cfcef

Given the Baidu work, I think we can safely say that Hinton et al's forecasts 4 years ago were on the money. Deep approaches are now clearly dominant and have yielded fantastic performance.

A 'further reading' is not a reference; 'further reading' means 'here are additional sources you might find relevant about material not covered here'. Which of 'Triumph of the City', 'The Ghost Map', and 'Guns, Germs, and Steel' did the stuff about animals come from, if you hadn't already read them? That's right - you don't know! Because he's not giving a reference.

The post reads like a (very poorly written) rehash and plagiarism of Diamond...

'Prosper'!='exist'. He claimed they existed. They did not, barring ancient people being some sort of bizarre near-super-human race which can laugh off protein and iodine deficiencies and parasite loads and early-childhood infections in a way modern people can only dream of.

This is the same problem you see when people solemnly pontificate about how many Einsteins are trapped in Africa/India and if only we would fund One Laptop Per Child we could unlock their potential...

They probably also benefited from low-hanging fruit. No one today has the opportunity to invent 'logic' or 'the Pythagorean theorem' or 'the tragedy'.

You can see underwater, but you don't see as well as they do:

> The kids had to dive underwater and place their heads onto a panel. From there they could see a card displaying either vertical or horizontal lines. Once they had stared at the card, they came back to the surface to report which direction the lines travelled. Each time they dived down, the lines would get thinner, making the task harder. It turned out that the Moken children were able to see twice as well as European children who performed the same experiment at a later date.

> Human intelligence has remained approximately the same for 50,000 years. The ancient world had its geniuses at the same rate as the modern world.

No, they didn't. Human genetic intelligence may be the same (although this is doubtful because as ancient genomes slowly become available for analysis, we see ever more signs of huge numbers of frequencies changing in soft selection sweeps when we go back only a few thousand years in Europe, so 50k years...?), but the environments are not nearly the same. The ancient world was absolutely grindingly dirt-poor compared to the modern world, and the negative environmental accordingly huge. (Even things like sanitation may not have made a difference: http://www.theatlantic.com/health/archive/2016/01/ancient-ro... )

The most comparable places to the ancient world right now would be somewhere like subsaharan Africa, where between the subsistence agriculture, parasites, poverty, and whatnot, despite the benefits of widespread literacy and vaccines, the average IQ is still quite low, somewhere around IQ 80, or at least 1 standard deviation below the West; with genius at a cutoff of IQ 140 or so, that implies a rate of geniuses much less than 1/8th the Western rate.

So no, the rate isn't going to be nearly the same. I would note that it's probably not an accident that when we think of geniuses of antiquity, we tend to think of people drawn from the urban elite of the capital city of empires at their peak (eg Athens, Rome)...

> The level of sophistication of science in medieval Europe may be way ahead of what we assume (have we even tried to measure it?).

I think he's not gesturing towards medieval Europe, but Rome. There was one empire which was extremely interested in the Mediterranean, the Black Sea, and England, and was bound together by ship, put a lot of effort into improving travel and logistics and measuring distances, much of which's science & math has been lost (leading to regular surprises like the Archimedes palimpsests and the Antikythera mechanism), whose maps could have lasted 800 years or so to be rediscovered by medieval Italians.

I wonder about the legal aspects. The reuse in one set of cases may be legal, but what about the others? If nothing else, you would think that the NYT or USA Today would have a clause in their contracts that puzzles are guaranteed new and unique...

And that expansion was done for the revenue it could deliver now, not in 2108. If the original Suez Canal had not been constructed in 1869, then this new one could still have been constructed now and it would be for the short-run gains, not gains a century-plus away.

> The only reason that things like canals get built is because people can charge money to recoup their costs.

In the case of the Suez Canal, though, it wasn't finished in 1869 for the income it could deliver in 2016, since discounted at 7%, even the headline figure of $350000 would've been worth in 1869 ~$17. (Which is just as well for the French investors in 1869, since they and the later UK investors would eventually be expropriated by Nasser in 1956 and would see nothing of any revenue after that.)

The Ainu barely exist anymore, and the whole thing about the Burakumin is that they aren't visible and all their descendants hide any connection. A better example would be the Korean descendant population, but that's still a tiny fraction of the population. (I think 1 million out of 130 million or <1%?) So there's no good comparison in Japan to the ~17% African-American population.

It's always 'climate change'. Dozens of megafauna disappear? 'climate change'. Whole tribes with fortifications suddenly disappear? 'climate change'. Wholesale population replacement? 'climate change'. It's our age's "peoples don't migrate, pots do". It's gotten to the point that when I read about the latest population genetics result show introgression or replacement and they invoke 'climate change', I no longer know if it's a case where it could be climate change or just the usual total refusal to speculate about pre-historic warfare and predation.

Sounds like 'all public Google+ photo albums'. (I say public because given what Facebook reports about daily photo uploads, a cumulative 126m seems like it would be way too small for G+ if it covered all Google Docs/Photos uploads including private ones.)

The point about ECC and accepting lower error rates makes a lot of sense. It's the end-to-end principle again: since all of these HDDs are going into a global pool in which each of them is disposable and written content is protected by FEC spread across multiple drives, there is no need for each drive to spend a lot of resources, going deep into diminishing returns, trying to make itself as resilient as possible. If a network transmission fails, it is retried and doesn't need to be swathed in huge numbers of elaborate checksums and ultra-reliable links; if a hard drive fails, the content is recovered from the FEC and re-written out to a new drive.

Some previous submissions: http://www.psmag.com/health-and-behavior/why-is-there-no-gen... http://www.psmag.com/health-and-behavior/why-is-there-no-gen...

When I was younger, I didn't understand the appeal of https://en.wikipedia.org/wiki/Georgism 's https://en.wikipedia.org/wiki/Land_value_tax ; proponents came off as cranks. But the more I learn about California and the Bay Area and watch the rents (in both senses of the word) rise, the more I think there may be something to it.

Ditto. I signed up for the free Prime offer, after a while decided it wasn't going to be worthwhile for me (a lot of the total delay between ordering and receiving seems to be on Amazon's warehouses' end of things, and not USPS/UPS physically moving it to my house), canceled, and kept using it for the rest of the month without problem.

> I know it because I grew up in one.

You did not grow up in North Korea. North Korea is not the USSR, and it operates differently. I laid out several ways in which it differed in explaining why being a tourist is only harmful.

> Everybody can be a tourist and do their part, though.

You know what part you can do which is even more helpful? Donate to malaria bed nets, or donate to the missionaries and other NGOs.

> And about the money — as a tourist, you'll hardly help the evil regime with more than $1000-$2000 (do you know how much a rocket costs? This is a small change).

And how 'small change' is maybe contacting a regular North Korean and maybe changing their view about something and this someday maybe having an effect?

Every dollar of hard foreign currency counts for a tiny impoverished country under sanctions. The NK economy is small and stagnant and regularly straining under the burden of the rocket & nuclear programs.

The NK regime believes that the tourism is very useful and effective, and it is not endangered in the least by the prospect of toilets with Bibles behind them, or the chance a tourist will see through 'a crack', and I agree with them.

> North Korean government is not doing it for your money.

They are absolutely doing it for the money. Just like they were doing Kaesong for the money, not to 'look good' based on some obsolete analogy to the USSR.

They feed westerners a propaganda machine to fill the time and provide a trip. You need paying customers for the hard cash, but you can hardly let them wander around in the general population. The propaganda is there, might as well use it. And it helps with domestic propaganda too: the spin put on the aid and payments made to NK, and trips to NK by high profile politicians like Bill Clinton, is the East Asian motif of tribute. What better tribute to the glories of the NK regime is showing your people how foreigners come from around the world to visit the shrines of the Kims?

That's possible. It's using Andrej Karpathy's char-rnn. Presumably it's doing something like running the trained model with 'th sample.lua -model something.t7 -primetext '$WORD, ' and taking everything after the comma as the definition. So to reverse this, you would take the dictionary corpus, remove '^$WORD, ', and suffix ', $WORD$'. Then it will be training to predict a final word conditional on the definitions, and you can do the same thing with the new model, feed in '-primetext '$DEFINITION, ' to get out a word.

That's absurd. We are talking about a totalitarian regime which does not care in the least what the West thinks, which carefully ushers tourists who feed it the hard foreign currency it needs through a well-honed propaganda machine (read the blog posts of people doing the tour - they all get bused to the same damn statues and propaganda museums and places) surrounded only by hand-picked minders & chaperones chosen for their loyalty with no contact with the general population, with most tours in a city, Pyongyang, whose populations is also chosen for loyalty to the regime. Try to go 'off script' and you'll see how they treat tourists. Thinking you'll change anyone's mind is as sad and pathetic as the people who go there and drop a Bible behind a toilet and think they have any chance of making a difference.

Real evidence of the world from the other side of the curtain is provided by Chinese & Korean smugglers smuggling in K-dramas and hard drives stuffed full of forbidden foreign media and balloon drops from missionary networks.

You aren't doing jack by going to NK as a tourist. You're propping up one of the most evil regimes with your money - no ifs, ands, or buts. At least when people waste money climbing Mount Everest, they're only endangering themselves.

I'm sure one could: https://news.ycombinator.com/item?id=11127534

But rolling out a paid service like that comes with a lot of expectations that being part of a free app doesn't.

You're being paranoid for several reasons. First, you will presumably be keeping the classifications you get from the API (if for no other reason than $2/1000 can really add up and why would you do that to yourself?). Second, Google has historically always had very generous end-of-life announcements. Third, there are already a number of competitors, including the trained CNNs Google has already released. Fourth, even if there were no competitors nor pre-trained models, deep learning is increasingly accessible and you could learn to imitate a past ImageNet-winning CNN in Torch/Theano/Tensflow/etc and train it within a few weeks. Fifth, paid Google services tend to hang around longer. Sixth, machine learning is a major part of Google and they are increasingly rolling it out to their services and may well be using this API themselves, making closing it not such a great idea.

So you're passing up what could be substantial benefits, Google isn't going to close it anytime soon, if they do you will have months of warning, and can easily replace it with a competitor or your own.

If you used Baidu's services, you might be more impressed. In any case, one barrier to rolling these things out to production is that it takes time to get them into production, and there are cost challenges: you represent too little ad revenue to justify a whole GPU devoted to you. Baidu has been experimenting with aggressive model reduction and batching up tasks and efficient GPU server design, but it's still a challenge to roll out the best deep learning results on fractions of a penny budgets.

Of course, Lem tried to top it with _Star Maker_, where multiple universes are too small for his canvas :)

So would OP be the 'fractal sundial'?

There are occasional torrents, AFAIK. Hope you have 10tb of free space handy.

Arxiv is a voluntary service where authors upload their own papers. It covers a few hundred thousand papers, probably, since authors have to know about it, want to use it, and have copyright service; it covers only a few areas of science where Arxiv use is in vogue, and only from the past decade or two.

Whereas, SH/Libgen acquire copies of tens of millions of papers from everyone everywhere everywhen.

It's the difference between a local library and the Library of Congress.