Martial arts training camp. Average sleep time was around 4 hours per day, also, guard shifts round the day, so sometimes it ended up being 2. So towards the end of the week I was quite... sleepy. And this seems to have an interesting effect on visual pattern recognition.
One day, me and another guy were standing guard, around 4 in the morning, the sun was just about to come up. Making circles around the countryside weekend house we were staying in, I noticed that some people appeared with a truck and started to pick grapes from the nearby field. I promptly...
k2pdfopt. It slices up pdfs so that you can read them without zooming on a much narrower screen, and since its output pdfs are essentially images, it eats everything up to (and including )very math-heavy papers, regardless of the number of columns they have. Also, it works with scanned stuff too.
(And even though the output is a bit bigger than the originals, I didn't encounter any problems with 600 page books... the result was about 50 megs tops.)
Possibly relevant: Sudbury schools, with the curriculum of "do whatever you want, as long as you're in school, surrounded by interesting stuff". Also, http://www.psychologytoday.com/blog/freedom-learn. It really seems that we are doing quite bad by default...
As it turns out, for example, kids are quite good at learning stuff from each other (including things like reading... "I can't always get the big kids to read me stories, so I'd better go and learn this <> thing from them"...)
Now, find a way to prevent that from happening. Sorting kids by age and separating the groups? Perfect.
Your second point is very important.
It's not just about lost opportunities to learn.. well, it mostly is, but consider socialisation. Kids are supposed to learn how to act in society from.. who, exactly? Obviously not the teachers, they aren't around enough of the time to do it, and socialisation isn't really in the curriculum anyway. So. Parents, typically, and hope you have good ones.
Historically, you'd learn from older siblings or friends, but families are smaller and age-sorting has almost entirely eliminated cross-age friendships. That's bad; eliminating that problem might be a good start towards fixing schools.
Your "running different code" approach is nice... especially paired up with the notion of "how the algorithm feels from the inside", seems to explain lots of things. You can read books about what that code does, but the best you can get is some low quality software emulation... meanwhile, if you're running it, you don't even pay attention to that stuff as this is what you are.
Aren't utility functions kind of... invariant to scaling and addition of a constant value?
That is, you can say that "I would like A more than B" but not "having A makes me happier than you would be having it". Neither "I'm neither happy or unhappy, so me not existing wouldn't change anything". It's just not defined.
Actually, the only place different people's utility functions can be added up is in a single person's mind, that is, "I value seeing X and Y both feeling well twice as much as just X being in such a state"...
You won. Aren't rationalists supposed to be doing that?
As far as you know, your probability estimate for "you will win the lottery" (in your mind) was wrong. It is another question how that updates the probability of "you would win the lottery if you played next week", but whatever made you buy that ticket (even though the "rational" estimates voted against it... "trying random things", whatever it was) should be applied more in the future.
Of course, the result is quite likely to be "learning lots of nonsense fr...
... this is the thing I've been looking for! (I think I had some strange cached thought from who knows where that posts do not have comments feeds, so I didn't even check... thanks for the update!)
Didn't they do the same with set theory? You can derive a contradiction from the existence of "the set of sets that don't contain themselves"... therefore, build a system where you just can't do that.
(of course, coming from the axioms, it's more like "it wasn't ever allowed", like in Kindly's comment, but the "new and updated" axioms were invented specifically so that wouldn't happen.)
Is there a nice way of being notified about new comments on posts I found interesting / commented on / etc? I know there is a "comments" RSS feed, but it's hard to filter out interesting stuff from there.
... or a "number of green posts" indicator near the post titles when listing them? (I know it's a) takes someone to code it b) my gut feeling is that it would take a little more than usual resources, but maybe someone knows of an easier way of the same effect.)
P(Luminosity fan | reads this comment) is probably not a good estimate... (count me in with a "no" data point though :)) Also, what is the ratio of "Luminosity fan because of Twilight" and "read it even though... Twilight, and liked it" populations?
(with "read Twilight because of Luminosity" also a valid case.)
Also, who is the target audience and what are the plans for reaching it? I don't think there are many people who are willing to invest time AND money into a book like this while still not having read the sequences (available freely on the web, and also in all kinds of e-book formats).
For the two use cases I imagine at the moment:
giving it as a gift as an introduction to rationalist stuff feels better with a physical book indeed. Yes, there is a difference between buying an e-book for yourself and downloading the same stuff for free, especially in terms o
The point is that it's not, but making it so is a design goal of the paper.
Example: Mario immediately jumping into a pit at level 2. According to the learned utility function of the system, it's a good idea. According to ours, it's not.
Just as with optimizing smiling faces. But while that one was purely a thought experiment, this paper presents a practical, experimentally testable benchmark for utility function learning, and, by the way, shows a not-yet-perfect but working solution for it. (After all, Mario's Flying Goomba Kick of High Munchkinry definitely satisfies our utility functions.)
I heard the opposite too: don't try to push your own research too hard, especially in the beginning, but try to find something the others in the lab group are working on, learn stuff from them, and after a while you'll end up with your own ideas anyway.
Pros and cons for both of the approaches exist, but "picking a thesis early on" might be hard as you don't necessarily know what the good problems are in your field. But that might depend on your field / advisor too.
You see that you won't get stuck: that you'll finish relatively fast.
Do you know of a way for estimating this? Every research problem (and the entire PhD itself) might look much easier before you start working on it (you don't have an outside view perspective before starting).
This thing looks more and more relevant as I think about it. What it does is not just optimizing an objective function in a weird and unexpected way, but actually learning it in all its complicatedness from observed human behavior.
Would it be an overestimation to call this a FAI research paper?
Wow. First thought: who is this guy who submits a really cool scientific result to a thing like SIGBOVIK? He could have sent this thing to a real conference! It's a thing no one has ever tried!
Then I checked out his website. The academic one. And the others.
Well, short description: "Superhero of Productivity". The list of stuff he created doesn't fit on his site. Sites. Also, see this remark of his,
One of the best things about grad school was that if you get your work done then you get to do other stuff too.
(I'm also at CS grad school, am happy if I have time to sleep, and my only productive output is... LW comments... does that count?)
In the class I TA for, the students can go to the professor's office hours after the midterm / final, and if they can solve the problem there, they still get... half of the points? I wonder how that one affects test-taking performance.
Also, this whole thing seems to be annoyingly resistant to Bayesian updates... "Every time I'm anxious I perform bad, and now I'm worried about being too worried for this exam", and, since performing bad is a very valid prediction in this state of mind, worry is there to stay.
Maybe if the tests are called "quizzes" the students end up in the other stable state of "not being worried"?
Do we need a realistic simulation at all? I was thinking about how educational games could devolve into, instead of "guessing the teacher's password", "guessing the model of the game"... but is this a bad thing?
Sure, games about physics should be able to present a reasonably accurate model so that if you understand their model, you end up knowing something about physics... but with history:
actually, what's the goal of studying history?
Have you played the Portal games? They include lots of things you mention... they introduce how to use the portal gun, for example, not by explaining stuff but giving you a simplified version first... then the full feature set... and then there are all the other things with different physical properties. I can definitely imagine some Portal Advanced game when you'll actually have to use equations to calculate trajectories.
Nevertheless... I'd really like to be persuaded otherwise, but the ability to read Very Confusing Stuff, without any working model, and ...
I have a similar experience... around two years ago, both my laptop and desktop power supplies died (power surge), leaving me a pII-300... with which I had some "let's be authentic nineties" fun previously, so Win98 and Office 97. Except for the browser (lots of websites didn't even load on IE4-ish browsers), so I ended up with Firefox 3.x (the newest that ran on win98).
It actually took long times with 100% CPU to render web sites. And then further time to scroll them.
My observation is the same as yours: there is nothing better to discourage rand...
The problem with no notifications is that because you're still in a room where interesting stuff is going on, of course you'll check the chat history and/or join the people already chatting. (Unless you use up willpower not to, but the whole point is using less of that.)
Having a 25 min work + 5 min chat cycle seems to be a good thing though; start working because everyone else went silent is so much easier as going back to the "library" while everyone else is still talking in the lobby. If you're working, don't go there, that's it.
I went through an Optimization course last semester (CS, grad), so it doesn't really qualify as an "out of class experience", nevertheless reading it was quite optional, and, actually, the questions I asked myself were very similar to yours.
Especially in the light of those small remarks textbooks tend to make along the lines of "we don't have any more space here, so if you're interested, the excellent book by X and Y is a very nice read". As if they were referring to some light and entertaining book if the one you were holding weren't r...
Good idea... especially given that the fact that the readers usually aren't an immortal vampire / pony / wizard makes it relatively complicated to emulate the protagonists.
Not to speak of the fact that much of the fictional rationalist awesomeness comes from applying existing stuff (common sense / science) in a setting where it's not expected to be applied. (See Harry's Gringotts money pump or Missy's sciencey superpowers). It's hard to extrapolate that to our world...
Counterexample: Cory Doctorow's Little Brother & Homeland. Although not really ration...
just finished reading.
It's kind of sad in a... grand way.
No one remembering anymore what exactly being "human" means. But... what do we expect? I don't see any human values that are not statisfied, it just does not "feel like home" that much. But still, orpbzvat bar bs gur yrffre fhcrevagryyvtraprf naq fgvyy univat n fcrpvny cynpr va zvaq sbe fgne gerx? It's as heart-warming as it gets, in a cold, dark and strange universe.
(If only we could do this well.)
Attach lots of sensors to lots of axons, try to emulate the thing while it's running... for me, it's the on-line method that sounds more plausible compared to the "look at axons with microscopes and try to guess what they do" approach. Nevertheless, imagining a scenario with non-destructive uploads... how many times would you allow people to upload? Ending up with questions like that, I think it's the destructive one that would generate less horrifyingness...
If you are trying to do X, surround yourself with people who are also doing X. Takes much less willpower to keep doing it.
On a related note make sure that they are people who are actively doing X, or at least making credible progress towards it not just professing a desire to X. This is an easy mistake to make.
As for this thread: wouldn't upvoting commens that you think are useful for someone else but not for you be actually an indirect case of other-optmizing?
Also consider mint.com. Draws awesome graphs. (It only works for US bank accounts only though...)
It's so nice that if you combine the words Alicorn + Twilight you get "let's make everyone else immortal, too" independetly of the universe in question.
Do we actually have an objective test for the quality of visual imagery? (as compared to subjective quality of it.) What I'm thinking of is something like the mental rotation experiments, proving that in fact there is a representation of images in our heads... but with somewhat more complicated images. Or scenes.
Otherwise... I think I have good imagination abilities (I was also once told so while solving math problems involving rotating cubes), but my subjective quality levels are similar: pictures are somewhat vague, especially compared the ones I can get...
Wow. Thanks for the idea. Just got instantly sleepy in around 20 seconds after installing it...
theoretically yes... supposing that the smaller and bigger light sources are of equal efficiency and have the same spectrum. The number of photons emitted adds up linearly after all (I think it even works if the spectra are different.)
In practice, as it turns out (... "let's read lots of wikipedia about lighting" project), a 100W bulb produces roughly 1.5 times more lumens per watt than a 40W one, so the "equivalence" is also somewhat questionable... as it's the lumens that count (that being "the perceived lightness that is radiated").
(This thing is called "luminous efficacy", by the way. Am I the only one who thinks that it would make a nice LW post title at first look?)
You're right in a sense that we'd like to avoid it, but if it occurs gradually, it feels much more like "we just changed our minds" (like we definitely don't value "honor" as much as the ancient greeks, etc), as compared to "we and our values were wiped out".
AIs that can't be described by attributing goals to them don't really seem too powerful (after all, intelligence is about making the world going into some direction; this is the only property that tells apart an AGI from a rock).
design an AI so that it can't self-modify
is there at all a clean border between self-modification and simply learning things? We have "design" and "operation" at two places in our maps, but they can be easily mixed up in reality (is it OK to modify interpreted source code if we leave the interpreter alone? what about following verbal instructions then? inventing them? etc...)
The latter is not necessarily a bad thing though.
... Accelerando by Charles Stross, while not exactly being a scientific analysis, had some ideas like this. It also wasn't bad.
I'm saying that although it isn't ontologically fundamental, our utility function might still build on it (it "feels real enough"), so we might have problems if we try to extrapolate said function to full generality.
ought to be canonically reducible
but it most likely isn't. "X computes Y" is a model in our head that is useful to predict what e.g. computers do, which breaks down if you zoom in (qualia appear in exactly what stage of a CPU pipeline?) or don't assume the computer is perfect (how much rounding error is allowed to make the simulation a person and not random noise?)
(nevertheless, sure, the SAUS might not always exist... but above question still doesn't seem to have any LW Approved Unique Solution (tm) either :))
One shouldn't confuse there being a huge debate over something with the problem being unsolved
sure, good point. Nevertheless, if I'm correct, there still isn't any Scientifically Accepted Unique Solution for the moral value of animals, even though individuals (like you) might have their own solutions (the question is whether the solution uniquely follows from your other preferences, or is somewhat arbitrary?)
generalize n-th differentials over real numbers
(that was just some random example, it's fractional calculus which I heard a presentation abou...
For the first read the question sounded like it didn't constrain anything in the real world (see the "tree falling in forest" question)... but, in fact, it is relevant because it impacts our moral judgments.
Which says something though about the consistency of our moral judgments... (recently discussed around here).
Although I've read the metaethics sequence, that was a long time ago, but I think I'll put reading it again on my todo list then!
My intuition behind thinking 1 unlikely (yes, it's just an intuition) comes from the fact that we are already bad at generalizing "people-ness"... (see animal rights for example: huge, unsolved debates over morality, combined with the fact that we just care more about human-looking, cute things than non-cute ones... which seems to be pretty arbitary to me). And things will get worse when we end up with entities that con...
This seems to be one of the biggest problems for FAI... keeping an utility function constant in a self-modifying agent is hard enough, but keeping it the same over a different domain... well, that's real hard.
Actually, there might be three outcomes:
I have images because it was super easy to cut them from the lecture notes for the class (press hotkey, draw rectangle, and either ctrl-v or drop file from dock to anki, depending on whether on mac or linux). Also, does latex display work on phones? (ankidroid specifically.) Nevertheless, using latex seems to be the nicer solution indeed, especially if you plan to publish the result (I didn't).
(aand I sent a PM with the link)
I've been experimenting lately with an Optimization deck, so far it's mostly about unconstrained optimization, Newton method, Conjugate Gradients, that sort of stuff, with some formulas added as pictures. If you're interested, I can upload it somewhere! (Note: it's for Anki2.)
Try k2pdfopt! I use it all of the time with scientific papers, with lots of formulas, and it works quite well. It practically converts the pdf to images and slices them up, outputting another pdf, but the size increase is not too significant (still usable file sizes with multiple-hundred page long books).
Oh, yes. I do Dropbox syncing, too (this is the other good thing about org-mode: plain text files). And there might be some truth in the statement that while org-mode is excellent for a single file, things start to be less seamless when it comes to more of them... inter-file links don't seem to be that reliable, for example. Is this the reason for your One Big Org File?
For white on black, it's just (setq default-frame-alist '((background-color . "black") (foreground-color . "white"))) in your .emacs.
Actually, it's kind of typical lesswr...
I tried the Android app just after I read your comment (it's a thing I've been putting off for a long time), well... it really doesn't include the "creating nested outlines easily" part I like org-mode for, and the synchronization part also seems to be kind of... strange. Just as you said.
What I really like about it is the minimum effort that it needs to, for example, create a todo item (compared with web-based solutions). Too bad that these todo items usually end up really unorganized. Would be indeed nice to have some interface between, e.g. No...
... I did my fair share too, Santa vs. thin threads spun across the way between where the presents were supposed to emerge and the door... "stand back, I'm going to try Science" for the first time I remember.
Actually, it was a really nice experience not only about Science but also about how compartmentalization feels from the inside. I definitely remember thinking both that it's my parents and that it's some kind of mystical thingy, the only new thing that year was that these two aren't supposed to coexist in the same world. Not surprisingly, it's the very same feeling that I felt after being exposed to a semester of catholic middle school. Didn't have a name for it then though...