I think that the author was using Anki incorrectly, and that led them to the spurious conclusion that "Anki is dead". I also have attempted to use Anki this way- using someone else's deck to try to force myself to learn something new. But that doesn't work, because it is just memorizing random symbols, as they noted in the article ("The enemy is the static card"). For example in maths learning, memorizing arbitrary terms, symbols, etc is useless. However, once I am introduced to a concept for the first time, then I add it to my Anki deck so I can make sure I remember it. The key is the context, and writing terms / definitions etc that speak specifically to me. I still need to work out different variations of the concept to understand it, and that's not something that Anki can help with.
I haven't used Anki for language learning, but I imagine that if I did, it would be to add some new vocabulary I had just learned from a book, conversation, film, etc. I don't think it would help me learn a language from zero though- that would require practicing it.
In summary, Anki is great for reinforcing something you've just learned, but you can't reinforce your way into the context that is necessary to truly understand something.
Using someone elses deck is such a siren song, and I honestly believe it's detrimental to properly using Anki. Making and curating your own cards is an essential step in learning the new concepts.
Yes, thou the main problem stated in the article ("memorizing the rectangles") is very real.
Now, I don't think you need much magic / LLM stuff to overcome that problem:
1. Instead of the usual "1 new word, 1 card" have 2 - 3 new words on each card, and have each new word in 3 - 4 different cards (ideally with different inflection, meaning, nuance, etc)
2. Review cards fast and very few times each. Like 5 - 8 times / card (max) instead of usual 15 - 20.
3. Don't punish yourself, keep moving on even if you just half-remember. First familiarize and then internalize language patterns instead of just memorize words.
Review intentionally, totally concentrated on the task. 10 mins / day well done >>> 30 mins / day mindlessly.
Outcome: more fun, more effective learning, no memorizing rectangles.
Forty years ago, my French teacher said the easiest way to learn a language is to go live there for a few months. He didn't mention the boy/girlfriend part.
Yeah. Anki (and flashcards in general) are great for helping you remember something _you_ learned (from a book, video, class, etc.). Not for transferring knowledge someone else learned.
Writing my own cards as I'm learning is the only way I've found it effective.
First learn the material. Then use SRS to schedule the practice of effortfully recalling the material just before you would have forgotten it.
I think too many people use SRS to learn the material instead.
As I recall, the creator of Supermemo had a list of 20 suggestions or so, in which he urged people to first comprehend the information; then to learn it; then to memorize it; and then to rehearse it (SRS) so as not to forget it.
So I definitely agree that this is 100% the best way to use Anki, that's why I wrote the line about "Writing cards that trigger memories of experiences I had in the real world always produced better cards."
I couldn't give you a percentage, but I made most of my own cards, including all of those 2000+ kanji cards. There's lots of debate in the language learning community about vocab cards or sentence cards, and generally the ideal is the sentence cards, as it provides the context that helps you use is naturally (as opposed to literal translations from your native language).
> I still need to work out different variations of the concept to understand it, and that's not something that Anki can help with.
I agree that sentences are generally superior to vocab. Vocab cards are extremely problematic once you go beyond beginner level, because words can (and often do) have multiple translations in both directions. This could be because there are multiple words meaning the same thing, or because a word has multiple meanings.
For example if the English prompt is "watermelon" - are you supposed to recall the Italian word cocomero, anguria, or melone d'aqua (all of which mean watermelon)? If the English prompt is "bank" - is that a place you deposit money, a river bank, to bank (turn) a plane, or to bank (count) on something happening? You end up having to build in messy hacks like giving clues in the prompt as to which translation is intended (which means you memorise the clue instead of the word) or having cards for bank(1), bank(2), bank(3), and bank(4) which becomes very tedious for recall. Sentences mitigate these problems somewhat.
I now only use vocab cards for object nouns where there's only one important translation, and mainly because I can put pictures on these cards so that I'm learning from e.g. the concept of an orange instead of the English word for orange (which saves you the step of mentally translating when you aren't yet fluent with the word).
I'm so tempted to try improving my language skills with Anki, both for my native language and my daily use language. But the commitment feels so daunting- I've barely missed a day in my reviews for the last two years, and only have 28,000 reviews total. I'm very impressed by your 98,000!
I guess the best way to start is just to create a new deck in it with one card and then go from there. I already have a daily review habit, which is the most important part.
I can't agree with this when it comes to language learning specifically. I've used all sorts of tools to learn the language of my in-law side of the family: books, audio lessons, videos, tutoring, apps, etc. I've never made as much progress as I have with Anki. My language skills have improved in leaps and bounds so much with Anki that I now rarely bother with anything else, other than for sentence mining to create more Anki cards (e.g. from grammar books or apps) or just for a bit of variety to make the process more enjoyable. I actually find classes/tutoring (traditionally seen as the "best" method) frustrating because of how slowly I learn with them compared to Anki. In a 1 hour lesson we might cover a handful of concepts and words/phrases that I will almost certainly have forgotten most of by the time the next lesson comes around. In that same hour I can create and review a ton of Anki cards and I'll remember most of them.
With tools like Google and Microsoft's neural TTS and Anki's AwesomeTTS add-on my cards have audio that is so realistic that I am also constantly exposed to near-native listening. I do 3-way cards (Writing only -> English, Audio only -> English, and English -> Other language) so I'm actually getting a reasonable simulation of real life practice (reading, listening, speaking) on an individual sentence basis. My process is: (1) find a high quality sentence from a book / app / website / ChatGPT (with verification from a native speaker); preferably one that is fairly simple apart from a single word or verb conjugation that I haven't learned yet, in keeping with the i+1 rule, (2) create an Anki card for that sentence using my own custom note templates, (3) add audio with AwesomeTTS. Creating a card like this takes me perhaps 10-20 seconds as its mostly just copy-pasting and clicking a few buttons.
Of course to become truly fluent you need practice. But when I practice I'm already able to follow the gist of conversations and I can stumble my way through speaking in most situations: I've got a huge head start thanks to all the latent vocabulary and grammar that my brain knows thanks to Anki, instead of having to constantly look blankly at the other person while I pull out Google Translate.
I think people underrate the utility of a premade deck when constructed well and used appropriately.
A use case I've found is if you can find a deck that corresponds to a book you're reading.
I found a deck for the Rust book and it's structured such that you can see cards about things in the order you read about them. You simply read the book as usual, learning from your reading and entering code into a terminal as instructed, and then test you understanding with the cards.
When you end up reviewing older cards, you end up getting the benefits of putting them in long term memory but you also get the opportunity to make more connections as you revisit concepts which has its own benefits for deepening understanding.
I've found this makes reading the book 10x more effective. I get so much more out of it.
This all depends on having a source from which you're learning and the deck is just for testing understanding.
But yes anytime you're using Anki to learn/understand instead of to remember, you're likely misusing it. Anki is a tool for memory.
> The enemy is the static card. It always has the same front, formatting, and font. After enough reps I would latch on to little cues that are irrelevant to the meaning of the card, meaning I would skip the very important step of thinking deeply about the content. A fairly common occurrence was that a word in the sentence would remind me about the meaning of the sentence, giving away the answer to the target word, which is quite different from piecing together the meaning of the target word in a brand new sentence.
Interestingly, that's the "trick" behind a lot of the seemingly magic skill of geo guessers. The best players have played so much, that they now "see" things that a regular person wouldn't even consider to look for, like the camera quality, what year the car was from, and so they narrow down the possible countries by those aspects, before even looking at the "picture".
> Interestingly, that's the "trick" behind a lot of the seemingly magic skill of geo guessers. The best players have played so much, that they now "see" things that a regular person wouldn't even consider to look for, like the camera quality, what year the car was from, and so they narrow down the possible countries by those aspects, before even looking at the "picture".
Slightly OT, but this happens constantly with ML classifiers on any highly multi-dimensional problem. At first it seems like magic, and then someone digs into the principal components of the prediction, and finds a mixture of a few highly specific factors that -- in the worst case -- is an artifact of the dataset itself (image blur or color bias, for example).
Also common is that the predictive factors aren't pathological -- they're just "boring" -- and therefore the performance of the model is dismissed by the practitioner ("oh, I'd have thought of that, since it's only using a few common traits that are well-understood.")
The one time I tried Anki u realised I started recalling the answers based on the shape of the words lay out on the card without necessarily reading them.
I used Anki pretty heavily during ~8 years of law school and it was a game-changer. Two things that helped me the most (and that I think apply to any kind of flashcards, digital or paper):
Hallucinations in LLMs when learning is dangerous; IF you have some background, you can usually tell with LLMs go off the rails, but It would be unfortunate for you to commit to memory an incorrect fact at such a vulnerable time. It will be difficult to "uncommit it" at that point.
I don't think the LLM value prop here is to build lots of cards. It's just to interact with the model conversaitonally. If the model is wrong in this one translation, I'm not going to exactly "commit it to memory", I'm just going to keep on carrying on. I don't know that the LLM mistakes are any particularly worse than the many and sundry other mistakes I'm already continuously making as a language learner anyhow.
If I could speak a foreign language as well as a 8B parameter LLM, hallucinations and all, I'd be immensely ahead of where I am now. It's not like second languages aren't themselves often broken in somewhat similar ways.
I've used anki before. Most of the decks you get are randomly downloaded from the anki website. I'm not sure that an LLM hallucination is that much more likely than a typo or error in a random free deck I downloaded that someone compiled for me.
I use Anki to study all kinds of subjects, and more than half of the value is processing what I learned (usually with pencil and paper) into good anki cards ("atomic", as per Michael Nielsen's definition), including the insights that I had when studying the subject (like "what's the comparison with X that I used to understand Y?")
I'm not sure if it is efficient, mind you, but I suppose it's effective because I can recall information later when relevant, and I believe that like exercising just being able to stick to a study routine ends up being more important than picking the best routine
I had the exact same experience. The more work I did up front on a card, the better I was able to remember. When I was learning the kanji cards, I would take the time to draw them out in a special notebook while thinking carefully about the different components, and it really helped with retention.
I did not do this with many cards though, hoping that they would eventually stick.
I think in general the more you engage with the thing you are doing, the better you remember. Even when reading or listening to a lecture or whatever. Maybe what I'm proposing here is that by making it dynamic you create a system where deeper engagement is necessary.
Always good to see more activity in the language learning space. I wish you luck now that you've moved past Anki.
A piece of feedback: one of the common issues I've found with AI-generated questions & answers based on an article is that they will often hone in on testing values. It looks like incontextlearning.com often suffers from this same issue (over half my comprehension questions were "how many"-style). I can easily answer these types of questions even if I don't know what the content is about.
I was thinking the same thing. Currently all my cards try to all follow the same pattern, making it necessary to really remember at based on differences between them. (For example, I would never ask "What happened in 1914?" but instead just "1914" and have "Start of WWI" on the back, as a reversible card). LLMs could take such cards and provide the user with an endless stream of questions: What year did WWI begin? What happened in 1914? Which event took place in 1914? And so on. For vocabulary this might of course be even better when the word/phrase itself will always be contextualized in practice.
I agree that they could be complementary, but I think there's a not-yet-made tool that goes even beyond this, where you are interacting with an LLM that has an Anki-like backend of some kind, keeping track not only the number of mistakes but of what kind of mistakes you made and when, so that it can later bringing up the card in a more natural way.
Readit News
I haven't used Anki for language learning, but I imagine that if I did, it would be to add some new vocabulary I had just learned from a book, conversation, film, etc. I don't think it would help me learn a language from zero though- that would require practicing it.
In summary, Anki is great for reinforcing something you've just learned, but you can't reinforce your way into the context that is necessary to truly understand something.
1. Instead of the usual "1 new word, 1 card" have 2 - 3 new words on each card, and have each new word in 3 - 4 different cards (ideally with different inflection, meaning, nuance, etc) 2. Review cards fast and very few times each. Like 5 - 8 times / card (max) instead of usual 15 - 20. 3. Don't punish yourself, keep moving on even if you just half-remember. First familiarize and then internalize language patterns instead of just memorize words.
Review intentionally, totally concentrated on the task. 10 mins / day well done >>> 30 mins / day mindlessly.
Outcome: more fun, more effective learning, no memorizing rectangles.
Writing my own cards as I'm learning is the only way I've found it effective.
I think too many people use SRS to learn the material instead.
As I recall, the creator of Supermemo had a list of 20 suggestions or so, in which he urged people to first comprehend the information; then to learn it; then to memorize it; and then to rehearse it (SRS) so as not to forget it.
It didn't. They wrote "Anki is dead" because it brings clicks.
I couldn't give you a percentage, but I made most of my own cards, including all of those 2000+ kanji cards. There's lots of debate in the language learning community about vocab cards or sentence cards, and generally the ideal is the sentence cards, as it provides the context that helps you use is naturally (as opposed to literal translations from your native language).
> I still need to work out different variations of the concept to understand it, and that's not something that Anki can help with.
But imagine if it could!
For example if the English prompt is "watermelon" - are you supposed to recall the Italian word cocomero, anguria, or melone d'aqua (all of which mean watermelon)? If the English prompt is "bank" - is that a place you deposit money, a river bank, to bank (turn) a plane, or to bank (count) on something happening? You end up having to build in messy hacks like giving clues in the prompt as to which translation is intended (which means you memorise the clue instead of the word) or having cards for bank(1), bank(2), bank(3), and bank(4) which becomes very tedious for recall. Sentences mitigate these problems somewhat.
I now only use vocab cards for object nouns where there's only one important translation, and mainly because I can put pictures on these cards so that I'm learning from e.g. the concept of an orange instead of the English word for orange (which saves you the step of mentally translating when you aren't yet fluent with the word).
I guess the best way to start is just to create a new deck in it with one card and then go from there. I already have a daily review habit, which is the most important part.
With tools like Google and Microsoft's neural TTS and Anki's AwesomeTTS add-on my cards have audio that is so realistic that I am also constantly exposed to near-native listening. I do 3-way cards (Writing only -> English, Audio only -> English, and English -> Other language) so I'm actually getting a reasonable simulation of real life practice (reading, listening, speaking) on an individual sentence basis. My process is: (1) find a high quality sentence from a book / app / website / ChatGPT (with verification from a native speaker); preferably one that is fairly simple apart from a single word or verb conjugation that I haven't learned yet, in keeping with the i+1 rule, (2) create an Anki card for that sentence using my own custom note templates, (3) add audio with AwesomeTTS. Creating a card like this takes me perhaps 10-20 seconds as its mostly just copy-pasting and clicking a few buttons.
Of course to become truly fluent you need practice. But when I practice I'm already able to follow the gist of conversations and I can stumble my way through speaking in most situations: I've got a huge head start thanks to all the latent vocabulary and grammar that my brain knows thanks to Anki, instead of having to constantly look blankly at the other person while I pull out Google Translate.
A use case I've found is if you can find a deck that corresponds to a book you're reading.
I found a deck for the Rust book and it's structured such that you can see cards about things in the order you read about them. You simply read the book as usual, learning from your reading and entering code into a terminal as instructed, and then test you understanding with the cards.
When you end up reviewing older cards, you end up getting the benefits of putting them in long term memory but you also get the opportunity to make more connections as you revisit concepts which has its own benefits for deepening understanding.
I've found this makes reading the book 10x more effective. I get so much more out of it.
This all depends on having a source from which you're learning and the deck is just for testing understanding.
But yes anytime you're using Anki to learn/understand instead of to remember, you're likely misusing it. Anki is a tool for memory.
* Confused where in the original series Spock goes to school.
* Watches the video and sees 2009 "Star Trek"
* "as a kid"...
* feels old
Interestingly, that's the "trick" behind a lot of the seemingly magic skill of geo guessers. The best players have played so much, that they now "see" things that a regular person wouldn't even consider to look for, like the camera quality, what year the car was from, and so they narrow down the possible countries by those aspects, before even looking at the "picture".
Slightly OT, but this happens constantly with ML classifiers on any highly multi-dimensional problem. At first it seems like magic, and then someone digs into the principal components of the prediction, and finds a mixture of a few highly specific factors that -- in the worst case -- is an artifact of the dataset itself (image blur or color bias, for example).
Also common is that the predictive factors aren't pathological -- they're just "boring" -- and therefore the performance of the model is dismissed by the practitioner ("oh, I'd have thought of that, since it's only using a few common traits that are well-understood.")
- The twenty rules of formulating knowledge (https://www.supermemo.com/en/blog/twenty-rules-of-formulatin...) ; old but really solid advice on how to actually write cards that stick.
- The book Fluent Forever; it’s meant for languages, but the general principles carry over to learning basically anything.
If I could speak a foreign language as well as a 8B parameter LLM, hallucinations and all, I'd be immensely ahead of where I am now. It's not like second languages aren't themselves often broken in somewhat similar ways.
I'm not sure if it is efficient, mind you, but I suppose it's effective because I can recall information later when relevant, and I believe that like exercising just being able to stick to a study routine ends up being more important than picking the best routine
I did not do this with many cards though, hoping that they would eventually stick.
I think in general the more you engage with the thing you are doing, the better you remember. Even when reading or listening to a lecture or whatever. Maybe what I'm proposing here is that by making it dynamic you create a system where deeper engagement is necessary.
A piece of feedback: one of the common issues I've found with AI-generated questions & answers based on an article is that they will often hone in on testing values. It looks like incontextlearning.com often suffers from this same issue (over half my comprehension questions were "how many"-style). I can easily answer these types of questions even if I don't know what the content is about.
For example I use LLMs to generate cards for me, and Anki's algorithm to make them stick.
Similarly a LLM plugin could easily present a fresh sentence each time you review a particular vocab