Skip to content
Off-topic for aiqbench

Off-topic for aiqbench

This story was gathered from a source we follow but falls outside AI and quantum coverage, so there is no aiqbench take for it.

OODing up your Anki game

LessWrong4 min read

TL;DR: You've been using Anki right for ages; why don't you start using it wrong? Using Anki for tasks it's suboptimal for has synergistic benefits for someone already using Anki quite a lot.[1] There are many great introductions to Anki, so I offer undiluted hot takes, and I add whimsicality to teach by example.[2] Using Anki right Language learning Don't. It works, but it's much more inefficient than learning languages the right way: using Input Hypothesis learning allows you not only to have much more fun, but also to use time more efficiently by learning other stuff at the same time. Isolated facts A use case One day, deep in a rabbit hole, you learn about a king who inherited/conquered an empire on which the sun never set. You want to remember the guy's name. He also happens to have a face so funny that its funniness deserves its own Wikipedia page. Coincidentally, that same evening you watch When Harry Met Sally... and laugh at the "white man's overbite". Human memory is proof God friendzones us and wants us to despair about his mixed signals For the rest of your life, you will never, ever, forget that "a funny face guy inherited land from his four grandparents some time in the 1300s or maybe the 1600s, and also for some reason this makes me think of that white dude's biting face quote from that funny what-was-it-called movie". For the rest of your life, if anyone asks you to say the name of that king, you won't remember. Kronecker's bitter lesson You will never be able to remember the things you want to remember in the same natural, effortless way in which you remember the things God wants you to remember. God created whimsical memories; all else has to be the work of man. Scaling up the the time or compute you use to try to infer the answers to your Anki questions won't substantially change how well you remember them. Deep learning with Anki is only possible on the basis of Kronecker's finitist understanding of human memory, as an intuitionistic approach to Anki that rejects all cards except those that can be constructed from intuitive whimsical memories. One finitist, Anki-enlightened ajahn puts it so: The art of Anki is finding the cards that specifically prompt you to autocomplete correctly; brute-force memorization and regurgitation is much more of a slog than casting spotlights on specific features of your mental structures. This is the whole Anki; all the rest is commentary. ε-machining your Anki cards The perfect Anki card should be maximally short but contain all the information from your past life which can be useful for the retrieval of a certain word/number in the future. "Maximally short" means that it should be objectively ambiguous/incomprehensible for others (otherwise you aren't being concise enough); in the next section I advocate for using up to 5 words. Containing all useful information for future retrieval means that, even if you might forget the answer, you should never forget the meaning of your question. If you don't immediately know what you're being asked when reading a question, the question is wrong. It might happen sometimes that you read a question, don't get it, and after some seconds (or after reading the answer) you remember the original God-given intuition and why you wrote your card. This is a good sign! It means the intuition is still present in your mind, but your maximally short question now is a bit too short. Add a couple of words to make it obvious to future you what they're being asked. Rarely it might happen that you have completely forgotten what the original connection was and you might have to change the question to a boring rote memorization on. This happens almost never to me, so maybe the key is rather to get an authentic feeling of when your brain has formed a monkey-brain connection, and avoid lying yourself into thinking two things are monkey-brain connected just because you wish them to be. Epitomize Laconically in 5 Your question shouldn't have more than 5 words, and the answer should be one single word or number. Some strategies: Avoid using grammar and syntax Develop a system of symbols and abbreviations. I use ⌛ for "year in which the following event happened" or sometimes "year I have chosen to associate with a vague phenomenon that doesn't have an objective start of end",⌛⌛ for "year in which the following event started and year in which it ended",[3] 🗭 for "author of the quote, or author of the concept, or "creator" in a possibly very vague sense of creating" ➡️ for "what happened after this, possibly in a very vague sense of happening". Use images instead of words Related: Save yourself some brain cents by always writing Charles 5 and 1500s, never Charles V or the 16th century. The specifics of a card that is perfect for you will depend on your idiosyncrasies, and you'll need to test to get a feeling of them, but here's a good starting point: 1500s 4 grandparents -> never sunset Some instrumental convergence theorems You can't mem

Read the full article at LessWrong

Related coverage