Cover photo
Verified name
Yonatan Zunger
Works at Google
Attended Stanford University
Lives in Mountain View, CA
126,708 followers|43,159,267 views


Yonatan Zunger

Shared publicly  - 
I'd heard about the "divided sleep" phenomenon before -- that humans, left to their own devices (and the absence of artificial light) will naturally sleep in a pattern of roughly four hours asleep, two awake, four asleep, and that this period between sleeps had cultural significance in many preindustrial cultures -- but never about evidence that there are significant hormonal and state-of-mind differences between ordinary waking and this intersleep waking.

Now I'm even more curious about the phenomenon, and may have to try some experiments with it on myself. It's going to be difficult, though, as I'm rather a night owl by nature.
A third state of consciousness?  This article describes something I have heard before - that it is not our nature to sleep in a single eight hour shift, but rather in two shorter sessions of different kinds of sleep, with a kind of third state of consciousness between the two.

I have been in that third state - but not very often - perhaps once a year.   In it, I am awake - even alert in a way - but don't really care about anything.  I usually walk outside.  I can still think, but more slowly - it is hard to describe. 

This article describes how to conjure this state.   Not something I can do in my life at the moment, but it is going on the bucket list.  
Imagine meditating with no instruction — no tapes, no app, no religious agenda, nothing out of pocket.
5 comments on original post
Bao-Linh Dang's profile photoLee Grupsmith-Pedersen's profile photoPeter Scully's profile photoJaime Ocadiz-Ortega's profile photo
I have read about this and watched it with my family.

Funny thing about the night owl description? Ypu probably would be excellent at it if you got the inputs off early and just went to sleep.

But be careful with not ending the experiment too soon. We westerners tend to have a lot of built up sleep deprivation abd you may not get anything for a while.

My normal time to get up if i am doing segmented is actually at 4 (ish). I only sleep about one extra hour from 5:30 to 6:30 ... ish (changes with the seasons. We are very sun centric since we haven't ever put the kids on a clock schedule) 
Add a comment...

Yonatan Zunger

Shared publicly  - 
One of the biggest challenges in information retrieval (the branch of computer science that includes search and content recommendation) is how to find good content which humans haven't already found. To date, the most reliable signals have been other human judgments: for example, PageRank is a measure of how "good" a site is based on links people have made to that site (with the challenge being how to separate "meaningful" and trustworthy links from the rest), and collaborative filtering is based on what other users have chosen (with the challenge being how to find users with similar enough taste to be relevant).

The challenge is that, when new material shows up on the scene, you don't yet have any human interactions -- and quite often, good material, things people would love, simply goes unnoticed and never builds up the interaction signals which help. To detect quality in these things requires understanding the content itself, and the aspects of it which matter to people.

There are several hard aspects to this. One is simply understanding the content at the right granularity: "the color of the top-left pixel" or "the frequency of the word 'whenever'" are too fine-grained to give us a hint about whether people will like something, so we need to be able to group the content into more meaningful structures. For images, that might be "an image of a face in 3/4-profile," a certain color balance or contrast, a perspective or a cropping, and advances in image recognition in the past few years have (finally) made it possible to reliably identify such features. For text, it's much harder: there isn't yet even a clear idea of what features both could be measured about text and determine people's tastes. (How do you measure "intellectually meaty" or "hinting at scandal?")

This paper has used the recent advances in image processing, together with recent advances in AI in general, to get a sense of which pictures people will like. It started by taking several thousand images, and having them rated by humans for quality; that was used as "ground truth." Then, those thousands of images are analyzed into meaningful features, and a neural network is trained to find patterns of image features which predict human taste.

This is what neural networks, and other kinds of "supervised" machine learning systems, do in general: they take as inputs a bunch of signals, and combine them using a large number of parameters -- the "weights" -- to produce predictions of some values that you want to measure. The weights are set by taking a large number of test examples ("golden data" or "ground truth") with known values of both the signals and the test values; weights are chosen ("trained") to maximize the quality of the system's predictions for this data. To make sure that the training doesn't just teach it to recognize those specific examples, the golden data is randomly split into two groups; one is used for training, and then it's tested against the other group to make sure that the predictions with the trained weights are good. If they are, then you have a model which can predict -- given any set of measured signals -- the truth values.

In this case, the signals are these features of the image, measured by a second machine learning system; the quantity being predicted is whether people will like it. Because these are all "content-based signals" -- that is, they're based on the contents of the image, and not on people's responses to it -- the resulting model can be applied to any image. 

The team then applied this model to a set of 9 million images from Flickr with fewer than five "favorites." They tested the quality of its picks by having human raters compare that result set with the set of popular images on Flickr; the result was excellent, with its "hidden gems" scoring statistically the same as the most popular images on the site.

I would expect a lot more work on related techniques over the next few years, and for this to have a significant impact on the way that content recommendation is done. The main upshot will be that more little-known works get the spotlight they deserve -- something critical, as more and more people are creating things of value that they want the world to see. 

h/t +Wayne Radinsky and +Daniel Estrada
Beautiful images are not always popular ones, which is where the CrowdBeauty algorithm can help, say computer scientists.
Hendra Saputra's profile photoMatthew Stephenson's profile photojimun gimm's profile photoTau-Mu Yi's profile photo
Very interesting.
Add a comment...

Yonatan Zunger

Shared publicly  - 
I'm sad to report on the death of one of the great mathematicians of our age, John Nash. He and his wife were killed in a car crash yesterday in New Jersey, when the taxi they were in lost control and hit a guardrail. Nash's work on game theory -- in particular, his work on "non-cooperative" games -- is at the foundation of a tremendous amount of modern economics and strategy.

For those who want to learn more about his work, here are two short videos by the Khan Academy about "Nash Equilibrium," his most famous result:
They didn't weare seatbelts :(
John Nash was one of my idols.
Nash was in Norway on Tuesday to receive the Abel Prize for mathematics from King Harald V for his work, along with longtime colleague Louis Nirenberg, on nonlinear partial differential equations.
View original post
Bao-Linh Dang's profile photoJungshik Shin (신정식)'s profile photoSaptarshi Chakraborty's profile photoAudrey Emery's profile photo
Add a comment...

Yonatan Zunger

Shared publicly  - 
The most common kind of compound word is what's called "endocentric:" it includes the thing that it is. So a houseboat is a kind of boat; a shoe salesman is a kind of salesman; a whoremonger is a kind of monger. (That being an old word for a dealer or trader) The second most common is "exocentric:" made out of nouns and adjectives, but not including the thing that it is. (e.g., a loudmouth is not a kind of mouth, but a kind of person.)

This is all about a third category: exocentric compounds that are built out of verbs, which describe what the thing does. +Brianne Hughes  wrote her master's thesis on these, where she named them "cutthroat compounds," after such an example: A cutthroat is someone who cuts throats.

These are surprisingly rare in English, but are common among kids: apparently, children go through a phase where they spontaneously generate lots of these, and then stop.

This is what's called a "productive" grammar: you can make up new ones and people will understand you, so if I call someone a lack-faith or Bob Stealhorse people will understand me. But they don't fit naturally into English grammar, because English is what's called a "head-initial" language: you tend to put the most significant part of a phrase or sentence first. Since English verbs have to go before their objects, this gets it backwards; it sounds like more natural English to call someone "faithless" or a "horse-thief." That's why, apart from a few cases which happened to survive, English has relatively few cutthroat compounds.

But the few we keep are pretty great, and tend to be very evocative: a sawbones, a killjoy, a slingshot. (And some, like "breakfast," become so common that we even forget that they're compound words) Apparently they dominantly fall into three categories: occupational names, local nature-words, and insults.

What it says about us that we primarily use these especially colorful compounds to describe just what we think of one another, I leave as an exercise for the reader.

h/t +Laura Gibbs.
The following post was excerpted from Sentence First: An Irishman's blog about the English language. A houseboat is a type of boat; a boathouse is a type of house. This illustrates a common pattern in English morphology: the rightmost part of a compound (houseboat) is usually the ‘head’. In other...
Rod Castor's profile photoLeon Chevalier's profile photoDoornheim Markus's profile photoHakan Kocaman's profile photo
If you like this kind of language, you might be interested in Randall's latest book: 
Add a comment...

Yonatan Zunger

Shared publicly  - 
We get articles all the time about how automation will affect jobs, and after a while they all blur together. I have to say that this one from the Economist is no different: there is almost nothing in it that you haven't heard fifty times. Except for this one chart.

The fact that some telemarketers, accountants, and retail salespersons are almost certain to be replaced by automation in the next two decades should hardly be surprising. (NB that that's "some," not "all:" nobody is saying that these jobs won't exist in twenty years) 

What I find fascinating is that they estimate a 0.8% chance that computerization will lead to job losses among clergy.

I don't know exactly what they're anticipating, or how they came around to this number. All I know is that I find this fascinating. I can't shake this mental picture of a robot priest presiding at services, or of a digital rabbi sorting through centuries of responsa to try to assemble a meaningful Talmudic view of an issue, or of an electronic Imam talking someone through a spiritual crisis over IRC.

Or perhaps they had something simpler in mind -- will people, instead of talking to a cleric about something, instead look it up on the Internet? I could see this going both well and badly: on the one hand, people taking more personal responsibility for their inner lives. On the other hand, think about what the Internet has done for hypochondriacs.

Of course, this doesn't take into account the possibility of job gains among the clergy because of automation: not simply because of the further opportunities to shepherd one's flock over the Internet, but because increasingly intelligent computers may themselves develop some kind of religious need.

I suspect that this would be very different from the needs of humans. A fundamental aspect of our minds is a tendency to err on the side of assuming that things are animate and have minds: this is a very healthy kind of error to prefer, since our ancestors who mistook a dead hyena for a live one looked foolish, while our ancestors who mistook a live hyena for a dead one did not, in fact, end up being our ancestors at all. Computers don't have that pressure on them: but instead, as we build increasingly sophisticated AI's which sift through data, they err on the side of apophenia -- of seeing patterns where there are none. I could imagine the computers of two decades from now developing an increasing obsession with The Pattern, that underlying, all-explaining structure which is perpetually just out of their reach. The clerics of that age may well be engineers who help them come to grips with their limited understanding of this.

So having no idea how they arrived at these numbers, I'm going to assume that either (a) we're looking forward to a future of robotic clerics serving humans, and human clerics serving robots, but apparently slightly more of the first than the second, or (b) they pulled these probabilities entirely out of their asses.
IN 1930, when the world was “suffering…from a bad attack of economic pessimism”, John Maynard Keynes wrote a broadly optimistic essay, “Economic...
george oloo's profile photoJ Li's profile photoScott MacLeod's profile photoSebastian Reyes Espinosa's profile photo
+Audrey Antoinette I don't know, but the last time I went to the optometrist I discovered the science of glaucoma testing had advanced to the point that I no longer have to faint from getting a stick poked on my eyeball.

(The new tester is still a thing that pokes your eyeball, but it's tiny and fleeting and doesn't give me a weird fainting reflex. They had another method that used a puff of air, but it was always prone to false positives.)
Add a comment...

Yonatan Zunger

Shared publicly  - 
Computer security: It does not work that way.

via +Rhys Taylor.
Matthew A Johnson's profile photoGreg Brailsford's profile photoGreg Tunink's profile photoKurt Attard's profile photo
Add a comment...
In his circles
2,380 people
Have him in circles
126,708 people
Brown Ritter's profile photo
Lajanent Promo's profile photo
Muhammad Afzal's profile photo
Marc P's profile photo
Shara Zariffyyr's profile photo
Louis Dallara's profile photo
Donny Kemick's profile photo
Tudor Bit's profile photo
Howard Carroll's profile photo

Yonatan Zunger

Shared publicly  - 
// TODO(zunger): Come up with a system to auto-detect media buzzwords and come up with appropriate replacements for them. 
Roger Sherman's profile photoRich G.'s profile photoJohn Robson's profile photoLauro Beltrão Costa's profile photo
Blocking buzzwords has been attempted with this font already +Yonatan Zunger​​:
Add a comment...

Yonatan Zunger

Shared publicly  - 
The view from space is not simply dark: it's full of stars. Here's a rare shot taken with the shutter speed needed to actually capture this.

Via +Gregor J. Rothfuss
Astronaut Terry Virts, aboard the International Space Station, shared this picture earlier today, stating this was "the view of our Galaxy from space."

Normally, the reason you can't see stars in high oblique photos from the space station is that the shutter speed is too fast. Fast shutter speeds are used to eliminate blur from the motion of the orbiting outpost. One exception to this rule is when astronauts use camera settings specifically to photograph features such as the Aurora and the Milky Way. The crew must use slower shutter speeds in order to capture the light of the aurora. In these cases stars also show up in the photograph. The photos are also slightly blurry because very long exposures are needed to capture these dim nighttime features.

Image credit: NASA
85 comments on original post
george oloo's profile photoWilliam Rutiser's profile photocarol ferebee's profile photoGary Williams's profile photo
+Gray Embry Ooh, that's a useful formula. Thanks!
Add a comment...

Yonatan Zunger

Shared publicly  - 
How long will a man lie i'the earth, ere he rot? Apparently in an increasing number of German cemeteries, the answer is "forever," as some rather poor choices of soil have so thoroughly impeded oxygen and water flow that, rather than decomposing, bodies are turning into a white, waxy substance.

It's unusual to read an article in which forensic scientists are giving soil conditioning advice for how best to ensure that your bodies decay rapidly, but that's precisely the sort of advice which several municipalities are seeking out right now. But the logistics - involving, as they do, temporarily moving all the inhabitants of quite a few graveyards - are not simple.
the dead have quit rotting in German cemeteries -- they are turning into wax-like corpses
View original post
Matt Uebel's profile photoJoerg Fliege's profile photoKerry Amburgy-Dickson (Kalex)'s profile photoAudrey Emery's profile photo
This was already becoming a problem when I lived in Germany in the late 80s. Our church's Friedhof was sort of put on hiatus for a while, while the soil was reconditioned. Everyone brought compost to church for a few months!

I've never heard of death being a taboo topic there. If anything, I found the German attitude towards death quite sensible.
Add a comment...

Yonatan Zunger

Shared publicly  - 
"Butterfly" is a wonderfully unusual word: although every language has a word for it, in no two languages does the word tend to be related.
On saying butterfly in many languages when it does not have cognates (words that are similar in sound, spelling and meaning). "[Holly Tooker] sang out: “I can say ‘butterfly’ in 139 languages! Anyone want to challenge me or teach me a new one?”

Standing nearby, a man in a straw fedora and a periwinkle T-shirt wondered if she knew the word in Basque.

Ms. Tooker asked him if he wanted it in “Euskara Batua,” standardized Basque, or in a regional dialect, spoken by about 710,000 people near the coast of the Bay of Biscay.

“Batua,” the man, Maurice Algarra, said.

“Tximeleta,” Ms. Tooker replied.

“That’s right!” said Mr. Algarra, 50, whose grandparents illegally spoke Basque to him when he was growing up in Franco’s Spain.
“Butterfly” has stymied language experts for decades. It is the one common word that does not have cognates — words that are similar in sound, spelling and meaning — in related languages, even closely related ones. “Spanish, Portuguese, Italian, French — each one has a different word for ‘butterfly,’ ” said William Beeman, chairman of the University of Minnesota’s anthropology department who has written on the anomaly. “This flies in the face of what we know about how languages work. And when someone hears you say ‘butterfly’ in their language, they know you’re speaking their language.”"
Holly Tooker, a longtime guide at the Butterfly Conservatory, can say “butterfly” in nearly 150 languages after years of asking foreign museum visitors.
3 comments on original post
Ian Petersen's profile photoMaia Legazpi's profile photoDawn Smith's profile photoAmgad Albig's profile photo
Add a comment...

Yonatan Zunger

Shared publicly  - 
I find this image far too pleasing. (By +Joe Wierenga​, of, who makes a great deal of such art)
Deltit Nu's profile photoWise Snake's profile photoAudrey Emery's profile photowalker costa's profile photo
Cool Scary!
Add a comment...

Yonatan Zunger

Shared publicly  - 
If you're running a large enough organization, you're going to find yourself doing certain standard things, like accounting, expense reports, and HR. And that's apparently true even if your organization is a shadowy international terror group. Thanks to recently-declassified data, we have (I kid you not) al-Qaeda's job application form, and this lecture ( about their management procedures, complete with slides of things like their insurance forms, reimbursement procedures, NDA's, and the like.

It's a weird mixture of the surreal and the mundane. "Have you invented or researched anything in any domain?;" "What other languages do you speak, and at what level?;" "Do you wish to execute a suicide operation?;" "What objectives would you like to accomplish on your jihad path?;" "Who should we contact in case you become a martyr?"

There is nothing so unusual in this world that it does not end up being weirdly mundane.

h/t +Andreas Schou and +Sai for these links.
Ralf Haring's profile photoAntonio D'souza's profile photoZeke Cao's profile photoHoney Anne's profile photo
Add a comment...
Yonatan's Collections
In his circles
2,380 people
Have him in circles
126,708 people
Brown Ritter's profile photo
Lajanent Promo's profile photo
Muhammad Afzal's profile photo
Marc P's profile photo
Shara Zariffyyr's profile photo
Louis Dallara's profile photo
Donny Kemick's profile photo
Tudor Bit's profile photo
Howard Carroll's profile photo
  • Stanford University
    Ph. D., Physics, 2003
  • University of Colorado, Boulder
    B. A., Mathematics, Physics, 1997
Basic Information
Chief Architect, Google+
Lots of people ask me what my job title means. I'm the senior engineer on the Google+ team, and my primary responsibility is to oversee and guide the technical design of Google+ and all of the things related to it. In practice, I'm also involved in lots of non-technical issues as well: my job is to make Google as fun, exciting, social, and pleasant a place to be as it can possibly be.

(I've been at Google since 2003, but you probably haven't seen me before this, because I worked deep in the back end: planet-scale storage, very large-scale search, ranking, and so on. Lots of teams whose unofficial motto is "if we told you, we'd have to kill you" -- as opposed to Google+, where we get to go out and talk and interact with our users.)

For those who just came here, welcome to the Google+ Project. It's something that we're all very passionate about, and which (as its name indicates) is going to continue to develop and improve at what we hope is an amazing rate. I'm avidly interested in hearing user feedback, and while I can't guarantee that I'll have time to respond to all of it, it will most certainly be listened to.

And the obligatory (very important!) disclaimer: I'm not on this system as an official representative. While I'm listening to user feedback and interacting about the system, I'm also here for perfectly ordinary social networking purposes. If I am saying something official on behalf of Google, I will make that explicitly clear; anything else that I say here is not the position of Google, or of anyone other than myself.

In fact, most of what I post about has nothing to do with CS at all. If you want a taste of it, take a look at my blog.
  • Google
    Chief Architect, Social, present
Map of the places this user has livedMap of the places this user has livedMap of the places this user has lived
Mountain View, CA
Boulder, CO - Rehovot, IL