Do you have a TLDR for that article? It's pretty long and wordy.

I have found out that I was bad at card making when reviewing the first cards I made after a lapse of time of not using Anki. My then self got frustrated and irritated by how vague and non-specific the cards I made myself were. I certainly learned something about my memory and cognition. The article is long, but it has some insights that resonated with me. I think the quality of the cards one makes is as important, if not more so, than spaced repetition.

I have made a few runs at adopting spaced repetition learning via Anki into my life. The failure point each time so far has been making cards that are worth anything.

The article discusses how to create effective prompts that aid in long-term learning using spaced repetition. Matuschak suggests that good prompts should be specific, clear, and concise, and should be designed to encourage active recall rather than passive recognition.

He also emphasizes the importance of interleaving, which involves mixing up different types of problems or questions to enhance retention and facilitate transfer of knowledge. Additionally, he suggests that prompts should be personalized and contextualized to increase engagement and motivation.

Okay, so was this written by chatgpt

This is from Kagi's summarizer:

  Anki is a powerful tool for improving long-term memory.
  Anki is best used in service to a creative project.
  It is important to ask good questions when using Anki.
  Anki is most useful for learning new fields.
  Anki can be used to develop virtuoso skills.
  It is important to avoid orphan questions and yes/no patterns when using Anki.
  Memory is complicated and we should be careful before putting too much faith in any given model.
  Distributed practice is important for maximizing retention.
  Anki can be used to remember non-verbal experiences.
  Anki can be used to memorize APIs and code.

I asked ChatGPT to summarize the article as Anki cards. I thought it did a good enough job to share, but prefer to use my memory for processes rather than facts; while it has its uses, training a skill with Anki may cause the human equivalent of overfitting.

What type of memory is virtually limitless? Long-term memory

What is the main bottleneck between working memory and long-term memory? The speed at which we can move memories from the working memory to storage

What is a "schema" in the context of long-term memory? A pattern of knowledge stored in our long-term memory that gives our thinking depth and understanding

What happens to a long-term memory when we bring it back into working memory? The memory evolves and forms new connections, updating the living schema

What is Anki commonly used for? Rote learning and memorization

How much time does it take daily to maintain a database of 10,000 Anki cards? 15-20 minutes

What are Michael Nielsen's rules of thumb for whether to Anki-fy something? First, if memorizing a fact seems worth 10 minutes of future time, and second, if a fact seems striking, regardless of whether it seems worth 10 minutes of future time or not.

Can Anki be used to understand complex concepts, such as those in research papers? Yes, by breaking down the concepts into basic questions and building up knowledge through multiple re-reads and increasingly complex questions.

> summarize the article as Anki cards

that's a very clever prompt mechanism you used there

Why is it useful? My job at least isn’t bottlenecked by me not memorizing enough things. Or maybe it is! But I’m not convinced

This could give you some clues: http://augmentingcognition.com/ltm.html#:~:text=How%20import... (see part II)

Don't think in terms of bottlenecks but rather in terms of missed opportunities.

Although already a power Emacs user, it was a significant step up when I started memorizing keystone for various modes via spaced repetition.

People have this weird disdain for memorization. Memorization is the same as knowing.

Can you imagine that there have been points in your job that would have gone smoother if you'd known more?

I don’t believe them to be the same. I have friends who repeat words to me GPT style, that I know them to be imprecise in the spirit of the discussion.

Then they didn't really memorize them properly, did they?

Understanding is the structure of knowing. Memorization is the paint on top. It ain't gonna hold your house up.

Memorization is actually the foundation whether you do it with intention or not. Everything in your working knowledge is something you memorized.

Memorisation is adding nodes to your existing network of knowledge

Understanding is efficiently arranging and coordinating those nodes in an efficient path for most situations

Wisdom is knowing in which contexts to actually apply that path of nodes

Tacit knowledge is similar to wisdom just it generally lacks memorisation

There's not a perfect mapping in English between Memorisation, Understanding and Wisdom and their most pragmatic corresponding analogues in nature, but I have found this distinction between terms useful in the whole memorisation v understanding debate

> Memorisation is adding nodes to your existing network of knowledge

Adding nodes to your existing network of knowledge is learning. Memorization is making the nodes persistent. Both are important, but they're not equivalent. You need to do both.

You can learn without memorization (you can learn a lot and forget everything you've learned by the next day). You can memorize without learning the subject matter (for example, you can memorize a counting song in Mandarin Chinese without knowing that what you've memorized are numbers).

Flashcards can be used for learning, but they are optimized for memorization. We shouldn't expect them to be the best tool for learning, because that's not what they're designed for. One might get better results learning from another resource, and then using flashcards to memorize the concepts they've learned.

You can memorize every note of a difficult piano piece, but that doesn't mean you know how to play it well.

Yes, memory is a necessary but not sufficient component of learning. "Piano playing" seems like a really poor example to argue against the value of memorization.

The post I was responding to said "Memorization is the same as knowing". I specifically chose a counterexample to suggest that knowledge is deeper than memorizing some facts about the world.

In fact, I would argue that most of the forms of knowledge that are truly valuable are exactly the kinds that aren't amenable to memorization. Think about any skill that's in high demand and takes decades to master. You're not going to get there by studying Anki cards.

Do you think it's possible that knowing more information that is amenable to memorization will help you cultivate knowledge that isn't amenable to memorization better or faster?

At some point in our education we have to memorize the rules for manipulating algebraic formulas, trigonometric identities, etc. It would be hard to learn calculus without being able to do these things unconsciously.

On the other hand, what a professional mathematician does is create new tools for proving theorems. This can't be memorized because, by definition, the knowledge didn't exist. It's true that they draw on previous experience having solved similar problems. But it's not clear to me that this knowledge is most efficiently gotten by drilling the same problems over and over.

I think this idea of using Anki for learning everything is misguided because it attempts to reduce complex topics and pragmatic know-how to a set of trivia questions. An example from the article: the author wanted to understand the AlphaGo paper, and some example flashcards were "who plays first in Go" and "where did AlphaGo get its training data". They probably could have acquired a much deeper understanding by playing a few games of Go and implementing a reinforcement learning algorithm for tic-tac-toe. Incidentally, that also sounds a lot more fun and motivating to me than studying flashcards.

Learning to play a song or to play from sheet music is, itself, a task that is well suited for rote learning. There's really no way out of repetitive practice if your goal is to learn to play a musical instrument.

Knowing is not the same as understanding though. For example, I do not see any value of winning the spelling bee in real life applications. This is actually widely accepted as the downrisk in Asian education systems - the emphasis on rote learning vs practical application.

> For example, I do not see any value of winning the spelling bee in real life applications.

You might if you were an editor. I guess your job probably doesn't have much to do with that so you see it as pointless but there is a different sphere of knowledge where having various things committed to memory would save you time. For instance, if we're interested in learning a new programming language, I guess most of us can understand a sentence like "a string is an immutable sequence of characters," quickly read it, and move on. If you weren't familiar with the concepts in that sentence you'd burn a lot more time before making heads or tails of it.

> “I guess most of us can understand a sentence like "a string is an immutable sequence of characters," quickly read it, and move on.”

Can we? In the context of learning a new unfamiliar programming language… is it saying there is a string type or is it telling you that there isn’t a string type but only a sequence of characters? Like APL a string is an array of characters not its own type but in C# [string] is a thing and if you iterate it you get [char] type. In Prolog a string is traditionally a linked list of ASCII code numbers, there isn’t a string type or a char type … but there is still a string to chars conversion function. So is the sentence telling you that there is a char type? In Python a string is iterable but you get strings of length one for the characters rather than chars. But wait what about Unicode and surrogate pairs and glyphs, where does “sequence of characters” fit on those things in this language?

What does ‘sequence’ mean in this unfamiliar language - is it saying strings are an array or a list, or that sequence is a trait or behaviour which generalises over many types? What is immutable specifically, sequences or strings or characters or all data? What are the consequences of that? In APL you can assign into an array even though they are immutable (it will be copied behind the scenes) so it seems like it is an implementation detail you can ignore but wait that means dropping one character from the front copies the rest of the string in memory (slow) whereas in other languages trying to change a string at all is an error - compile time or runtime? You might think Prolog can do it if strings are linked lists but surprise, Prolog lists are immutable.

So if you read that sentence and move on, what have you really learned about the new language? In the context of this thread how would you Anki-fy the concepts immutable or sequence or string?

Do you think all those questions are someone who looks at the sentence and has no idea what it means would even think to ask? I don't know how you'd convert the knowledge to flash cards most effectively; my point is you have a ton of knowledge committed to memory about programming concepts and that allows you to more easily assimilate new knowledge, because you aren't having to spend time wrestling with the basics to make heads or tails of the topic you actually meant to learn about. It's a kind of yak shaving for the mind that you get stuck doing otherwise.

The context was about whether rote learning was helpful, and I read your comment as suggesting that it was - not just that background knowledge helps but that rote learning without understanding can help someone gain understanding later. By asking about Anki flashcards I was wondering what someone might rote memorise without understanding which would then help with learning about that programming sentence (I can't quickly think of anything but maybe you had something in mind).

And I'm also saying that rote learning and background knowledge can hinder progress and require time unlearning, like the beginner needs to spend time learning. e.g. a mathematical sequence like "even numbers" can be infinitely long and the sentence probably doesn't mean that strings can have infinite characters in them. I guess, if you don't know the meaning of the words the sentence doesn't help you, if you know the words from a different context but don't know the programming language then the words don't help you, and if you already know the language then the words add nothing for you. Learning is impossible, Q.E.D ;-)

I don't agree that there is a meaningful distinction between these types of learning. To understand a topic you must simply commit a lot of details about it to memory. If you wanted to make flash cards perhaps you could do terms and definitions.

I live in Taiwan and this is a constant presumption by other immigrants here - oh, the taiwanese aren't good at critical thinking, their schools just have them memorize everything, there's no imagination, they have no ability to self start or self direct.

There's some grains of truth to aspects of the culture resisting individualism (which isn't inherently negative I don't think, as distasteful as that may sound to Americans), but not only is this a mischaracterization, it's also missing the upsides of having a population that all has stashed in their heads a whole shitload of math and science that enables them in all sorts of interesting ways.

For example, on a construction team, any single member is capable of more geometric reasoning than the same team in the usa (in my limited experience working with construction in both countries), making for faster, more accurate, and cheaper builds. Or, when bringing a software engineer up to speed with a new framework or whatever, I find they retain mundane details remarkably quickly, probably by having a well honed memorization method. That's really nice for when you're trying to think of how to implement something and basically the entire library documentation is just at hand in the noggin.

You are completely correct. It is very rarely worth memorising information you do not understand, doubly so if it's information divorced from the rest of your knowledge (e.g. memorising the first ten digits of pi is useful perhaps in extreme niche cases, but is vastly more useless if you've done it without even knowing what pi actually is). The creator of Supermemo notes this: https://supermemo.guru/wiki/Do_not_memorize_before_you_under...

On the other hand, memorising things you do understand is a time saving measure at the point of memory access as well as something that allows you to play with concepts/ideas/facts in your head while doing anything else, which I've found to be quite useful. You do have to be selective about it, and software like Anki shouldn't be used to build memories for everything (for one, it can be so boring that you'll be turned off of SRS entirely if you overuse it).

> Memorization is the same as knowing.

They are not the same which is why they are two different words. For example, memorising first 1000 prime numbers is not same as knowing prime numbers. The former person will not be able to figure out 1001st prime number because it's not in their memory bank but the later can calculate it. Sure it takes time to calculate but it's not impossible.

Now, I'm not belittling memorisation. A baseline memory is needed otherwise one'll be wasting time and energy figuring out everything from first principles all the time. This is akin to rainbow tables saving time. But as with all things in life finding right balance is key. As you pointed out, it's useful to memorise things at work to save time and energy.

You can memorize things like "Definition of prime number" or "Main application of prime numbers". I think memorizing concepts is much more important than memorizing raw data

Completely agree here. Memorization is actually quite important.

Try learning Spanish from scratch with a "bad memory" and let me know how it goes.

Nonsense. Memory is a necessary component to acquire knowledge and mastery. But knowledge is not memory.

I am fluent in three languages. Learning words is the easy part, fluency unlocks when you are able to think, or feel, in that language. When you have thoughts that you cannot losslessly translate back to your native language.

Knowing Spanish words is not knowing Spanish. Memorization is the first and easiest 5% of the whole process.

Memorization is not comprehension.

In problem solving there can be a balance between creativity and knowledge. Creativity can help you link different domains to create novel solutions, and knowledge can help with avoiding blind spots. Knowledge can also be a blunt force for problem solving, it is most likely the case that someone has already encountered your problem, and knowing even a bit about that solution is very helpful.

I don't personally use this method, but it seems to me like this would help with all the cases where you search for something on stackoverflow, use the solution, and 3+ months later you encounter the exact same issue. You remember that you already had this issue, and you remember that you searched on SO and found a solution, but you don't remember what it was. If you wrote a flash-card and learned it, you'd have that knowledge available immediately. Or, let's say you're learning a new language/framework that you want to use professionally, but don't use frequently yet. So you start a project in that language/framework, learn some of the things that make it different, and every time something surprised you or you had to search for a solution, you write a flashcard. Then you can learn these aspects separately from having to work on a project, and once you do have a use for it, you have the knowledge available.

I think this could be less likely if you're very specialized, but if you're a consultant or freelancer and have to work with lots of different technologies, locking down the differences and unique aspects for each could save a lot of time.

There are (arguably) better ways to handle this, though. When you get enough cards, you won't conceivably review them all within a reasonable enough timespan to encounter concepts often enough to really remember.

The use case you mention is more suited to a personal knowledgebase with good search capabilities and/or networking, like Obsidian, Tana, Roam, org-mode, whatever. That way you have the knowledge somewhere, you can search it (of course, this can be lossy - like writing good cards this requires writing good notes), and you can build up all the context you need.

I think the article misses the mark on the dichotomy here - it's not search vs. flash cards, it's owning your knowledge vs. relying on external, potentially ephemeral sources. Flash cards and spaced repetition are great for certain use cases (Anki is incredibly popular among language learners for a reason), but not for all of them.

It seems to me that pretty much anything has better ROI then 10-20 minutes of pure rote memorization a day. Plus, keeping in 10-20 minutes a day requires separate skill on itself. Anki has a way of taking over and expanding the time it requires.

> of pure rote memorization

I wouldn't describe my use of Anki as "rote memorization". Though certainly some people use it that way. There is a big difference between blindly putting raw and unfamiliar facts into an SRS, and using an SRS to remember things you have learned and understand. It's the second case that pays large dividends over time.

Anki certainly is rote memorization. SRS is just the most efficient way to successfully do it.

There is uh, a reason the program's name is almost literally "blind memorization". The connotations are not quite as negative as the phrasing sounds in English, buut the meaning is not a coincidence.

Strong disagree. The pivot completely away from rote memorization in public schools is a mistake. Being able to immediately conjure math facts for example without having to use higher level concentration makes the next level of thought easier to access. Rote memorization is currently undervalued.

Oh man I could not disagree more. “Math facts” don’t require any sort of conjuring, in fact cease to seem mystical at all, once you have a deep understanding of the material. Take your favorite math textbook—do you seriously think the author wrote all that down by rote memorization? Of course not: it came naturally as a byproduct of study. Memorization for the sake of memorizing is the worst possible way to learn mathematics.

The 2 aren't mutually exclusive. If you give 2 middle schoolers a test and one knows all of his time tables by rote, he's going to fly circles around the "kid with deep understanding" that's writing out the multiples manually.

Is what you are measuring actually meaningful?

Also you dont need to write them out, you can still calculate it in your head "manually".

I think second student may have easier time to apply knowledge on things like 4 * 2,75.

If the test is purely about multiplying small numbers and the other kid don't know multiply, maybe.

If the test is designed to test understanding, then the kid with deep understanding will win.

By your definition, multiplying kid is faster at multiplying, but don't really know what he is doing outside of memorized situation. And at other is better at everything else.

This applies to lots of areas though, the multiples is just one tiny example. Like rote memorization of an api for a language. It's far easier to see and plan ahead when you don't have to constantly stop to figure out something else. More of your mental energy can go toward higher level thinking when you don't have to continually downshift to figure out the low level pieces.

Now let’s ask both of those kids to multiply 226*148, and see who comes out on top.

If you give them a typical school tests designed to test a typical school scenario, then the kid best at school will get the highest score.

But the question is, if school tests are at all related to reality. I don't think so.

How do you have a "deep understanding of the material" without committing any of it to memory?

Would it be better if memorisation could be achieved through distributed practice, vocabulary learned through reading and listening, spaced repetition achieved through well written textbooks, curriculum, mentorship, etc? Absolutely, but these factors aren't always a given and often are often outside of one's own control. If failing to keep good Anki hygiene is considered in the ROI calculations, these issues should be included to.

Maybe your textbooks just don't distribute problems well enough, the curriculum changes drastically such that what you've just learned isn't touched on again before you forget it, or maybe you just don't have the time or material to consistently practice foreign languages for 1-2 hours a day. In these cases, the 10-20 minutes spent reviewing cards serves as a useful crutch in preventing that information being forgotten entirely, which undoes all the time spent trying to learn it in the first place.

If you have 20 minutes a day to use anki for memorization in foreign language, then you have 20 minutes a day to consume foreign language media. You just swap activity. You do not need perfectly designed media for that, you need any media slightly harder than your level. Internet has them.

That is exactly setup where anki makes less sense.

Anki makes great sense when you have a lot of time for that foreign language activity and are spending portion of it in anko vocabulary.

I find that Anki, by showing up the cards at certain (optimised) intervals, delivers distributed practice. Out of the efficient learning methods as evaluated by this paper: "Improving Students’ Learning With Effective Learning Techniques: Promising Directions From Cognitive and Educational Psychology" [1] (scroll to Table 4 for results) Anki delivers both distributed practice and practice testing in a very natural way.

[1] https://journals.sagepub.com/stoken/rbtfl/Z10jaVH/60XQM/full

This is where I've found ChatGPT sooooo effective in searching.

I ask it a question much more naturally vs google - meaning less cognitive load and the answer it spits back is usually a result of multiple sources distilled down to the smallest amount of information - basically all the "fluff" removed. This means less time spent reading, filtering, finding the answer.

It's been wrong a number of times which costs me 5-30 minutes at a time, but I've severely benefited in time overall by orders of magnitude.

The other high ROI activity is writing/re-writing and writing these prompts yourself will make it more effective.

Could you share some examples of the type of stuff you’ve added to this system that’s had that kind of professional pay off? I’m familiar with the approach, but I suspect part of the reason I’ve not been able to maintain a long term habit is I’ve always used it for things like learning a language. Which is great, but doesn’t have the kind of process re-enforcing payoff given I rarely get to apply it.

And I’m drawn a blank on the kind of professional stuff I’d put in here.

Here's an example: statistical methods. I'm a software engineer and was going to start helping with analysis tooling. Having characteristics and example use cases for many statistical methods helped me work faster and make better choices than needing to look things up.

Generalized, cases where you can choose from one of a few options/technique and this choice is an expression of your professional opinion are good. What you memorize are the attributes of the options that you tend to make the choices on. It's hard to get across (and impossible if people don't want to believe it) the difference in speed/quality/accuracy that comes from having 3 attributes of 10 options at instant recall versus a few that come to mind right away and then more if you remember to try.

Some examples from my life:

- Where are the log files for $service stored?

- What do the values of $one_letter_code in this message format/log line mean?

- What do I grep for to diagnose $problem?

More generally, when I find that I have memorised how to look something up (search slack for $foo to find the instructions, type $bar in the browser address bar to bring up the wiki page with the host name I need), I try to just memorise the fact directly, to reduce friction.

I have the exact opposite opinion. My wife is a doctor.

Anki was insanely popular in med school. It was certainly helpful for passing exams, but it's all stuff that can be easily looked up. It's all simple, basic facts.

It's far better to learn the theory, then fill in the details with easy reference.

In theory. But in practice having a lot of facts memorized not only saves you from having to look things up but actually makes you get more out of the things you do look up! See: https://www.aft.org/sites/default/files/LookItUpSpring2000.p...

Besides that, the original task Anki was designed for was language learning, and you can't have much of a conversation if every sentence requires multiple dictionary lookups.

Ha! I used it in law school. Actually wrote my “cards” down on a palm pilot during classes using the funny stylus it had. Professor later told me he had wondered what I was doing with this odd device. Until it dawned on him that I was taking notes! Other people paid hundreds for exam prep courses. For me, SuperMemo (the “app” as it was called back then, before “apps” were a thing) was all I needed.

I am completely self taught (started programming at 14 and dropped out of high school) yet I have never user Anki and spaced repetition once.

I am curious to know how it can make my life better, and I'm not being sarcastic, but yours is a tall claim and I am very used to learn everything I do by myself.

What can Anki do for me in 10-20 minutes a day? Finding or preparing good flash cards is a lot of effort for someone used to ADHD-fueled information skimming to learn something breadth-first.