= 27d567bc2d48d9f40e9ccc20ea370b15)
I'm so glad that this exists now. Finally, an organized place for all the Summit videos, all on one page. Will it be updated in the future?
Adding on to that, these three links seem to be broken because of a stray %0A:
Alexander Wissner-Gross. Planetary-scale intelligence.
Tyler Cowen & Michael Vassar. Debate on the Great Stagnation.
Dileep George and Scott Brown. From planes to brains: building AI the Wright way.
In response to
Michael Lewis on Kahneman and Tversky! [link]
Here is a quiz from that article.
I think that the answer to 2 is incorrect.
I think that the answer to 2 is incorrect.
Yeah. According to Google, the experiment did find that people neglect base rates:
Kahneman and Tversky (1973) observed that the mean [estimate of the probability that Jack is an engineer] in the two groups, one receiving the base rate information 30 to 70, the other receiving 70 to 30, were for the most part the same
Ironically, when analyzing the experiment, the Vanity Fair writers failed Bayes theorem in the opposite way: neglecting evidence, thus making the posterior equal to the prior.
We conclude that people don't understand Bayes.
Whaa? How do you even manage to get signed up for conflicting classes- is MIT's registration system set up such that it allows you to do that?
How is this done? When I was at university, scheduling conflicts would start becoming a problem at 5 classes and be almost insurmountable at 7.
Students manage conflicts by simply skipping class sessions. Last semester, I often skipped two thirds of my class sessions. As long as you read lecture notes, do the work, and show up to tests, you're fine.
In response to
MIT Challenge: blogger to attempt CS curriculum on own
At MIT, some students take 8+ classes over ~15 weeks. This involves lots of busywork and an expectation of getting the highest grade (an A). [They also often do side projects.]
Scott Young aims to complete classes at the same rate. But he's skipping much of the busywork and requiring merely passing grades. I wouldn't be surprised if he pulls it off.
I'm an MIT student and currently spend 60-100 hours/class. Taking Young's approach, I could probably average 30 hours/class, which for 33 classes might be doable in about 2 months... Maybe doing 33 MIT classes in 1 month is something for a Tim Ferriss.
In response to
comment
by
on
Another treatment of Direct Instruction getting more into the technical details of the theory
So, if I were to make you a bet that #1-and-#2 is true that you should rationally take if you believe that 99% estimate, it must be set up so that gjmgain*0.01>owengain*0.99...
So unless I'm making some embarrassing simple math mistake here, if I put up say $2000 (Canadian) for "gjmgain" (wish I had more to play with, but unpaid intern, no work visa here, etc), you should be willing to put up anything less than $20.20...
Ah! But what if rather than money you had to put up that you would read the entire Theory of Instruction and the entire Research on Direct Instruction?
In response to
comment
by
on
Another treatment of Direct Instruction getting more into the technical details of the theory
So, if I were to make you a bet that #1-and-#2 is true [then you should rationally accept it if] gjmgain0.01>owengain0.99..
If you seriously made such a bet, then gjm would probably update on that evidence and revise his 99% probability upwards.
[But as gjm says the bet is impractical anyway because it's too hard to resolve]
In response to
Rationality and Relationships September 2011
I think this is a very bad idea, considering the record of past discussions about sex, gender, and related matters on LW. I've seen quite a few of those, and almost inevitably, the result is either an awful death spiral or, in case someone tries injecting a serious dose of reality, quarrels and internet drama. If the recent discussions superficially look better than usual, this is only because nobody has bothered much with trying to steer them closer to reality, and the death spirals have been able to drift away happily and undisturbedly.
For whatever reason, this forum has shown to be incapable of conducting rational discussions about these topics. This is a sad verdict, but I'm afraid it's realistic.
nobody has bothered much with trying to steer [discussions] closer to reality
Feels like you have forbidden knowledge. Not coincidentally, I want to know what it is.
What is it roughly? That innate differences across the sexes play a strong role in causing statistically different mating behaviors to develop? That these differences end up somewhat resembling "females want high-value sex and a devoted father while males want sex and sexually faithful partners"? That females are often attracted to high value behavior (e.g. PUA stuff)? That many people have some, possibly very vague, estimate of how sexually valuable they are, and act upon this belief? Is there any way you can quench my curiosity? It seems obvious that if you answer in general terms you won't offend anyone, as meta thought doesn't really push the emotional buttons.
PS: It has been suggested that general statements can cause worse beliefs in a group, since they're very simplified. But there should be some way of pointing to an area of the map without degrading that region of the map.
Previous discussion on the same video: http://lesswrong.com/lw/2xe/interesting_talk_on_bayesians_and_frequentists/
I think we should refrain from trying to publicize this, for that reason. People perceiving cryonics as "that weird thing Simon Cowell signed up for" are unlikely to sign up themselves.
Simon Cowell is known as the brutally honest [singing contest] judge. He worked himself up in the music industry and doesn't seem to be a nutjob. Personally, I think his name is good for the reputation of cryonics.
View more: Next
= b715d02e85f7b3b4ae65ca3d3e450868)
Subscribe to RSS Feed
= f037147d6e6c911a85753b9abdedda8d){kind=logo}
I think what they're doing is doing statistical inference for the fraction upvotes/total_votes. I'm not sure this is the best model, possible but it seems to have worked well enough.
I suspect they're taking the mean of the 95% confidence interval, but I'm not sure. There's actually a pretty natural way to do this more rigorously in a Bayesian framework, called hierarchical modeling (similar to this), but it can be complex to fit such a model.
Edit: However, a simpler Bayesian approach would just be to do inference for a proportion using a 'reasonable' prior for the proportion (which approximates the actual distribution of proportions) expressed as a Beta distribution (this makes the math easy). Come to think of it, this would actually be pretty easy to implement. You could even fit a full hierarchical model using a data set and then use the prior for the proportion you get from that in your algorithm. The advantage to this is that you can do the full hierarchical model offline in R and avoid having to do expensive tasks repeatedly and having to code up the fitting code. The rest of the math is very simple. This idea is simple enough that I bet someone else has done it.
If you use the Bayes approach with a Beta(x,y) prior, all you do is for each post add x to the # of upvotes, add y to the # of downvotes, and then compute the % of votes which are upvotes. [1]
In my college AI class we used this exact method with x=y=1 to adjust for low sample size. Someone should switch out the clunky frequentist method reddit apparently uses with this Bayesian method!
[1] This seems to be what it says in the pdf.