If it's worth saying, but not worth its own post (even in Discussion), then it goes here.
New Comment
149 comments, sorted by Click to highlight new comments since:
Some comments are truncated due to high volume. (⌘F to expand all)Change truncation settings

Recently I've been carrying textbooks and reading them over lunch, which is not my normal pattern. A friend asked me about it. A brief discussion of AI risk ensued, resulting in this quote:

Wow. Ten minutes ago, if you had given me a button and said "This button will put a random AI onto your computer", I would have pushed it. Now that seems like a really bad idea.

This startled me at first, and worried me greatly (as we both work at Google, and our peers are working on AI.) Upon reflection, I realized that I shared the same sentiment up until last year.

This episode really drove the False Consensus Effect home for me.

Speaking of False Consensus Effect, when you say this, are you mentioning something which is already public knowledge, like self-driving cars or Google now, or are you talking about some other project that perhaps people outside of Google haven't even heard of and you can't talk about in detail? There are a lot of things that someone might call to mind when they hear someone else say "our peers are working on AI." But I wouldn't be surprised if I was getting the entirely wrong impression from this sentence and I should try to avoid that.
I was referring to things like the youtube-trained visual learner and some other public knowledge.

A few weeks ago I started assessing my own calibration, using tools such as the CFAR calibration game. I got fairly good and concluded that I am relatively well calibrated.

When given a question, my instincts would immediately throw out a number. I'd unpack it and adjust it in accordance to known biases. (Avoid representativeness, start from base rates, treat the initial number as a degree of support, factor in strength of evidence, etc.)

Yesterday, an assessment of probability came up in conversation. Immediately, my instincts threw out the number "80%". My thoughts went like this:

My gut says 80%. I'm well calibrated, so 80% is probably right.

I opened my mouth to speak.

Then I shut my mouth.

I understand Löb's theorem on an intuitive level now.

I don't understand. What's wrong with 80%?

I achieved good calibration by paying attention to evidence and avoiding known biases as well as I was able. Once I had established reliable calibration, I experienced temptation to justify my first intuitive instinct by asserting my own calibration.

My calibration was based upon mediating my intuitions with reason. I can't invoke it to trust any old estimate that comes out of my mouth: if I did then I could say whatever I want and trust it to be calibrated (which would yield probability estimates out of touch with reality). Hence the parallels to Löb's theorem.

TIL this app exists. Thank you. After using it for half an hour it turns out that I am well calibrated at probabilities except 80% and 90%. Weird. Anyhow, could this be combined with a program such as Anki? Meaning, that you place the answer in your head and indicate how certain you are in percent. If correct, the card will be placed accordingly into the future. This should work splendidly with learning vocab for linguistically close languages.
Hey, thanks for mentioning this. I hadn't heard about it. I've tried my hand at this app (50 questions or so), and it seems like the correct strategy, for me, is to go 50% for anything I have a little doubt on, and 99% for that I'm sure about. Maybe 5% of the questions fall into the 60%-90% range. I'm still working to understand the tutorial and how to interpret my results.
:-) It's not particularly hard to "perfect" your calibration in that game -- if you're over/under on a certain percentile, you can throw questions where you're confident into percentiles where you're "poorly calibrated" in order to spoof a good calibration curve. The trick to that game, if you actually want to asses your calibration, is to play for points rather than for a good curve. Being well-calibrated means that when you play for points, you have a good curve automatically. (I wish that they'd show you your curve less often, perhaps only when you leave the game. It's hard to resist cheating the curve. Then again, I'm not sure of a better way to provide the necessary feedback.)
I'm not strong enough in math to figure out how the scoring actually works without spending some time with it, and I wouldn't "throw" questions anyway. But I do like seeing that, say, on my 60%s I'm actually right 70% of the time. So when I'm feeling "60%" I should actually go with 70% more often. I think I'm afraid of getting questions wrong because the score penalty appears so high relative to the score bonus (I know that's likely appropriate, even though I don't understand the actual log bits, etc of scoring ).
The scoring is done so that if you have 70% of your answers right, then you get the best average score by guessing 70%, not 60%. The increased penalty you get for getting 30% of those answers wrong is smaller than the increased gain for getting 70% of them right. But that's true only as long as you really get 70% of them right; so changing your answer e.g. to 80% while being only 70% correct would decrease the average score, because then the increased penalty for getting 30% of those answers wrong would be greater than the increased gain for getting the 70% right. Without understanding the log bits, you can easily verify this in a spreadsheet calculator. Make a formula saying how many points you get if you report probability R and if you really get P answers right. Playing with numbers, you will find out that for a given P, you get the highest average score for R = P.

Debating etiquette: If you assert that X is "obvious", "blatantly obvious" or "so fucking obvious you'd have to be an evil, low-status person not to see it" then you are not allowed to provide any arguments or evidence for X unless you first concede that X is not obvious. None of this "Anyone with half a brain can see why X is true, but here's for paragraphs arguing for X anyway" nonsense.

People who do this are trying to double dip. By claiming their point is obvious they can shame their opponents for being too stupid to understand the plain truth. But since what they're saying isn't actually obvious, they layer arguments to make X look more obvious. If X is obvious, you should say "X" and everyone will know you are right. But if you find the need to defend X with arguments, don't try and shame the people who disagree that X.

That seems like a great norm... under assumption that the goal of the debate is to speak meaningfully, as opposed to trying to get more power to one's group. Because, if the true goal is more power to the tribe, then declaring all opponents stupid or evil and repeating the party line in a situation where it cannot be opposed is obviously the winning strategy. But this could be a funny game -- be the first person to declare that some applause light of the given group is "obvious", and then support it with a wrong argument. The goal is to calibrate the level of wrongness precisely so that someone within the group will recognize the wrongness, but they will be a very small minority and afraid to speak loudly against the applause lights. (And if they speak, then attack them and let them feel the wrath of the group.) Uhm, to excuse this evil game, I should probably find some positive purpose. So... let's say that it teaches people to examine the arguments critically even if they already agree with the conclusion. Here is an (imaginary) example: At a meeting of Amnesty International: "Everyone knows that racism is evil and unscientific. I don't want to hear any more of this bullshit of innate racial differences in IQ. Cognitive science has proved that intelligence is not a function of brain, but of kidneys, and all races have exactly the same kidneys. So I ask all racists to please shut up." A wave of applause. Then, a silent trembling voice: "Are you sure intelligence is in the kidneys? I never heard about such research and I am a cognitive scientists..." Acting offended: "Shut up, you nazi, how dare you!"
Clearly you have not spent much time on teh interwebs... :-D

A few days ago I was talking with two people who had "experienced" total brain inactivity for some period of time before (one for a few days on ice!). I tried asking what they thought about the discontinuous experience/identity and associated philosophical issues that e.g. cryonics or other forms of not-being-alive for a while entails, but I couldn't get them to interpret the question as anything besides "what do you think of the personality changes that might happen from such an event?".

I found this miscommunication highly informative.

This sounds exceedingly interesting. Please clarify on the "few days on ice," does that mean they were cryogenically frozen, or...
She had cardiac arrest and they cooled her down to prevent organ damage. Now that I research it more, probably not enough to completely stop brain activity, though that was the premise of the question and they seemed to understand that part.
I've wondered if that sort of dramatic cooling is more of an optimal path to cryonic preservation than the current method of waiting for "natural" death and then immediate perfusion. E.g. since it's an official medical treatment, get cooled significantly "to prevent organ damage" and kept in that state until some measure of clinical death occurs, at which point it may be more likely to avoid ischemia. I looked into cold-water drownings for similar reasons but it appears to me that what actually happens is just a preservation of oxygen in the brain and heart and lungs by the mammalian diving reflex and not necessarily any reduction in risk from ischemia during subsequent cryonic suspension. But being pre-cooled may still have some advantages.
Cooling to a few degrees below normal body temperature for a few days after ishchemia does help keep down reperfusion injury and after-the-fact inflammatory brain damage and is indeed now the standard of care at some heart centers around the US now...
Maybe a naive question, but you've got me curious: why is it the standard of care at some heart centers, rather than most or none? Is it a matter of cost, or are the benefits you mentioned not well-established, or are heart centers slow to change their standard of care? Or is it some fourth thing I didn't think of?
It was recommended as potentially useful all the way back in the 2005 American Heart Association guidelines for CPR, actually, and there's been spotty literature even further back. It isn't so much about preserving function during hypoxia (that damage is already done by the time they get to the hospital) as preventing the subsequent damage that happens both immediately after and over several days after restoration of bloodflow. This is due to reperfusion injury as cells poison themselves due to metabolisms disrupted by the hypoxia, and slower issues caused by the immune system reacting via inflammation against very slight brain damage and thus greatly exascerbating it (a useful response to physical damage in peripheral tissue but problematic if you manage to restart someone's heart). I suspect the reasons for slow adoption consist of it having a focus that is very different than the ususal 'get the blood flowing again' aspect, where the bloodflow and proper immune function is actually the problem, and the fact that you need the equipment procedures and institutional coordination to integrate another step of care for a day or more after the acute treatment. I don't think its so much a question of cost as setting up all the procedures and conditionals and institutional experience to do it reliably and automatically. Course that leaves institutional inertia or laziness as additional prime obstacles. See http://emedicine.medscape.com/article/812407-overview When a friend of the family had a heart attack resulting in temporary cardiac arrest recently this aspect of treatment consisted of keeping them in a chemically induced coma for several days after the incident and cooling them to somewhere around ~34 C (I'm estimating based on what I heard), then slowly warming them and withdrawing the drugs.

There's a new chapter of HPMOR and it's pretty fun. The terrorists have won and now all citizens must go to the toilet in groups of three.

Hopefully someone will create a new discussion thread for it.

What are the reasons to go to family meetings, and meet the subset of family members who happen to be (1) Stupid (2) Religious (3) Non-rationalists (4) Absolutely clueless about reality (5) Pushy about inserting their ideas/ideals/weltenshaaung/motifs into you?

Of the top of my mind: [1] Avoid losing inheritance money [2] Avoid losing reputation with related family members who are not so silly [3] Avoid losing reputation with other people who may give you inheritance money [4] It is an investment in the far future, if you break a leg, or something, family members are more likely to take you to a hospital

These 4 reasons do not seem sufficient to me. Personally I don't think (a) You owe them, for their past good deeds towards you (b) Sharing genes is an important property or (c) One should love one's relatives.... have any truth in them. So do you have any extra arguments besides the 4 above that may be more convincing? Or should I abdicate the meetings alltogether?

It's a hard decision, I feel for you.

Having observed several people in a similar situation, I saw them go through the reasoning you describe. If you discard the virtue-ethics non-consequentialist reasons, like "One should love one's relatives" (regardless of how bad they are), or "You owe them, for their past good deeds " (despite all the poisonous and mean stuff they inflicted on you), you are left with enumerating options and calculating utilities.

At least one person I know had decided that the emotional damage of maintaining contact outweighs any potential financial benefits and severed her connections with one part of the family entirely, instead relying on her friends for socializing and emotional support. When her parents passed away some years later, they left their millions to some church charity and nothing to her, but that was already factored in her decision and so was not a big upset.

Another managed to learn to detach himself emotionally from whatever is going on at the meetings, by treating his family as low-level NPCs who simply follow their faulty programming and are no more worthy of being upset at than a wordprocessor program with a bug in it. ... (read more)

Do you know where I might find information about implementing this technique? It sounds really useful. Did your friend follow some methodology for accomplishing this?
Keep in mind that the definition of a sociopath is more or less "one who treats other people as low-level NPCs".
Indeed, and people would do well to remember that there may be situations wherein you are in fact the relatively "low-level NPC".
Also known as "the mark". The good news is that you are rarely aware of being one.
I am not sure this is good news from the standpoint of consequences...
Point well taken! However, this still seems like a potentially useful skill to have when you must interact with someone but wish to defend yourself emotionally.
I don't know of any sources he used. This is one of those hard self-modifications that require highly developed emotional intelligence and introspection skills. I know that when I tried to do something like that (not getting annoyed at a person for constantly bringing up the same settled point over and over for years), I failed. Basically, the feeling of annoyance flares up before I have a chance to consciously deconstruct it. I managed to quell it quicker, but not prevent it from happening. I tried preparing myself for the situation in advance, but that only made it worse, as I would get annoyed and upset during the simulation, as well. Actually alieving that a person close to you is basically a moist robot is hard.
Might it help to think of the person as running on habit about a particular subject or in response to a particular stimulus rather than them being pseudo-conscious in general?
Brian Tomasik gives some tips in a recent essay: Such "intrapsychic" strategies seldom work for me, however. I find the "extrapsychic" approach of just avoiding irritating people much more effective. (This may require terminating a relationship which, if maintained, would expose you to such people on a regular basis.)
Over time, they can be reprogrammed to some extent, if you are not just straightforwardly responding to their actions or ignoring them. Raise your own status in their eyes and then teach them skills that enable more accurate control (i.e. more efficiently changing their minds as opposed to facing pointless arguing or deep wisdom). Finer control can be used for further skill development and for making them more useful or pleasant to be around (including developing their cynicism, so that they become capable of not responding negatively to things like this comment).
Ignoring them is actually an extremely powerful tool for reprogramming people, particularly when it comes to the kind of toxic emotional interaction habits that can be failure modes in family relationships. I'd go as far as to say it is one of the best.

They keep you well calibrated about the level of insanity in the society. They allow you to not only have a good model of the society, but to also feel it on a gut level.

Upvoted because it's true and relevant, but there are much, much cheaper ways to achieve that goal.
Facebook, for example.

Advice that you probably don't need to hear, but might be useful to someone in a similar situation:

If you do cut family ties, try to do so without hurting anyone or burning any bridges. The outside view says you're likely to reconcile in 5-10 years, and that's a lot easier if you drifted apart than if you had a big cathartic fight.

I'll just vanish slowly, no fight intended. I don't think fighting is worth it most of the times it occurs, do people disagree?
Strong agreement. In one's personal life, fighting is occasionally better than no action at all, but (effectively) never the best action.
meet potential allies who are by default more loyal to you than other people. if you can act nice to family they will like you way more than strangers will and a network of people you can rely on and who can rely on you is a useful and awesome thing to have. meeting the few members of your family that agree with you can also be positive and fun.
4Ben Pace
Don't forget to add status quo bias. Most people think its important to care about horrible family members, and not doing so because you obviously shouldn't just seems weird.
Deciding that you don't really care about continuing your relationship with those family members can be freeing because it means that you can do anything in their presence. It can be fun when you talk with a religious person to just argue a more fundamentalist position then them for one meeting and see how they will deal with it.

What (not necessarily LW-related) things do people find useful to Anki? (Or have Ankied but they turned out not to be useful, etc.)

Some things I have:

  • the NATO phonetic alphabet

  • mass of Earth/Moon/Sun, radius of Earth/Moon

  • log_2 of 1.25, 1.5 and 1.75, and log_10 of 2 through 9

  • The 68-95-99.7 rule

  • Some things that I noticed I had to keep looking up: which is which between SQL left and right joins; the argument order to python's datetime.datetime.strptime function; the spellings of irrelevant and separate

I think only the latter group have had any use worth speaking of so far, though the third and fourth are things that I have more than once wanted to know and not known. The first two may just be almost-useless (though I like knowing them, so not necessarily worthless).

Things I kind of want to remember but suspect they wouldn't be worth it include other alphabets, and locations of countries/US states/UK counties/London underground stops (the aggregate may be useful, but there's an awful lot of cards there).

If you are going to memorize any logs, you should only bother memorizing logs of primes, since you can add those together to get other numbers.

This took me a minute. Just to unpack for the dense ones like me:

log(a * b) = log(a) + log(b)

so if you memorize log(7) and log (11), log(77) is easy as pi.

Pharmacy stuff Brand & generic name pairs for prescription drugs. Classes & mechanisms of action for prescription drugs. 1st line therapies for various diseases. Etc. Mandarin Chinese *Mostly just doing vocab at the moment, but have used it for listening (MP3 clips), writing, & grammar. Misc work stuff Names of new employees (they're Chinese names, so difficult to remember) Who is the contact person for what (eg if you want a new email account, you need to contact Mrs. Wu YiJun for approval)
I don't really see much use for Anki for everyday life, honestly, because I don't have too many things I need to memorize right now. However, some people at the LessWrong DC group have started learning Lojban, and Anki is just the biggest possible stick to hit the "learn words in a new language" problem with. I've learned a lot of words just doing that for a week. I still have to learn grammar concepts, but having a bigger vocabulary at my fingertips is pretty great.
If you live in London, the tube map can become suffiiciently memorised to serve for pegs in itself.
London cabbies probably have a great foundation for building memory palaces.
Mnemonics Using anki to memorize a complete 3 digit peg list allows you to do some pretty impressive memory feats, including recalling long lists in order, and memorizing really long numbers.
What kind of images do you use for the pegs?
Many people use the Major Method to come up with peg words/images so that the number to word relationship is nonrandom and easier to commit to memory. Here is an example of how it should work. I choose to use peg words that are objects and not abstract verb/ideas. The number 201 can be 'nest' or 'incite', both of which satisfy the major mnemonic method, but using 'nest' as your peg word is better because it is an object that you can better picture in your head and create interactions between it and other peg objects, to create a string of events.
I've got a major system peg list of a 100 English words, but even there it was hard to come up with a word for a concrete object for all the combinations. Can you really find concrete object names for all of the 1000 sound combinations? There's an online list, but it has words that aren't concrete objects and several of the words don't seem to actually encode right (eg. its word for 55 is 'hello', which has only a single L-sound, and isn't an object. Mine is 'lily')
This is what I did, and its far from complete, I took the rememberg list and copied it to a txt file and then uploaded it to an anki deck, then I went through each peg word to make sure I like it, if not I replace it. Rememberg uses the typographic system which is why 'hello' is is 55. I personally prefer the phonetic system. Also, I wouldn't like 'hello', as 'lily' is so much better.
Background: I've been using Anki for about 2.5 years. I have done the following: * (+3) assorted unusual English vocabulary (English is my first language) * (+1) the NATO phonetic alphabet * (+2) hiragana and katakana * (0) phone numbers of family and friends * (+2) the streets of San Francisco * (+1) assorted technical concepts, some LW-related The numbers in parentheses are my rough impression of usefulness and/or enjoyment on a possibly familiar scale of -10 to +10. When I was first getting used to Anki and only using it for English, the usefulness was around (-1), for reasons I can get into if anyone's interested. My biggest problems with Anki are first that it's a pain to input cards in a useful way, and second that for some things (e.g. hiragana and katakana) a more structured format would be strictly better.
I'm currently using anki just for English, so I'd be interested in what you found harmful.
It was a few different things, and it only lasted for the first month or so of using Anki. During that time I occasionally had moments in conversation when I grasped for one of the words in my deck, when normally I would smoothly talk around the idea with simpler words. Sometimes I succeeded in incorporating a new word into my speech, but the usage was awkward. Sometimes my interlocutor didn't know what the new word meant, and not only did I have to explain in simpler terms, I came off as out of touch and a bit of a know-it-all. It was a little uncomfortable at the time, but the harmful effects did fade as I became more aware of which words were ok to say in what contexts. The primary benefit has been a better understanding of the written word rather than a larger productive vocabulary.
Using Antisuji's system: * (+3) Emacs Keybindings + Listing good usecases for the bindings * (+1) Git commands * (+2) Compound Kanji * (+1) Basic Unix Command Line * (+0/+0.5) C I/O Function prototypes * (+3) Gaussian Integrals * (+4) Addresses * (+1) GRE Vocabulary words (All of it from taking the GRE, not from general usage) I've considered adding all of my family's birthday's to the list but 1) I'm too embarrassed to ask 2) Calenders are an easier solution. Has anyone else done something similar? Also, indirectly, I teach a class of about 25~ students every quarter and while I don't put them in a deck, I make sure that I'm exposed to the entire classes' names in a roughly spaced repetition way (First class I attempt to say everyone's name twice, grade different assignments at the appropriate spacing and 'reset' my schedule for mistaken names). This has caused my students to respect me as a teacher much more (No other Teaching Assistant knows everyone's name!) and slightly deters people from being quiet when they don't understand something (as I can just call out their name).
Next time (right now?), why don't you try the students using anki and see how it compares? Does the school give you their pictures ahead of time?
There's no pictures and the first time I get the dossier is on the day I teach my class. It's slightly premature optimization to start an anki before the first week of TAing, because about 5 or so students shuffle in an out during the first two or so weeks. Currently though, I'm applying for a physics major only class where there would be pictures and the class size is much more static. Thanks for suggesting an out and out comparison. It hadn't really occurred to me to do this if I do land the other job.
My more recent failed attempt to get an Anki habit going involved using Vim and a text file to input cards, instead of the tedious Anki GUI. I write lines of three tab-separated sections into the deck text file, with the first being the question, the second the answer and the third the card tags. Anki's import will ignore lines that start with # as comments. This lets me do batch editing of the cards using the macros, search-and-replace and block editing functions in the text editor. Problems with this approach is that I need the Anki software to preview my Latex formatting and I need to write raw HTML into the import if I want to use formatting or images. Anki should support card text with newlines if it's enclosed in quotes, but I don't seem to have cards using this in my example deck. Another problem is that Anki uses the question field of the card as a primary key and keeps the old deck when importing, so if I edit a question in one of the cards, I'll either need to completely regenerate my deck from the txt source or manually delete the card with the old question from the Anki database. As a Vim-specific tweak, my text file has the modeline This disables physical autowrap of long lines, ensures that pressing tab emits physical tab characters and makes long physical lines visually wrap at the word break and indent the continuation of the physical line on the next visual line by two spaces so it's easy to tell apart from the next item. The text file approach does not store the review data for cards, to if I should lose the Anki database, I could reimport my deck but would have to go through all the cards with zero review data. More experienced Anki users can maybe tell how big a problem this would be. For all this interest into making the thing technically nice to use, still haven't found a suitably big and growable set of stuff I want to memorize to bother with the habit.
The street grid layout for the city you live in, if relevant. Not very hard and allows you to understand locations much more easily if you live in a city where people speak in terms of cross streets. Don't go overboard trying to remember places you'll never go and other exceptional situations.
Some stuff I used Anki for: * Japanese (I have decks for kana, vocabulary, grammar rules...), that's the one I've used the most consistently for the past six months or so. * Paris subway map (which color is line 8? Which is the green subway line? What lines pass at Montparnasse-Bienvenue? What is the east terminus of line 1?); not particularly useful but should help consolidate my mental map of Paris; I haven't been reviewing that very consistenly but know it pretty well by now. * A few misc. facts in AI and robotics; I made the mistake of putting too much stuff I don't really care about, and despite a few cleanups haven't been reviewing this consistently A couple years ago I had used Anki with some pre-made decks (German, Lesswrong sequences), but reviewing it started feeling like a chore, and I stopped. It's easier to stay focused with cards I made myself. When I'm in Japan I jot down all the phrases I learn on a piece of paper, and then later on enter them into Anki (I also have a running list of "things I would like to be able to say").
A related question: What sort of time per card have long-term Anki users experienced? As a first "project" with Anki, I've been learning english vocabulary. Half are new words, and half are words I recognized but understood only vaguely. I have 2300 cards, and spent around 18 hours reviewing and 8 hours collating cards, which is around 40 seconds per card. Three months out, I'm reviewing ~30 cards a day in ~2 minutes. I feel like a fool when I think that I've spent 26 hours studying vocabulary this summer, but I'm pretty pleased with the equivalent "40 seconds to learn a new word". More generally, the cumulative time spent on any habit will sound pyschotic when quoted over long time periods: 4 minutes a day is 24 hours over a year.
CFA exam: There are many dry facts to be learnt. Driving Theory Test: What different roadsigns mean and so on. Sports teams: I don't care about sports but it's handy to be able to understand colleagues' conversations. Genocides: Because it is sad nobody even remembers most of them. Latin phrases: A part of my education I missed. The Lord's Prayer: Because going to hell is a bad idea. Facts from books I have read: Because otherwise I'd just forget them in a year. Small exponents: e.g. 6^3.

I'm tired of people never, ever, ever, EVER stopping 2 hours to 1) Think of what their goals are 2)Checking if their current path leads to desired goals 3)Correcting course and 4)Creating a system to verify, in the future, whether goals are being achieved. I'm really tired of that. Really.

I do 1-3 on a regular basis, though path 2) is ridiculously prone to bias. Part 4 is probably where I'm stuck right now...

Does anyone know of any job opportunities with constraints reasonably like this?

  • In the US (ideally California, for the climate)
  • Software development, ideally on GPUs (not graphics, but massively parallel number-crunching)
  • Reasonably low risk (no two-person startups)?

My qualifications are, extremely briefly, a physics PhD and two years' experience as architect and lead developer of the GooFit framework for maximum-likelihood fits.

Edit to clarify: I'm not asking people to google for me! I was thinking more in terms of networking, as in "do you kno... (read more)

The NERSC suoercomputing facility, which I believe is adjunct to Lawrence Berkeley National Laboratory, would be worth looking into. Or the Computational Research Division also at LBNL. I know other California national labs are heavily into supercomputing as well. It's a government job so the risk is pretty low.
Come to think of it, I have a network connection there; I will shake it and see if anything falls out.
I've seen a few similar positions recently. Haven't been paying that much attention to them, since they're usually looking for deeper graphics experience than I have, but I expect you ought to be able to find something. This Monster posting seems representative.
I'll edit my post to clarify, but my experience is not with graphics per se, it is with use of GPUs for general processing, ie number-crunching.
D. E. Shaw Research hires algorithm and software developers: They are in NYC. I've heard that they offer outstanding salaries (considering the wealth of the founder, they are unlikely to run out of funds).
Thanks, I will give it a shot.
If you do manage to get a job at Shaw, I'd be grateful if you could get me numbers on how much energy their Anton supercomputer uses. (Their publications don't seem to say, and my emails have gotten no responses.)
Sorry, they turned me down.
Thanks for counterfactually asking for me.

My wife has constant pain. She describes most if it as "joint pain" and some in her right arm as "nerve pain". The joint pain has been around for about a decade and the right arm pain for about 3 years, ever since what certainly looked like a repetitive stress injury at work (lots of mousing with a desk that was too high).

Doctors have checked her out and haven't found a cause for either. Our possible next steps are:

  • Get 2nd (actually 3rd-4th) opinions from local doctors. This feels futile but may still be helpful.
  • Hire a doctor whose j
... (read more)

A doctor friend's sister-in-law suffers similar pain. My friend and I discussed her condition, and he told me in private "yeah, she's screwed". From what I recall of the conversation, doctors have just no idea what do to with a patient like this.

I'm not sure how to go about this but it feels like it should be possible for some price.

That, in a nutshell, is the reason we spend so much on healthcare for remarkably little effect.

MetaMed? (See also.) (And also.)
My wife has similar-sounding pain. She was diagnosed with Fibromyalgia (which, as far as I can tell so far, appears to be in many cases a diagnosis of exclusion - we don't know what causes this, so we'll put it in the Fibromyalgia bucket) and Ehlers-Danlos Syndrome, which makes her connective tissues weaker than normal. We have tried quite a few things with varying degrees of success. * Trigger Point Therapy - A type of massage therapy that focuses on treating a muscular phenomenon named (poorly, in my opinion) "trigger points". In brief, these are small regions of muscle that become constantly contracted and unable to relax. I'm not aware of the specific mechanism that causes this. Trigger points can have strange effects, including pain appearing in different parts of the body than where the muscular problem exists. The best resource we have found on this is this work by Travell and Simons. Here is the volume for the lower extremeties. We have had good success with this sort of treatment, but we have to keep treating. It's worth investigating, even if the problems seem like joint or nerve pain. Many of the referred pain patterns appear in joints, for example. I don't know how well-researched this phenomenon is - many doctors seem to be unaware of it's existence. * Other forms of massage therapy give her some temporary relief. * Aquatic physical therapy - This involved exercises and stretching in a pool kept at about 80F, and it seemed to have a large positive effect. I'm not certain of the exact exercises done, but if you're interested I can find out. * We have explored possible dietary factors, and have found that removing wheat has some effects on other conditions, but we can't correlate it strongly to pain. * Stress reduction - Stressful events cause her pain levels to rise quite a lot, and it takes quite some effort to bring the pain back down after an event. It appears that the pain and stress work in a positive-feedback loop to make things worse. * Al
Have they checked for rheumatoid arthritis (and not just with a blood test, it doesn't always show)? It took many doctors vists for them to get the correct diagnosis for my wife (despite a history of it in her family).
Rheumatoid arthritis: Blood tests negative, specialist declared after the second visit that RA was very unlikely. Blood tests did show a slightly positive ANA.
Is the joint pain constant or on movement? Does it react to heat/cold? IANAD (I Am Not A Doctor) but I'd check a variety of inflammation markers (the more the better) with special attention paid to possible autoimmune issues. I assume the usual suspects (arthritis, gout) were ruled out? If I were you I'd try more doctors (third, fourth, fifth, etc.) hoping to hit one of three: (1) someone highly competent; (2) someone lucky to order the right test; (3) someone with determination to figure it out.
See also Scott Sonnon -- he's got a system of joint mobility exercises that seems to be gentle and effective. I find that the series that starts here is the easiest to follow.
A friend of mine suffered from undiagnosable pain in her joints for about a decade as well. About a year ago she discovered it was lime disease. Have any doctors you've talked to already eliminated that possibility?
Blood tests were negative for Lyme.
Nitpick: Lyme disease.
I thought it was spelled that way, but Firefox said it was wrong :/
Firefox is correct: "lyme disease" is wrong. What's wrong is not the spelling but the capitalization. When I right-click on "lyme" in firefox, it suggests capitalization, just like it does for "firefox."
I didn't know about right-clicking on words that a spell-checker says are wrong.
You can't trust spell-checkers-- their vocabulary lists tend to be incomplete. I recommend at least going to a search engine if you have a strong intuition about how a word should be spelled. It's not that intuition is entirely reliable, either. It was a hard fight to convince me that it should be irrelevant rather than irrelevent. Now they both look wrong.
I would try some form of bodywork. The Taoist standing meditation that NancyLebovitz suggest would be one way. If you don't want to do something that comes from a religion Feldenkrais would be a good choice.
I'm not sure if you have already tested for this. Please have the test for hyperthyroidism done. My wife had a problem with a finger ache and after many tests, we eventualy zero-ed in on hyperthyroidism.

Ah good, I forgot :-)

How much would it cost to get Eliezer to reveal the logs of his AI box games?

As we don't know the content of the logs it's difficult to assess the value that conceiling the logs has to Eliezer. Without a person who would plausible pay real money to him for revealing the logs I don't think it makes sense for Eliezer to give you a $ number.
that doesn't make any sense at all. ELIEZER knows how valuable they are to him. He's rational enough that if he thinks about it he would realize there is an amount of dollars someone could pay him that would get him to reveal it, and it's in his interest to be paid those dollars since by definition it's a profitable transaction. And it can only take place if he says what the price is.

Any tips on journal keeping, specifically the format or style of writing?

I've been keeping a journal but I am unsatisfied with the result. I have no methodology, some days I write as if speaking to a future me, some I write as if to an audience, and some days I write essays, and others I just list what I have done. As a result, my journal is difficult to read, it doesn't have a consistent feel.

are your expectations for more structure useful for your purposes in keeping a journal?
I write paragraphs beneath headings, to prevent rambling.
My journal is primarily is revision record for my life, something Future Me can look up and cross reference against memory to refresh on why certain choices were made or how landmark events made me feel. The second purpose is to brainstorm on and explore upcoming choices and current events, to be honest with myself on how I feel. As a result my journal is a shameless full-disclosure letter to Future Me, and that works for me. Ask yourself why you keep a journal? What is the goal? Then write to achieve your goal. If that doesn't work, split your existing entries into similar groups. * Notes to future self. * Messages to an audience. * essays. Now work backwards from those groups to assess your motivations. Perhaps you want to split your journal into three sections, one to address your future self, one to vent into and get stuff off your chest, and one to explore and summarize ideas. Please provide feedback.

The septic tank is full: the official unfiltered dump thread has been archived 'cos it's six months old; need to start another one (or not).

I haven' t seen anything about human evolution and lullabies, and I'd really like to.

Is it effective to study by writing down all the information you want to learn? When I'm faced with the task of studying from some source, say a textbook, my first impulse is to write all of it in a notebook. But recently I've been worrying that this makes me shift my focus away from learning and towards doing a scribe's work -- and also I think I'd be a little more motivated to start studying if my mental image of the activity involved less writing and more reading.

How does this method compare, in terms of information retention per time spent, with other study methods?

When I was a student I spent a lot of time in class writing down what was being said, and reading my notes later on, there's a lot of stuff I don't remember learning. When I look stuff up on Wikipedia because I'm curious, I don't think I remember less than when I was writing everything down. Now I tend to only write down a very short summary of the most interesting/surprising bits, ideally with a diagram or silly drawing, but then I haven't been reading much textbooks recently. If I was back in "I nead to learn all this new information" mode, I'd probably do a mix of exercises, and Anki.
Depends how well you retain it. Note-taking is a highly effective mnemonic technique for some people; not so much for others. If you're going to be doing the work of transcribing it anyway, it might be a good idea to put it into an Anki deck or other spaced repetition framework, so as to make later reinforcement easier if you need it.
Question: Answer: Gosh, do you really think so?
Thing is, there's no better answer than that, because it's too variable. You have to experiment for yourself.

I have a moral question.

Is it better for the last million people of a certain population to die, or for two million people all around the world, randomly selected and evenly distributed, to die? For the first group, their death would not just result in loss of human life, but potentially loss of a lot of cultural information; their language, their religion, their mythology and folklore, their music. I feel like this cultural information has value.


I would generally value a million lives over cultural information. We're always producing more culture, anyway; it's kind of what we do, as a species. Any particular set of in-jokes, songs, and stories is, I think, less valuable than a million people who will make more.

I would kill the million, every time. (I can imagine populations of size 1,000,000 which I would value more than 2,000,000 random humans, but I don't think any have yet existed.) What about 1,000,000 versus 1,000,001? I'm not sure. I think that could depend on the population in question.
I'd like to preserve the culture, but not at the cost of a million lives.
Depends on whether that culture was going to disappear anyway (quite a few cultures seem doomed today), and how valuable and unique it is. It's sad that we lost a lot of information about the ancient greeks, but information about say the various central asian seems less regretted.
That depends on how you define the population. Killing the worst 1 million people (people who have caused the most harm to other people, and would continue to cause significant harm) instead of 2 million random people would be a very large net benefit. There have probably been few or no traditional populations (nations, cultures, political movements, etc.) that would be worth completely eradicating, and probably never an entire 1 million people in such a population worth killing out of hand, but if I was forced to choose, I think I could find examples in the 20th century.
That depends. Is knowledge of a culture worth a million people's lives?
What do you mean with "certain population"? Any selection of one million people describes a population.
people are mostly interchangeable. i'd save the million.
Fortunately, it's not the kind of choice we're generally offered in the real world.

Yes it is. We can focus on preserving culture, or we can focus on preserving life. We only have a finite amount of money, so we have to decide which is more important, or if something else altogether is.

To give a concrete example, it is compulsory for school-children in Wales to learn the Welsh language, even if they are not ethnically Welsh, and even though most people in Wales do not speak Welsh. This public policy choice is justified on the grounds of preserving Welsh culture. Whether or not you approve of this decision, it's clearly an allocation of limited time and money which could be put to other uses.
My first guess on reading Bill_McGrath's thought experiment was indeed that he had in mind something about Irish-language-related public policy choices.
Generally it's good to have choices. :)
You might find the last couple of paragraphs of this quote useful.
In order: 1:This kind of thing does not come up. 2:If it did, the moral obligation is to find a third option. 3: The million would be the lesser evil, but is exceedingly unlikely to to actually die, as an identifiable group faced with the prospect of a loosing that high a number of its membership is going to exert a lot more leverage than the world at large is going to exert over a one in 3.500 chance of death.

Is there anything in the works for a mobile-friendly version of Less Wrong?

I seek help on a problem that I stumbled upon when thinking about a rational teleporter's story.

As typical of such protagonists, he finds that he can teleport and teleport a human's mass sideways with him, seemingly unharmed. As befits a rational protagonist, he experiments and finds out that he can teleport animals and after a demonstration to a very reluctant brother, he realises that he can teleport a human being, unharmed. After a crazy week of teleporting, he realises that he needs approximately 3 minutes to recuperate after a teleport to really do th... (read more)

One option is to have clients enter their source, destination, bid, and the range of times when they'd be willing to travel, and to have an algorithm for selecting the highest revenue path (which won't always mean accepting the highest bid, if slightly lower bids can chain together destinations and sources). (There could also be a secondary market to fill any otherwise-empty legs in this path.) The operations problem of creating this algorithm seems related to the traveling salesman problem (it's a different problem, but may involve similar math). But I expect that, to maximize revenue, satisfying the highest-paying clients will be more important than efficiency in maximizing the number of paying clients per hour. Prices of trips are likely to vary by more than an order of magnitude, and trips with a fixed source, destination, and time would tend to get lower bids. He might even want to give some of these otherwise-empty trips away for free, (e.g., as "upgrades" to airline passengers who were scheduled on that route, in order to build goodwill with airlines who are now sharing the airport with him, and who may be able to do him favors like transporting passengers who he has to cancel on). There are complications in terms of the timing of booking (regardless of whether there are secondary auctions), because some high-paying clients will want to book at the last minute and get a trip ASAP, while others will want to book in advance and have a guarantee that the trip will go as scheduled. So the revenue-maximizing strategy would probably involve a mix of trips booked in advance and trips booked closer to the departure time, with some preference for clients who are more flexible about timing or cancellations.
I think the correct solution probably involves hiring secretaries to manage the whole affair instead of bidding on a website.

Tutoring vs Classes

I want to learn American Sign Language. As a kid I grew up know some Deaf people and am attracted to the culture. It is also very relevant to my current job and my work will pay for me to learn (both my time and expenses).

I spent last semester taking classes at the local community college and while I learned the expected amount of ASL, I also learned I don't like going to classes. It feels like a good portion (>50%) is a waste of time. I'm conversing with other hearing students who don't know the language either; some things I already... (read more)

But that's a good thing, isn't it? I mean, that's what rational Harry Potter would do. Try finding a tutor that will let you make short videos of them doing the signs, and then you can put those videos into Anki. (At least I hope it is possible to put videos in Anki, never tried. Or maybe photos.) Could be more efficient than usual learning. Decide in advance how much is acceptable for you. Find out the minimum hourly wage in your area. Those are your limits. You can start by offering something in the middle of this interval.
I paid $25/hr for an ASL tutor in Washington DC. I found it to be a lot worse than taking a community college class. My tutor was Deaf, but didn't have a lot of teaching experience. She wasn't very good at steering the convos into new vocabulary ("Tell me about what you did last week" involved a lot of repetition). And she was much better at helping me pick up vocab (which I was decent at on my own) than grammar/expression. I ended up stopping the lessons and looking for a better option.

Any examples of total recursive functions that are not primitive recursive and do not violently explode?

The set of primitive recursive functions is interesting because it is pretty inclusive, (lots of functions have a primitive recursive implementation) and primitive recursive functions always terminate. I'm interested in trying to implement general purpose machine learning by enumerating primitive recursive functions. Which raises the question of just how general the primitive recursive functions really are.

Ackermann's function gives an example of what yo... (read more)

The inverse of the Ackermann function grows very slowly but is not primitive recursive. That is, the function f where f(n) is the smallest m such that A(m) exceeds n. Of course, this is hardly a different method of proof.

Is Friendly AI or more specifically CEV predicated on Eliminative Materialism being false? To what extent is FAI predicated on folk psychological theories of mental content turning out to accurately reflect human neurobiology?

From the article:

Modern versions of eliminative materialism claim that our common-sense understanding of psychological states and processes is deeply mistaken and that some or all of our ordinary notions of mental states will have no home, at any level of analysis, in a sophisticated and accurate account of the mind. In other words,

... (read more)
Let me see if I can unpack this idea a bit more. CEV is based on the idea that there is an algorithm that can look at the state of my brain, filter out various kinds of noise, and extrapolate what sort of desires and values I'd want to have if I lived in a kinder more benevolent society, wasn't subject to nearly as many serious cognitive biases, etc. The problem I'm seeing is that the origin and meaning of terms like 'desire' and 'value' are in prescientific culture - folk psychology. they were created by people in absolute ignorance about how brains work, and it seems increasingly plausible that these concepts will be totally inadequate for any accurate scientific explanation of how brains produce human behaviour. It seems to be common sense that desires and values and the like are indispensable theoretical posits simply because they are all we have. Our brains' extremely limited metacogntive abilities prevent us from modelling ourselves as brains, so our brains invent a kind of mythology to explain their behaviour, which is pure confabulation. If these ideas are right, by asking CEV to consider folk psychological ideas like desires and values, we would be committing it to the existence of things that just aren't really present in our brain states in any objective sense. In the worst case, running CEV might be somewhat analogous to asking the AI to use Aristotelian physics to build a better airplane. What we perceive as the fragility and complexity of human based values might not map onto brain states at all - 'values' as we wish to conceive of them may not exist outside of narrative fiction and philosophy papers. My recent thinking on these topics has been heavily influenced by the writings of Scott Bakker , Daniel Hutto and Peter Watts' Blindsight I hope I'm wrong about this stuff, but I don't have the training to fully analyze and debunk these ideas by myself - if it's even possible. I hope LW and MIRI have some insights about these issues, because I am
2Wei Dai
You may be interested in Yvain's Blue-Minimizing Robot sequence, which addresses these concerns. To read it, go to http://lesswrong.com/user/Yvain/submitted/?count=25&after=t3_8kn, and read the posts from "The Blue-Minimizing Robot" to "Tendencies in reflective equilibrium".
Thanks! I've read some of the stuff by Yvain but not these posts.

So, question - has anyone written a proof that Bayes' Theorem works on log-odds? It's a pretty simple proof, but I couldn't actually find it anyone on LW, which seems odd.

If it hasn't already been written I might toss it up as a back-to-basics post.

I think it's in chapter 3 of Jaynes' *Probability theory". Also (google is your friend), check here or here.
Oh, I see. Thanks!

This seems appropriate, considering the recent post on reality being weirdly normal: http://www.youtube.com/watch?v=ritaljhhk7s

Open comment thread:

If it's worth saying, but not worth its own top-level comment in the open thread, it goes here.

(Copied since it was well received last time.)

ETA: Recur further at your own risk. :)

Open subcomment thread: If you think it's worth saying, but not worth its own top-level reply in the open comment thread, you're too much of a nitpicker and seriously need to recalibrate your expectations of either what's worth saying or how much a comment costs.