Every now and then I like to review my old writings so I can cringe at all the wrong things I wrote, and say "oops" for each of them. Here we go...
There was once a time when the average human couldn't expect to live much past age thirty. (Jul 2012)
That's probably wrong. IIRC, previous eras' low life expectancy was mostly due to high child mortality.
We have not yet mentioned two small but significant developments leading us to agree with Schmidhuber (2012) that "progress toward self-improving AIs is already substantially beyond what many futurists and philosophers are aware of." These two developments are Marcus Hutter's universal and provably optimal AIXI agent model... and Jurgen Schmidhuber's universal self-improving Godel machine models... (May 2012)
This sentence is defensible for certain definitions of "significant," but I think it was a mistake to include this sentence (and the following quotes from Hutter and Schmidhuber) in the paper. AIXI and Godel machines probably aren't particularly important pieces of progress to AGI worth calling out like that. I added those paragraphs to section 2.4. not long before the submission deadline, and regretted it a couple months later.
one statistical prediction rule developed in 1995 predicts the price of mature Bordeaux red wines at auction better than expert wine tasters do. (Jan 2011)
No, that's a misreading of the study.
On September 26, 1983, Soviet officer Stanislav Petrov saved the world. (Nov 2011)
Eh, not really.
in the U.S., the administering charity need not spend from the donor-advised fund as the donor wishes, though they often do. (Jul 2012)
Silly. Donor-advised funds basically always fund as the donor wishes.
On September 26, 1983, Soviet officer Stanislav Petrov saved the world. (Nov 2011)
Eh, not really.
The Wiki link in the linked LW post seems to be closer to "Stanislav Petrov saved the world" than "not really":
Petrov judged the report to be a false alarm, and his decision is credited with having prevented an erroneous retaliatory nuclear attack
...
His colleagues were all professional soldiers with purely military training and, following instructions, would have reported a missile strike if they had been on his shift.
...
Petrov, as an individual, was not in a position where he could single-handedly have launched any of the Soviet missile arsenal. ... But Petrov's role was crucial in providing information to make that decision. According to Bruce Blair, a Cold War nuclear strategies expert and nuclear disarmament advocate, formerly with the Center for Defense Information, "The top leadership, given only a couple of minutes to decide, told that an attack had been launched, would make a decision to retaliate."
A closely related article says:
Petrov's responsibilities included observing the satellite early warning network and notifying his superiors of any impending nuclear missile attack against the Soviet Union. If notification was received from the early warning systems that inbound missiles had been detected, the Soviet Union's strategy was an immediate nuclear counter-attack against the United States (launch on warning), specified in the doctrine of mutual assured destruction.
That he didn't literally have his finger on the "Smite!" button, or that the SU might still not have retaliated if he'd raised the alarm, is not the point.
previous eras' low life expectancy was mostly due to high child mortality.
I have long thought that the very idea of "life expectancy at birth" is a harmful one, because it encourages exactly that sort of confusion. It lumps together two things (child mortality and life expectancy once out of infancy) with sufficiently different causes and sufficiently different effects that they really ought to be kept separate.
I'm going to do the unthinkable: start memorizing mathematical results instead of deriving them.
Okay, unthinkable is hyperbole. But I've noticed a tendency within myself to regard rote memorization of things to be unbecoming of a student of mathematics and physics. An example: I was recently going through a set of practice problems for a university entrance exam, and calculators were forbidden. One of the questions required a lot of trig, and half the time I spent solving the problem was just me trying to remember or re-derive simple things like the arcsin of 0.5 and so on. I knew how to do it, but since I only have a limited amount of working memory, actually doing it was very inefficient because it led to a lot of backtracking and fumbling. In the same sense, I know how to derive all of my multiplication tables, but doing it every time I need to multiply two numbers together is obviously wrong. I don't know how widespread this is, but at least in my school, memorization was something that was left to the lower-status, less able people who couldn't grasp why certain results were true. I had gone along with this idea without thinking about it critically.
So these are the things I'm going to add to my anki decks, with the obligatory rule that I'm only allowed to memorize results if I could theoretically re-derive them (or if the know-how needed to derive them is far beyond my current ability). These will include common trig results, derivatives and integrals of all basic functions, most physical formulae relating heat, motion, pressure and so on. I predict that the reduction in mental effort required on basic operations will rapidly compound to allow for much greater fluency with harder problems, though I can't think of a way to measure this. Also, recommendations for other things to memorize are welcome.
Also, relevant
In my experience memorization often comes for free when you strive for fluency through repetition. You end up remembering the quadratic formula after solving a few hundred quadratic equations. Same with the trig identities. I probably still remember all the most common identities years out of school, owing to the thousands (no exaggeration) of trig problems I had to solve in high school and uni. And can derive the rest in under a minute.
Memorization through solving problems gives you much more than anki decks, however: you end up remembering the roads, not just the signposts, so to speak, which is important for solving test problems quickly.
You are right that "the reduction in mental effort required on basic operations will rapidly compound to allow for much greater fluency with harder problems", I am not sure that anki is the best way to achieve this reduction, though it is certainly worth a try.
Some names familiar to LWers seem to have just made their fortunes (again, in some cases); http://recode.net/2014/01/26/exclusive-google-to-buy-artificial-intelligence-startup-deepmind-for-400m/ (via HN)
Google is shelling out $400 million to buy a secretive artificial intelligence company called DeepMind....Based in London, DeepMind was founded by games prodigy and neuroscientist Demis Hassabis, Skype & Kazaa developer Jaan Tallin and researcher Shane Legg.
I liked Legg's blog & papers and was sad when he basically stopped in the interests of working on his company, but one can hardly argue with the results.
EDIT: bigger discussion at http://lesswrong.com/r/discussion/lw/jks/google_may_be_trying_to_take_over_the_world/#comments - new aspects: $500m, not $400m; DeepMind proposes an ethics board
Robin Hanson on Facebook:
Academic futurism has low status. This causes people interested in futurism to ignore those academics and instead listen to people who talk about futurism after gaining high status via focusing on other topics. As a result, the people who are listened to on the future tend to be amateurs, not specialists. And this is why "we" know a lot less about the future than we could.
Consider the case of Willard Wells and his Springer-published book Apocalypse When?: Calculating How Long the Human Race Will Survive (2009). From a UCSD news story about a talk Wells gave about the book:
Larry Carter, a UCSD emeritus professor of computer science, didn’t mince words. The first time he heard about Wells’s theories, he thought, “Oh my God, is this guy a crackpot?”
But persuaded by Well’s credentials, which include a PhD from Caltech in math and theoretical physics, a career that led him L-3 Photonics and the Caltech/Jet Propulsion Laboratory, and an invention under his belt, Carter gave the ideas a chance. And was intrigued.
For a taste of the book, here is Wells' description of one specific risk:
When advanced robots arrive... the serious threat [will be] human hackers. They may deliberately breed a hostile strain of androids, which then infects normal ones with its virus. To do this, the hackers must obtain a genetic algorithm and pervert it, probably early in the robotic age before safeguards become sophisticated... Excluding hackers, it seems unlikely that androids will turn against us as they do in some movies... computer code for hostility is too complex... In the very long term, androids will become conscious for the same reasons humans did, whatever those reasons may be... In summary, the androids have powerful instincts to nurture humans, but these instincts will be unencumbered by concerns for human rights. Androids will feel free to impose a harsh discipline that saves us from ourselves while violating many of our so-called human rights.
Now, despite Larry Carter's being "persuaded by Wells' credentials" — which might have been exaggerated or made-up by the journalist, I don't know — I suspect very few people have taken Wells seriously, for good reason. He's clearly just making stuff up, with almost no study of the issue whatsoever. (On this topic, the only people he cites are Joy, Kurzweil, and Posner, despite the book being published in 2009.)
But reading that passage did drive home again what it must be like for most people to read FHI or MIRI on AI risk, or Robin Hanson on ems. They probably can't tell the difference between someone who is making stuff up and an argument that has gone through a gauntlet of 15 years of heated debate and both theoretical and empirical research.
In this article, Eliezer says:
Bad argument gets counterargument. Does not get bullet. Never. Never ever never for ever.
Recently, a similar phrase popped into my head, which I found quite useful:
Confusion gets curiosity. Does not get anger, disgust or fear. Never. Never ever never for ever.
That's all.
I've been systematically downvoted for the past 16 days. Every day or two, I'd lose about 10 karma. So far, I've lost a total of about 160 karma.
It's not just somebody just going through my comments and downvoting the ones they disagree with. Even a comment where I said "thanks" when somebody pointed out a formatting error in my comments is now at -1.
I'm not sure what can/should be done about this, but I thought I should post it here. And if the person who did this is here and there is a reason, I would appreciate it if you would say it here.
A quick look at the first page of your recent comments shows most of your recent activity to have been in the recent "Is Less Wrong too scary to marginalized groups?" firestorm.
One of the most recent users to complain about mass downvoting also cited participation in flame-bait topics (specifically gender).
Reason #k Why I <3 Pomodoros:
They really help me get over akrasia. I beemind how many pomodoros I do per week, so I do tasks I would otherwise procrastinate if I can do 20 minutes of them (yes, I do short pomodoros) and get to enter a data point at the end. Often I find that the task is much shorter/less awful than it felt in the abstract.
Example: I just moved today, and didn't have that much to unpack, but decided I'd do it tomorrow, because I felt tired and it would presumably be long and unpleasant. But then I realized I could get a pomodoro out of it (plus permission from myself to stop after 20 min and go to bed). Turns out it took 11 minutes and now I'm all set up!
Even if you know that signaling is stupid, it doesn't escape the cost of not signaling.
It's a longstanding trope that Eliezer gets a lot of flack for having no formal education. Formal education is not the only way to gain knowledge, but it is a way of signaling knowledge, and it's not very easy to fake (Not so easy to fake that it falls apart as a credential on its own). Has anyone toyed around with the idea of sending him off to get a math degree somewhere? He might learn something, and if not it's a breezy recap of what he already knows. He comes out the other side without the eternal "has no formal education" tagline, and a whole new slew of acquaintances.
Now, I understand that there may be good reasons not to, and I'd very much appreciate someone pointing me to any previous discussion in which this has been ruled out. Otherwise, how feasible does it sound to crowdfund a "Here's your tuition and an extra sum of money to cover the opportunity cost of your time, I don't care how unfair it is that people won't take you seriously without credentials, go study something useful, make friends with your professors, and get out with the minimum number of credits possible" scholarship?
Has anyone toyed around with the idea of sending him off to get a math degree somewhere?
I think the bigger issue w/ people not taking EY seriously is he does not communicate (e.g. publish peer reviewed papers). Facebook stream of consciousness does not count. Conditional on great papers, credentials don't mean that much (otherwise people would never move up the academic status chain).
Yes it is too bad that writing things down clearly takes a long time.
Somehow I doubt I will ever persuade Eliezer to write in a style fit for a journal, but even still, I'll briefly mention that Eliezer is currently meeting with a "mathematical exposition aimed at math researchers" tutor. I don't know yet what the effects will be, but it seemed (to Eliezer and I) a worthwhile experiment.
I don't think you understand signaling well.
Eliezer managed signaling well enough to get a billionaire to fund him on his project. A billionaire who fund people who drop out of college systematically in projects like his 20 Under 20 program.
Trying to go the traditional route wouldn't fit into the highly effective image that he already signals.
Put another way, the purpose of signaling isn't so nobody will give you crap. It's so somebody will help you accomplish your goals.
People will give you crap, especially if they can get paid to do so. See gossip journalists, for instance. They are not paid to give boring and unsuccessful people crap; they are paid to give interesting and successful people crap.
4 years (or even 1 year if you are super hard-core) of time is a pretty non-trivial investment. I was 2 classes away from a second degree and declined to take them, because the ~100 hours of work it would have taken wasn't worth the additional letters after my name. I also just really don't know anyone relevant who thinks that a college degree or lack thereof particularly matters (although the knowledge and skills acquired in the course of pursuing said degree may matter a lot). Good people will judge you by what you've done to demonstrate skill, not based on a college diploma.
I think IlyaShpitser's comment pretty much nails it.
People often ask why MIRI researchers think decision theory is relevant for AGI safety. I, too, often wonder myself whether it's as likely to be relevant as, say, program synthesis. But the basic argument for the relevance of decision theory was explained succinctly in Levitt (1999):
If robots are to put to more general uses, they will need to operate without human intervention, outdoors, on roads and in normal industrial and residential environments where unpredictable physical and visual events routinely occur. It will not be practical, or even safe, to halt robotic actions whenever the robot encounters an unexpected event or ambiguous visual interpretation.
Currently, commercial robots determine their actions mostly by control-theoretic feedback. Control-theoretic algorithms require the possibilities of what can happen in the world be represented in models embodied in software programs that allow the robot to pre-determine an appropriate action response to any task-relevant occurrence of visual events. When robots are used in open, uncontrolled environments, it will not be possible to provide the robot with a priori models of all the objects and dynamical events that might occur.
In order to decide what actions to take in response to un-modeled, unexpected or ambiguously interpreted events events in the world, robots will need to augment their processing beyond controlled feedback response, and engage in decision processes.
My meditation blog from a (somewhat) rationalist perspective is now past 40 posts:
Did someone here ask about the name of a fraud where the fraudster makes a number of true predictions for free, then says "no more predictions, I'm selling my system."? There's no system, instead the fraudster divided the potential victims into groups, and each group got different predictions. Eventually, a few people have the impression of an unbroken accurate series.
Anyway, the scam is called The Inverted Pyramid, and the place I'd seen it described was in the thoroughly charming "Adam Had Three Brothers. by R.A. Lafferty.
Edited to add: It turned out that someone had asked at Making Light.
John_Maxwell_IV and I were recently wondering about whether it's a good idea to try to drink more water. At the moment my practice is "drink water ad libitum, and don't make too much of an effort to always have water at hand". But I could easily switch to "drink ad libitum, and always have a bottle of water at hand". Many people I know follow the second rule, and this definitely seems like something that's worth researching more because it literally affects every single day of your life. Here are the results of 3 minutes of googling:
http://www.sciencedirect.com/science/article/pii/S0002822399000486:
Dehydration of as little as 1% decrease in body weight results in impaired physiological and performance responses (4), (5) and (6), and is discussed in more detail below. It affects a wide range of cardiovascular and thermoregulatory responses (7), (8), (9), (10), (11), (12), (13) and (14).
The Nationwide Food Consumption Surveys indicate that a portion of the population may be chronically mildly dehydrated. Several factors may increase the likelihood of chronic, mild dehydration, including a poor thirst mechanism, dissatisfaction with the taste of water, common consumption of the natural diuretics caffeine and alcohol, participation in exercise, and environmental conditions. Dehydration of as little as 2% loss of body weight results in impaired physiological and performance responses. New research indicates that fluid consumption in general and water consumption in particular can have an effect on the risk of urinary stone disease; cancers of the breast, colon, and urinary tract; childhood and adolescent obesity; mitral valve prolapse; salivary gland function; and overall health in the elderly. Dietitians should be encouraged to promote and monitor fluid and water intake among all of their clients and patients through education and to help them design a fluid intake plan.
The effect of dehydration on mental performance has not been adequately studied, but it seems likely that as physical performance is impaired with hypohydration, mental performance is impaired as well (62) and (63). Gopinathan et al (29) studied variation in mental performance under different levels of heat stress-induced dehydration in acclimatized subjects. After recovery from exercise in the heat, subjects demonstrated significant and progressive reductions in the performance of arithmetic ability, short-term memory, and visuomotor tracking at 2% or more body fluid deficit compared with the euhydrated state.
So how much is 2% dehydration? http://en.wikipedia.org/wiki/Dehydration#Differential_diagnosis : "A person's body, during an average day in a temperate climate such as the United Kingdom, loses approximately 2.5 litres of water.[citation needed]" http://en.wikipedia.org/wiki/Body_water quotes Arthur Guyton 's Textbook of Medical Physiology: "the total amount of water in a man of average weight (70 kilograms) is approximately 40 litres, averaging 57 percent of his total body weight." So effects on cognition become apparent after 40l*2%=800ml of water has been lost, which takes roughly 800ml/(2.5l/24h) = 8 hours. Now, this assumes water is lost at a constant rate, which is false, but it still seems like it would take a while to lose a full 800ml. Which implies that you don't have to make a conscious effort to drink more water because everybody gets at least mildly thirsty after, say, half an hour of walking around outside on a warm day, which seems like it would be a lot less than 800ml.
http://freebeacon.com/michelle-obamas-drink-more-water-campaign-based-on-faulty-science/ : “There really isn’t data to support this,” said Dr. Stanley Goldfarb of the University of Pennsylvania. “I think, unfortunately, frankly, they’re not basing this on really hard science. It’s not a very scientific approach they’ve taken. … To make it a major public health effort, I think I would say it’s bizarre.” Goldfarb, a kidney specialist, took particular issue with White House claims that drinking more water would boost energy. ”The idea drinking water increases energy, the word I’ve used to describe it is: quixotic,” he said. “We’re designed to drink when we’re thirsty. … There’s no need to have more than that.”
http://ask.metafilter.com/166600/Drinking-more-water-should-make-me-less-thirsty-right : When you don't drink a lot of water your body retains liquid because it knows it's not being hydrated. It will conserve and reabsorb liquid. When you start drinking enough water to stay more than hydrated your body will start using the water and then dispensing of it as needed. Your acuity for thirst will be activated in a different way and in a sense work better.
Some thoughts:
- More frequent water-drinking makes you urinate more often, which is probably a bad thing for productivity.
- There might be negative effects with chronic mild dehydration at levels less severe than in the studies above.
- There might also be hormetic effects. (As in, your body functions best under frequent mild dehydration because that's what happened in the EEA, and always giving it as much water as it wants will be bad.)
Thoughts? Please post your own opinion if you're knowledgeable about this or if you've researched it.
A year ago, I was asked to follow up on my post about the January 2013 CFAR workshop in a year. The time to write that post is fast approaching. Are there any issues / questions that people would be particularly interested in seeing this post address / answer?
Is there a reasonably well researched list of behaviors that correlate positively with lifespan? I'm interested in seeing if there are any low hanging fruit I'm missing.
I found this previously posted, and a series of posts by gwern, but was wondering if there is anything else?
A quick google will give you a lot of lists but most of them are from news sources that I don't trust.
Has anyone had experiences with virtual assistants? I've been aware of the concept for many years but always been wary of what I perceive to be the risks involved in letting a fundamentally unknown party read my email.
I'd like to hear about any positive or negative experiences.
One problem with searching for information about the trustworthiness of entities like these is that one suspects any positive reports one finds via Googling to be astroturfing, and if one finds negative reports, well, negatives are always over-reported in consumer services. That's why I'm asking here.
The MIRI course list bashes on "higher and higher forms of calculus" as not being useful for their purposes and calculus is not on the list at all. However, I know that at least some kind of calculus is needed for things like probability theory.
So imagine a person wanted to work their way through the whole MIRI course list and deeply understand each topic. How much calculus is needed for that?
Not much. The kind of probability relevant to MIRI's interests is not the kind of probability you need calculus to understand (the random variables are usually discrete, etc.). The closest thing to needing a calculus background is maybe numerical analysis (I suspect it would be helpful to at least have the intuition that derivatives measure the sensitivity of a function to changes in its input), but even then I think that's more algorithms. Not an expert on numerical analysis by any means, though.
If you have a general interest in mathematics, I still recommend that you learn some calculus because it's an important foundation for other parts of mathematics and because people, when explaining things to you, will often assume that you know calculus after a certain point and use that as a jumping-off point.
Last night we had meetup in Ljubljana. It was a good debate, but quite a heretical one for the LW standards. Especially when organizers left us. Which was unfortunate. We mostly don't see ourselves particularly bonded to LW at all. Especially I.
We discussed personal identity, possible near super-intelligence (sudden hack, if you wish), Universe transformation following this eventuality, and some lighter topics like fracking for gas and oil, language revolutions throughout history, neo-reactionaries and their points, Einstein's brains (whether they were lighter or heavier than average - I am quite sure they were heavier but it seems that the Cathedral says otherwise).
We discussed Three Worlds Collide, IBM brain simulations, MIRI endeavors and progress, genetics ...
More than 5 hours of an interesting debate.