* Adding too much. Either from pre-made decks or just making cards for absolutely everything.
* Adding things you haven't learnt yet. SRS works best when you basically know something and want to still know it in 6 months time. Make cards at the end of a math study session when you have completed multiple problems. Don't make cards when you first encounter a piece of information.
* Treating SRS as a goal in and of itself. I see this in language learning a lot. "If I just memorise these 6000 words I will be fluent in this language". If you want to learn baseball you can't just swing the bat all day, you're gonna have to play some baseball games.
* Keep trying. If you're having a bad time with it it's because you haven't figured out how to use it well. SRS must be useful for at least some of your knowledge. Don't add too much and get overwhelmed. Add fewer cards and try to figure out what makes a good card vs bad.
* Stop adding cards if you get overwhelmed. Do all your reviews for a while and the daily number will reduce to something more manageable. Let the magic of SRS work itself out. You get the real benefit when it starts spacing things in years.
* Start slow. Develop your skill of making good cards by making a few and observing your mistakes then ramp it up as you gain confidence.
* Delete or fix bad cards aggressively. Flag them if you can't fix them immediately.
* Use cloze deletion. This is like "[Paris] is the capital of [France]" and you get two cards where one or the other is blanked out. Much easier to create cards this way. Be careful not to make too many.
* Use image occlusion. It's like cloze deletion for pictures. Really excellent for visual stuff.
I’ve used Anki for language learning for several years, and I’m consistently surprised by what I can recall, and how that directly translates to reading content in that language.
For books, I’ve been using Readwise for at least a year as well — it will surface highlights to you in an SRS-like manner each day in a more lightweight way.
I agree with some of the other comments I’ve seen that we don’t need to remember everything. The author says they tried Anki, but dropped it because of the sheer amount of review it would require. But this is the point, in some sense. Studying with spaced-repetition is effective because it is tedious. So I’d only recommend it for information you really want to internalize, or languages.
I’ve sunk countless hours into Anki and I don’t feel like I have a lot to show for it.
If I could do it over I would use Anki less and read more out loud, and write notes on paper as if trees didn’t matter.
I feel like we are often met with the biggest success stories on the web with regards to tools and it can feel very discouraging when things don’t work as well for yourself. So my point is just to chime in and say “some of us also fail using these tools”
I also think there is a real skill to formulating good flash cards when using SRS that I for one don't think I've yet mastered.
The underlying "atoms" you embody in the cards matter, and possibly matters a lot.
A lot of vocabulary decks based on textbooks used in class, e.g.:
人 -> person
I also did a large premade deck of example sentences from a grammar book, which I actually found to be very helpful for general language understanding.
I've also failed trying to complete Heisig's Remembering the Kanji, using Anki.
So I might put in: ハンバーガーを食べる
Then there's two good ways to practice this. One is just reading/recognition. Can you understand the sentence?
The other way is to make it into a card to practise output:
The part in brackets is a hint to help you get the correct word. Your hints at the start will often be direct translations of words, but you can also use images, emoji, dictionary definitions in the L2. Anything is ok, or combinations.
I often make my output cards very easy:
While it may look too easy, I still have to remember the word I'm studying.
In general words taken from real books, TV shows, youtube videos tend to stick in my mind a lot better than words from textbooks, wordlists and grammar books. Those tools have their place but they're not the be all and end all.
Good luck! And remember it's a long road. I've made 6000 flashcards over 10 years to learn a second language and I'm still far from "done".
disclaimer: not affiliated, and WK is not a perfect product (they have some weird quirks), but I've found it to be rather effective. You can also the first three levels for free.
Personally, after 10+ years of consistent SRS, I focus only on vocabulary/terminology and fundamentals. Trivia is never worth it.
There are some way to get things into long lasting memory that may be less documented academicaly but do work : learning by doing for instance. No one learns Java by putting all the elements of the language into Anki and learning them with Anki. One learns Java by writing Java programs.
I have tried many times using SRS and that's not something for me. I need context and enjoyment. Flashcards are too articial a learning environment for me.
The catch is that, in my experience, SRS won't help you learn material you've never seen or used before. For programming, I make Anki cards for a given language/library/algorithm during or after the process of actually implementing something with it; then I can add my own code snippets into the card. For natural languages, I make a card only after I've seen or had to look up a particular grammatical construction or vocab word; then I can add my own example sentence, or the context where I first saw the term, into the card. So long as I'm regularly reviewing, then even for rare idioms or vocab, I can remember things I last saw years ago.
SRS is not as effective if you're studying prebuilt material by rote without any other interaction with the content. Rather it should be one part of a broad portfolio of activities you do to learn. For coding it should complement things like implementing projects, learning new libraries, reading open source code, reading technical books on the subject and so on. For languages it should complement things like talking to native speakers, reading news articles in the target language, listening to podcasts etc. where you use content from your other things and turn them into Anki cards.
Spaced repetition is an evidence-based learning technique that is usually performed with flashcards. Newly introduced and more difficult flashcards are shown more frequently, while older and less difficult flashcards are shown less frequently in order to exploit the psychological spacing effect. The use of spaced repetition has been proven to increase rate of learning
Have you been able to do this ?
For this author, visualizing worked well, for others maybe not.
The sad fact is that all memories fade over time unless recalled.
My opinion: commit only the most important ideas to long-term memory. The rest can be looked up.
If you're not remembering something your mind's "garbage collection" system has come to the conclusion that it's not relevant enough for active recall.
I'm for changing approaches to the material to change it's relevancy score, but forcing my mind to remember something's turns out usually to be pointless.
Also... A mind full of irrelevant visual associations from some mnemonic techniques seems littered and distracting.
For more scarcely needed information, the latency of recall (200ms vs 10s) doesn't matter, and so it is often preferable to store it outside of the brain. It might still be important to store a pointer to that information with some metadata in the brain so that it can be used in conceptual reasoning. But the actual data can be stored elsewhere (e.g a phone number on a smartphone).
Anyway, I use a PO Box ... that I also can't remember.
Mine's similar: By default, take notes and figure out a way to search them quickly whenever needed. Only memorize those things where lookup adds too much latency/is impractical.
For note taking I've found Vimwiki to work well, and fzf.vim for searching. However effectively taking notes is something to be learned, and it's definitely not just "summarize everything you've learned".
One particularly useful form of this is keywords and jargon: If you memorize the formal names of nebulous concepts, you’ll be in a much better position to look up the details when you need them. For example:
“Encoding scheme derived from a tree of symbol frequencies” => “Huffman Coding”
An internet or literature search for the latter term will turn up relevant results much more quickly than searching for the former directly.
Unfortunately, fun and ease is not synonymous with effective use of learning time. Novelty can be a good motivator for initial learning, but without spaced repetition retention is lower. People love the feeling of mastery, so doing the reading and homework in quick succession is a great form of dopamine. IMO you can either bite the bullet and recognize most of what you read will be forgotten in a year, or recognize that most of learning is boring.
When I ran through a stats textbook last summer, the authors clearly had spaced repetition in mind. Each chapter was divided into subsections with small problem sets as you read, and an end-of chapter question bank covering chapter contents (and then a cliff-notes chapter summary at the end). Each of the three major sections included a comprehensive question bank that are basically an exam. End result is that with a pace of about 10 pages per day, you'd do some reading and some problem solving, but every few days you'd need to reach back in your head to cover pages you read a few days ago. The end of chapter homeworks were often much slower going than the question sets sprinkled in, and the exams were equally rough.
To supplement that, every weekend I'd go through chapters completed that week and add some Anki cards. I don't necessarily need to know how to rapidly calculate correlation, as much as feel confident I can re-learn it later if I need to because I did so once already.
That's the method I recommend. It takes time, though, that I sadly no longer have. There are probably ten textbooks on my shelf waiting that I can't tackle for lack of time.
Often, the answers would appear later in a chapter. Sometimes, re-reading the chapter would allow me to catch answers to questions I'd missed. If questions remained after reading and re-reading, it was time to talk to someone else who had already studied the material.
Records transmit knowledge. Not all human knowlege is equally facilitated by explitic (spoken or written) transmission. Different domains and topics exercise different skills: memorisation in the case of history and a substantial portion of practice lore, structured knowledge in the case of most of the sciences, and what might be considered cerebral skills development in areas such as maths, logic, rhetoric, management, negotiation, sales tactics, and relationship management. These are learnt in different ways, and books play different roles in their education.
Numerous HN submissions: https://hn.algolia.com/?dateRange=all&page=0&prefix=true&que...
Notably: https://news.ycombinator.com/item?id=12209446 https://news.ycombinator.com/item?id=177214 https://news.ycombinator.com/item?id=1002360
One thing I tried that I liked was reading the entire textbook out-loud into a set of bluetooth headphones chapter-by-chapter. I then process the raw audio file using Audacity's 'truncate silence' tool, as well as increasing the tempo of the audio file by 1.5-2x speed. I read it mechanically, straight-through as quickly as I possibly can on the first run. Then I make 1-2 passes back through the entire book using the audio file of my own recorded voice sped up in order to pace myself.
That way I go through the entire book like 3 times. Once slow, twice fast. I might do my second or third pass months or years later. Doing it mechanically means that I can grind through nonfiction technical books that are hundreds of pages long pretty quickly, and I know exactly how long my reviews will take, since it's the length of my audio file. The largest book I've pulled it off with was 900 pages. It was a sufferfest, but if you smoke a little pot while you're doing it, you can get into a nice rhythm and it's kind of fun.
I don't remember every single detail of the book using this method, but I've found cranking through them relentlessly with some time in-between to be very, very effective from a practical perspective. It scratches the 'done is better than perfect' itch, and when I need a specific detail I know exactly where to look.
Memorizing a text book has very little to do with learning a text book.
I just think the categorization of "textbook" is nonsense to start with. Of course you want a book with text as opposed to a picture book.
You have to start with the right book. The variability between books on a subject is so huge and probably even different person to person based on prior knowledge.
However, this is not strictly necessary because when you relearn a subject, you often relearn it faster due to previous experience(memory) with the subject matter.
If your goal is to retain skill/knowledge over time, than that's a different matter.
Learning is gaining the "wisdom" of the material. If all you're doing is rote memorization, you're missing the forest for the trees.
Having the knowledge is difficult if you don't have the information.
I don't completely subscribe to Bloom's taxonomy of learning as the complete picture, but has at least some kernel of applicability.
The idea that there are different interdependent levels to mastering some field sounds true to me.
It’s a critique of AI but gives a thorough take on the progress from novice to expert.
Having a Grey's anatomy book in front of me, all available to read, does not make me able to perform anatomy at an advanced level.
It's similar to having data in secondary storage, that still needs to be loaded into a process's working memory, before it can be worked on. This is specifically true of condensed terminology that's used to compress further explanations.
Following the anatomy example - if I don't have the terminology to describe the different positional views available without doing an external lookup first, following a basic description becomes very difficult and involves so many lookups as to become impossible to follow.
I can't find the article/paper, but I remember reading something where someone calculated the trade-off for a specific API between knowing a percentage of the API method signatures of by heart versus looking up each one using the IDE documentation each time. It made a very big performance difference in the end.
BTW: knowing what you don't NEED to remember and can just reference when required, is probably part of the skills making you an expert.
I'm reading a bunch of textbooks and adding notes as I go to anki. I find that so long as I'm equally as willing to _drop_ a card from anki, it works well. However if I focus too much on this then I lose the flow of just reading and absorbing.
* Crafting Interpreters
* Real-World Cryptography
* The Algorithm Design Manual
* Some materials science book but I've forgotten the title
* Introduction to Probability, Statistics, and Random Processes
* (just ordered) AI: A Modern Approach
But when I say 'reading' I really mean that I've read the first few chapters and have gotten stuck on most of them.
I kind of doubt they'd even make sense if they were because they're crafted to the way I think, and mostly terrible.