嘘とワンダーランド

Well, I mentioned a while ago I was trying modifying 3h everyman sleep to make it work for me. I tried extending the core, but that made it worse. Anything except 3 hours is pretty much useless as a core length. 90 minutes works occassionally, but it’s not really long-term.

Anyway, I then tried a second core and soon found myself ignoring the naps and just going for biphasic straight away. Well, it works. But it kinda sucks. It works way better than monophasic for me and I will now consider it my default to fall back on. I aim for 3h + 90min, but on average I get more like 5h in total, oversleeping either on the short core or falling asleep 2 hours too early on the long core. And I need caffeine again (low dose though, 1-2 weak cups of green tea per day) to make it.

It’s not so great, so I will go back to everyman, past experiences be damned. I bought a sleep mask and I will get my naps, dammit. But I’m switching right away because there’s an important exam in 5 days and switching right before that would be suicide. (I already did this once and almost failed it.)

Anyway, the real reason I’m going back is a simple statistic. It matches many of my own memories, but now I have reliable data. Here is a list of time I spent each day on several study topics in total. (This counts stuff like learning for an exam, studying vocabulary, but not just watching a movie.) Guess until when I was monophasic, when I switched to everyman and when to biphasic sleep. I was equally busy on all days (i.e. not at all).

study time (in hours) per day for the last 40 days

Polyphasic sleep doesn’t just give you more time – it changes how you use the given time as well. In most cases, for the better.

There is a major philosophical schism in language study. Well, learning any skill in general. Let’s explore them via an example that I just today made a final decision on. If the following sounds to specific, then this is a misconception. It applies to any process of learning.

What’s the problem? Say you learn vocabulary via a SRS. That means adding the given item (be it character, word or sentence) as a two-sided card, with a question on the front and the answer on the back. In the case of Chinese characters, the question would be a keyword (or short phrase) containing the meaning of the character (e.g. “to guard” or “にち”) and the answer would be the actual character (e.g. “守” and “日”). You would see the meaning, remember the character and write it out (mentally or with a pen) and check the answer. So far, so good.

However, here comes the schism: Should you add hints to the question side, i.e. a picture, additional sound, any mnemonic you use or whatever kind of hints you can think of short of giving the answer away? There are two schools of thought here:

One group believes that you should study thoroughly and face hard goals, so that you will be able to get good results in the wild. In this case that means not giving any hint at all. You should have internalized these hints already and rely only on your memory. You wouldn’t have access to these hints in a real texts after all. The alleged advantage is a better retention rate, with the drawback of being slower and being slightly less fun. But in the end, it pays off. I will call this group the Nietzscheans because “what doesn’t kill me only makes me stronger.”

The other group thinks you should make the process of studying as easy and fast as possible. In the given example they would always give their mnemonic and maybe additional material on the question side. The alleged advantage is the ability to cover more material in the same amount of time and less exhaustion. Your real retention rate might be lower, but it will pay off anyway. I will call this group the LaFargueans because the exemplify the Right to be Lazy.

So, which group is right? Let’s make the example more concrete. I will be intensively studying a total of roughly 3500 characters so I can get up to speed on all recent Japanese literature. I learn about 100 new characters per day and do daily reviews, so this will take a significant of my time in the near future. So I am very much interested to decrease my workload.

I have read several reports and my own statistics to derive three basic approaches:

a) Use no hints. This make the card a bit harder to get right at first (80% retention rate), but after you have one good answer down, you will be fairly consistent (90% retention rate).
b) Use hints. The first review is somewhat easier (85% retention rate), but still involves some work to get it right. After this, you will almost always be right (95% retention rate).
c) Use hints at first, but drop them later. This is similar to b), but after 3 correct answers you switch off the hints. This will make you fail a lot more cards at first (70% retention rate), but you improve fast.

Now, how much reviews would you have to do daily with each approach? Graphs to the rescue!

daily reviews

same data, logarithmic scale

As you can see, using hints some or all of the time makes almost no difference with regards to reviews, but both approaches are quite a bit better than using no hints. How much better? Using no hints, the Nietzscheans will do about 47,000 reviews in 2 years. The LaFargueans will do about 37,000 reviews. That’s 10,000 reviews less, i.e. 20% less work! And because they are easier, they will also be faster. In my experience up to twice as fast.

How significant is this? Well, 20% less work means you can do 20% more in the same amount of time. The LaFargueans could learn 700 additional characters, not spending a minute more than the Nietzscheans. Of course they would argue that the LaFargueans won’t learn them as well. That’s true. Their real retention rate is lower. Let’s be generous and say the Nietzscheans have a retention rate of 90% after a year, while the LaFargueans have only 75%. But because the LaFargueans could cover a lot more material, they will know the same total of characters (3150), but their knowledge will be a lot more diverse. And they will be done faster (and have more fun). Instead of studying more, they could also stick with 3500 characters. That means they will have trouble with about 525 characters more. They could then spend their 20% more free time on them and would still be done faster (learning 525 cards separately would involve about 5000 repetitions, still 5000 less).

And this totally reflects my experience. Those that go the easier, more playful way always outperform the stricter ones. They can learn a lot more and see results earlier. And that’s why you should always cheat.

Let’s do some blatant propaganda for Spaced Repetition Software aka SRS.

Say, you want to learn something. Something big, like, Japanese or Chinese. Japanese uses 4 different writing system, but the one that stands out are the 漢字, i.e. the thousands of funny symbols. To be literate in Japanese, you need to now about 3000 of those. How would you learn something that huge?

To learn anything, you need two things. First, the information must be sticky. That means it must be represented in a form your brain can actually remember. What that means is: Ever tried remembering a long number? Like, 20 digits long? Impossible, unless you break it down. But ever remembered the whole plot, including all scenes, of a great movie? Totally easy. Your brain can remember pictures and narratives (related things, both by time and cause) easily, but abstract information is very hard. So you need to transform the 漢字, or whatever your learning, into pictures and stories, aka mnemonics. Fortunately, they were designed with that in mind, so that’s very simple.
Second, you need to review regularly. Your memory is leaky and needs constant reinforcment. Fortunately, every time the memory is refreshed, it will stick around a lot longer – roughly 2-3 times as long if you review just on the brink of forgetting. If you know some math, you’ll recognice this as an exponential progression. What does that mean? You only need to review about 7-8 times and the memory will stay for decades! So, that’s manageable. Unfortunately, the brain is a little faulty, so you will forget a few things anyway. The good thing is, though, that with very little effort, you can already reach a retention rate of 90-95%, so on average you only need around 10 reviews per fact to make sure you’ll remember it for a very long time.

That sounds pretty nice already, but still, 3000 漢字? Isn’t that a lot of work? No. That’s 3000 facts, meaning about 30,000 reviews. A review takes 10 seconds, at most. On average, it will take only about 5, but let’s assume 10. Worst case scenario, you know. In total, that’s only about 3.5 days of work. If it were not spaced out so much, you could finish it in a week. Sweet!

Have a look at those graphs.

3000 facts, 20 new facts a day

3000 facts, daily reviews

That’s your work over 10 months. The first shows how much reviews you will be doing per month in total. Yellow is the amount of new (or unseen) facts, red are reviews (or reps) of old facts. Below that is the amount of reviews per day for each month. As you can see, the daily workload is at most 20 minutes and goes does down rapidly. After 5 months, you know all 漢字 and will only be refreshing. And that’s only for a moderate amount of work with 20 new facts per day. You can easily do 50, or even 100 if you are determined. Pretty good, right?

Let’s look at what I’m doing right know. I’ve already learned around 2000 漢字 quite a while ago, but I’ve neglected a lot of them and I found I could only read them, not write them. Which sucks. So I started anew, but while I’m at it, I might as well do some more. :)

Chinese uses the 漢字 exclusively (almost), so you need more to be literate, around 4000 or so. A lot of those overlap (at least 60%) with Japanese, so it’s not a bad idea to learn the superset of 漢字 that both languages use. That would add up to about 5000-6000. I also want to read somewhat older literature and just love obscure sources, so I decided to totally rock the 漢字 and go for over 6000. A realistic upper bound is 8000 or so. After that you’ll have trouble finding any actual sources outside of taxonomy. I will also add 50 per day, on average. Right now, for the first 2000, I’m doing 100, but as they become more obscure, I’ll slow down to focus on more important aspects of the language. Still, 50 a day is maybe an hour of work. Let’s show some graphs.

8000 facts, 50 new facts a day

8000 facts, daily reviews

Less than an hour per day on reviews for half a year, then only up to 20 minutes. Half of them done in 2.5 months, meaning already pretty much 95% literacy. A total of 9 days of work for the reviews and 6 days for the initial learning. And that’s why I love my SRS.

I have been on a 3h everyman schedule for a month or so, but I will now have to accept defeat. You see, I’d have a week of near-to-perfect adherence. I’d fall asleep almost every time, never miss or oversleep and I remember dreams most of the time. But I get more and more tired and after one week, I’m on uberman levels of sleep deprivation and simply have to sleep at least 4h or so more or I’m dead. I start again, same thing. 3x in a row and I accept my schedule and my body disagree with each other.

*sigh*

Still, during the first 4 days I feel great. And I was able to maintain uberman sleep, so I’m not completely incompatible. That’s why I’m not giving up on polyphasic sleep per se, but I do not some fix here and moving naps is not gonna be it. None of my naps seems flawed, so I just don’t see anything to tweak in the first place. I have the perfect everyman schedule, I eat right, take no drugs at all, am busy all the time, but I still have a total deficit.

I’ve thought about my options. I won’t add a nap because this makes it too impractical. I will try a longer core first. I’ll add one hour (maybe 1.5 if necessary) and see if this is enough too account for the deficit. On average, it should be, but it doesn’t work that way. :)
If I can’t maintain this core or if it screws with my naps too much (I can only nap when sufficiently tired – give me too much sleep and I can’t stay polyphasic.), I’ll try an intentional oversleep / 2. core every 3 days or so. Basically do regularly what I would do anyway, but early enough so I never go through the bad phase of sleep deprivation. I know that catching up with sleep is bad for your health, but it’s what I’ve been doing for almost all my time in school, so I’m kinda used to it. (笑)

Let’s see.

I must be insane. I’ve finally decided to learn Chinese (Mandarin, specifically), too. As if I wasn’t busy enough. Oh well, at least it makes me feel better about my Japanese. And more 漢字 is always fun.

Alas, let’s learn some Prolog first. I’m not sure if I’m intrigued or disgusted by logic programming. During the last few months, I have come to dislike mathematics (and especially mathematicians). It is not scientific (because the axioms are arbitrary and not chosen to reflect a particular world) and, worst of all, I feel it’s useless. Ok, that’s too harsh a statement. I will have to elaborate on that, but I don’t have any time right now. In a nutshell, pure mathematics seems to me to produce only barely useful results (outside of mathematics, of course). What good is Fermat’s Last Theorem in the first place anyway?
The biggest problem I have with logic programming (and mathematics, and prescriptivists in linguistics) is their worldview. They believe that their ontology, their categories and grammars actually apply to the world, or anything at all. They don’t. There is no consistent description of the universe. Pretending there were one isn’t a good idea, even though it can get you a few good results in the short-term.
(You only need to look at the great results of computer linguistics and that there isn’t a single grammar book for one language that can hold up 5 minutes in a normal dialog among 12-year-olds. After all those years you’d think they’d give up. Alas, I hope no one listens to them anymore once we have hardware-implemented neuronal networks and machines passing the Turing test.)

It all reminds me of String Theory. Maybe it’s elegant, maybe not, one thing is sure – it is useless. 40 years and not a single experiment. That’s not science, that’s wanking.

For the last 3 weeks, I’ve been logging how I spent my time. The last week I’ve also taken screenshots every 10 minutes to facilitate this. I’m really surprised by the results. Let me break it down. I’m currently attempting to adapt to everyman sleep, which means I should have 20 usable hours per day. It’s not going as well as I had hoped, so it’s really only about 15 right now, but it’s getting better and I hope to tackle this soon. Anyway, out of those I’ve spent, on average, 1 hour on Japanese, about 20 minutes on programming, 40 minutes on studying for university and about 20 minutes studying anything related to AI. Generally, I spent longer periods on one topic, like 2-3 hours, and don’t spend any time a few days later. Also, I’d watch Japanese TV or listen to Japanese music most of the time as well. This is completely inaedequate. If I sum this up, I generally spent only up to 3 hours on something meaningful. Where does all the other time go? An hour reading random blogs here, another hour reading up snails on Wikipedia there, and so on. I’m really disappointed in myself.

Now, I’m also a hacker. I know that it’s useless to get frustrated or angry with a system and instead you just have to sit down, start from scratch and get things right. I have a very strong belief that anything can be fixed, even though it might take a lot more resources and a very different path than what you might have expected or hoped for. So, how can I solve this mess?

As 勝元 would probably say, tail needs to be busted. I have been at this point several times in my past and I have tried a multitude of ways to fix it. I know that punishment is useless, as is well-known among animal trainers. However, I have tried making the right thing easy and rewarding, but that alone was not enough to fix it. It certainly improved it, though, and I have had a few good runs. Simple goal-setting doesn’t work so well, either. Ironically, tracking progress by paper works magnitudes better than doing it digitally. (I do both, currently. One for motivation and one for statistics.)

So it’s time for another solution. Let’s start this whole thing from scratch. How do you get good at something? Time. You need to put in at least >1,000 hours to become decent and >10,000 hours to become great. To be able to put in all this time, you need focus and motivation. If you don’t know where you are going or are getting distracted all the time, you will never arrive. If you are not obsessive about your goal, you will give up before you are there.

In the spirit of “fail often”, I will start a simple experiment. How would an ideal day look? If I could just snap my fingers and everything would be perfect, what would I do on such a day, then?

I get up at 23:00. I eat a little and take a shower, waking me up effectively. I sit down with Anki and complete my daily reps. I have 2 decks right now, one for university and one for Japanese. On most days, I need 15 minutes each, but I need up to 30 on a hard or bad day. I then check my mail. If I’m done early, I study a few more items in Anki.
It is now 0:00. I read some Japanese text (for science!), marking interesting or unknown sentences along the way. [0] To avoid getting tired, I do this for 30 minutes.
It is 0:30. I swith the material to something related to university. I might read an important text or do some exercise, again for 30 minutes.
It is 1:00. I switch to programming. I work on one of my projects for 30 minutes. Although I could easily work longer, limiting it to 30 minutes encourages me to keep the complexity down and get results as early as possible.
Now the whole thing starts all over again. However, instead of only reading Japanese, I will also listen to it and shadow it about 1/3 of the time. Also, instead of studying for a course, I will study something AI related  in an alternating pattern.
This goes on until 5:00, when I take my first nap. After waking up, I meditate for 15 minutes, then eat something and go for a run while listening to some Japanese material, be it music or audiobook.
I’m back and ready at 7:00, where I start the cycle of AI/uni, programming, Japanese again basically up until 20:00, when I go to bed. I take two naps, one at 10:00 and one at 15:00, and eat with my parents at 12:00 (or 14:00, depending on the day) for half an hour. Every third day, I do an additional 30 minutes of muscle training (while watching Japanese TV, of course).

That’s about it.[1] That vision sounds great, like a lot of fun and productive. So, to make this experiment complete, I’ll make a detailed schedule and adhere to it for the rest of the week. Let’s see how practical it all is.[2]

[0] Currently, I have set up a hotkey to save a small selection of the screen as an image so I don’t have to worry about typing while reading. This make thes process unintrusive and often I just put the images straight into Anki, with some additional material.
[1] “Don’t you work?” I’m a student. “Don’t you have to go to university and shit?” I don’t like lectures most of the time. I do go occasionaly, but continue this cycle on the train and so on as much as possible. I don’t tend to let the outside world disrupt me.
[2] Of course it’s a trap. I know it is practical and I want to adhere to it totally, but small plans that get extended constantly are easier to stick to.

Optimal Brain Damage

(it’s a form of backpropagation used in neural networks)

All the problems I encountered and how I solved them. Except for those that puredoxyk has already answered.

Problem: I wake up groggy after my nap.
Solution: First, make sure you are not simply adapting to naps in general. During the first few days it’s inevitable to feel really tired after a nap. However, the exact length of a nap is important. Contrary to what you might think, shorter is better. Start with 20 minutes and stick to it. Once you have adapted enough that you can fall asleep in a minute or so, you might fiddle with nap length.
If you tend to wake up during dreams and still feel tired, the nap is too short. If you wake up groggy and barely concious, it’s too long. I found a length of 22-23 minutes to be ideal.

Problem: No matter how long I make my core, I never feel refreshed afterwards.
Solution: First of all, a core should be (more or less) a multiple of 90 minutes. Anything else isn’t gonna work in the first place. But more importantly, the time of the day you go to sleep and wake up at matters a lot. If your core just doesn’t seem to work at all, try a completely different time. The early evening and early morning seem to be the most common ideal times. You will generally find your time by just disabling all alarms and sleeping whenever you are tired for a few days.
When in doubt, try sunset. This seems to work for most people. I can only sleep well for a long time when I go to bed at 20:00 or around 05:00. Anything else is a waste of time.

Problem: I can’t fall asleep, even though I’m tired!
Solution: Are you drinking caffeine, even tiny amounts? Stop that. Caffeine totally screws up your sleep and once you have a good, regular schedule down, you won’t need it anymore at all.
Also, make sure you are comfortable, it’s (reasonably) dark and quiet. Regular noise, especially white noise (be it artificial or from a fan) works for many people.
And of course, don’t fiddle with your schedule. Your brain needs a long time to get used to it, so if you go to bed at a different time every few days, you will never sleep well.

Problem: But I need caffeine to wake up!
Solution: Try fresh fruit or a shower instead. Works well enough for me and I was on the mental level of a plant without caffeine for at least 30 minutes after getting up. :)

Problem: I like caffeine / alcohol / other drugs, can’t I ever have them anymore on a polyphasic schedule?
Solution: Most likely not. Rare consumption isn’t too bad (after you have adapted), but it tends to screw you up every time. Believe me, I have tried. Find out what’s more important and stick to it. As I’m writing this FAQ, you might have a guess what my decision was. ;)

Problem: I find it hard to get up and often tend to oversleep or just go right to bed again.
Solution: Your bed is probably too comfortable. ;) A low room temperature makes it really hard to leave the bed. Light also tends to wake one up pretty well, especially sunlight.
During the adaptation phase, you might also resort to sleeping in normal clothes or on the couch / floor. Having something cool or important to do right after you get up helps, too.

Problem: I always oversleep during this one nap, but never anywhere else!
Solution: Is it too close to another nap, i.e. less than 2 hours? That won’t work, ever. Furthermore, try moving the nap. There are times during the day you will have trouble getting up, no matter what you do. Try to not sleep during them.

Problem: I like sleep and/or lying in my bed. Of course this conflicts with rare and short (i.e. polyphasic) sleep. What can I do?
Solution: This was one of the hardest and least expected problems I had. Everyone how hated sleep and was an insomniac, oh you had it so easy. ;) Let me tell you that trying to make sleep less comfortable won’t work. The human brain doesn’t react well to punishment. It’s futile, so don’t even try it.
There were only two things that worked here. Make it more rewarding to be awake. Do a lot of cool stuff, always have something lying around you can’t wait to get done and reward yourself often. Sweets are nice. If your life is boring and horrible, you will find it hard to convince yourself to be awake. Certainly you have played that one game, read that one book, studied that one topic, etc. that you just couldn’t put down, that you skipped sleep over and just had to do – try to make your life that awesome all the time. Stay in this mental state.
On the other hand, there is one realization I found to be very important. If you are putting pleasure first and going the way of instant gratification, understand what you will become. What’s the difference between you and this?

rat with electrodes wired to its pleasure center

This is a rant. Don’t take it too seriously. I’m sleep deprived and needed something to do, ok?

Neo 2 claims to be an ergonomic keyboard layout. I have been using it until about a month ago when I forked it and created my own layout. (I’ll upload it here once I get around fixing the last few bugs.) I don’t contend its claim as far as the letters are concerned. However, level 3 aka Mod3, containing the punctuation characters, is totally fucked up. How fucked up? Let me demonstrate.

First, let’s get some data. What’s a typical use for Mod3? Programming and normal texts. I decided on analyzing C (representing curly-bracket languages), Python (representing pseudo-code and list languages), my shell history and shell files (representing Unix usage) and Perl (representing line noise). Additionally, I included my local Usenet cache to represent normal texts. I took several thousand files each and in the end averaged them all together. (data: freq-c freq-pl freq-py freq-history freq-news average)

Second, we do a little experiment. Say, if Neo 2 is actually ergonomically optimized, we would expect the most frequent characters to be on the easiest to reach positions. Of course, it depends a little on personal taste, keyboard form and hand size how easy each position is, but let’s go with the arrangement the Neo devs chose for their letters. If the layout is optimized, the most frequent letters and the most frequent punctuation characters should be on the same physical keys.

Let’s see if this is true. Here is a little table, showing the letters sorted by (German) frequency and the corresponding Mod3 character. I’ve put the keys with “.” and “,” according to their left-hand equivalent. Green means 5% or more, yellow down to 1.5% and the rest is red (which is often around .01%). Have a look.

Great fucking job, guys. The 7 best keys get only 2 frequent characters. 6 frequent characters are in the middle of fucking nowhere. You decided to place an obsolete character, the long S, that no one in my fucking news cache even used at all, in a better position than ,, ., “, ‘[3] and #, which together make up a good 40% of all punctuation. [0] The one character that is more frequent in many cases than any letter except the friggin’ E, and more frequent than 2/3 of the letters in general, the underscore, isn’t even anywhere near a good position.[1]

Way to go, Neo, way to go.[2]

[0] Some of the reasoning behind this is that any Mod1 key is easier to reach than any Mod3 key. If you think that Mod3+N is harder to type than Ö, then I wouldn’t let you design a wooden stick because it seems your brain is barely able to use 2 fingers at the same time.
[1] “But I don’t program! I rarely need an underscore!” But you constantly use curly brackets, more often than any other bracket?!
[2] To be fair, it’s impossible to optimize for punctuation-heay curly-bracket languages and normal text at the same time. However, this half-assed mess makes it hard for both groups. That’s clearly not a good solution. I chose to optimize for the programming languages because a) I use them more and b) they tend to be so rich in punctuation that they simply overshadow normal text.
[3] Which are not even normal punctuation, if the Neo devs weren’t so inconsistent in their design. If you want to follow official guidelines[4], you should use the German quotation marks „ and “, and the proper accent ’. Which are even harder to reach. Which makes the whole thing even more retarded.
[4] Prescriptionist dipshits are the topic of another rant.