Start Up No.1,008: YouTube’s paedophile problem, Nest’s secret mic, picking real people, Samsung folds, and more

Machine learning predicted lots about film popularity – but this one surprised it. CC-licensed photo by AntMan3001 on Flickr

»You can sign up to receive each day’s Start Up post by email (arriving at about 0700GMT each weekday). You’ll need to click a confirmation link, so no spam.«

A selection of 11 links for you. Use them wisely. I’m @charlesarthur on Twitter. Observations and links welcome.

Pinterest blocks vaccination searches in move to control the conversation • WSJ

Robert McMillan and Daniela Hernandez:


Pinterest has stopped returning search results for terms relating to vaccinations, a drastic step the social-media company says is aimed at curbing the spread of misinformation but one that demonstrates the power tech companies can exert to control the conversation around hot-button issues.

Most shared images on Pinterest relating to vaccination cautioned against it, contradicting established medical guidelines and research showing that vaccines are safe, Pinterest said. The image-searching platform tried to remove the antivaccination content, a Pinterest spokeswoman said, but has been unable to remove it completely.

Pinterest described the search ban—which the company hasn’t previously publicly discussed but went into effect late last year—as a temporary but necessary measure until it can develop better strategies to sift through what it calls “polluted” content. The company made a similar decision last year to block searches for dubious cancer therapies.

Users can still pin vaccine-related images to their online boards, which could lead to suggestions to more similar content, but the posts no longer show up in searches. “It’s better not to serve those results than to lead people down what is like a recommendation rabbit hole,” said Ifeoma Ozoma, Pinterest’s public policy and social impact manager.


They’re being responsible. It’s weird. Quite a contrast.
link to this extract

Samsung’s foldable phone is the Galaxy Fold, price $1,980 • The Verge

Tom Warren:


Samsung has built a sturdy backbone to the device, with a hinge system that has multiple interlocking gears. All of these gears are hidden at the rear of the device, and allow the Galaxy Fold to transform from tablet to phone modes. At the rear of the device there’s also a triple-camera system that will be used for both tablet and phone modes. There’s a 16-megapixel ultra-wide camera, alongside 12-megapixel wide-angle and telephoto cameras at the rear, and a 10-megapixel cover camera for selfies. Samsung is also creating four different colors for the Galaxy Fold, but it’s the main tablet display that’s key here.

Samsung is allowing the Galaxy Fold to run three apps at once on this Android device, and it’s using an app continuity system to adjust these apps when you move between tablet and phone modes. Apps like WhatsApp, Microsoft Office, and YouTube have all been optimized for the new display and modes, and Samsung has been working with Google to ensure Android 9 Pie fully supports this display.

Samsung demonstrated a variety of apps running in this mode, and the switching from phone to tablet and vice versa. It looks rather smooth in the software right now, but it’s fair to say that the Galaxy Fold looks far better when it’s folded out than being used as a traditional phone. The phone display is clearly designed to be used with one hand, but it’s flanked by large bezels that aren’t found on the tablet mode. We’ll need to get a closer look at the Galaxy Fold to find out exactly how this impacts the device usability, though.


Sooo.. an iPad you can fold up?
link to this extract

Oracle claims a fighter of pirated apps is a front for ad fraud • Ad Age

Garett Sloane:


A company that claims to combat app piracy is a pirate itself, according to a report Oracle released on Wednesday. Oracle claims the company, Tapcore, has been perpetrating a massive ad fraud on Android devices by infecting apps with software that ring up fake ad impressions and drain people’s data.

Based in The Netherlands, Tapcore works with developers to identify when apps are pirated and then enables developers to make money from those bootleg copies by serving ads. Oracle says that Tapcore’s anti-piracy code was a Trojan horse that was generating fake mobile websites to trick ad serving platforms into paying them for non-existent ad inventory.

“The code is delivering a steady stream of invisible video ads and spoofing domains,” Dan Fichter, VP of software development at Oracle Data Cloud, tells Ad Age. “On all those impressions it looked like the advertiser was running ads on legitimate mobile websites. Not only were they not on a website, they were on an invisible web browser.”

On its website, Tapcore says it works with more than 3,000 apps, serving 150 million ad impressions a day. The apps whose pirated versions it has worked with include titles like “Perfect 365,” “Draw Clash of Clans,” “Vertex” and “Solitaire: Season 4,” according to Oracle’s report.

Tapcore’s scheme works like this, according to Oracle: the app developer signs up with Tapcore and is given code to put in its software. After the app is downloaded by a consumer, hours, even days later, the code updates with new functions—what’s known as sideloading—that turn a device into a fake ad generator.


Hope Oracle is certain about that, or Tapcore is going to have a big lawyer’s letter.
link to this extract

Zittrain and Zuckerberg discuss encryption, ‘information fiduciaries’ and targeted ads • Harvard Law


“The idea of us having a fiduciary relationship with the people who use our services is intuitive,” said Zuckerberg [interviewed by Jonathan Zittrain].  “[Facebook’s] own self-image of ourselves and what we’re doing is that we’re acting as fiduciaries and trying to build services for people. … Where this gets interesting is who gets to decide in the legal sense, or in the policy sense, of what’s in people’s best interest.”

The conversation segued into another topic area involving competing sets of interests: the use of end-to-end message encryption to make private communications inaccessible to eavesdroppers. End-to-end encryption has come under criticism for making it difficult in some cases for law enforcement agents (with the proper warrants) to access evidence locked up on devices. Zittrain raised the possibility that governments not embracing the rule of law might use their legal and technical capabilities to peek into unencrypted private communications at will. “The modern surveillance states of note in the world have a lot of arrows in their quivers… they’ve got a plan B, a plan C, and a plan D,” he said.

Zuckerberg said he is inclined to implement more end-to-end encryption. “I basically think that if you want to talk in metaphors, messaging is like people’s living room, and we definitely don’t want a society where there’s a camera in everyone’s living room,” he said.

Zittrain pointed out that people are happily installing Facebook’s own smart camera–the Portal–in their living rooms. Zuckerberg laughed. “That is I guess… yeah. Though that would be encrypted.”


link to this extract

How 20th Century Fox uses machine learning to predict a movie audience • Google Cloud blog


Understanding the market segmentation of the movie-going public is a core function of movie studios. Over the years, studios have invested in high-level data processes to try to map out customer segments, and to make predictions for future films. However, to date, granular predictions at the segment level, not to mention at the customer level, have remained elusive because of technological and institutional barriers.  

Miguel and his team have been able to lift some of those barriers by working with partners like Google Cloud. Together, we’ve built privacy-robust data partnerships to better understand moviegoers, and have developed in-house deep learning models that train on granular customer data and movie scripts to identify the basic patterns in audiences’ preferences for different types of films. In the span of 18 months, these models have become routine considerations for important business decisions, and provide one of their most objective, data-driven, and effective barometers to evaluate the tone of a movie, its affinity with core and stretch audiences, and its potential financial performance.

When it comes to movies, analyzing text taken from a script is limiting because it only provides a skeleton of the story, without any of the additional dynamism that can entice an audience to see a movie. The team wondered if there was some way to use modern, advanced computer vision to study movie trailers, which remain the single most central element of a movie’s entire marketing campaign. The trailer release for a new movie is a highly anticipated event that can help predict future success, so it behoves the business to ensure the trailer is hitting the right notes with moviegoers. To achieve this goal, the 20th Century Fox data science team partnered with Google’s Advanced Solutions Lab to create Merlin Video, a computer vision tool that learns dense representations of movie trailers to help predict a specific trailer’s future moviegoing audience.


This is entirely predictable, though also slightly weird. However, notice what happens: the right-hand column is what it forecast, the left-hand what happened. The ones to pay attention to are the unexpected, grey ones – particularly Deadpool.

link to this extract

Which face is real?

Jevin West and Carl Bergstrom at the University of Washington:


while we’ve learned to distrust user names and text more generally, pictures are different. You can’t synthesize a picture out of nothing, we assume; a picture had to be of someone. Sure a scammer could appropriate someone else’s picture, but doing so is a risky strategy in a world with google reverse search and so forth. So we tend to trust pictures. A business profile with a picture obviously belongs to someone. A match on a dating site may turn out to be 10 pounds heavier or 10 years older than when a picture was taken, but if there’s a picture, the person obviously exists.

No longer. New adverserial machine learning algorithms allow people to rapidly generate synthetic ‘photographs’ of people who have never existed.

Computers are good, but your visual processing systems are even better. If you know what to look for, you can spot these fakes at a single glance — at least for the time being. The hardware and software used to generate them will continue to improve, and it may be only a few years until humans fall behind in the arms race between forgery and detection.

Our aim is to make you aware of the ease with which digital identities can be faked, and to help you spot these fakes at a single glance.


So now we’re using humans as the adversarial network (which calls out the generative network).
link to this extract

This Cat Does Not Exist • TCDNE

It’s thispersondoesnotexist but for cats. Though the adversarial network (the yin to the yang of the generative network, which creates the non-existent cat) needs some work; it still lets too many obviously non-cats through.
link to this extract

Is the insect apocalypse really upon us? • The Atlantic

Ed Yong:


In their review, Sánchez-Bayo and Wyckhuys found 73 studies showing insect declines.

But that’s what they went looking for! They searched a database using the keywords insect and decline, and so wouldn’t have considered research showing stability or increases. The studies they found aren’t representative either: most were done in Europe and North America, and the majority of insects live in the tropics. This spotty geographical spread makes it hard to know if insects are disappearing from some areas but recovering or surging in others. And without “good baselines for population sizes,” says Jessica Ware from Rutgers University, “when we see declines, it’s hard to know if this is something that happens all the time.”

It’s as if “our global climate dataset only involved 73 weather stations, mostly in Europe and the United States, active over different historical time windows,” explained Alex Wild from the University of Texas at Austin on Twitter. “Imagine that only some of those stations measured temperature. Others, only humidity. Others, only wind direction. Trying to cobble those sparse, disparate points into something resembling a picture of global trends is ambitious, to say the least.”

For those reasons, it’s hard to take the widely quoted numbers from Sánchez-Bayo and Wyckhuys’s review as gospel. They say that 41% of insect species are declining and that global numbers are falling by 2.5% a year, but “they’re trying to quantify things that we really can’t quantify at this point,” says Michelle Trautwein from the California Academy of Sciences.


Useful nuance, but it’s still concerning unless we’re going to train cockroaches to fertilise all the crops.
link to this extract

On YouTube, a network of paedophiles is hiding in plain sight • Wired

K.G Orphanides:


Videos of little girls playing Twister, doing gymnastics, playing in the pool and eating ice lollies are all routinely descended upon by hordes of semi-anonymous commenters, sharing time codes for crotch shots, directing other people to similar videos of children and exchanging phone numbers along with a promise to swap more videos via WhatsApp or Kik. On some videos, confused children who have uploaded videos of them playing in the garden respond to comments asking them how old they are. On one video, a young girl appears to ask another commenter why one of the videos had made him “grow”. The video shows the child and her friend doing yoga and is accompanied by pre-roll advertising from L’Oreal. The video has almost two million views.

“We’re absolutely horrified and have reached out to YouTube to rectify this immediately,” a Grammarly spokesperson said. “We have a strict policy against advertising alongside harmful or offensive content. We would never knowingly associate ourselves with channels like this.”

A spokesperson for Fortnite publisher Epic Games said it had paused all pre-roll advertising on YouTube. “Through our advertising agency, we have reached out to YouTube to determine actions they’ll take to eliminate this type of content from their service,” the spokesperson added. A World Business Forum spokesperson said it found it “repulsive that paedophiles are using YouTube for their criminal activities”. A Peloton spokesperson said it was working with its media buying agency to investigate why its adverts were being displayed against such videos.


Remarkable investigation – but this is terrible for YouTube’s reputation. (Is there ever good news for YouTube?) This will surely get pickup by the big news organisations. And then, YouTube has a gigantic problem. Not only that; exactly the same story was reported in August 2017. YouTube has failed.
link to this extract

Nest Secure had a secret microphone, can now be a Google Assistant • CSO Online

Ms. Smith:


When announcing that a software update will make Google Assistant available on Nest Guard, Google added, “The Google Assistant on Nest Guard is an opt-in feature, and as the feature becomes available to our users, they’ll receive an email with instructions on how to enable the feature and turn on the microphone in the Nest app. Nest Guard does have one on-device microphone that is not enabled by default.”

Nest Secure owners have been able to use Google Assistant and voice commands, but it previously required a separate Google Assistant device to hear your commands. I suppose it depends upon your outlook on if you are happy or creeped out that your security system secretly had an undocumented microphone capable of doing the listening all along.

Google didn’t really focus on the “surprise there was a microphone hidden in the Nest Guard brain of your Nest Secure” angle, preferring a take on how Google Assistant and Nest Guard can help you out.


This is not something you accidentally include. It’s not something you accidentally forget to tell people about either, because your engineers know that it’s there, because they’re going to enable it in the future: it’s on the schedule.

Surprising that a teardown by iFixit et al didn’t find this. But it’s bad for Google not to tell people, because that’s how you undermine trust.
link to this extract

Grand Canyon tourists exposed to radiation, safety manager says • AZ Central

Dennis Wagner:


For nearly two decades at the Grand Canyon, tourists, employees, and children on tours passed by three paint buckets stored in the National Park’s museum collection building, unaware that they were being exposed to radiation.

Although federal officials learned last year that the 5-gallon containers were brimming with uranium ore, then removed the radioactive specimens, the park’s safety director alleges nothing was done to warn park workers or the public that they might have been exposed to unsafe levels of radiation.

In a rogue email sent to all Park Service employees on Feb. 4, Elston “Swede” Stephenson — the safety, health and wellness manager — described the alleged cover-up as “a top management failure” and warned of possible health consequences.

“If you were in the Museum Collections Building (2C) between the year 2000 and June 18, 2018, you were ‘exposed’ to uranium by OSHA’s definition,” Stephenson wrote. “The radiation readings, at first blush, exceeds (sic) the Nuclear Regulatory Commission’s safe limits. … Identifying who was exposed, and your exposure level, gets tricky and is our next important task.”…

…Stephenson said the uranium threat was discovered in March 2018 by the teenage son of a park employee who happened to be a Geiger counter enthusiast, and brought a device to the museum collection room…

…The report indicated radiation levels at “13.9 mR/hr” where the buckets were stored, and “800 mR/hr” on contact with the ore. Just 5 feet from the buckets, there was a zero reading. The abbreviation, “mR” typically stands for milliroentgen, a measurement roughly equivalent to a millirem, according to the Nuclear Regulatory Commission.

The NRC says average radiation exposure in the United States from natural sources is 300 millirems per year at sea level, or 400 at high altitude.

The commission lists a maximum safe dosage for the public, beyond natural radiation, is no more than 2 millirems per hour, or 100 per year.


At no point explained: what the hell the buckets were doing there.
link to this extract

Errata, corrigenda and ai no corrida: none notified.

4 thoughts on “Start Up No.1,008: YouTube’s paedophile problem, Nest’s secret mic, picking real people, Samsung folds, and more

  1. “Sooo.. an iPad you can fold up?”.

    Erm, not everything is about Apple. That’s a brand new category and format. I don’t think an iPad or tablets in general are a useful frame of reference here, the thing is rather tiny even unfolded, has a very square screen, doesn’t make sense vs a tablet financially (you can get a flagship phone *and* a flagship tablet for less, at least on the Android side). Aside from show-offs, I’d think the market is users who have a professional reason to want a large, square screen always at hand esp. on the go.

    Samsung’s press material mostly show 3 tiled apps, that’s the way to keep them in the usual aspect ratio, though well-written app should adjust gracefully to 4:3… But I’ve tried to tile 2 apps on my 7″ 18:9 Mi Max 3… It’s not really worth it compared to just switching back and forth, those screens are too small to multitask, and my use cases don’t really require it, or I take out my 2nd phone or a tablet.

    4:3 and 2-way split 4:3 doesn’t work for video consumption, not for video while socializing which I’d assume to be the most probable case for consumer multitasking.You’ve got to go either full-screen or 3-ways, unless you like weird formats or wasted space, which is a weird reason to spend an extra $1-1.5k. Dual-messaging works though, and reading + messaging ?

    I like 4:3 for Office or graphics/video work, but is it going to happen on the go on a 7.3″ screen ?

    Oh well, last time around Samsung came up with something that different, it was the pen-enabled Note series which ended up being 5 yrs ahead of its time, finding a few devoted niches but mostly selling on being the pre-christmas flagship refresh, and then being upstaged by the iPad Pro and Surface on the tablet side. This time around it seems competition will be quicker to come, and there’s a bit more uncertainty about what the final product formula will be: it’s not just “add a pen”, you can tweak to what the screen unfolds. Maybe photographers will use the screen for previews, and the storage for backup, or kids will go hog-wild live commenting viral videos…

    We sorely need some competition in AMOLED screens, even more so in flexible ones, because those prices do seem unworkable.

    • There’s also a niche of sysadmins who want to be able to remote into servers at all times. That screen format is nice for it, and Android works well for that.

  2. re. Paedophiles on YT, I think that’s the same broad issue as unscience and fake news: repugnant and harmful, but piling on new-media companies feels like shooting the messenger, not thoroughly addressing the issue.
    I mean, TV is broadcasting beauty pageants with 6yo in swimsuits and more paint on than a stolen car. Schools are presenting Evolution as a debatable theory. Fox is still unsure about Obama’s birth certificate… Can we really expect Youtube to be better than that, especially when they don’t even produce the content ? Why are we blaming them for exploiting the same base content that old media gets away with ?
    I think we need legislation, not self-regulation (also, education). At least in the long run. In the mean time, every little bit helps, and recommendations sure should stop making things worse.

  3. The bit about Movie AI is interesting, though it reads a bit like an ad, not really a white paper.

    What strikes me is that the analysis is purely physical. I decide to go see a movie mostly on whether it is fiction or not, whether it’s period/contemporary/SciFi/fantasy, whether it’s drama/suspense/horror/action/comedy, who’s in it and who wrote and shot it, what I’ve seen around the same time…

    I’m sure that AI is an OK tool in the toolbox. I’m not sure how relevant it is in the grand scheme of ticket sales.

Leave a Reply to stormyparis Cancel reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.