alextp

> I agree with this, but I haven't actually been to a country that wasn't like that. Certainly here in Denmark, many Danes are quite certain that Denmark is run better than any other place. ;-)

You should visit the third world more often. At least here in Brazil many people believe everything is run by incompetent crooks and if they could just get out to foreign countries they would see what actual management and order are. (this is only partially true, specially as the country seems to be quickly decrooking in many levels)

> For these reasons, most every educated person in the world knows more about America than almost any other country (excluding their own and a handful of neighbors)

At least in brazil people seem to know more about america than the rest of the country, specially in the richer and more educated regions in the south and southwest. Everyone I talk to is far more likely to know who is the president of the US or the governor of california than who is the governor of Piauí or the vice-president of brazil.

At least here in brazil, where punctuality is generally disregarded, classes start on time (at most 5min tolerance) and people arriving 45min late to a scheduled appointment shouldn't expect to find anything waiting for them.

At the same time, social events in general happen always an hour or so later than announced, except for small gatherings and things with a fixed timeframe such as going to the movies. As a naturally punctual person I've arrived in birthday parties thinking I was being fashionably late and found everybody still pre-shower and the place a mess.

By voting patterns, what do you mean? Features like how many times the user has voted in this thread/hour/day? How many times the comment has been voted, the replies have been voted, and the story has been voted? I didn't try those; maybe they'd work, but I have no intuition saying why they should.

Using ML to provide feedback is a bad idea. Most ML techniques latch on to surface features of the text rather than the deeper structure, so it'd just make it really easy for people to reword their mean comments ("this is just stupid" becomes "What an incoherent piece of gobblydegook" or something like this, which might make things funnier but I doubt it would help).

They can do something HBGary-style and hack sony's servers to gather confidential information that would embarass the company and bring the stock price down. That would hit where it hurts, can be doable with some social engineering, and any company that big is bound to have some skeletons in its closet.

Not that I condone these actions, I'm just pointing out that they have done finer things in the past than just using LOIC.

The problem is that word features are not really that good predictors of quality.

I have done some research on this (unpublished), and I got a really good performance on predicting hacker news votes by just counting how many new words (not stopwords, not very-high-frequency words) a comment was adding to a thread. Just using a few variations on this theme predicted better than word counts or bigram features.

Fundamentally, though, I disagree with machine learning- based approaches as they can only _reinforce_ present behavior, and we'd like to shape voting behavior.

How about penalizing submissions by how often the title keywords and domain have been posted in the last week or so? This should push a lot of repetitive content down, and encourage the discussion to continue on the older submission.

Whenever there's some important piece of news like the japan earthquake we have highly redundant discussions on many different articles. This is wasteful.

How about adding a star to the top log(n) comments in each post?

Except with sufficiently snarky/provocative comments trying to shame the commenter might look from the outside like something very similar to petty disagreement.

I think some people here are learning that hugely self-confident, strongly opinionated, obvious writing tends to spark a strong agreement reaction on the readers, who quickly upvote a comment that adds nothing to a discussion. These are a problem, as they encourage snark and posing over effectively arguing things out, but they are very hard to treat as disagreeing with them is likely to cause knee-jerk reactions in many upvoters.

Also, the bird's eye view in bing maps always makes me go back to it, it's great to see what a neighborhood look like and how it would feel to actually follow a route. Google street view is just too zoomed in and too painful to navigate in to get a broad idea most of the time.

Except science and engineering is as much about people as it is about machine. Look at the most succcessful scientists and engineers: a lot of what they accomplished they owe to the other people who collaborated and worked with them. The amount of politics and interaction that goes in the higher levels of these professions is also something that can't be ignored.

I think it came with the economic boom, but I'm at a loss at explaining how or why. Also, part of my perception can be because I moved from a less to a more "civilized" place in Brazil (from Salvador to São Paulo).

I have anecdotal reports, from members of my family who used to work in law, that a couple of decades ago buying and selling real estate, for example, required a lot of palm-greasing. Maybe due to pressures from the industry this is not true right now in the big cities (I rarely go to the countryside, so I don't know how things are over there, although in brazil the coutnryside tends to move more slowly).

It seems, however, that even at its peak Brazil never had as bad a bribing culture as India seems to have, from the news.

All of the bribing I hear about these days are teenagers and young adults bribing cops to overlook possession of small amounts of drugs (and also some very high profile cases in the government), but bribing to get things you should have by right is something I or anyone I know in my generation has never had to do.

Except bribes undermine social fairness significantly. Notice that as the clerk presumably knows how much money he should give you for your tax returns there is a strong incentive for him to ask for a bribe that covers a significant share of your tax returns, effectively making the entire tax return process useless for the citizen.

Also, if you think of a slightly different scenario, imagine the office is only allowed to issue X tax returns a month (maybe the money comes in packets), and the bribe is required to cut the line. Now it's really unfair and wrong, and a lot more so for the briber than for the bribee.

You should have in mind that it's not actually that hard to bribe people indirectly. Any place with a bribing culture tends to have people you can hire to unofficially bribe some government employee to do something for you. In Brazil these are called "despachantes" (dispatchers), and even though in the last decade or so they have been getting progressively more useless as the country moves away from a bribing culture, you should keep in mind that a lot of despachantes are legally incorporated and even pay taxes. One can't be blamed for hiring that sort of person.

I'm in the same situation, but unhiding friends to recover sense is just not practical, with all the app-spam going around facebook these days.

I might be the only one in here who can't stand viral-app-spam, but in my social circle there's a sharp division between people who can and who can't stand it, and I think this is in part responsible to the huge drop in interaction in facebook.

It's not about never quitting, though. It's about quitting when you close the last document, or something similar, and breaking the distinction between closing and quitting.

Heavy-duty apps should quit in the background after you close what you were working on, but this shouldn't necessarily matter to the user.

Visual classification sounds interesting. So you render the page and use location-based features to extract content?

Actually I've created custom styles by modifying existing ones, and it isn't nearly as hard as it seems. Indeed, most academic styles are not more than a few hundred lines of code.

The issue is that generic machine learning algorithms work ok enough as black boxes, but to squeeze top performance out of them you need to do feature engineering, architecture/model structure futzing, method selection, etc, and in practice there are far too many of these meta-hyperparameters to tune with cross-validation or something similar.

While the generic ML tools work really well, it takes domain knowledge to find the best way of applying them to the problem at hand, specially since it almost never fits into the classification/regression from IID training training data model that most algorithms are designed based on. At first this might seem counter-intuitive to you, but I've seen dramatic reductions in the error rate just from picking good features or a reasonable model structure in a way that's not easy to automate. And while deep learning or structure learning tries to address these problems, there are issues with nonconvexity and really long training times that make these algorithms unrealistic in many situations (and, consequently, make them underperform simpler methods with clever domain engineering).

Please don't implement this sort of system---it's a headache, and can hurt users more often than not (important email from many important sources is unlikely to be authenticated, say a recruiter, or a contact for a new job, or a distant great-aunt who can't use computers), plus it's really annoying when, for example, mailing lists get this sort of annoying check back almost every time they email a new user.

The fact that transit visas are something you need to apply for, pay fees, and risk being refused (and that going into a permanent record) makes no sense for me.

UML brings back good memories from OS classes. We had to patch ext3 to add some triviality, and I did it first in UML so I could more easily test and make sure it was working correctly. It felt a lot faster and easier to setup than virtualbox or qemu (my options at the time), although I guess with hardware support and things like hypervisors it's no longer practical.

The problem with standardized testing is not that they don't measure the progress well but that they measure all other sorts of confounding variables. Each standardized test has a small number question templates, and if the students familiarize themselves with the question templates they will do better on the tests, for example. Another important factor is timing: some standardized tests have a fixed amount of time allocated to a student, and hence there are many different ways of pacing yourself that lead to better or worse scores. Yet another important factor is the choice between open-ended and multiple-choice questions, and even the balance between these can be learned.

So every standardized test has some confounding variables, and effort spent teaching the students to do better on the tests by dealing well with these variables is effort that does not go towards helping the students learn the material. If the standardized tests are known and used to evaluate the teachers the incentives suddenly get really strong to teach to the test by focusing on the idiossincrasies of specific testing schemes over teaching the actual subject.

Standardized tests are a great way to measure performance and improvement but only if the results won't be used for decision-making, otherwise it creates an instant incentive to game the system.

Spielman and Teng are very famous, and have received even a Godel prize for their work in smoothed analysis. Currently they're doing some very interesting things with efficient algorithms on graphs laplacians and linear systems, you can see Spielman's talk at last FOCS, it's very enlightening: http://weyond.com/www/focs/2010/downloads/1335.f4v

This is very interesting and very true. I think the main points of this article are:

1. Constants do matter, specially most programs end up with an asymptotic that looks like a N^b, and a and b are often not what you'd expect. This matches my experience a lot, and it's interesting that this is one level below O notation but one level above "counting how many operations should happen, roughly", which is what you have to do in practice to make algorithmic code go fast.

2. The examples around which an intro to CS course is structured are really uninteresting. I couldn't agree more, and never cared for computing factorial, fibonacci, etc, or for writing silly games or dumb data-processing programs; Simulations and numeric algorithms are far more interesting, and can get students interested more easily when they realize that other classes' work can be made far easier with computers.

3. Everyone should know CS. Even my philosophy-major wife often has the need to run short scripts to automate some tasks, and I think that if a philosophy major can benefit from CS concepts anyone else should as well.

All these things one should find out eventually, and I agree that exposing people to these concepts as soon as possible is a very good idea, far better than what is usually taught as CS-for-non-CS-majors.

When I didn't own a kindle I used to buy and read a lot of kindle books on my iPod touch. Even now that I own a kindle I share it with the wife, and we might both be on a bus/airplane without internet wanting to read.

I really hope this is not the way forward.

There are loads of duplicate papers. Searching, for example. for "latent dirichlet allocation" shows the same JMLR paper about four times in the top results.

I'm not sure about the market, as it doesn't seem to be explored, but the culture is certainly different.

First and foremost, we don't have the concept of a date in the same way, and we don't have the concept of a pick-up bar/pub either. As far as I can tell, most people would be really unhappy with posting their face and names on a butcher's shop-like website, in the same way that most people wouldn't want to admit they're out there looking for dates (it makes you seem lame) while flirting a lot.

The interaction patterns between people and specially couples are different, so it's hard to say what would work over here, but outside of the gay market it certainly doesn't seem to be a traditional dating site.

As a brazillian, yes, there are lots of points in common. Also, orkut is pretty much dead now among my social circle; facebook has pretty much got all users of orkut and some more people who wouldn't have joined orkut (like parents and grandparents of the orkut generation).

As far as I know when orkut came around here there was simply no other social network with a significant number of brazillian members. Myspace was just too complex to understand, I think, and part of the annoying things of early orkut's design (you had no news, so had to keep checking your friends' and comunities' pages out to see what was going on, which fostered a whole cyber-stalking culture of checking people out) really clicked with a lot of the demographics.

I think more than by analogy that the dating market over here is also underdeveloped, and there is definitely demand for something that caters more closely to the brazillian forms of dating (which are different from the european forms of dating, and from the indian).

> Anything that's free but is limited gets abused. You need look no further than the NHS in the UK were most GPs are employees of the NHS. Office hours of 9-4 (ie no incentive to provide a better service) are typical, the system clogged up with people who either don't need to see a doctor or, in the case with many old age pensioners, they simply go to the doctor for someone to talk to.

Actually, my experience in Brazil seems to provide counter-evidence to that. Higher education here can be either public or private. There are federal and state universities, but also it is reasonably easy to open a private college and get federal certification for it. In the public universities, in general, there is a far higher pressure for excellence at all levels, from admissions to lecturing and research (indeed, it is usual that the more interested undergrads will work on research projects during their course), and research specially is almost lacking in the private universities.

It is generally accepted here that if your objective is to make money the first thing that goes is demands on research (which is expensive and doesn't make much money for the university), closely followed by standards (as the more people you let in and the higher grades you give them the better-oiled the assembly line is and you can effectively print money as long as you don't lose your government certification).

In the public realm the people are kept motivated by: (1) the promise of very early tenure (effectively one year after you're hired by a public university), (2) the not-so-high salary, with availiability of extra funds in the form of productivity grants (for a researcher's personal use, not necessarly to hire students, etc), (3) the decoupling of the adviser's money from the students' funding (which ends up giving the students more freedom).

With my mother this failed miserably. After I gave up on her nagging for someone to do the dishes, she started nagging that the dishes weren't coming out as clean as they should. After I started scrubbing a lot longer and harder she started nagging that the drain and the sink weren't always pristine. After I started cleaning the drain and sink every time I did the dishes she complained the the countertop got wet. After I started drying the countertop, then the kitchen table had crumbs over it. After that, I broke down and had a really bad fight with her, and got her to shut up and just clean whatever it was that bothered her so much in the kitchen.