Hacker News new | past | comments | ask | show | jobs | submit login
Want to Remember Everything You'll Ever Learn? Surrender to This Algorithm (wired.com)
172 points by naish on Apr 22, 2008 | hide | past | web | favorite | 61 comments



I will make a shameless, shameless plug.. (disgusting)

I have been working on a simpler spaced repetition system (the Leitner System relies on a simple yes/no response) implemented as a wepapp: http://flashcarddb.com

The article mentions the popularity of spaced repititon software in Poland. And sure enough a large number of my incoming users are from Poland and a few other European countries that are familiar with the concepts.

Another open source Supermemo alternative is Mnemosyne.

I have thought of adding the option of a more complicated rating system for grading answers but most new users tend to be relatively new to the idea of spaced repetition. So, for now, I am keeping it relatively simple.


(edit) Sorry. I misread your post. Here's the rest of my original comment:

I've been using SuperMemo off and on for the last ten years. It is by no means an elegant or user friendly app. However, there is a lot of science behind it and it's worth checking out by anyone interested in memory. Readers of yc should have no problem with it.

Another point worth mentioning is that Wozniak has shared his knowledge with the world. He publishes each algorithm and explains it in detail. He also writes extensively about memory and learning and it's all available on his website. He also provides free downloads of older versions of the app.


Dude, I meant my reference was a shameless plug. I was doing the plugging. Sorry if that was unclear!

I couldn't agree more with your points about Wozniak and Supermemo. Wozniak, Leitner and others were my inspirations. I used their systems while studying myself.

(Edited the original to make it more unambiguous!)


Likewise. Sorry about that.


I think r7000 (a significant improvement over the earlier r6000 model) was referring to the shamefulness of plugging his/her _own_ software. ;)

So no Mac version? The site mentions an in-progress Linux version but it sounds like it's not happening.


I just starting using Anki ( http://ichi2.net/anki/ ) which runs on OS X, Windows, Linux, and is open-source.

Accepts Mnemosyne decks, too.


I'd suggest having something which explains that the idea is you think of the answer and then tell the computer whether you were right. It took me a while to randomly click "Show answer". I figured that was just for when I gave up.


There seems to be a lot of these flashcard applications. Another I just came across is active-flashcards.com.


Strange - I was on the verge of discovering this for myself for my own language site. I'm already storing the data, but hadn't started with the number crunching yet.

On another note - isn't it scandalous how inefficient education is? Cramming all those mind-numbing details in your brain the night before an exam, and then forgetting 70% of them after a couple of months. Ridiculous and wasteful.


That would be remarkably good retention. I guess it's way worse than that on average, at least for stuff crammed once with no more interest than passing an exam.

Back to the main point: yes, school is very inefficient, but I think the problem of student apathy and disinterest for the subject matter is in more urgent need of fixing than retention itself.


Perhaps it's not so wasteful. The stuff you remember is the stuff that you keep on using, while the stuff you forget is the stuff that was never really important. Perhaps learning a vast amount of stuff and then filtering out everything except the stuff you need to use again is the best way to fill your finite amount of brain space with useful knowledge.

That said, I'd be happy if my brain had retained a bit more of the undergraduate mathematics, and a bit less of the lyrics to theme tunes from 1980s sitcoms, but never mind.


Spaced repetition is killer. It seems to take the work out of remembering things, leaving nothing but determination. I've been using http://kanji.koohii.com for awhile, which is an excellent implementation of the technique for remembering the meaning of Japanese characters using a particular method book. A niche site, to be sure, but an excellent implementation overall.

I also recommend http://ichi2.net/anki/, which has a Japanese focus but is by no means exclusive to it.


This is good for me; I'm studying for the JLPT 1, and writing a simple web application to assist in vocabulary drills. The application is part of my final project, so I'm not stopping development, but I'll definitely look into adding some of these timing methods into the system.

Also, just to let you know, I've found that studying kanji on their own is generally a waste of time; without context, they tend to lose their meaning quickly. Even if you ask Japanese people to explain kanji, they usually do so by referencing compounds in which that character is present.


Second reply: another interesting use of spaced repetition, in a somewhat different context, is the AJATT method: http://www.alljapaneseallthetime.com. It uses complete sentences as the cues, thus ensuring valid context all the time.


I have found the same thing. The Heisig method seems to serve mostly to bridge the gap between learning a character and learning a word that uses the character. While I can learn character meanings with the memorization technique, the really don't stick until I've got vocab to go with them.


Anki looks great, thanks for the link.

I was just thinking today how it's time to start studying Japanese again. I studied there for a year when I was in high school, but have forgotten a lot since then. Time to relearn :)


Wozniak is an interesting and respectable guy; I have the feeling that he tries to remember everything, or something close to it. This strategy works well for declarative memories, but does not generalize with the same efficiency for other memories.

SuperMemo is an ingenious program with a rotten UI. I have been looking for better ways to do this and would be interested in speaking with anybody else with similar interests.


Maybe this technique could apply to other kinds of memories when combined with lifelogging. I could video tape all the time and then choose certain portions of video to retain in memory by review.


Absolutely.

However, a video taped portion of your life contains memories encoded in many different modalities (as opposed to, say, information on flashcards, which is, for the most part, factual). This mixture of memories makes an algorithm specifically designed for a certain type of memory a suboptimal one. That's not to say it's not useful, but there is a lot of improvement to make in this area (and I would very much like to be a part of it! :-) )


Pimsleur language CDs use this technique; I believe they call it "graduated recall" or some such. It's been very effective for me. Of course, it's not fine-tuned for the individual, but it's still pretty good.


On the surface this appears to be one of the best things ever posted here. I'd like to see some other references.

The article says the forgetting point is different for everyone, and is different for different sorts of info. Does the software somehow find the right point? Or is the variance not enough to worry about?


The software finds the right point. The variance is also not enough to worry about. Over time, you get an approximation, which is not perfect, and I would go as far as to say that it will never be. Nevertheless, the approximations are good.


The inventor of the ideas behind SuperMemo has a fascinating site in which he explains how he uses the same principles to organize his life. http://www.supermemo.com/english/contents.htm

For instance, he states that "lifetime minimum cost of remembering a single well-formulated piece of knowledge can easily be estimated to fall into the range from 30 seconds to 5 minutes (minimum cost is the cost that is achieved using the optimally spaced repetition schedule). This estimation can be used in cost-benefit analysis in deciding which mission-critical pieces of knowledge should be subject to learning based on spaced repetition."


I love this article and wish there was more like it--thanks for posting.

I also wish Piotr Wozniak would write a book (even though he'd probably consider it a waste of time). Work like his is so important, and positive, and humanitarian. Even if he's just doing what he enjoys, I believe we're all the better for it.


Those 7 pages cannot be necessary. Could someone sum it up in a paragraph with wikipedia references?


I found it a fascinating article, and also skipped 2/3 of it.

Summary: memory decays exponentially, (e ^ -t) but slower with each repetition (e ^ -0.5t). So there is a sweet spot for timing repetitions, and it can be calculated if you store past data, which makes memorization much more efficient.


Solve this problem automatically, and you have a viable startup!


While we're waiting, zapreader (http://zapreader.com) solves half the problem.


Also check out dictator, a more robust form of zapreader, but not for browsing. I think it helps me read alot faster.


Neat, I've been looking for something like that. Thanks.


I agree. Those 800 pages of Anna Karenina couldn't have been necessary either. Fortunately, I was able to aquire the Cliffs Notes.


For those who don't want to read the book, it was something about Russia.


I didn't read it, I was too paranoid that I'd get to the last page and it would be a rickroll.



The article actually discussed this very issue.


If you're too lazy to either read it or use google, then why shouldn't we be?



(on reading..)stopping the instant his mind begins to drift or his comprehension falls too low and then moving on to the next item in the queue.

brilliant idea in itself. (bought a copy at $19)


The logic behind this, "acquire a piece of knowledge, then don't forget it" is similar to the description one of my university professors gave for learning how to fly: "fall over and don't hit the ground."

I suspect that physics may remain a limiting factor for the professor's suggestion, but SuperNote is an intriguing application. I believe the Pimsleur language courses use a similar method of 'staggered reminding' to encourage your retention of knowledge.


In case you were curious of the source of that comment, it's from "The Hitchhiker's Guide to the Galaxy".

"There is an art, it says, or rather, a knack to flying. The knack lies in learning how to throw yourself at the ground and miss. Pick a nice day, [The Hitchhiker's Guide to the Galaxy] suggests, and try it."

Douglas Adams was awesome.


Ah, but the way to not forget it is to simply refresh it right before it's about to be forgotten. The idea behind SuperMemo, the software described in the article, is an attempt to do just that. Learn when it is your likely to forget and bring it up again around that time. It's a cleverly ingenious idea.


fascinating article. wonder if it also applies to correcting wrong memorization -- ie to break a mental habit, space it out according to this formula.


As opposed to creating one stimulus-response association, correcting a wrong association would involve two (or more processes). One, reinforcing the correct association; two, not enforcing (or inhibiting) the wrong association.

Reinforcing the right one can be spaced, but breaking the habit is harder. One study involving OCD patients achieved this via a staged process. I forgot the details, but essentially, you train yourself to use your bad habit as a cue for another behavior, which diverts yourself from the result of your bad habit. Something like, if I wanted to stop washing my hands obsessively, I would put a message on the faucet that said "take a walk!" Then, every time I get the urge, I would take a walk, etc. It's obviously not this simple, but that should illustrate how reshaping old behavior is different from creating new ones.


Isn't this really good for route learning? Which outside of a limited few situations (vocab, exams) is the most useless type of learning there is?


Given that the vocabulary (by a loose definition of "vocabulary") of any new skill is a significant portion of the barrier to becoming proficient, I would say that making that aspect a solved problem would be a miraculous improvement in learning efficiency.

For example, learning a new programming language (once you know how to program in general) is at least 50% vocabulary. It's learning the keywords and syntax, library functions, etc. Likewise for most kinds of skills. Sailing is largely knowing the names, the knots, and the laws (plus practice). Driving is laws, and a few basic automotive concepts. Law is lots and lots of cases. Medicine is lots and lots of anatomy. All of these things are memorization, and that memorization is a barrier to expertise.

Memorization is clearly not the most useless type of learning. It's core to almost every skill.


Touched on slightly before, with no answers - however, I feel that the question as to different peoples memory retention is a valid one. I know myself that being Dyspraxic, I have quite a low retention rate when it comes to short term memory. Could be of use to set a base level of memory retention and then use that to define the length of time between repetitions?



It will put America and silicon valley to shame if this guy can't get some capital to take this to a web-based system with a better interface. This deserves to be used and available.


Is there a Mac version?


"Genius" is an excellent Mac flashcard program with spaced repetition.

http://lifehacker.com/software/featured-mac-download/memoriz...

http://web.mac.com/jrc/Genius/

A lot of Mac people sing its praises.


Should I be suspicious of those glowing reviews, given as the article says, "[the] problem here, from the psychologists' perspective, is that the user's sense of achievement is exactly what we should most distrust"?



Thanks for sharing, very interesting article.


update: here is the online version! http://www.supermemo.net


does Rosetta stone employ something similar?


Nope. The article specifically states that Rosetta Stone violates most of the principles the learning research has shown.


i don't know... the algorithm killed jeeves


That's because the algorithm is from New Jersey. If you know what I mean.


whoo... tough crowd :)


I would venture that most of us visit for insight, not repartee.


Touché




Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: