I used Anki very differently than you do, and I used it consistently for many years, totalling more than 30,000 card reviews. (We’ll get to why I stopped.) My learning was focused on foreign languages, which is a special case.
If I had to boil down my advice:
Make card creation as easy as possible. This may mean building or using special tools.
As you said, limit new cards to 10-20 per day.
It is almost impossible to make a card too easy to be useful. For example:
Put an interesting snippet of foreign language text on the front. Boldface one word. If you understand roughly what the boldfaced word means in that context, pass the card. You’re not obligated to read all the text. (Credit to Khatzumoto for this idea.)
Alternatively, cloze (hide) one word on the front. Or better, just half the word. If you can recall the word, pass the card.
Use automatic tools to convert subtitles & audio into cards. If you can understand at least 80% of one subtitle, pass the card.
Most importantly, once you make it easy to create lots of cards, and once you make the cards easy to answer, then the final Anki skill is to DELETE DELETE DELETE cards at the slightest provocation. Cards are cheap! If a card annoys or frustrates you, delete it! If you need to learn that information, you’ll eventually encounter it in a better context and turn it into a new card.
In fact, consider deleting cards instead of failing them. Or set cards to auto-suspend if you fail them twice.
Most of the high-value learning occurs in the first 30-40 days. It’s OK to keep older cards if you like them! But if you’re not learning any new cards, and you’ve just been reviewing old cards for a couple of months, it’s fine to just suspend entire decks.
This approach to Anki may seem unfamiliar! That’s because it’s based on a different philosophy:
The normal way to use spaced repetition software is as a database of facts to memorize. The goal is to learn almost everything.
The philosophy I describe above is using Anki as a memory amplifier. The goal is to “turn up the volume” on interesting knowledge, particularly linguistic knowledge. But even with the volume cranked, many signals will still be too weak. And that’s OK!
I strongly suspect that selective forgetting is closely tied into how the brain generalizes. Trying to remember everything feels slightly unnatural. Trying to remember more and to generalize better can work quite well.
Ohh I love this and it is indeed very different from what I do! I’d be super interested in a wee writeup on what non-language learner Anki users can learn from this approach, if you ever have the time for that. Maybe there’s a hybrid approach with the best of both worlds? (Also lowkey interested in why you stopped now!)
Oh, sorry! I stopped because for the language I cared the most about, I had reached a point where natural use of the language was enough to maintain at least 90+% of college-level reading skills. If I go too long without doing enough reading, then I start to miss obscure vocabulary in difficult texts. So when doing Anki reviews on old decks became tedious, I followed my advice and suspended my decks!
Adapting to non-language areas. If I were going to try to adapt this language-focused “memory” amplifier approach to other areas, I would start by experimenting with new card formats, looking for new embarassingly easy formats. I don’t know exactly what would work. But let’s try an experiment!
Keeping with your example, I asked ChatGPT to summarize a Wikipedia article about Isaac Newton. It gave me this (I have manually removed the citations and header/footer text):
Birth and death. Isaac Newton was born on 4 January 1643 (Julian calendar: 25 December 1642) in Woolsthorpe‑by‑Colsterworth, Lincolnshire, England, and died on 31 March 1727 in Kensington.
Polymath and scientific impact. He was an English polymath—mathematician, physicist, astronomer, alchemist, theologian and author—and a central figure in the Scientific Revolution and Enlightenment.
Classical mechanics & Principia. Newton formulated the three laws of motion and the law of universal gravitation in his 1687 work Philosophiæ Naturalis Principia Mathematica, unifying terrestrial and celestial mechanics.
Independent invention of calculus. He developed infinitesimal calculus (his “method of fluxions”) around 1666, independently of Leibniz.
Contributions to optics. Newton conducted experiments showing that white light is composed of a spectrum of colors via prism refraction, and published Opticks in 1704.
First reflecting telescope. In 1668 he built the first practical reflecting telescope—now called the Newtonian telescope.
Turning this into a card. This feels like a good time to use cloze cards, because they’ll let us autogenerate a bunch of easy cards from the base text. Let’s split this into two cards, and add Anki cloze markers. Here’s how I might mark up the second:
Isaac Newton (2/2)
Independent invention of {{c1::calculus}}. He developed infinitesimal {{c1::calculus}} (his “method of fluxions”) around {{c2::16}}66, independently of Leibniz.
Contributions to optics. Newton conducted experiments showing that white light is composed of a spectrum of colors via p{{c3::rism}} refraction, and published Opticks in 1704.
First ref{{c4::lecting}}telescope. In {{c2::16}}68 he built the first practical ref{{c4::lecting}} telescope—now called the Newtonian telescope.
For those unfamiliar with cloze cards, this will generate 4 cards, each hiding the information marked with c1 through c4, respectively.
When reviewing each card,
You are in no way obligated to read anything other than what you needed to fill in the blank.
If you can fill in the blank, mark the card correct.
Note that these are all stupidly easy:
c1: “calculus” is a gimme, especially thanks to the mentions of infinitesimals and Leibniz.
c2: I only hide the first two digits of the year. And “17” is left unhidden elsewhere. Usually getting people in right century is good enough, so why stress over it? Or you could hide the decade digit if you were focusing on a specific century.
c3: “prism” is pretty obvious from context, but just in case, I leave the “p” visible.
c4: I leave “ref” visible, and you need to remember it’s “reflecting”, not “refracting”, but that doesn’t require more than basic physics knowledge.
Predictions. My hypotheses:
This style of card can be created very cheaply, especially since cloze gives you 4 cards for the price of 1.
You will retain a surprising amount of information about Isaac Newton.
You will (and should!) delete a bunch of these cards within a month or two, and not really feel the loss.
Reviewing these cards can be very low stress, especially since you should have a hair-trigger delete.
Possible objections:
“Doesn’t seeing the same information on multiple cards mess with the spaced recognition algorithm?” Surprisingly, this seems to matter less than I expected. We’re focusing on amplification, not flawless memorization.
“Would you actually use an LLM to generate initial summaries?” Possibly, at least if I could get the hallucination rate low enough. I might also copy-paste interesting bullet points from various places.
“Aren’t these cards too easy?” The more Anki reviews I did, the more I started to suspect “too easy” wasn’t actually a thing.
Anyway, like I said, this is mostly guesswork! I would be fascinated if someone wanted to create (say) 10-20 base cards this way, with 3-5 clozes each, and review them for at least 40 days, then report back. But I hope this provides some useful ideas to someone!
The problem with these cloze cards is that you tend to link the shape to the information rather than the words themselves. After a few goes you’ll basically stop reading the words entirely. It’s not very effective for recalling the facts irl, since usually you’ll be trying to recall the answer to a specific question (prompt), not fill in the blanks. I find that the things the OP talks about in the above guide are much better for actually recalling info when it counts.
As an aside, the philosophy of impulsive deletion/suspension from your main comment seems like a promising idea. I typically take the opposite approach and don’t even suspend leeches, with that being exclusively for useless or obviously-defined words. I might try it out if I go about learning another language though, it definitely has potential (though it also seems far more suited for high-immersion learners which isn’t something I’m good at).
Huh, I suppose. My card formats (see sibling comments) are easy and fun enough that reviewing 50-100 cards takes 15-30 minutes, and it’s not especially high stress.
But then again, like I noted above, I delete cards pretty ruthlessly if they’re obnoxious to review.
Thank you for an interesting guide!
I used Anki very differently than you do, and I used it consistently for many years, totalling more than 30,000 card reviews. (We’ll get to why I stopped.) My learning was focused on foreign languages, which is a special case.
If I had to boil down my advice:
Make card creation as easy as possible. This may mean building or using special tools.
As you said, limit new cards to 10-20 per day.
It is almost impossible to make a card too easy to be useful. For example:
Put an interesting snippet of foreign language text on the front. Boldface one word. If you understand roughly what the boldfaced word means in that context, pass the card. You’re not obligated to read all the text. (Credit to Khatzumoto for this idea.)
Alternatively, cloze (hide) one word on the front. Or better, just half the word. If you can recall the word, pass the card.
Use automatic tools to convert subtitles & audio into cards. If you can understand at least 80% of one subtitle, pass the card.
Most importantly, once you make it easy to create lots of cards, and once you make the cards easy to answer, then the final Anki skill is to DELETE DELETE DELETE cards at the slightest provocation. Cards are cheap! If a card annoys or frustrates you, delete it! If you need to learn that information, you’ll eventually encounter it in a better context and turn it into a new card.
In fact, consider deleting cards instead of failing them. Or set cards to auto-suspend if you fail them twice.
Most of the high-value learning occurs in the first 30-40 days. It’s OK to keep older cards if you like them! But if you’re not learning any new cards, and you’ve just been reviewing old cards for a couple of months, it’s fine to just suspend entire decks.
This approach to Anki may seem unfamiliar! That’s because it’s based on a different philosophy:
The normal way to use spaced repetition software is as a database of facts to memorize. The goal is to learn almost everything.
The philosophy I describe above is using Anki as a memory amplifier. The goal is to “turn up the volume” on interesting knowledge, particularly linguistic knowledge. But even with the volume cranked, many signals will still be too weak. And that’s OK!
I strongly suspect that selective forgetting is closely tied into how the brain generalizes. Trying to remember everything feels slightly unnatural. Trying to remember more and to generalize better can work quite well.
Ohh I love this and it is indeed very different from what I do! I’d be super interested in a wee writeup on what non-language learner Anki users can learn from this approach, if you ever have the time for that. Maybe there’s a hybrid approach with the best of both worlds? (Also lowkey interested in why you stopped now!)
Oh, sorry! I stopped because for the language I cared the most about, I had reached a point where natural use of the language was enough to maintain at least 90+% of college-level reading skills. If I go too long without doing enough reading, then I start to miss obscure vocabulary in difficult texts. So when doing Anki reviews on old decks became tedious, I followed my advice and suspended my decks!
Adapting to non-language areas. If I were going to try to adapt this language-focused “memory” amplifier approach to other areas, I would start by experimenting with new card formats, looking for new embarassingly easy formats. I don’t know exactly what would work. But let’s try an experiment!
Keeping with your example, I asked ChatGPT to summarize a Wikipedia article about Isaac Newton. It gave me this (I have manually removed the citations and header/footer text):
Turning this into a card. This feels like a good time to use cloze cards, because they’ll let us autogenerate a bunch of easy cards from the base text. Let’s split this into two cards, and add Anki cloze markers. Here’s how I might mark up the second:
For those unfamiliar with cloze cards, this will generate 4 cards, each hiding the information marked with c1 through c4, respectively.
When reviewing each card,
You are in no way obligated to read anything other than what you needed to fill in the blank.
If you can fill in the blank, mark the card correct.
Note that these are all stupidly easy:
c1: “calculus” is a gimme, especially thanks to the mentions of infinitesimals and Leibniz.
c2: I only hide the first two digits of the year. And “17” is left unhidden elsewhere. Usually getting people in right century is good enough, so why stress over it? Or you could hide the decade digit if you were focusing on a specific century.
c3: “prism” is pretty obvious from context, but just in case, I leave the “p” visible.
c4: I leave “ref” visible, and you need to remember it’s “reflecting”, not “refracting”, but that doesn’t require more than basic physics knowledge.
Predictions. My hypotheses:
This style of card can be created very cheaply, especially since cloze gives you 4 cards for the price of 1.
You will retain a surprising amount of information about Isaac Newton.
You will (and should!) delete a bunch of these cards within a month or two, and not really feel the loss.
Reviewing these cards can be very low stress, especially since you should have a hair-trigger delete.
Possible objections:
“Doesn’t seeing the same information on multiple cards mess with the spaced recognition algorithm?” Surprisingly, this seems to matter less than I expected. We’re focusing on amplification, not flawless memorization.
“Would you actually use an LLM to generate initial summaries?” Possibly, at least if I could get the hallucination rate low enough. I might also copy-paste interesting bullet points from various places.
“Aren’t these cards too easy?” The more Anki reviews I did, the more I started to suspect “too easy” wasn’t actually a thing.
Anyway, like I said, this is mostly guesswork! I would be fascinated if someone wanted to create (say) 10-20 base cards this way, with 3-5 clozes each, and review them for at least 40 days, then report back. But I hope this provides some useful ideas to someone!
The problem with these cloze cards is that you tend to link the shape to the information rather than the words themselves. After a few goes you’ll basically stop reading the words entirely. It’s not very effective for recalling the facts irl, since usually you’ll be trying to recall the answer to a specific question (prompt), not fill in the blanks. I find that the things the OP talks about in the above guide are much better for actually recalling info when it counts.
As an aside, the philosophy of impulsive deletion/suspension from your main comment seems like a promising idea. I typically take the opposite approach and don’t even suspend leeches, with that being exclusively for useless or obviously-defined words. I might try it out if I go about learning another language though, it definitely has potential (though it also seems far more suited for high-immersion learners which isn’t something I’m good at).
I think you might have misread that part, the post is advocating 10-20 review(!) cards per day and only 1-2 new cards
Huh, I suppose. My card formats (see sibling comments) are easy and fun enough that reviewing 50-100 cards takes 15-30 minutes, and it’s not especially high stress.
But then again, like I noted above, I delete cards pretty ruthlessly if they’re obnoxious to review.