Start Up: Close (Tiny) Encounters, China’s slowing smartphones, WhatsApp’s age puzzle, and more

When an AI was given the task of landing a plane on an aircraft carrier, it found an unexpected solution. Photo by Official U.S. Navy Page on Flickr.

You can sign up to receive each day’s Start Up post by email. You’ll need to click a confirmation link, so no spam.

A selection of 12 links for you. Something for the weekend. I’m @charlesarthur on Twitter. Observations and links welcome.

A ‘Close Encounters’ example of forced perspective • Trailers From Hell

Glenn Erickson:


The most complex miniature was not a full model, but a ‘reflections’ model of an auto toll booth, in which Director of Photography of Special Effects Richard Yuricich was tasked with adding the effects of a glowing saucer flying through. On a shoot in San Pedro, Steven Spielberg suddenly decided he wanted a new shot, and had Yuricich line up his 65mm camera at a bridge toll booth to film the action of the cop cars racing left to right. The shot wasn’t in the schedule or the budget, which didn’t please Columbia production head John Veitch. Yuricich had to brainstorm to figure a way to do the shot for next to nothing, and the solution came to him on his drive home.

All Gregory Jein had to work with was a clip of a 65mm shot of the booths. He painstakingly constructed little panes of glass in depth, aligned so that when double-exposed onto the existing shot, reflections would be added of the passing flying saucer (the glowing red ‘Tinkerbell’ dot) as it chased the highway patrolmen: “Hey, that’s Ohio. It costs a quarter.”

Richard Yuricich lined up the shot, and I assisted Greg in plotting it out – mainly, I held up little pieces of plastic while he directed me through the viewfinder, as if I were a surveyor’s assistant. The ‘model’ ended up being this little set of boxes and windows. It worked amazingly well.

Storyboards called out for a wide shots of the crossroads where Power Company employee Roy Neary runs into his first saucers, glowing hot-rods that are apparently in love with our road system. Spielberg wanted a night-for night look where one could see to the horizon, which meant that the shot couldn’t be made in a real location. Bright lights from the saucers were meant to blast down on the road, throw shadows, etc.


It’s amazing – I’d never have thought, watching the film, that any of it wasn’t real landscape.
link to this extract

Chinese smartphone market suffers a hard-landing, shipments decline by 21% in Q1 2018 • Canalys


Smartphone shipments in China suffered their biggest ever decline in Q1 2018, down by more than 21% annually to 91m units, a number first passed some four years ago in Q4 2013. Eight of the top 10 smartphone vendors were hit by annual declines, with Gionee, Meizu and Samsung shrinking to less than half of their respective Q1 2017 numbers.

Huawei (including Honor) managed to grow shipments by a modest 2%, maintaining its lead and consolidating its market share to about 24% by shipping over 21m smartphones. Second-placed Oppo and third-placed Vivo bore the brunt of the overall decline, with shipments falling by about 10% to 18m and 15m respectively. Xiaomi was the only company to buck the trend, growing shipments by 37% to 12m units, and overtaking Apple to take fourth place.


That’s quite an ominous downward trend on the right. Counterpoint Research also sees China down (8% year-on-year), though it reckons Huawei, Oppo, Xiaomi and Apple all grew; it puts Apple ahead of Xiaomi.
link to this extract

LG mobile division sees big jump in losses, blames new strategy • Android Authority

Hadlee Simons:


The [overall] firm reported a 20% increase in operating income, making for the “highest first-quarter profit and revenue in company history.” The performance also translates to the “highest quarterly profitability” since Q2 2009, LG added.

It’s a different tale for the LG mobile division, however, as the unit posted sales of 2.16trn won ($2bn) and an operating loss of 136.1bn won ($126.85m) for the quarter. By comparison, Q1 2017 saw the unit deliver an operating loss of 200m won ($185,000 by today’s conversion rate).

“Sales declined from the same quarter last year due to a revised smartphone launch strategy,” the company explained.

That strategy was first revealed back at CES 2018, when LG Electronics CEO Cho Sung-jin said the company would release new phones “when it is needed.” The executive said it would also reveal more variants of the G and V series flagship phones. We saw the first major variant at MWC 2018 in the form of the V30s ThinQ, packing more RAM/storage, as well as improved AI-related camera features compared to the V30.

The company is forecasting improved sales in the second quarter of the year, citing the upcoming LG G7 ThinQ flagship and its software upgrade center as two major drivers. The new flagship will be officially revealed on May 2, but we already know it’ll pack a notch, a super bright LCD screen, and a dedicated AI button.


The only, tiny chance that LG’s mobile business has of ever coming back to profit is if Huawei and ZTE get banned from the US (and maybe UK) markets. So far it has lost money for 12 straight quarters – a tidy $2.6bn.
link to this extract

Ray Ozzie’s proposal: not a step forward • Steve Bellovin

Bellovin is well known in the computer security field; if you saw yesterday’s set, it was one of them:


[Ray] Ozzie presented his proposal at a meeting at Columbia—I was there—to a diverse group. Levy wrote that Ozzie felt that he had “taken another baby step in what is now a two-years-and-counting quest” and that “he’d started to change the debate about how best to balance privacy and law enforcement access”. I don’t agree. In fact, I think that one can draw the opposite conclusion.

At the meeting, Eran Tromer found a flaw in Ozzie’s scheme: under certain circumstances, an attacker can get an arbitrary phone unlocked. That in itself is interesting, but to me the important thing is that a flaw was found. Ozzie has been presenting his scheme for quite some time. I first heard it last May, at a meeting with several brand-name cryptographers in the audience. No one spotted the flaw. At the January meeting, though, Eran squinted at it and looked at it sideways—and in real-time he found a problem that everyone else had missed. Are there other problems lurking? I wouldn’t be even slightly surprised. As I keep saying, cryptographic protocols are hard.

The other point is that security is a systems problem. To give just one example, the international problem alone is a killer issue. If the United States adopts this scheme, other countries, including specifically Russia and China, are sure to follow. Would they consent to a scheme that relied on the cooperation of an American company, and with keys stored in the U.S.? Almost certainly not. Now: would the U.S. be content with phones unlockable only with the consent and cooperation of Russian or Chinese companies? I can’t see that, either.


link to this extract

Here’s what you need to know about the new Gmail • The Washington Post

Geoffrey Fowler:


Truth be told, when Google first showed me the redesign, I thought: Who asked for this? It’s like Gmail got a facelift to look cuter. Buttons are rounded and its bubbly new font could have been ripped from a 1992 school yearbook.

But I came around after living with the new Gmail for a week. It’s cleaner and the stuff you rely on hasn’t moved far. What’s more, it has some good ideas to keep you from missing important emails and to make them more secure. Google is getting much more deeply involved with our messages, and the result moves email in the right direction.

My favorite new thing: You can now send emails that self-destruct.

Your Gmail experience won’t change immediately, unless you tap the gear icon (for settings) and turn on the new website look. But it’ll come to you eventually: in the coming months, Google will bring the new design and features to everyone, including people with corporate accounts…

…But the AI isn’t all-knowing (yet). Gmail doesn’t know if you actually replied to a message in real life, on the phone, or in a text. It also doesn’t let you identify certain senders as top priority.

To keep annoyance to a minimum, Google promises to send a max of three nudges per day. It’s on by default, but you can turn it off if you’d like.

The new Gmail applies intelligence to another information-overload problem: notifications. Previously, Gmail apps for smartphones and tablets sent an alert every time you got a new email in your primary inbox. Turning on notifications would cause your phone to buzz all day long. When new Gmail apps for iOS and Android debut in the coming months, they’ll be able to alert you only when you’ve got an email Gmail actually thinks is “high priority.”


Lots more emphasis on AI. Let’s see how that could go.
link to this extract

Don’t buy the MacBook Pros even on sale, in my opinion • The Outline

Casey Johnston:


Since I wrote about my experience, many have asked me what happened with the new top half of the computer that the Apple Geniuses installed, with its pristine keyboard and maybe-different key switches. The answer is that after a couple of months, I started to get temporarily dead keys for seemingly no reason. Again.

I still had my 2013 MacBook Pro around, so I sold my 2016 MacBook Pro back to Apple’s refurb program, and now I just use the 2013 as my laptop (I used the recovered money to build a PC, lord help me). This old MacBook Pro is still fine, and most importantly, all the keyboard keys work. The new MacBook Pro is gone. When I started working at The Outline, I was offered a choice of a new MacBook Pro or a MacBook Air for my work computer, and I chose the MacBook Air, with its good keyboard that doesn’t break from dust. I’m fully committed to this bit.


I wonder if Apple is keeping the MacBook Air, with its old keyboard model, going because (1) it’s the best-selling laptop (2) there are no significant problems with the keyboard. A lot of people who have heard about the keyboard problems are holding off on buying new machines.

Apple’s fortunate that the wave has moved on from laptops. If this had happened 10 years ago, it would be in trouble. Similarly, if some equivalent to this happened on an iPhone model or iPad, it would take rapid action: look at how quickly it moved when the whole “#bendgate” stuff when the bigger iPhone 6 and 6 Plus came out.
link to this extract

MOSISO clear keyboard skin cover for Touch Bar Models 2017 & 2016 release newest version macbook pro 13 (A1706) and MacBook Pro 15 (A1707) with Touch ID (EU Layout) •

Ian Fogg had problems with his new Mac keyboard; swears by this. It’s effectively the same thing as the “cloth” cover that the iPad Pro keyboards (which are the same key system) use. And cheap. Except doesn’t it stop air escaping around the keys, so you get a very hot device?
link to this extract

WhatsApp plans to ban under-16s. The mystery is how • The Guardian

I wrote an op-ed:


WhatsApp didn’t explain in its blogpost how it will enforce its policy. No social network – in fact, no ad-supported service – is keen on truly finding out the age of its users, because then it might have to turn them away, which means less chance to show them adverts, collect data about them, and tie them into longer-term use. Teenagers are the ideal customer: they have long lives ahead of them, can be brand-loyal, and are very susceptible to peer pressure to join. And online age verification really is difficult. Phone companies such as O2 demand photo ID at a shop (effective, but no use for online-only services) or a credit card (easy to defeat by stealing details from parents). Quizzes can be beaten by a search. Perhaps playing the mosquito alarm through the speakers constantly? Or maybe a question asking about a new meme; if you get the answer correct, you’re too young.

Nobody knows, as the government’s 2016 consultation demonstrated. Plans to enforce it for sex sites starting this month were quietly buried in March. Though it insists the scheme will be introduced this year, don’t hold your breath. The subtler, but bigger, problem is that the big social media companies (which includes YouTube) are flattening out the teenage years. To Facebook, YouTube, Google and so on you’re either over 18 or over 13; no other user ages exist.

Yet that period between 13 and 18 is one of enormous intellectual and social development, and you wouldn’t throw such disparate ages together in a youth club and just leave them to it. But in their indifferent rush to corral the world, social networks do, with the results that Ofcom noted: young children coming online are confronted by much older ones who already own the space [and report encountering hateful or sexual content].


link to this extract

Snapchat to stop retaining location data on under-16s in Europe • FT

Tim Bradshaw:


Snapchat plans to stop retaining certain data about users under the age of 16 in Europe, including precise location history, to comply with new EU privacy regulations.

The move is intended to allow the California-based company to meet the requirements of the General Data Protection Regulation without having to cut off millions of younger teenagers from using its app. 

Shares in Snapchat’s parent company Snap fell 7% on Wednesday after WhatsApp, a rival messaging app, said it would raise its minimum age for users in Europe to 16 in response to GDPR. 

The move by Facebook-owned WhatsApp is the most drastic yet by a popular mobile app to mitigate any risks that might arise from GDPR. Any breach can result in fines of as much as 4% of annual global revenue.

Snapchat’s steps might allay fears among some users and investors that it would have to follow WhatsApp’s example and cut off under-16s altogether. Teenagers are among Snapchat’s most loyal and active users.


link to this extract

UK healthcare startup said to have posted fake reviews online • Bloomberg

Giles Turner:


The two-year-old company [Cera Care] matches patients with in-home health-care professionals capable of providing everything from support for elderly clients to live-in assistance for people with dementia, and has garnered positive press from UK publications. Its website also boasts sky-high ratings on customer-satisfaction sites and partnerships with 10 UK National Health Service organizations.

But according to three people with knowledge of the matter, Cera Care doesn’t in fact have partnerships with at least seven of those groups, and up to a dozen of the reviews on third-party sites were crafted either by Cera Care employees or people close to them, rather than unbiased customers. These people asked not to be identified because of the sensitivity of the information.

In a widely publicized announcement in March 2017, Cera said it has partnered with a range of public health groups in London, including St. Barts, one of the U.K.’s largest. It later listed 10 of these groups on its website. However, the startup regularly works with only three of these groups – Lewisham CCG [Clinical Care Group], Haringey CCG, and Tower Hamlets CCG – according to people with knowledge of the matter.

“We note that our website was not fully up to date with these materials and are rectifying it,” Cera said in an emailed statement. The company had previous partnerships with health care groups including Brent, Harrow and Hillingdon and East London Foundation Trust, the company said…

…Patients and their families considering private in-home care often look closely at reviews on sites such as Trustpilot AS, HomeCare UK and Google Reviews before choosing a provider. Of the 104 reviews on Trustpilot, a number were fakes created by Cera Care employees, people with knowledge of the postings said. Of them, 12 have been taken down in recent days. “Good customer service. Great care from care workers as well. Happily continuing to do business with Cera,” one review said.

“We take any allegations of false reviews extremely seriously and these will be investigated thoroughly and dealt with strictly,” the spokeswoman for Cera, said. “We have no tolerance for this.”

Trustpilot said it has been investigating Cera Care and has been removing several reviews.


link to this extract

Letting neural networks be weird: when algorithms surprise us • Ai Weirdness

Janelle Shane:


machine learning algorithms are widely used for facial recognition, language translation, financial modeling, image recognition, and ad delivery. If you’ve been online today, you’ve probably interacted with a machine learning algorithm.

But it doesn’t always work well. Sometimes the programmer will think the algorithm is doing really well, only to look closer and discover it’s solved an entirely different problem from the one the programmer intended. For example, I looked earlier at an image recognition algorithm that was supposed to recognize sheep but learned to recognize grass instead, and kept labeling empty green fields as containing sheep.

When machine learning algorithms solve problems in unexpected ways, programmers find them, okay yes, annoying sometimes, but often purely delightful.

So delightful, in fact, that in 2018 a group of researchers wrote a fascinating paper that collected dozens of anecdotes that “elicited surprise and wonder from the researchers studying them”. The paper is well worth reading, as are the original references, but here are several of my favorite examples.


There are so many, but I think my favourite is:


In one of the more chilling examples, there was an algorithm that was supposed to figure out how to apply a minimum force to a plane landing on an aircraft carrier. Instead, it discovered that if it applied a *huge* force, it would overflow the program’s memory and would register instead as a very *small* force. The pilot would die but, hey, perfect score.


“OK, engaging machine learning autopilot for landing…” But truly, the paper itself (linked above) is hilarious. Some of the interactions between humans and programs truly are like captors and wild animals.
link to this extract

Snap’s second-generation Spectacles are more grown up — and more expensive • The Verge

Casey Newton:


Given Snap’s struggles in hardware, it’s fair to ask why the company is trying again. The reason is that hardware has the potential to be a much better business for Snap than software does. It enables new services, it can be sold at healthy profit margins, and it’s harder for competitors to copy. Hardware is extremely difficult to do well — but if Snap can pull it off, hardware could help the company chart a profitable path forward. (It lost $720m last year.)

To build a sequel to Spectacles, Snap talked with customers and drew up a list of employees’ own complaints about the original, a product designer told me this week. The company ultimately made three major refinements.

First, it created a slimmer design. Spectacles should now fit more easily in a pants pocket, or dangling from the front of your T-shirt. The case is smaller, too, and will take up less space in your bag. All those small tweaks added up to a product that feels a bit more like sunglasses and a bit less like a face camera.

Second, Snap improved the transfer speed between Spectacles and your phone. If you owned the first-generation Spectacles, you likely remember this as the worst part of using them. Transferring videos off your device means opening Memories, tapping “import,” accepting a dialog box that switches your phone to the Spectacles Wi-Fi network, and then waiting for several minutes while they download.

The good news is the download speeds are now three to four times faster, Snap says. The bad news is that the rest of the process, down to the awkward connection to a Spectacles Wi-Fi network, remains the same.


Spectacles v1: cost $150, sold 150,000, wrote off $40m.
Spectacles v2: cost $170, but now the excitement of “Snap making SPECTACLES?!” has worn off, it’s hard even to see them hitting the first version’s sales. This reminds me of the WC Fields quote: “If at first you don’t succeed, try, try again. Then quit. No point being a damn fool about it.”
link to this extract

Errata, corrigenda and ai no corrida: yesterday’s picture of French Uber icons on the map didn’t show France. It showed Bethnal Green, which has never been French. Thanks, Andrew B.

2 thoughts on “Start Up: Close (Tiny) Encounters, China’s slowing smartphones, WhatsApp’s age puzzle, and more

  1. That’s interesting in terms of, if an algorithm is supposed to “recognize sheep” – well, what is a “sheep”? It seems that a “sheep” is being deemed something along the lines of “discrete ground lump around a meter long, usually white but not necessarily so”. From that perspective, the picture above being called a “herd of sheep” makes sense. Just realize that “sheep” isn’t being given the meaning you think it should have. But really, why should it? The algorithm is doing exactly what it was asked – finding “similar” structure. But – and I see the paper does make this point – the structures it is finding are rather simplistic. Note this has implications for any idea that AI moderation will solve the presence of bad speech on social media.

    Remember, man is a plucked chicken, err, featherless biped.

Leave a 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.