Cloze the Ultimate (Anki Reflections)
Table of Contents
In the middle of the spring 2024 semester, I began an experiment. I had read Gwern's article on spaced repetition and a bunch of LessWrong articles on Anki earlier in the semester, during one of my periodic searches for workflow optimizations1. I had used flashcards in high school, on Quizlet2, but had gone off them as I hadn't gained much benefit from them during that time. I suspected that my studying methodology was sub-par3, and wanted to upgrade it to make it easier for me to keep my 4.0 GPA4.
Now, anyone who has read the FAQ will know that I majored in business, and will be wondering why I'm bothering, since you can usually get an A in a business course just by showing up.
Well, asshole, there's a difference between "I got an A without paying attention in the business gen-ed I had to take" and "I want to make certain that I get an A in all my business courses through my major." While it may be true that using Anki for business school curriculum is like calling in a tactical nuclear strike on an unruly hedgehog, the promise of guaranteed memorization enabling easy tests without long grind sessions in the library or cheating with ChatGPT5, was alluring enough that I took a gamble and used Anki as my primary studying tool for the rest of the semester.
It worked out, of course6, and I continued to use it for my final two semesters7, during which I never scored below a high B on any final, and got As on all8 of my midterms. I have open-sourced my card decks, and will now discuss my workflow, the study load I incurred, and how well it worked.
1. Workflow
Essentially, Take notes in class or from book → grab most relevant parts → find a way to turn them into flashcard form
One interesting point to note is that Wozniak's 20 rules are not set in stone. There were a lot of things that either didn't really fit his "single factoid per card" format that I still needed to remember9, so I ignored the guidelines and freeballed it. It all worked out pretty well; as it turns out, poor formulation can be countered with enough repetitions, and I always worked a week or more ahead, so time crunch never became an issue.
1.1. Cloze the Ultimate
I mixed card types at first, mainly Basic and Cloze, since I was primarily memorizing text (definitions, lists, etc.) However, it turns out that Anki's Cloze format can pull double-duty as a basic card and a cloze deletion10, and can also enable the memorization of more-complex facts, event sequences, and equations11, and thus can serve as a multitool for the more efficiency-oriented user.
1.2. Emacs
Speaking of efficiency, Emacs made my Anki life considerably easier. I already used it to take notes12, and the text-munging facilities paired with its macro and scripting capabilities made my workflow much more efficient and lower-friction. At various points, I was able to chop and screw a lecture into flashcards as it was occurring.
This made the marginal cost of creating a card to drop to \~0, and I promptly created a shit-ton of them13. The primary macros I used were anki-cloze-def, which converted a word: definition sequence into {{c1::word}}: {{c2::definition}}, anki-cloze-list, which did a similar thing for a dashed list, and cloze-region, which just inserted curly brackets around the highlighted region14.
;; If these don't work for you, it's probably because of my weird movement minor mode.
(fset 'anki-cloze-def
(kmacro-lambda-form [?\{ ?\{ ?c ?1 ?: ?: ?\C-s ?: return ?\C-h ?\} ?\} ?\C-l ?\C-l ?\{ ?\{ ?c ?2 ?: ?: ?\C-e ?\} ?\} ?\C-l] 0 "%d"))
(fset 'anki-cloze-list
(kmacro-lambda-form [?\C-l ?\C-l ?\{ ?\{ ?c ?1 ?: ?: ?\C-e ?\} ?\} ?\C-l] 0 "%d"))
(defun cloze-region ()
"Make a region an Anki cloze"
(interactive)
(let ((pt (region-beginning)))
(goto-char (region-end))
(insert "}}")
(goto-char pt)
(insert "{{")))
Obviously, this was helped by only using Cloze, since I could tailor my workflow to creating cloze cards and only cloze cards. This is one of the reasons why I didn't use Gwern's pre-written scripts; the other reason is that the non-standardized nature of my classes15 meant that ad-hocery was required and I might as well embrace it16.
2. Card & Time Load
I started out with 20 new cards per deck per day, and used that up until my final semester, where I switched to 50 new cards per deck per day. I didn't notice much of a drop-off in retention, just that I was front-loading effort whenever I added new cards17. However, this may be a personal thing, so find what works for you.
The amount of time I spent reviewing varied considerably. I never went past an hour and a half, and my longer sessions usually combined a fresh batch of cards in multiple decks, along with the backlog I had generated over the weekend18. Most of the time, my sessions were 20 minutes or less, averaging around 4.5-5 seconds per card, and, as Gwern discusses, declined over time (my final non-cram study session was about 9 cards total).
3. Benefits
So, did Anki help? Almost certainly.
Qualitatively, I noticed that tests and quizzes went by considerably faster, and I was more certain of my answers on them. Quantitatively, my test grades improved19, especially on final exams, which I had previously struggled with20.
From a mental load standpoint, it was definitely a boon. The requirement that you do your flashcards every single day counteracted any tendency to procrastinate on my part21, and having a strong central technique eliminated the "what should I do to study?"22 and "what if I don't get a good grade?"23 questions that inevitably arise in the mind of the underconfident neurotic24. Cramming becomes trivial, since you just use Anki's own cram mode, and have your skill demonstrated to you by the ease with which you demolish a randomly-shuffled deck of 100 cards25.
You also amortize your studying over the semester; 5 8-hour days in the library during finals week spreads itself over 4 months with no effort on your part26.
Overall, a hearty recommend from me; it turns out the med students had something useful all along27. Just don't use it as an excuse to bikeshed your studying workflow and avoid actually working…
Footnotes:
This tends to occur in the early springtime. I don't know why.
Making my own, thank you very much.
It was.
And thus my scholarship.
Through an innovative method my roommates devised that worked around the rootkit browser we were forced to take tests with.
This would be a very different post if it hadn't .
As well as in my personal life.
?
And I was too lazy to put effort into making fit.
Simply cloze-deleting the part of the card that would be the back of a Basic card
As it turns out, the basic structure of "what you need to remember" can be reduced down to a handful of keywords, with the correct conjunctions spawning from the aether. I had previously leveraged this during competitive debate, but never really considered it deeply.
Primarily in org-mode, with a brief diversion into Perl's POD format as another experiment.
Some of you may have already looked at my Git repository and noticed that certain decks are significantly larger than others. This is partially because I had a number of classes that could not be successfully flashcarded or whose grading was based on things unrelated to what was taught in class (like the one where I had to write 9 papers and talk in class), and partially because I went whole-hog on Ankifying everything this past semester to counteract senioritis.
I kept the macros simple both because I couldn't be bothered to add more functionality, and because I would edit and re-edit facts as I clozed them, so something like auto-numbering wouldn't have helped a whole lot.
In how material was presented and tested, as well as how I recorded it.
Also, because interactive > batch. LISP mafia, baby.
This was by design.
I refuse to do any form of schoolwork after-hours, on weekends, or during breaks, unless I absolutely have to. It helps with procrastination and compartmentalization.
This is confounded by the fact that I had previously studied through re-reading my notes, covering up bits of them to remember vocabulary, and unhinged ranting to my parents, to my friends, in the shower, and on the Internet. The third technique was almost certainly the most effective.
I.e., I'd end up with a B of some form, rather than an A.
Since all I'd get out of it would be spending double the time doing flashcards the next day.
"Your fucking flashcards"
I.e., me.
This is operating under the assumption that the cards are good representations of what is on the test. That one's on you.
This is especially useful if all your class material is accessible at the beginning of the semester, so you can ankify it all at once and front-load a semester's worth of effort into a month.
Then again, CS students have the same thing going with Git, good text editors, and scripting. Hum-ho.