Start Up No.883: fertility app woes, Roku’s ad truth, four-day week yay!, hot MacBook Pros, ‘vomit fraud’, and more

Is Google Translate being trained on Bible content – and getting a bit religious nutty as a result? Photo by George Redgrave 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.

PLEASE NOTE that this week will have a shorter posting schedule: Monday (this one), Tuesday, and Wednesday. After that I’m on a fortnight’s break.

A selection of 14 links for you. Re-up. I’m @charlesarthur on Twitter. Observations and links welcome.

‘I felt colossally naive’: the backlash against the birth control app • The Guardian

Olivia Sudjic tried an app called “Natural Cycles”:


One paid-for post I saw featured a still life of a puppy, a pair of on-trend headphones, a self-help book and a thermometer, with a 250-word caption starting with “5 things I need in the morning. Cuddles from Bee [the dog], tea, music, positive quotes and the first thing I do when I wake up – my Natural Cycles thermometer.” But I found that taking your temperature regularly is not so easy. The number of times I leapt out of bed bleary-eyed and needing to pee, then realised I hadn’t first taken my temperature, meant I started waking up in the middle of the night to pre-emptively urinate, panicked about missing my measuring window in the morning. On the pill, it didn’t matter if I’d just woken up, was lying down or standing up when I took it. With Natural Cycles, the slightest motion seemed to count. It was comedic until it became tragic; I got pregnant when the predictions of fertile and infertile changed back and forth in one day, turning from green to red, after I had unprotected sex.

I now know that the ideal Cycler is a narrow, rather old-fashioned category of person. She’s in a stable relationship with a stable lifestyle. (Shift-workers, world-travellers, the sickly, the stressed, insomniacs and sluts be advised.) She’s about 29, and rarely experiences fevers or hangovers. She is savvy about fertility and committed to the effort required to track hers. I could add that her phone is never lost or broken and she’s never late to work. She wakes up at the same time every day, with a charged phone and a thermometer within reach.


Tech is no match for the female reproductive system.
link to this extract

Roku is in the ad business, not the hardware business, says CEO • The Verge

Chris Welch:


CEO Anthony Wood was frank and open about his company’s evolving business strategy in an interview on this week’s Vergecast. “We don’t really make money… we certainly don’t make enough money to support our engineering organization and our operations and the cost of money to run the Roku service,” he said. “That’s not paid for by the hardware. That’s paid for by our ad and content business.”

If you’ve got a Roku TV or streaming gadget, you’ve no doubt seen advertising for shows and apps plastered on the home screen. That’s some prime real estate. The shortcut buttons on the remote control — you know, with one or two that you’ll never use — are also paid placement. “It’s kind of an exchange of value. We help content distributors find customers, sign up customers, and promote their content, and we get paid for that.”

But it goes much deeper. Roku is learning fast as it hulks up its advertising operation, and now partially controls the ad infrastructure for some apps on its platform. So if you’re using an app like Crackle, some of the ads you’ll see are sold by Roku itself. Business Insider recently reported that “in some cases, Roku insists on selling 30% of a publisher’s ad inventory for an app if they want to be distributed on Roku devices.”

Netflix, Hulu, and other major streaming services are big enough that they don’t let Roku directly sell ads for their apps, but many smaller players do.


Roku went public earlier this year, so it needs a message that it’s got a reliable income stream. Certainly the hardware isn’t priced for profit.
link to this extract

A four-day workweek? A test run shows a surprising result • The New York Times

Charlotte Graham-McLay:


A New Zealand firm that let its employees work four days a week while being paid for five says the experiment was so successful that it hoped to make the change permanent.

The firm, Perpetual Guardian, which manages trusts, wills and estates, found the change actually boosted productivity among its 240 employees, who said they spent more time with their families, exercising, cooking, and working in their gardens.

The firm ran the experiment — which reduced the workweek to 32 hours from 40 — in March and April this year, and asked two researchers to study the effects on staff.

Jarrod Haar, a human resources professor at Auckland University of Technology, said employees reported a 24% improvement in work-life balance, and came back to work energized after their days off.

“Supervisors said staff were more creative, their attendance was better, they were on time, and they didn’t leave early or take long breaks,” Mr. Haar said. “Their actual job performance didn’t change when doing it over four days instead of five.”

Similar experiments in other countries have tested the concept of reducing work hours as a way of improving individual productivity. In Sweden, a trial in the city of Gothenburg mandated a six-hour day, and officials found employees completed the same amount of work or even more. But when France mandated a 35-hour workweek in 2000, businesses complained of reduced competitiveness and increased hiring costs.


link to this extract

Video raises concerns about excessive thermal throttling on 2018 MacBook Pro w/ Intel Core i9 • 9to5Mac

Chance Miller:


YouTuber Dave Lee, a respected and popular reviewer, shared his hands-on with the 2018 15-inch MacBook Pro this evening, showcasing the top-of-the line model with the 2.9GHz 6-core 8th-generation Intel Core i9 processor. Apple offers the processor exclusively with the 15-inch model in the form of a $300 upgrade.

In his video, Lee explains that after a “few seconds” of high-intensity work, such as editing in Adobe Premiere, throttling begins to kick-in and limits the clock speed. In Lee’s testing, the average clock speed while under load for the MacBook Pro is around 2.2GHz.

Lee compares the 15-inch MacBook Pro’s performance to that of a 2018 Aero 15X, which uses an Intel i7 processor with a base clock of 2.2GHz. That machine is able to secure an average clock speed of 3.1GHz thanks to Turbo Boost.

This i9 in the MacBook can’t even maintain the base clock speed. Forget about Turbo Boost, it can’t even maintain the 2.9GHz base clock speed, which is absurd. This CPU is an unlocked, overclock-able chip, but all of that CPU potential is wasted inside this chassis, and the thermal solution inside this chassis.

Somewhat humorously, Lee ran a render time test using Adobe Premiere (which is rather poorly optimized for macOS) with the MacBook Pro in his freezer in an attempt to cool the machine as it rendered. In doing this, the render time dropped from nearly 40 minutes to 27 minutes.


Apple Insider reckons it knows why:


Apple’s 2016 MacBook Pro chassis was designed more than two years ago. We got the first glimpse of it in a photograph in May of 2016.

At the time, Intel was promising smaller and smaller dies, with lower and lower TDP to go with it. The company didn’t make its own die-shrink projections. Even the processor in the MacBook Pro currently is well over 18 months late, according to Intel’s ever-shifting timetables.

Odds are, Apple was counting on this when it developed the enclosure.

Apple is hardly the only vendor dealing with i9 thermal conditions, and like we said, Premiere performs far better with Nvidia GPU silicon than AMD Radeon gear which explains most of the Dell ripping through the test. However, while related, this isn’t really the meat of the matter given that Lee put the MacBook Pro in the freezer and got better completion speeds out of it.

Video producer Lee suggested that the entire MacBook Pro cooling solution, an Apple-designed heatsink and fan module, is insufficient for the beefy (and hot) i9 Intel silicon as it stands.


One has to observe that Apple keeps designing itself into a thermal corner.
link to this extract

Why Is Google Translate spitting out sinister religious prophecies? • Motherboard

Jon Christian:


On Twitter, people have blamed the strange translations on ghosts and demons. Users on a subreddit called TranslateGate have speculated that some of the strange outputs might be drawn from text gathered from emails or private messages.

“Google Translate learns from examples of translations on the web and does not use ‘private messages’ to carry out translations, nor would the system even have access to that content,” said Justin Burr, a Google spokesperson, in an email. “This is simply a function of inputting nonsense into the system, to which nonsense is generated.”

When Motherboard provided Google with an example of the eerie messages, its translation disappeared from Google Translate.

There are several possible explanations for the strange outputs. It’s possible that the sinister messages are the result of disgruntled Google employees, for instance, or that mischievous users are abusing the “Suggest an edit” button, which accepts suggestions for better translations of a given text.

Andrew Rush, an assistant professor at Harvard who studies natural language processing and computer translation, said that internal quality filters would probably catch that type of manipulation, however. It’s more likely, Rush said, that the strange translations are related to a change Google Translate made several years ago, when it started using a technique known as “neural machine translation.”

In neural machine translation, the system is trained with large numbers of texts in one language and corresponding translations in another, to create a model for moving between the two. But when it’s fed nonsense inputs, Rush said, the system can “hallucinate” bizarre outputs—not unlike the way Google’s DeepDream identifies and accentuates patterns in images.


Another theory: Google did some training using the Bible, as translated into different languages. Notice that bit where the weird translation disappears when notified to Google PR. This is either (a) preventing others confirming it or (b) improving the system when notified by users. Pick your preference.

link to this extract

Farewell, Google Maps • In der Apotheke

Bartłomiej Owczarek and Tomasz Nawrocki run a startup which helps people find medicines at p[hysical pharmacies; they’ve previously used Google Maps, but suddenly found the prices rising dramatically:


After a conference call with Google Maps customer service (who, contrary to the email, offered no discounts or credits whatsoever) we realised that price increases are huge:

• Current free usage limit of 750k requests monthly turns into ca. 28k requests (almost 30 times less)

• Current $0.5 for commercial usage becomes $7 (14 times more), $5.60 with high volume

Importantly, prices are the same from US to the Africa, despite the fact that revenue generation is vastly different in most developed countries compared to the others. We know it well from comparing Polish market to Germany, as we expand there. 

Comparison of Google Maps monthly bill before and after price hike

If we maintained current monthly usage of both maps and Places (ie. location search), the cost of Google Maps would be multiple times higher than the total cost of all other infrastructure.


They are going with MapBox and MapTiler – but also swapped in some code so that they can quickly swap between providers.
link to this extract

Vomit fraud could make your Uber trip really expensive • Miami Herald

Catalina Ruiz Parra:


The next time you use Uber, check your bill. The trip could turn out to be expensive — not just for the distance but for a type of fraud that is on the rise.

It’s called “vomit fraud,” a scam repeatedly denounced in social networks yet still taking place around the world.

And Miami, of course, is a common spot.

What is it? Passengers request Uber cars, which deliver them to their destination. So far so good.

But soon the passenger receives a note from Uber reporting an “adjustment” in the bill and an extra charge that can range from $80 to $150, depending on the driver’s degree of crookedness.

If you think that’s frustrating, you’re right. But the worst is still to come.

The passenger, unaware of what’s happening, tries to contact Uber. The only way to do that is through the “help” button on the company’s app or internet page.

The first reply usually goes something like this: “I understand that it can be disconcerting to receive adjustments to the tariff after your trip ended … In this case, your driver notified us that during your trip there was an incident in the vehicle and therefore a cleanup fee of $150 was added.”

The message is accompanied by photos of the alleged incident — vomit in the vehicle. The Uber driver had sent the images to the company, which considered them sufficient evidence to add the cleanup charge to the bill.


I’d imagine the drivers just have a stock or multiple pictures that they send. (Does Uber check the EXIF data for the photo?) Or perhaps they throw some vegetable soup over it? Either way, Uber is caught in the middle – and regulators say it’s not up to them.
link to this extract

Evolutionary algorithm outperforms deep-learning machines at video games • MIT Technology Review


Many genomes [of evolving code, where “good” code is reused] ended up playing entirely new gaming strategies, often complex ones. But they sometimes found simple ones that humans had overlooked.

For example, when playing Kung Fu Master, the evolutionary algorithm discovered that the most valuable attack was a crouch-punch. Crouching is safer because it dodges half the bullets aimed at the player and also attacks anything nearby. The algorithm’s strategy was to repeatedly use this maneuver with no other actions. In hindsight, using the crouch-punch exclusively makes sense.

That surprised the human players involved in the study. “Employing this strategy by hand achieved a better score than playing the game normally, and the author now uses crouching punches exclusively when attacking in this game,” say Wilson and co.

Overall, the evolved code played many of the games well, even outperforming humans in games such as Kung Fu Master. Just as significantly, the evolved code is just as good as many deep-learning approaches and outperforms them in games like Asteroids, Defender, and Kung Fu Master.

It also produces a result more quickly. “While the programs are relatively small, many controllers are competitive with state-of-the-art methods for the Atari benchmark set and require less training time,” say Wilson and co.

The evolved code has another advantage. Because it is small, it is easy to see how it works. By contrast, a well-known problem with deep-learning techniques is that it is sometimes impossible to know why they have made particular decisions, and this can have practical and legal ramifications.


link to this extract

24% of Tesla Model 3 orders have been canceled, analyst says • CNN

Jordan Valinsky:


Cancellations for Model 3 orders have picked up in recent weeks. Refunds now outpace deposits for Tesla’s new mass-market electric car, according to Needham & Co. analyst Rajvindra Gill. Tesla disputes that.

In an analyst note delivered to clients Thursday, Gill cited extended wait times for the car, the expiration of a $7,500 tax credit, and the fact that Tesla has not yet made the $35,000 base model of the car available for purchase yet.

About one in every four Model 3 orders is canceled, Gill said, double the rate from a year ago. Customers have to put down a refundable $1,000 deposit to reserve a Model 3, then pay another $2,500 to choose their specific version. They pay the rest when the car is delivered.

The wait time for a Model 3 is about 4 months to a year, and base model customers could wait until 2020, Gill said.

A Tesla spokesperson denied that Model 3 cancellations exceed new orders. The spokesperson also said the wait times that Gill cites are outdated. Tesla’s website currently lists wait times from 1 month to 9 months.


There are signs of stress at Tesla, and none of them being leavened.
link to this extract

WhatsApp to limit message forwarding after rumor-led violence in India • WSJ

Krishna Pokharel:


Facebook Inc.’s WhatsApp messaging service is making it harder for users world-wide to forward content, after the spread of rumors on the app led to mob violence and the killing of more than 20 people in India.

False messages about roaming child-kidnapping gangs spread through WhatsApp—one of the most widely used apps in India with over 200 million monthly active users—have triggered a spate of lynching as panicked groups attack strangers they find suspicious, Indian authorities have said.

WhatsApp’s announcement Thursday came a day after Facebook under a new policy said it would begin removing misinformation that could spark violence. The initiative will start in Sri Lanka and Myanmar, which have also struggled with violence fueled by false reports spread on social media.

The Indian government earlier this month asked WhatsApp to take immediate action to stop the misuse of its platform, saying rumors circulated on the messaging service had led to deadly attacks.

The Menlo Park, Calif.-based company said in a blog post that it is putting restrictions on the number of groups to which a message can be forwarded.


The things Facebook touch turn to crap, don’t they?
link to this extract

Risky Thailand cave rescue relied on talent, luck—and on sticking to the rules • Ars Technica

Chris Peterman is a professional diver with 16 years’ experience, and understands the dangers of cave diving; he nearly died once, as he recounts, and so can explain the challenges that the Thai football team faced:


Cave diving has five rules. These sum up the hard-won wisdom of the cave-diving community, as conducted through the analysis of cave-diving accidents and fatalities. Though the exact wording of each will differ from instructor to instructor, the rules are:

• Be well-trained and do not dive beyond your certification level
• Never use more than one third of your breathing gas to enter the cave—reserve one third for exiting and one third for emergencies
• Maintain a physical guideline back to the cave entrance at all times
• Never dive below the appropriate depth for your breathing gas mixture
• Carry at least three lights per person—one main and two back-ups

Since these rules were introduced in the late 1970s (first as only three rules, later expanding to five), fatalities per number of dives have dropped among the cave-diving community. Today, the largest segment of fatalities in underwater caves comes not from certified cave divers but from divers not specifically trained by a professional cave instructor to be in that environment.

The first of these rules is therefore simple, and one that I broke badly: never dive beyond your certification level.


Rock climbing has similarly made itself safer and safer though the accretion of experience and technology, though it doesn’t have rules in quite the same way.
link to this extract

The best thing about Samsung’s exciting new tablet might also be a fatal flaw • BGR

Chris Smith:


just like the iPhone X, the Galaxy Tab S4 will sacrifice the fingerprint sensor, a feature many people love on a smartphone or tablet.

Apple replaced Touch ID with Face ID, a secure 3D facial recognition system that’s a first for the industry. Samsung doesn’t have that luxury, however.

In lack of a 3D front-facing camera, Samsung will employ the Intelligent Scan feature that’s already available on the Galaxy S9.

In case you’re not familiar with that, that’s a mix between the iris scanner and facial recognition system that Samsung has had for years. SamMobile discovered a video from the official Galaxy Tab S4 firmware that plays when you’re configuring Intelligent Scan on the tablet.


It’s not as secure as Face ID, which means you’d want to go with a passcode/password, which is a retrograde step. I’m already interested to see what the second generation of Apple’s Face ID is like; the first is pretty good, but there was a huge difference between the first generation of Touch ID and the second.
link to this extract

Registry of Open Data on Amazon Web Services


This registry exists to help people discover and share datasets that are available via AWS resources. Learn more about sharing data on AWS.

See all usage examples for datasets listed in this registry.


This is pretty amazing. Landsat pictures (zoom in on your house!), IRS 990 filings, satellite data labelled for machine learning, 5bn web pages from web crawling, a global database (from broadcast, print and online news) from every country identifying key events, OpenStreetMap, bourse data from the German stock market, Hubble telescope data… it’s such a colossal reservoir of data waiting to be made use of. Sure, many others have got there first, but what could come from cross-matching Landsat data with OSM with bourse data with key events? And then applying machine learning to that?
link to this extract

Congress to leave Trump’s deal with China’s ZTE untouched • Bloomberg

Jenny Leonard and Erik Wasson:


Negotiators from the Senate and House of Representatives late Thursday agreed to abandon efforts to reinstate harsher sanctions against the Chinese telecommunications-equipment maker as part of the defense policy bill, the people said. Both chambers are expected to vote on the National Defense Authorization Act next week.

Draft language advanced in the House earlier this year focused on a procurement ban for ZTE products, whereas the Senate approved language that would reinstate the sales ban for US companies to sell to ZTE. The White House strongly opposed any efforts by Congress to block its deal for ZTE to resume business.

The Trump administration in April announced a seven-year ban on US exports to ZTE after it said the company violated sanctions agreements by selling American technology to Iran and North Korea. The move forced ZTE to announce it was shutting down.

Trump reversed course in May, saying he was reconsidering penalties on ZTE as a personal favor to Chinese President Xi Jinping. Later that month, his administration announced it would allow the company to stay in business after paying a new fine, changing its management and providing “high-level security guarantees.”

Following through on the promise, the Commerce Department last week lifted a ban on American firms selling products to ZTE after the company paid the final tranche of a $1.4bn penalty by placing $400m in escrow at a US bank. Congresspeople from both parties had blasted the Trump administration for helping ZTE.


This is the sort of revival to make Lazarus whistle in admiration.
link to this extract

Errata, corrigenda and ai no corrida: none notified

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.