Pleased to see this posted here again. Nielsen's writing on Anki and memory systems more broadly has been nothing short of incisive, inspiring and paradigm-shifting in contributing to the ongoing "hermeneutic turn" within the community. Whilst this is not as evident as with his later essays, ATLM lays the foundations for some of the most important points for beginners of using the tool: a focus on coherence, disambiguation, and crystallisation as virtues of an Ankification practice; and above all shifting the conversation from it as a miserable prosthetic to memorise arbitrary associations to a genuinely beautiful tool that you can use to shape and change what you know, are and do.
The former interpretation's popularisation does unfortunately seem to be the outcome of it being embedded within the epistemic ecologies of language learning and medical school, two situations where Anki is almost inevitable if one is to succeed, but also where its unique power of meaning making is most neglected. As he writes in "Building a better memory system" [1], targeting the audience of the creative expert seems to be where the most tractability in expanding and improving a mnemonic practice lies. I hope such progress continues in the vein of current open-source and community-oriented advancements such as FSRS - the best feature of which is making switching from SuperMemo all the more obvious ;)
I don't get where the "miserable prosthetic" thing comes from. Teen me discovered spaced repetition around 10 years ago and considered it absolutely revolutionary even then - an actual way to combat having a bad memory!
It definitely is! It's unfortunate that what seems to have developed to be the consensus view on SRS within most language learning/high-load traditional academic circles (>90% of the userbase, I'd imagine) that it is, at best, a very useful but unfortunate inevitability, or at worse a torturous and exhausting experience. That obviously doesn't apply to everyone; factors like card design, course content/interest (the passion element which Nielsen rightly identifies as integral) and good old mindset and habituation all vary within those communities and beyond; but I think it is fair to say it's the majority opinion, particularly for those just starting to use the software. Some of that difficulty can be overcome by being patient with the obtuseness I think, but at the end of the day it helps to at least somewhat enjoy what you're studying! Anki can help you love a subject you like, but it usually can't help you like one you hate.
I'd switch to FSRS, but the short-term workload increase is too much for me. I know it can be avpided by only enabling FARS for new decks and then gradually shifting over the entire collection, but that I don't think I'd like my collection fragmented over multiple spacing algorithms.
For now, Anki's v3 algorithm is good enough for me. The only thing I didn't like about v2 was the lack of buzz factor which meant that a lot of closely related flashcards showed up in the same reviewing session, making review of such cards slightly ineffective.
If you enjoyed the OP, then Andy Matuschak’s article on how to write good prompts (cards) should also be interesting for you. And if you want to learn more about the research behind spaced repetition, then Gwern’s article is pretty good.
20 rules for formulating knowledge [1] are also very useful. They were written by the creator of SuperMemo, which to my knowledge developed a lot of the spaced repetition algorithms.
For example Anki, the software talked about in the article, uses the SM-2 algorithm by default which was developed by SuperMemo.
While SuperMemo is at SM-18 in meantime and also much ahead of Anki in other features like incremental reading (which in my experience is its real super power, not the spaced repetition as such). Although there might be plugins for Anki that offer something similar.
That's one of my major gripes with Anki, it's just a bit too brittle for my taste. SuperMemo is not without its quirks (I mean, it's a quirky niche software to begin with), but all in all, I never had any major problem with it and I've been using it for more than a decade now.
I think for most people who want to get started with SRS, Anki is just fine. A lot of people like to hate its UI but I personally find it perfectly fine (and even good before the recent update where they forced drop shadows on random UI elements). It being available on every major platform is also a major plus point for when one needs to review cards with no access to their workstation.
That being said, the $40 is probably what holds most people back from using SuperMemo. If I had $40 to spend on SRS without feeling guilty I'd 100% use SM over Anki.
For the past few months, I've spent most of my nights and weekends working on an LLM-based learning management system called Reasonote, which aims to hybridize the SRS strategies of Anki with the smoothness of Duolingo. It focuses on a curiosity driven approach, where users can aggregate "Skills" into their library as they participate in Activities. Activities can be anything -- flashcards, quizzes, games, chatrooms -- and can be dynamically generated, or manually created if you prefer.
I'll be releasing more information soon, but since this article's content seems aligned with my mission (build the Young Lady's Illustrated Primer, or something better), I wanted to give it a mention.
If you'd like to beta test, or collaborate, please send me a note directly, explaining what skills you most want to learn -- luke (at) lukebechtel.com
Can you give some example usecases of your application? I wonder how it scales to complex (in terms of structure) information processing, e.g. digesting scientific topics
Anki is great. I am gigging asking 33,000 cards among all of my decks right now, with about 6000 in active rotation. I built what one person called the highest quality Finnish flashcard deck on the Internet, and it was the project that finally convinced me that software was my calling.
Some very scattershot hacker's notes.
- The underlying Anki database is "just" an SQLite database. You can copy it, run Datasette on it, turn on full text search, stream it with Litestream, the works. Anki itself can be very poorly described as a tiny local only web browser-server chimera running one cronjob per entry in this database, written in Rust and JS, which is just phenomenally weird and cool.
- Image Occlusion Enhanced is S-tier. If you don't believe me, use it to memorize the names and locations on the map of your 5 closest restaurants.
- If you can turn it into a CSV, you can import it into Anki, no questions asked. This is the bottom floor for automating Anki stuff.
- If you have what you want to memorize in an HTML table, take a look at running my table2anki (https://github.com/hiAndrewQuinn/table2anki) or writing your own scraper for it. This isn't useful as often as you would think, but sometimes you see a blob of information already ready to go and just want to dive in.
- Don't obsess about the "20 rules of formulating knowledge", or anyone else's way for that matter. They're helpful, but you know what's more helpful? Actually using Anki every day and figuring out for yourself which cards are helpful and which aren't through ruthless practice.
- Once you get good enough at making cards, IME it can basically replace having Obsidian or whatever for taking notes. I have somewhere around 30 books' worth of notes in Anki, and I attribute probably 20% of my overall career's success so far to them, whereas I'd probably attribute like 5% to traditional notes, because I would never actually review them of my own accord if I did that. If those both sound low, I consider the top ten EE and math degree I got to hover around 10%.
- Making time for Anki is the most annoying part of it. Sucking on a nicotine lozenge while doing reviews helps a lot with habit formation once you have crested the difficulty curve of making cards that are actually interesting and helpful to you personally, but you have to make sure to take the lozenge out or chew it up and swallow it the instant you're done. Pavlov knew what he was doing!
- Anki + LISS cardio is a similar match made in heaven. The hardest part is figuring out how to get a screen in a comfortable position for you to watch while you review. Anki actually has a "Controller Mode"you can rig up to help a lot with this.
- This is extremely niche, but: Having trouble remembering a step in a math problem? Do the work out on paper. Take a photo. Image Occlude just the step you're blanking on. Review this regularly. Boom - instant, targeted, low effort high reward deliberate practice in mathematics, tuned to what your own brain reports it is having trouble on.
And some things I have studied successfully with Anki:
- Language acquisition, natch. Many of my software side projects are related to lowering the activation energy of this for myself.
- Abstract algebra theorems and derivations. I ended up getting a perfect score on my final exam and walking out after 20 minutes in college because of this, up from a C- on the midterm.
- Haskell! I made 3 separate attempts in my life to learn Haskell, but only when I committed from the beginning to put every weird symbol and linguistic turn into Anki did it actually stick. I actually get monads now!
- LeetCode problems. Might be the highest ROI 2 week vacation of my life, considering I went from no DSA knowledge whatsoever to having the whole Grind 75 on mental tap. I did the dumbest thing that could possibly work - problem on one side with a link, open the link -> resolve the problem -> mark Good if successful.
Which controller do you use for Anki? Also, I assume the combo with LISS needs to be done on a treadmill? I heard of somebody who used an audio plugin, but my cards are not so audio-friendly.
I actually do them on an exercise bike at home. For the time being I just use an Xbox controller, which I hold while pedaling. LISS cardio is pretty easy to stay in the groove of, so I don't need to actively hold onto the handles to stabilize myself.
Would you care to expand on how Anki helped you with Leetcode problems?
Did you just think of a solution in your head and then compared it with the back of the card?
I put an href on the front side of the card to the problem, and most of the time actually clicked through and redid the problem from scratch (making sure to avert my eyes while I deleted LC's copy of my last attempt). If I felt particularly lazy or confident, I would just mentally sketch out the solution.
Worth noting I also became much more adept at things like list comprehensions and using Counters and DefaultDicts in Python this way. When you type out a solution over and over like that, your brain naturally wants to chunk things down as much as possible.
No. In fact, I had never solved a Leetcode problem before. Every time a card came up that I had never attempted before, I simply flipped it over and typed in whatever answer was on the other side - but only the first time. Each other time I would try to remember as much of the solution as I could.
Eventually I got good enough that I started naturally thinking up solutions to entirely new problems without having to check the answer first, which is what one would expect if they were building new crystallized intelligence by practicing earlier problems.
Having trouble remembering a step in a math problem? Do the work out on paper. Take a photo. Image Occlude just the step you're blanking on. Review this regularly. Boom - instant, targeted, low effort high reward deliberate practice in mathematics, tuned to what your own brain reports it is having trouble on.
This actually might be super useful for helping homeschool my dyscalculic daughter, thanks!
> Why are some memories maintained without repetition?
They aren't. It's just that the repetition might not come from the outside but from within. Or just by way of daily exposure to something. Or by trauma (which you probably want to avoid of course).
But generally speaking, if there is no repetition at the neural level, you will forget a memory.
A lot of that also happens during sleep. E.g. if you go through something again in a dream, this also counts as a repetition.
The former interpretation's popularisation does unfortunately seem to be the outcome of it being embedded within the epistemic ecologies of language learning and medical school, two situations where Anki is almost inevitable if one is to succeed, but also where its unique power of meaning making is most neglected. As he writes in "Building a better memory system" [1], targeting the audience of the creative expert seems to be where the most tractability in expanding and improving a mnemonic practice lies. I hope such progress continues in the vein of current open-source and community-oriented advancements such as FSRS - the best feature of which is making switching from SuperMemo all the more obvious ;)
[1] https://michaelnotebook.com/bbms/index.html