Mine is a weird one: I started taking every other day off. Even as deadlines loom, I am committed to doing no work today. I can cook and read and surf the Internet and hang out on Less Wrong and chat with friends and take a nap and do art (but not art for my webcomic), but at all costs I will do no work. Tomorrow, I will do work (in my case, papers for school and art for my webcomic and editing some fiction), and unless something breaks the trend that’s been working nicely for a week and a half now, I will do more work than I could have expected to do in three or four days before I started this. (I make exceptions for time-dependent things like class meetings.)
I have a few hypotheses for why this works for me:
It prevents the low-level burnout that used to plague me. I can decompress from whatever heavy mental lifting I do regularly and for a large chunk of time.
I actually enjoy most of my work when I actually do it, so obliging myself not to do it lets me get through the akratic aversion during my downtime. By the time I wake up on my work day, I’ve worked up a fair amount of antsiness about wanting to do something productive. Also, my creative ideas accumulate over time, not over effort; I have more interesting work-related ideas by the time I fire up Word when I’ve set the project aside for a day.
I can goof off more efficiently. Instead of spending all day on Stumbleupon because I can keep telling myself “one more site and then really, I’ll do something”, I can read an entire novel or bake a cheesecake or watch half a season of Buffy the Vampire Slayer. These things require significant time commitments, so if I feel like I ought to be working I don’t do them, but when I do them, they are more relaxing than the same amount of time in two-minute bursts spent obsessively refreshing Google Reader or checking my website stats or bothering people on IM.
I stopped after several months. Juggling it with a class schedule became intractable, and new techniques have a tendency to stop working for me after a while even if they start working very well. It was really useful while it lasted, though, and I still try to take time off in day-long chunks when feasible.
I’m seriously thinking about asking my boss about that one. With a pro-rata decrease in salary, of course.
The extra money just doesn’t seem to be worth the constant struggle with myself. Plus I think it would be good to start at a level I’m comfortable with and build on that. By forcing myself to work at a rate I’m clearly incapable of, I’m losing out on all the positive feedback that comes from small successes.
To draw a crude analogy, air pollution modelling is as hard a problem for me as say, AI is for EY. And if he needed to take every other day off once upon a time,...
EDIT:
PS I have been reading OB/LW for a while but have started commenting here only recently. Hello everyone!
You would probably like Ferris’s Four hour workweek, has an example of how to get your boss to let you work from home and stuff like that. Not the same as above, but similar enough to help you.
My actual AI work is extremely hard mental work, harder even than writing, which Harlan Ellison once called the toughest labor he ever performed (way harder than being e.g. a truck driver, which I myself have never done). There was no way I could do it two days in a row—though you’ll note that I say ‘was’ not ‘is’ since it’s important to keep in mind that these things often change over time.
I can attest to that; programming is complex because you formalize solving a problem over just solving it. AI is doubly complex because you’re formalizing how to do that.
AI is way more than twice as complex as ordinary programming. I have written plenty of programs that write programs, dealing with two layers of formalizing solutions, that is not anywhere near AGI. For one thing these programs only generate a certain class of programs. And much more importantly, they are not more powerful than I am so I can actually detect mistakes and fix them after I execute them.
Meh. AI conceptual work can be hard. But in reality, on any programming project where you’re both the brains and the brawn, you’re going to spend 90% of your time doing stupid stuff like writing hundreds of boring little subroutines; investigating different libraries, data sources, and data standards; figuring out which database software gives you the best performance; profiling and optimizing SQL queries; and of course DEBUGGING.
“Exponential” refers to how a quantity relates to another. For example, we would say that (until environmental limits are encountered) a population’s size is exponential with respect to time, and mean, that there is an initial population size P0 at a time t0, and a doubling time T, such that the population at a given time, P(t) = P0 * 2^((t—t0)/T). In computer science, we might say that the time or memory requirement of an algorithm is exponential with respect to the size of a list, or the number of nodes or edges in a graph, which could be represented by a similar equation, assigning different meanings to the variables. (Often, we really the mean the equation to be an approximation, or an upper or lower bound on the actual quantity.)
But if you say that designing and programming an AI is exponentially hard, you have not identified a variable of the problem that is analogous to the time in population growth. “Exponential” is not a vague superlative, it has a precise meaning. If all you mean to say is that AI is much harder than conventional programming, then just say that. Yes it is vague, but that is better than having your communication be more precise than your understanding.
Targeted commenter doesn’t really deserve being hit that hard, but voted up anyway.
The thing I really despise is when people use “exponential” as a superlative to describe fast-growing quantifiable processes that are not known to be exponential.
Thanks Alicorn. This sounds like a brilliant idea. I have been thinking of something along these lines but hadn’t quite thought of day chunks—makes a lot of sense to me too.
I’ve effectively used a variation of this: setting a daily deadline for when all work must be finished, following which I will do something (specifically planned in advance) that I enjoy. It’s like being about to go on vacation.
These ideas could both be considered variations on Neil Fiore’s concept of the “unschedule”, and there are other authors/speakers who’ve promoted the idea of clearly separating work/non-work days, e.g. Eben Pagan’s “Altitude” and “Wake Up Productive” programs.
For some reason, I find it hard to negotiate with myself over chunks of time shorter than a day. If I tell myself I’ll goof off after dinner, I’ll have dinner early; if I tell myself I’ll goof off after six p.m., I’ll dither until six p.m. But a day begins when I wake up and ends when I go to bed. Conveniently, my desire to stay up late persists even if I’m staying up late doing work.
If I tell myself I’ll goof off after dinner, I’ll have dinner early; if I tell myself I’ll goof off after six p.m., I’ll dither until six p.m.
Two important distinctions:
I frame it as, “I have to stop working at X pm”, not “I’ll goof off at X pm”. This presupposes that I’m going to be working and what’s more, that I don’t want to stop working (otherwise, I wouldn’t “have to”).
I don’t “goof off” (an unspecified activity), I have a book that I’ve planned to read, a show to watch, etc. -- thus it is a specific thing that I “have to stop work” for at that time.
This is a good example, btw, of how self-help techniques easily go awry, as there are often many subtleties to why/how something works.
That’s not to say that these changes will definitely make it work for you; as I’ve commented before, it’s trivial to defeat a technique simply by expecting something else to happen or thinking that it’s probably not going to work!
But you’ll notice that what makes it work (or not work) in both our cases has a lot to do with what we expect our behavior to be, and how we frame those expectations. And those expectations tend to hinge on fine details, rather than abstract concepts.
I’m a little concerned about other people adopting my idea, because it doesn’t seem that it would work for everyone. I recommend trying it for the first time when you’re on vacation for a week and you want to use part of that vacation to work on a nonessential personal project you’re being akratic about, like writing a novel or something.
Mine is a weird one: I started taking every other day off. Even as deadlines loom, I am committed to doing no work today. I can cook and read and surf the Internet and hang out on Less Wrong and chat with friends and take a nap and do art (but not art for my webcomic), but at all costs I will do no work. Tomorrow, I will do work (in my case, papers for school and art for my webcomic and editing some fiction), and unless something breaks the trend that’s been working nicely for a week and a half now, I will do more work than I could have expected to do in three or four days before I started this. (I make exceptions for time-dependent things like class meetings.)
I have a few hypotheses for why this works for me:
It prevents the low-level burnout that used to plague me. I can decompress from whatever heavy mental lifting I do regularly and for a large chunk of time.
I actually enjoy most of my work when I actually do it, so obliging myself not to do it lets me get through the akratic aversion during my downtime. By the time I wake up on my work day, I’ve worked up a fair amount of antsiness about wanting to do something productive. Also, my creative ideas accumulate over time, not over effort; I have more interesting work-related ideas by the time I fire up Word when I’ve set the project aside for a day.
I can goof off more efficiently. Instead of spending all day on Stumbleupon because I can keep telling myself “one more site and then really, I’ll do something”, I can read an entire novel or bake a cheesecake or watch half a season of Buffy the Vampire Slayer. These things require significant time commitments, so if I feel like I ought to be working I don’t do them, but when I do them, they are more relaxing than the same amount of time in two-minute bursts spent obsessively refreshing Google Reader or checking my website stats or bothering people on IM.
By the way, are you still practicing this? What was the long-term dynamic of it?
I stopped after several months. Juggling it with a class schedule became intractable, and new techniques have a tendency to stop working for me after a while even if they start working very well. It was really useful while it lasted, though, and I still try to take time off in day-long chunks when feasible.
Good to know!
Can you tell my boss about that one?
I’m seriously thinking about asking my boss about that one. With a pro-rata decrease in salary, of course.
The extra money just doesn’t seem to be worth the constant struggle with myself. Plus I think it would be good to start at a level I’m comfortable with and build on that. By forcing myself to work at a rate I’m clearly incapable of, I’m losing out on all the positive feedback that comes from small successes.
To draw a crude analogy, air pollution modelling is as hard a problem for me as say, AI is for EY. And if he needed to take every other day off once upon a time,...
EDIT: PS I have been reading OB/LW for a while but have started commenting here only recently. Hello everyone!
You would probably like Ferris’s Four hour workweek, has an example of how to get your boss to let you work from home and stuff like that. Not the same as above, but similar enough to help you.
Thanks. I’ll check it out.
I had to take every other day off during the year I was able to work on AI with Marcello.
Had to? Why had to?
My actual AI work is extremely hard mental work, harder even than writing, which Harlan Ellison once called the toughest labor he ever performed (way harder than being e.g. a truck driver, which I myself have never done). There was no way I could do it two days in a row—though you’ll note that I say ‘was’ not ‘is’ since it’s important to keep in mind that these things often change over time.
I can attest to that; programming is complex because you formalize solving a problem over just solving it. AI is doubly complex because you’re formalizing how to do that.
AI is way more than twice as complex as ordinary programming. I have written plenty of programs that write programs, dealing with two layers of formalizing solutions, that is not anywhere near AGI. For one thing these programs only generate a certain class of programs. And much more importantly, they are not more powerful than I am so I can actually detect mistakes and fix them after I execute them.
Meh. AI conceptual work can be hard. But in reality, on any programming project where you’re both the brains and the brawn, you’re going to spend 90% of your time doing stupid stuff like writing hundreds of boring little subroutines; investigating different libraries, data sources, and data standards; figuring out which database software gives you the best performance; profiling and optimizing SQL queries; and of course DEBUGGING.
Not that my programs ever have bugs, of course.
If I had to guess, I’d guess that you’re spending your time in Java or C (++|#) :)
True. I should’ve said exponential.
That does not really mean anything.
“Exponential” refers to how a quantity relates to another. For example, we would say that (until environmental limits are encountered) a population’s size is exponential with respect to time, and mean, that there is an initial population size P0 at a time t0, and a doubling time T, such that the population at a given time, P(t) = P0 * 2^((t—t0)/T). In computer science, we might say that the time or memory requirement of an algorithm is exponential with respect to the size of a list, or the number of nodes or edges in a graph, which could be represented by a similar equation, assigning different meanings to the variables. (Often, we really the mean the equation to be an approximation, or an upper or lower bound on the actual quantity.)
But if you say that designing and programming an AI is exponentially hard, you have not identified a variable of the problem that is analogous to the time in population growth. “Exponential” is not a vague superlative, it has a precise meaning. If all you mean to say is that AI is much harder than conventional programming, then just say that. Yes it is vague, but that is better than having your communication be more precise than your understanding.
Targeted commenter doesn’t really deserve being hit that hard, but voted up anyway.
The thing I really despise is when people use “exponential” as a superlative to describe fast-growing quantifiable processes that are not known to be exponential.
Thanks Alicorn. This sounds like a brilliant idea. I have been thinking of something along these lines but hadn’t quite thought of day chunks—makes a lot of sense to me too.
I’ll give it a try. And yes, I’ll be careful.
I’ve effectively used a variation of this: setting a daily deadline for when all work must be finished, following which I will do something (specifically planned in advance) that I enjoy. It’s like being about to go on vacation.
These ideas could both be considered variations on Neil Fiore’s concept of the “unschedule”, and there are other authors/speakers who’ve promoted the idea of clearly separating work/non-work days, e.g. Eben Pagan’s “Altitude” and “Wake Up Productive” programs.
For some reason, I find it hard to negotiate with myself over chunks of time shorter than a day. If I tell myself I’ll goof off after dinner, I’ll have dinner early; if I tell myself I’ll goof off after six p.m., I’ll dither until six p.m. But a day begins when I wake up and ends when I go to bed. Conveniently, my desire to stay up late persists even if I’m staying up late doing work.
Two important distinctions:
I frame it as, “I have to stop working at X pm”, not “I’ll goof off at X pm”. This presupposes that I’m going to be working and what’s more, that I don’t want to stop working (otherwise, I wouldn’t “have to”).
I don’t “goof off” (an unspecified activity), I have a book that I’ve planned to read, a show to watch, etc. -- thus it is a specific thing that I “have to stop work” for at that time.
This is a good example, btw, of how self-help techniques easily go awry, as there are often many subtleties to why/how something works.
That’s not to say that these changes will definitely make it work for you; as I’ve commented before, it’s trivial to defeat a technique simply by expecting something else to happen or thinking that it’s probably not going to work!
But you’ll notice that what makes it work (or not work) in both our cases has a lot to do with what we expect our behavior to be, and how we frame those expectations. And those expectations tend to hinge on fine details, rather than abstract concepts.
These techniques seem to be related to the idea of an “unschedule”.
Incidentally, is http://htht.comicgenesis.com/ your webcomic?
Yes, that’s my current one.
When you first tried it, did you start with an on day or an off day?
I don’t remember. Probably an on day.
Fascinating, I so want to try this. I do alternate-day calorie restriction, so this is basically what I do with my diet, except w/ willpower.
Best productivity tip ever!
(Seriously, somehow this actually does seem to make me more productive.)
I don’t normally have problems with akrasia, but this idea seems awesome. I’ll have to try to implement it next time I can consider taking a day off.
I’m a little concerned about other people adopting my idea, because it doesn’t seem that it would work for everyone. I recommend trying it for the first time when you’re on vacation for a week and you want to use part of that vacation to work on a nonessential personal project you’re being akratic about, like writing a novel or something.