorize and not to find the best pipeline. You may think that the optimal pipeline leads to optimal learning, but (1) what kind of difference are we talking about compared to the simplest method of building cards?, (2) if setting up the pipeline means having the pipeline as the goal (as in ML the platform and not the predictions become the goal) or abandoning the study because the whole process becomes too cumbersome, is that something we should aspire to?
One of my favorite novelists, Arturo Pérez-Reverte, when asked what software he uses to write his novels, replied: "I only use Word, but it's true that I don't know any other [software]".
I found this post fascinating. It makes me wonder is there an equivalent superpower that's more software engineering focused rather than coding focused like leetcode questions? I know everyone here likes to dump on Gang of Four, but perhaps design patterns and their usecases?
If I'm being honest, I went overboard. Like I said, I literally memorized entire books. Entering cards into the system was the most difficult part. But it absolutely works! Had I stayed in academia, I would have loved to developed a set of flash cards to go along with course material.
To really understand something first you blindly memorize - but that is not enough, once you do examples and exercises using what you blindly memorized you get to understand things quicker, a lot quicker.
Not memorizing stuff and figuring things as you go is mostly recipe for disappointment.
Like in chess - people think that chess players are somehow super intelligent - but being super intelligent without rote memorization of loads of chess settings will not help winning grand master title.
There's actually an interesting section of the book "Moonwalking with Einstein" where it talks about a study was done on chess masters where they showed them the board in positions that would be impossible under the rules and suddenly the master chess players didn't do all that much better than random people.
The suggestion in the book was that really Grandmasters have spent so much time practicing, that they have memorized the game and the board to a certain extant that allows them to more easily handle the board and all the pieces on it cognitively.
Off-topic, but I dislike chess and adore chess960/fischerrandom chess for essentially this reason. It's almost disheartening enough to want a different hobby when you review an online game and realize what you thought was a clever solution to an interesting "puzzle" of a board position was really just one you learned by experience a week ago. More variety in piece arrangement makes playing feel far more like doing chess problem solving than remembering the last time you messed up the same chess problem.
Memorization only works well if your approach leads to chunking. They have done studies on chess players specifically, which supports what you are saying. Masters are no better than a beginner at memorizing a random chess board. This critical mass of memorization, when finally encoded into chunks is what I think a lot of people are describing as the “aha” of memorizing to conquer complex topics. So it isn’t really the blind memorization, but the process of learning and getting information and concepts into coherent and related chunks.
It similar to the question, "if you could do just one exercise, what would you do?" And you have all these answers, it is the squat, no, it is the deadlift, please guys it is the power clean.
It is not a useful question because you will never be in the position of choosing just one exercise to be done for the next month, year or decade.
Memorization is important and understanding is important, and the two are not in any conflict whatsoever.
You are making a very important assumption that doesn't hold in most situations.
Chess and math are regular domains. They are bounded. There is right and wrong. Most domains are not like this. Ex: art, software engineering, investing, etc.
You can't memorize your way to understanding in an irregular domain. It also doesn't tend to do well in non-knowledge work domains. Ex: sports.
I disagree becuse rote repetition is the only way to learn how to draw. One cannot simply draw something complex without trying out again and again to draw something.
For sports rote repetition is basis to excel or at least getting good at any sport.
I agree one cannot "memorize to understanding" because one has to actually apply what he memorized in practice.
I've picked up a lot of (human) languages over my life, and I've used SRS to great effect (I started on JMemorize, a now defunct Java app.) I have a simple strategy:
1. Read a text
2. Lookup unknown word/grammatical pattern
3. Create a flashcard for it
4. Apply SRS
and it works great. I've also tried to use other people's Anki decks and they've never worked particularly well. Personally engaging with the material is still the prerequisite for memorizing it, but memorizing means you don't need to struggle to figure out basic concepts constantly and can instead move onto the higher level of meaning.
I've used SRS over the years for many, many things. I've used it for memorizing divisibility rules, used it for annoying math lemmas, used it for data structures, and more. Each of the time I've attempted the material by myself and then turned my knowledge/engagement into a flashcard. I've even considered using it to learn tools like Blender so I can dial in workflows.
All of the courses that I did really well in I memorised the entirety of the book that was the core of the course. Everywhere where I didn't feel like doing it, I was strictly mediocre. If you think you can understand something without knowing all about it, I must say I think you are mistaken.
My (entirely uninformed… though I am a “doctor” har har) theory is that some sort of compression occurs in the brain. I also wrote out the answers to my flash cards by hand. I feel like these actions accelerate the compression process.
I've really understood a lot of stuff that I've since completely forgotten. For example, if you don't speak your native language for 20 years, you'll be surprised how hard it is to find any of it when you need it.
I imagine it's possible that they are related. In my life, I've only witnessed it as a required for being an expert at any topic: experience and memorized knowledge, of a topic.
The qualifying exams were difficult until I reached some "critical mass" of knowledge. Then I could regurgitate proofs and even attack novel problems easily.
There's an analogy here somewhere to the "leetcode" style of software engineer interview. On one hand qualifying exams and leetcode questions are a stupid gatekeeping mechanism, but on the other hand the best researchers/engineers I know have a huge number of facts and examples memorized and ready at their fingertips. I didn't think I needed to do so, but perhaps there is something to suffering through the rote memorization phase to make what comes next that much easier.
[0] https://www.wired.com/2008/04/ff-wozniak/