alextp

By non-intrusive I meant that I have no idea what kind of ML is really going on behind the scenes. In this sense it's very different from interacting with, say, Netflix's recommender, or youtube's ad predictor (a friend "liked" a vodka ad once and is now assumed to be an alcoholic), where you really do need to build a mental model of what the ML is doing to effectively use the site. With prismatic I don't seem to need to care.

Tweet at me (atpassos_ml) and I can spend one of my three invites on you.

As a recent daily user, prismatic is awesome.

The way they leverage twitter is pretty interesting, mixing in "endorsements" to articles from both people close to you, famous people, and random people, in a way that keeps it interesting.

There's also fairly heavy machine learning going on behind the scenes, in a way that is non intrusive. Congratulations to those guys.

The population is strongly clustered on some big cities in the coast, the roads aren't super safe, and the terrain around the coastline is not flat, so it's probably just cheaper to lay them out in the ocean.

Jorge Luis Borges in a book of memoirs told that when he worked in the library in Buenos Aires as a clerk one day one of his colleagues barged into his office to show him an encyclopedia (or list of famous people, or something like that) that showed another argentinian named Jorge Luis Borges, a famous writer at that, who was born on the very day that Borges was born! (of course, it was an article about himself)

I don't think you can answer these questions with theorems or experiments about models of computation (which is how TCS answers questions), so I don't think they are TCS questions.

They are, however, central questions in the philosophy of computing.

As an owner of a shattered iPod and a pristine (but much more aggressively dropped) nexus S, the solution is really curved screens with protective plastic on the edges.

The title is kind of a bait. The article is an interesting-enough discussion of Amdahl's law and how it can limit efforts to parallelize stuff at google.

You can look for #<programming language> or #<library> on freenode, for example.

Regarding [1], each bus company has a different color. The companies won different government contracts for different parts of the city, so the east side is roughly red, the south blue, the north-west green, etc.

Indeed, after living in são paulo for a while going anywhere else in brazil lead to a slight culture shock.

In a sense this didn't hurt advertisers at all, as the demand for advertisement stayed exactly the same, they just had to change the available media. Think of it as a prisoner's dillema-like situation, where defecting is plastering a huge sign in front of people. It's better for everyone if nobody does that, but whoever breaks the law and does it can get an advantage.

Preventing people from defecting in these situations is exactly the point of a government (regardless of how much I disagree with Kassab's actual politics and would never vote for him).

> Why does this happen?

I'd guess it's because promoting this possible causal link as a cure for diseases is pretty much the easiest way to get it tested---where else will you find 25k volunteers?

As always, when people are involved, the reason for failure seems to be in the incentive structure.

This suggests that it's worthwhile to educate the young in not just doing the same things with computers that everybody knows how to do (Office-style and web browsing) but inculcating habits (versioning, say) which will make life much easier in the long run.

This, however, would make the education system not agnostic, and drastically favor some companies.

This is an interesting question. I'd say it's a mix of technical and social factors. Technically it's a solid abstraction (versioned directories), it's ridiculously fast (specially when compared to most alternatives at the time, even local ones), and somewhat space efficient (if you're not keeping binaries around). Because the format has been specified from the beginning it's easy to build all sorts of tools on top of it and add as much knowledge as you want, including line-by-line diffs, rename tracking, etc. It also ignores some hard problems people were focusing on at the time (merges, for example).

Socially, it first had a big userbase because the kernel is a large enough project and the switch was made top-down as far as I remember (not everybody was forced to do it but it was much easier if you could). This means a lot of people had to get acquainted with it, and this set of people turned out to be somewhat competent at changing it to make it faster, more stable, and easier to use (up to a point). Linus's personality played an important role here, as he entertainingly made a case for why it was good, which created a certain "cool factor". Second, there was the github factor which allowed this user base to explode exponentially by making it really cheap and easy to use.

> sometimes high intelligence or high ability, can be detriments to greatness.

I'm always bothered when I read this kind of conclusion. From the rest of your comment the detriments come from high standards, not high intelligence. Choosing a better overall strategy at the expense of precise execution is not evidence of lower strategy, just of lower standards when it comes to the execution.

The problem is more complex. Some errors will damage the scientific conclusions, others will show up as fractions of a percentage point inside a much larger confidence interval, and will ultimately not matter much.

In machine learning research, for example, most evaluations consist of running a new program on soem data, getting results back, and from these results computing some aggregate measure of performance. A bug on the code that computes this measure of performance is _really bad_ and can invalidate all your conclusions. If that code is right, however, a but on the code that trains your model is completely meaningless, because as long as your results are good you can argue that you actually meant to write a paper about the model actually implemented rather than the model you were supposed to implement.

I'm sure other scientific areas have similar distinctions, and a naive code reader might fail to notice that a bug is harmless (and there's also the fact that scientific code carries within it a lot of assumptions about the data which, if broken, can be buggy, but are not broken by real data).

I think the Berkeley starcraft AI used some similar concepts to move units around.

Except these days enrollment/retaining rates have a lot to do with perceived difficulty.

The worst thing is many investors who hold shares of one company also holds shares of the competitor, and hence they are spending money wildly to make their portfolio as a whole worth less.

At an internship at MSR I recently used windows 7 for real, and after I got through the first week the general experience was maybe more pleasant than ubuntu with unity. Even things that annoyed me at first (grouping windows by program in the task bar, for example) grew slowly.

Which is a pity when you think about how crappy and unfinished the unity UI in ubuntu 11.04 is. In years this is the first time I've stopped using a standard ubuntu (now I'm using Gnome 3, which is almost as bad).

Search is about taking you to the place you want to go to; facebook really wants to be the interesting place.

Their whole strategy seems to center on making sure everything in the web is a gateway to facebook but that facebook itself is where the fun is at. See for example those boxes websites can use saying "such and such friend liked such and such article"; they take you to facebook, never out of it. I really don't see entering search as something they should even want to do.

Indeed. Living in a third-world country makes it really obvious that while beauty is rooted in nature it can very easily be bought if you have enough money. As you age, the niceness your skin is inversely correlated to solar exposure and manual labor, and positively correlated with money spent in moisturizers and other products. How fit your body is also depends on the money/time spent on a gym and on the quality of the food you eat. Your hair will also look as nice as the money you spend on it (even if you're bald), and even teeth can be fixed if you have enough money at a young age. Finally, beauty is socially defined, and over time the standards will shift to mirror other aspects of society (in Brazil, for example, white people are generally better off, so all standards of beauty encourage "looking white", by hair straightening, makeup, etc. In places where rich people don't do manual work and stay indoors all day being pale is considered pretty, while in places where rich people can afford to go to the beach looking tan is considered better).

If all providers start actively trying to not touch these ads the expected incentive of posting them will go down; hopefully enough so that they end up having to negotiate directly with websites or something like this.

Indeed, I agree. I just hope they follow through and use this case as a precedent to shut down or severely harm the other ad networks that do essentially the same thing.

It's very relevant, as selective enforcement of laws is the bread-and-butter of corrupt and dictatorial governments everywhere. As the former de facto dictator (known anecdotally as "colonel" even though he was a civillian) of the place where I was born (Salvador, Bahia, Brazil) used to say: "to my friends, everything. To my enemies, the law"

Part of my point is that even engagement might not be the best metric. For example, here in hn, I'm far happier if people don't post than if they post something trivial or uninteresting (your post and comments are interesting as I think this is a discussion worth having).

I find this kind of thing funny, but misguided.

Karma in websites is not necessarily about accuratly reflecting some ground truth upvote probability, mean chance of liking a comment, expected future vote ratio of the commenter, etc. It's an incentive-design mechanism, in that a karma system is good if and only if it leads to the desired behavior when people use the website. When you ask yourself "how should I compare 5 upvotes and 5 downvotes versus 1 upvote versus 1 downvote versus no action at all", the answer is not weighting one of these situations higher/lower because it will better approximate one of the criteria above in expectation or something like this, but instead weighting these high/low depending on, for example, if you want to encourage activity, agreement, controversy, etc.

Ideally, you should have some other behavioral metric in mind (say, mean comment quality, top comment quality, bottom comment quality, engagement, etc) and try to tune the voting system to maximize this quality over time. (this tuning can either be done intuitively, as pg tries to do, or algorithmically, using something like the technique behind Gmail's priority inbox or bandit algorithms)

Do not "define away" a social problem with mathematics, use the mathematics to help you solve the actual social problem.

Also, hiring in third-world countries might be a problem. I'm brazillian, and unless you're in São Paulo you're likely to find that locals have a different sense of time, hurry, priorities, and civility. In certain parts of the country (Salvador, or Rio, for example), you should be ready for everything to happen at least an hour late, except when it doesn't. People will expect to be able to be late to meetings and appointments, and will miss deadlines for no reason.

Another limiting factor is traffic. As an american, you probably love driving, but here in Brazil (again excepting São Paulo and maybe parts of the south) long drives should be avoided if at all possible, as highways are full of potholes or randomly dangerous. Also, most of the economy (and hence good employees) will be in the big cities, and these have horrible traffic in the level of London or Manhattan for nothing close to the economic output.

The cooperative multitasking model of F#'s async makes good progress in solving your first point. There are essentially two ways of assigning the result of code blocks to variables: doing "let a = foo bar" which runs the code without a yield and "let! a = foo bar" which can yield as required inside foo. So you always know which library functions are safe (and you can even write unsafe ones yourself) and can write cooperative threads very easily (just as easily as usual threads).