At the beginning of graduate school, I decided that I was going to take a new approach to studying. While my undergrad was rewarding in some ways, I felt as though much of the information I learned each semester was totally lost by the time the next round of classes, or even finals, rolled around. This couldn’t be the correct way to educate myself. There had to be another way. Anki was that way.
Anki is flashcard program that uses a spaced repetition algorithm to determine when its cards should be reviewed. The review schedule follows an exponential distribution that’s based on the research of Hermann Ebinghaus on the mind’s natural forgetting curve. Rather than cramming material which you soon forget, or haphazardly reviewing a random set of cards everyday (which ends up being a lot of extra work) Anki allows you remember nearly anything for approximately ~10 minutes of work spaced out over your entire lifetime. I’ve already reviewed anki once, and my opinion of it remains pretty positive. Below I’ll share what I’ve learned over the past year and what my plans for the software are going forward.
Stats and Workload
I’ve reviewed/made almost 5000 cards this year, although many of those are on track to be suspended/deleted (more on that later). These cards are divided into three decks: books (for deep reading of non-fiction books, as well as anki sentence mining), graduate school (which contains cards made from classes and papers I have read for my thesis), and personal which contains birthdays, recipes, the greek alphabet, matching between lexicographic and numerical order and a trivia deck that I am about to retire. Yes, many sources, including Gwern and Michael Nielsen, say that you should combine your decks for maximum benefit. This may happen at a later date, but as I am currently cramming 30 new cards a day of biology material before my oral exam next Thursday, that day seems at least a week distant.
Reviews were most extreme during last fall semester where I was learning new cards in both a trivia deck (more on that later), and thirty new cards a day for graduate school. There was a large drop off in reviews after I stopped looking at new cards from the trivia deck. Falling for the sunk cost fallacy, I continued to review the trivia cards that I already had, thinking that I had already spent most of the time I would ever spend on those cards, meaning it was worth it for me to keep reviewing them. However, I have recently realized the utility of me knowing the voice actors of the Simpsons, or the home stadiums of the teams in the Champion’s League is near zero, making any additional time I spend reviewing these facts a waste. Also many of these cards, because of their lack of connectedness to other knowledge, kept coming up for review, actually eating up a fair bit of my time. Thus, I decided to ax any trivia card that I did not want to know when it came up for review by suspending it and then periodically deleting any suspended cards.
I stopped making new cards altogether for a few months in the spring for similar reasons that I stopped Spanish immersion (Rotations starting, personal issues, etc). I also missed reviews for a couple weeks, as you can see by the 35 missed days noted on the graph. Because it’s active recall, and many of the cards that you see everyday are things you didn’t learn properly or have forgotten, I often wake up dreading my Anki reviews. It’s kind of like taking a pop quiz everyday. Recently, with my studying for my oral exam, I’ve been averaging about an hour a day for Anki reviews and a couple more for making new cards (which doesn’t feel as stressful as the actual review process). This is too much, and I think after my exam is over, I am going to hopefully dial back my new cards to 5–10 a day to keep my daily reviews under 100 cards and hopefully time spent to 15–30 minutes.
This is the closest metric I have to effectiveness outside real test results. Compared to my result from last time, overall mature card mastery has gone down, learning has gone up and young has stayed about the same. I credit the increase in learning percentage to dropping the Trivia deck, and making more effective cards. The mature decline is probably an artifact of the fact that my initial update was only after four months and I didn’t have very many mature cards. Law of Large Numbers means that the distribution now probably more closely matches reality.
One thing that I want to emphasis about the effectiveness of Anki is that you really get out what you put in. Card design really matters, as well as the ecosystem around that card. Consider the example I gave in my last update for a “bad” card.
I’d like to upgrade my rating of this card from bad to shit tier. Not only is this card not really that useful information (identify a specific piece of DNA repair machinery in an abstract diagram), but I have no context for the information even if I do manage to remember that this the CAAK subcomplex. What does the CAAK sub-complex do? How is it recruited for nucleotide excision repair? What is the importance of this process for the cell? I couldn’t tell you, and I don’t have other cards that could because in my naiviety and laziness I thought that merely covering up images on a diagram would help me to understand biology.
No, to really understand a concept you want a large amount of possibly redundant cards covering that concept from multiple levels of granularity. For example, if I’m trying to remember how the Burrows-Wheeler transform works, I’ll want a card for each step in the process, a card justifying each of those steps, a couple cards for a high level overview and a few more for potential use cases. Coming up with these cards while studying is a lot of work, and often requires periodic card editing/addition of new cards/review of the topic until you’ve got the right setup. But the alternative is the knowledge not sticking.
I’ve still got a long way to go in terms of mastering Anki, but I have certainly noticed the benefits of this more in-depth approach for my understanding of biology. For the first time, I am starting to feel as though I have a holistic and accurate view of the field as a whole. This is a great feeling, and makes me want to shake my undergraduate self for not paying more attention in class/discovering Anki earlier. Almost nothing beats actually feeling competent at something.
What I’ve Learned Since Last Time
- You need more cards than you think. The best Anki card requires that you only remember one piece of information. This gives your review algorithm maximum modularity to target specific pieces of information that you may be missing. For even a single concept in biology or a single algorithm, this usually means many more cards than you think it means as there are many levels of granular detail.
- Anki is not free. When you decide to study a concept, you are committing potentially hundreds of cards and tens of hours to understanding and remembering it. While this is a good trade off for things that you are likely to see in the rest of your career (biology and algorithms for me) or life (Spanish, paradigm shifting books), it is not worth it for trivia. Just because you can Ankify something doesn’t mean you should.
- Be careful what you Ankify. During the fall/winter I was Ankifying Daniel Kahneman’s Thinking Fast and Slow. I have since stopped reading about 2/3 of the way through the book and discontinued my Ankifying because most of the conclusions of the book have not passed the replicability crisis. Even though I couched all the cards that I made from the book as being from the book, all these untrue facts about social science are stuck in my brain as if they were true. I think it may be a better bet to stick to more hard sciences (math, physics, biology, etc) when ankifying books. Or alternatively, when ankifying something from history, to draw from multiple opposing sources.
- Graduate School Studying I have my qualifying exam in 8 days. Between then and now I will be cramming in as much biology knowledge as I can. After the exam, I plan to use Anki to remember the three weekly papers that I read, as well info from seminars and lab meeting.
- Spanish Sentence Mining See my Refold posts for more information about this. Basically this is a strategy to enhance uptake of lower frequency vocabulary.
- Deep Reading Books While Thinking Fast & Slow was a failure, I think the general principle is still sound for ankifying books. However, it may be better suited for books I have already read, and VERY low frequency (it is strenous to ankify a book), perhaps 1–2 books a year.
Hope this was enlightening, and if you have anything to add to the discussion, leave a comment below or on reddit.
Deus ex Vita