sqrt17

A point that has become more obvious over the years is this: there are "soft" instances of NP-complete problems where you can find good approximations in reasonable time, and there are "crunchy" instances of the same class of problem where you don't find a good approximation.

Commercial ILP solvers (e.g. Gurobi, CPlex) profit from the fact that quite often it is possible to formulate NP-complete problems and find good solutions in acceptable time. Similarly, many of the "physical" ways of solving NP-complete problems work ok for easy instances and get unwieldly fast if confronted with difficult instances.

usually, we assume that a meat grinder does not "understand" meat, that a rice cooker does not "understand" food, and that a telephone or a CD player does not "understand" human language.

Wiktionary terms "understand" as (i) be aware of the meaning of something (so we've just reduced it to the meaning of awareness and meaning), or (ii) to impute meaning that is not explicitly stated.

To put it in more colorful terms, a blind person can understand the difference between green, yellow and brown bananas, but they usually cannot understand the visual aspect of color.

By the same rationale, a vector does not understand the word it describes any more than the telephone book understands the people listed in it or the city that they live in.

I'd argue that NNs (i) have to have a strong bias towards sensible structure to work well (e.g., use image topology in convolutional networks), and that (ii) a trained NN models one particular mode of the distribution, similar how Variational Bayes fits to one particular mode.

Which is arguing for ensembles of neural networks, which - as people have found out - are even more effective than single NNs.

The article is reporting about a scientific study (http://arxiv.org/pdf/1504.00680v1.pdf discussed in https://news.ycombinator.com/item?id=9399090) that specifically talks about "future banned users" without asking why these users would be banned; The rest of the economist article is journalistic speculation.

The study authors do report that there's a propensity for polarization, i.e., that "we find a community bias against FBUs: posts written later in an FBU’s life are more likely to be deleted, regardless of whether they are actually worse".

I don't see at all why nominal interest rates have to be superior to the interest that people get through other assets (including cash).

The real problem is that the base of the reserve requirements are based on demand deposits, which means that customer-centric banks are subject to reserve requirements while investment banks and (e.g.) bitcoin-based deposits are not. Which makes sense to some extent, because customer-centric banks are expected to have some cash on hand, but not to the extent that they would be the (only) ones who suffer from lower nominal interest rates. Alternatives would be, e.g.:

a value tax on all investments (don't see how that could be unpopular)

devaluing the currency, by printing more money

In particular, the Apache License 2.0 is the most appealing license for "commercial-friendly" open source, whereas the GPL with its "developer-friendly" license correlates highly with FSF-hosted, Gnome/KDE/... or independent projects.

The Java/C(++)? divide also seems to correlate with "commercial-friendly" (Hadoop, Lucene) versus "developer-friendly" (Blender, Linux) software.

I'm not always a fan of modelviewculture, but I think what they say implcitly is very important: start with culture.

Have a culture where men and women alike consider themselves functional and important parts of your business.

Let this culture speak through your actions (specifically: offer benefits that are attractive to the kind of person you want to attract), and let both the culture and you actions shine through in your public appearance (general website, HR-specific website, job ads).

Also: pre-recruiting activities (community involvement, generally getting in touch with people) can help your visibility especially with people who are not well-connected members of the majority group. Get involved, and, as mentioned, let your culture shine through.

As they further mention, some of the recruiting practices that are popular tend to be systematically biased towards a single, relatively homogeneous group of people. See this as an opportunity to perform arbitrage on that market by talking to suitable candidates that some of the other players are ignoring.

This gives me a strange impression of deja vu ... oh right, RedHat Linux and the not-quite-standard gcc 2.96 version and the solution of having a separate "kgcc" program that you could use to compile the programs that the default gcc broke. I feel slightly old now.

The article does seem to imply that they want to ship the new gcc but keep using the old ABI for a while to keep things compatible. Did I read that right?

Germany in the 1700s/1800s had similar shoulder poles for carrying water: http://static3.akpool.de/images/cards/65/655879.jpg They added a thicker, slightly regressed bit in the middle to avoid the some of the pressure on the shoulder/neck area.

You're arguing from a frame where there is a demand for a constant set of things ("jobs") and people capture the money that can be exchanged for that set of things. This assumes that a number of parameters in the market are fixed, and that you can "create" or "destroy" jobs with economic actions that are partly independent of the non-economic world.

Looking at historical economics, another way of looking at this is to see an economy as things going in (sunshine, non-renewable resources consumed, loot acquired in wars) and things going out (people eating, garbage produced, etc.), and the ways of distributing the flow between these.

It turns out that you can have a "surplus", when you suddenly have more money/resources than you had before. There are different means of getting rid of surplusses, including wars, social programs, or simply population growth (consider how New York's area used to sustain about a hundred people).

Looking back at the demise of weaving and spinning as jobs, you see that the immediate effect of that was a higher concentration of wealth in the hands of some individuals, which (in Europe) led to counter-movements involving democratization and social issues.

The real problem is not that it's possible to find more effective ways to do some jobs than employing people for it, the real problem consists in the readjustment of power relationships that becomes necessary when you remove resposibilities from lots of people and put them in the hands of relatively few people.

Maybe some governments institute a robot tax, which augments the cost of robots by a percentage not high enough to consider other alternatives (e.g. moving off-shore, which has been possible with production jobs but not with manufacturing jobs), and use these funds towards things that society as a whole regards as beneficial (such as building nice parks, everyone getting a university degree, or every city having an opera and a theater, or war or terrorism or spying on your citizens, if those are what a society thinks it wants).

Standard "capitalist" economic doctrine wants us to believe that surplusses always belong to the innovators or those taking associated risks, and both historically and currently this has only been partly true. Maybe those surplusses will just be destroyed in the next "AI" bubble, maybe they will be used to create a post-democratic oppressive state, maybe they will be used to create a modern-day utopia. We don't know yet what will happen, but we sure should not let others reframe the issue as technocrats-vs-luddites, because that is not what it is.

There's a thought error in here: "the economy", as an unmovable construct, does not really exist. As an example, a good shirt does not cost thousands of dollars anymore, because we have automated weaving looms and a whole textile industry. We're not poorer from it, though, because we shifted from weaving shirts to doing construction work or working in finance.

If we can automate simple manual labor, people will shift to different tasks; the non-beneficial changes is that now the energy needed for that task is consumed directly by the machine, and also the investment that has been done to get the machine there in the first place. Beyond that, people who did simple tasks before may perform tasks that require conscience or risk-awareness (for example, machines injuring or killing humans cannot be fired or fined, so someone has to be responsible), that require empathy and human attention; or people will resort to being burglars or robbers or revolutionaries or salesmen.

The article specifically made the distinction between service dogs - which are well-trained and, as a result, well-behaved and very seldomly a bother to anyone - and untrained animals that are held only for reasons of mental comfort.

The idea of the article is that, although some people with disabilities benefit from an untrained animal to the point that one wants to prevent others (e.g. landlords) from doing undue harm by not being accommodating at all, uncertainty about these provisions (not the provisions themselves!) are being abused by some egocentric pet owners.

There's nothing special about latin.

If you teach people the grammar of a language throroughly, and expose them to all sorts of historic, philosophical and poetic writings, they will also learn some abstract thinking.

For most foreign languages, it's considered more useful to be able to do trifling things like buy food, find shelter, or write business letters.

Romans had some of these woes, but their empire-building efforts and the ensuing plundering and enslaving of half of Europe (and a corresponding part of Northern Africa) allowed them to employ philosophers and poets that were not caught up in the rat race that most such people in the 19th and 20th Century were subjected to.

The answer in other countries has been to train the police force so that they can accurately assess risks to themselves and others and use the amount of caution and, if needed, force necessary to fulfill their task.

Police should value their own lives higher than that of people behaving aggressively towards them, but that's speaking in a degree. And they shouldn't behave like dangerous paranoics (because of lack of training, or because they are dangerous paranoics).

Not trying to be cynical about this, but: You want nice things, not bitcoins and not dollars.

Currently, more market actors accept dollars (or Euros, or JPY, or UKP) than bitcoin, hence you're closer to the things you desire if you have dollar liquidity than if you have bitcoin liquidity.

Investments, on the other hand, should be neither in dollars nor in bitcoins, because dollars have inflation and bitcoin is too volatile.

Quoting a symposium with Chomsky talking about statistical AI: http://languagelog.ldc.upenn.edu/myl/PinkerChomskyMIT.html

"I think there have been some successes, but a lot of failures. The successes, to my knowledge at least, are those that integrate statistical analysis with some universal grammar properties, some fundamental properties of language; when they're integrated, you sometimes get results [...] On the other hand there's a lot work which tries to do sophisticated statistical analysis, you know bayesian and so on and so forth, without any concern for the uh actual structure of language, as far as I'm aware that only achieves success in a very odd sense of success. There is a notion of success which has developed in computational cognitive science in recent years which I think is novel in the history of science. It interprets success as approximating unanalyzed data."

The model that has become dominant in statistical AI -- positing a basic grammar that is strongly underconstrained and eliminating spurious analyses not through "universal grammar" (i.e. presupposed innate structures) but through learned parameters, would be something that Chomsky has been very much against; Simultaneously, work that models grammar with enough precision that you could derive predictions from it (e.g. Ed Stabler's grammar implementation) are seen as nice-to-have but not central to the undertaking of generative grammar.

And I think Chomsky put his thumb right on the difference in goals: Chomsky defines progress in linguistics as work that posits the right ("universal") structures, and argues that these are cognitively real and innate, whereas statistical AI is more interested in predicting useful things with structures that may or may not correspond to anything that is cognitively real.

To people nowadays, the whole notion of constrained "universal" models with few statistics versus underconstrained "statistical" models seems to be a very minor one, since today's statistical models have a lot of structure, and people doing generative grammar aren't totally opposed to using statistics or optimality theory to select most-plausible structures. But, back in the day, when the most expressive statistical models people used were HMMs [hidden Markov model - a probabilistic regular grammar] and PCFGs [probabilistic context-free grammars], the gap was much wider, whereas nowadays the models are a bit more similar while the goals are still different.

It's important to make a distinction between (i) Chomskyan linguistics, (ii) 90s style symbolic systems, (iii) 90s/early 2000s style statistical systems and (iv) 2010s style statistical systems.

Chomskyan linguistics assumes that statistics and related stuff is not relevant at all, and that instead you need to find the god-given (or at least innate) Universal Grammar and then everything will be great. 90s style symbolic systems adopt a more realistic approach, relying on lots of heuristics that kind of work but aim at good performance rather than unattainable perfection; 90s style statistical models give up some of the insights in these heuristics to construct tractable statistical models.

If you look at 2010s style statistical models, you'll notice that machine learning has become more powerful and you can use a greater variety of information, either using good linguistic intuitions (which help even more with better learning algorithms, but require a certain expressivity as well as some degree of matching between the way of constructing the features and the classification) or unsupervised/deep-NN learning, which constructs generalizations over features.

The main reason that you won't ever see people talking about systems with great machine learning and great linguistic intuitions is that you normally want to treat one of them as fixed and focus on improving the other, i.e., it's more a practical/cultural difference than an actual limitation.

Maybe it's because Perl was really popular at the time I discovered Unix and its tool, but why would you use sed and awk instead of a Perl one-liner? (Or even :s// in vim or M-x query-replace-regexp in emacs, if it's just regex munging)

The author of the post started by getting into programming/CS by wanting to earn more/better money, by picking up an Access book and working it out from there, and now that he's more established, he looks down at the young hungry people who picked up an Access book in the hopes of more/better money, running through all the tropes that people who got where they are through knowledge will use.

Yes, it's useful to understand things better, and to know math. And as always since the first CS degree started, CS people gripe that people should Know More Math. Sure it helps. Other, less prestigious, things also help but you don't hear people griping about it. TDD allows the idiots in. Yes, that's effectively why you want TDD, you want to get more mileage and solving more complex or more bug-sensitive problems using the same people. Building software is not about being smart (although that helps on occasions), it's about getting stuff done.

Yes, machine learning and AI are the new kids on the block, and like Web programming, they will see a bloom of increased customer demand, and like Web programming, we'll get a progression from bespoke boutique software to frameworks that make people's lives easier to frameworks that allow any person with the intelligence of a pet rock to do simple stuff productively. Why is that? Because building frameworks is the only way that the smartest people can earn money faster than programming the (N+1)th variation on that theme everyone follows -- frameworks are what make people more productive, or allow you to use a workforce that's more accessible.

As a Wizard With a Pointy Head (aka academic), I'd say that the need for Wizards With Pointy Heads in production work is often overestimated and/or idealized. There is a large number of PhD graduates, and the market happily gobbles them up (indeed, realizing that you can hire PhDs and have them do productive work is one of the things that made Google successful as a company back in the early 2000s).

The point is not (merely) the technology behind it, it's the agencies' model of selling the customers "rich internet sites" whose main point is to show off to other people that you have pockets full of money.

A good intro can help to sell your site project to a customer (and help them selling it to their superiors), but no longtime user would protest if you took it away.

Single page apps are a way to fit in an interaction model that doesn't fit the idea of "one address, one page, one text" that was the idea behind the Web in the first place. So, yes, you give up some of the benefits of the Web (navigation with bookmarks, back button) or have to fight hard to get them back, while keeping others (mostly, platform independence and a built-in client-server architecture)

The point is: most of these applications (say, Photoshop, Gimp, Eclipse, OpenOffice, probably others) are things that you load once to work with them for a couple hours, and don't mind when they consume a lot of your system resources.

If you look at web applications, Google's music app ("My Music" in Google Play) uses a load screen (complete with progress bar) because it's seriously bloated, while Amazon's Cloudplayer just loads and does its thing (which is, playing music).

Needing a load screen is a sign that your app is not, or aims at not being, a Web app anymore - in the sense that you'd also want to loosen the other restrictions - give it its own window, possibly have local file storage or FS caching, etc.; In that case, something like TideSDK ( http://www.tidesdk.org/ ) may come in handy to deliver a desktop app to power users that is based on the same codebase. And, maybe you'd still want a web app that is functional and lean. (As an example for two apps with non-identical possibilities, Google's My Music allows you to change song metadata on desktop browsers but not in the Android app - yet another sign that My Music is meant as a desktop-y app).

In many cases it can be a sign of hubris or lazyness that you cannot get rid of the bloat. Be real and think whether your users actually enjoy waiting that time, since they'll be doing it more often, on machines with less resources and/or slower connection, than yours.

IEEE may be more prone to precision errors (letting bad papers in) while NIPS may be prone to recall errors (throwing good papers out). With the way reviewing is done (no one can take a week off to read and fully comprehend the four papers they are given) you cannot achieve perfect separation - even if that were possible.

If you look at Nordstrom's interviewing site (linked in the article), you see that they require people to have some technical background but also general communication skills. All the talk about "t-shaped" people boils down to the fact that in their line of work, understanding requirements (including non-technical descriptions given by a customer) is a non-negotiable requirement of the job, whereas purely technical excellence isn't as useful to them.

In that sense, having good communication skills and some programming experience, on top of being good at "getting stuff done" is better than what they'd get out of an average CS graduate. Put in other terms, current CS curricula don't serve these companies well, because they yield people who are deeply technical, but bring neither good communication skills nor necessarily familiarity with the tools used by them.

Stopping a course because the instructor behaved in inappropriate ways outside the classroom (which is a fuzzy notion with EdX courses) is one thing, which would be an undrstandable step for MIT to defend their reputation.

Taking down existing video content, which is only incidentally connected to the problematic behaviour (Lewin seems to have his role as the course instructor rather than the videos as an excuse to make contact with course participants), to me would be more akin to burning the books written by an unliked person than to any act legitimately defending MITs or the course participants' interests.

That seems like the difference between object-based (early VB) and object-oriented.

XSLT is referentially transparent (no setf for you) but withholds from you most of the goodies that people take for granted in functional or logic programming.

You could see it was written by well-intentioned FP enthusiasts. IMO the best alternative at the time when XSLT was developed would have been XMLPerl - embedding an imperative language in something that deals with the XML-specific parts appropriately. But Perl was never enterprisey enough, and XmlPerl died a painless death.

I know DrRacket from back when it was called DrScheme and you had to manually set it from "extremely limited scheme-like language that won't allow you to do anything" to "actual R5RS scheme".

I found DrScheme and its cousin DrJava nice for beginners, but I think within the context of Python, IPython Notebook is just so much better because you can wire in some pretty pictures and formatting. Then again, the separation between "code that is meant to stay" and the REPL may be a useful concept that you want to teach beginners.

Kind of like spider? Or Emacs? Or PyDev?

Hi, I'm a computational linguist and I would find it really great if some people could share their traces of their typing/editing process.

In programming, we have editors that have strong support for writing because we know exactly what the semantics of code is, and what good operations for editing/refactoring are. With writing prose, the best we currently have is edit histories from wikipedia articles, which are on a much larger timescale and full of things that should not be part of the editing process (vandalism, NPOV wars, etc.)

From the other side of the curtain, most companies have "salary bands" that they apply semi-uniformly, i.e., as a "junior developer" you can make from X1-10% to X1+10%, as a "senior developer" you can make from X2-10% to X2+10%, and as a "project manager" you can make from X3-10% to X3+10%.

The invisible hand of the market will lead prople to companies with the highest value for X2, which are usually also those where the productivity (i.e. contribution to company revenue) per person is highest.

Much research in corpus linguistics or computational linguistics occurs on data that naturally occurred. It's part of the traditions in psychology to manipulate your subjects because that's the only way you can get more direct observation of suspected cause/effect relations, especially when those are subtle enough to go under in the noise in "in the wild" data.

Given a big enough sample size and a multi-factorial model (i.e. a model that models all confounding factors as well as the variable you're interested in), the importance of manipulation should decrease, and many astronomers, economists, or computational linguists would be able to draw meaningful inferences in such a study. Psychologists may have gotten too used to the idea of "get experimental subjects, present with manipulated input, do comparison statistics" to consider a purely observational study as the default.

tl;dr: It's not just astronomers, many fields do empirical studies without manipulating their subjects in a Skinner-ian fashion.