This is the last Overspill of the year. It will return on Monday January 15, 2018.
Charity time: ahead of Christmas, I’m encouraging readers to make a donation to charity; a different one each day.
Shelter, the UK charity for the homeless. It’s a difficult time to be homeless.
(If you’re not in the UK, and want to donate to a charity nearer home, please search on “homeless charity [your country].)
• Thursday’s charity was Wikipedia.
• Wednesday’s charity/ies was/were:
– UK readers: The National Deaf Children’s Society
– US readers: American Society for Deaf Children
– Australian readers: Deaf Children Australia
(In other countries try a search on “deaf children [your country]”.)
• Tuesday’s charity was The Internet Archive, which preserves web content that might otherwise be lost (or conveniently scrubbed). It’s in the middle of a $6m funding drive, and is presently at $3.6m. (The average donation is $41.)
• Monday’s charity was BookTrust: give £10 and a child in social care will receive books for Christmas.)
Please donate as you see fit.
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 10 links for you. Use them wisely. I’m @charlesarthur on Twitter. Observations and links welcome.
Anna Ridler’s Fall of the House of Usher unspools, rooms and bodies spreading half-seen across the frames of this 12-minute film like gossamer. A woman appears to walk down a hallway, then melts into a moonlit sky. A face appears in the dark, contorts into shapes. The animation is based on a 1929 film version of Edgar Allen Poe’s story, but its inky and strange visuals are the result of something altogether more modern: machine learning.
Each moment of Ridler’s film has been generated by artificial intelligence. The artist took stills from the first four minutes of the 1929 movie, then drew them with ink on paper. These versions were then used to train a generative adversarial network (GAN), teaching it what sort of picture should follow on from another. The GAN uses this information to create its own procession of stills, based around a pair of networks that work in competition with each other — one as a generator, one as a discriminator, evaluating the work of the former like an algorithmic critic.
‘Fall of the House of Usher,’ by Anna Ridler. Photo: Anna Ridler
The result is an AI-generated animation based on drawings that are based on the opening minutes of a 1929 film, which is based on an 1839 short story about a decaying lineage. It is a project that uses machine learning techniques not to showcase the technology, but as a way to engage with ideas of memory, the role of the creator, and the prospect of degeneration. It is primarily an artistic work, leveraging artificial intelligence as a medium in a way another artist may use acrylics or videotape.
I like this. On the other hand…
algorithms, unlike humans, are susceptible to a specific type of problem called an “adversarial example.” These are specially designed optical illusions that fool computers into doing things like mistake a picture of a panda for one of a gibbon. They can be images, sounds, or paragraphs of text. Think of them as hallucinations for algorithms.
While a panda-gibbon mix-up may seem low stakes, an adversarial example could thwart the AI system that controls a self-driving car, for instance, causing it to mistake a stop sign for a speed limit one. They’ve already been used to beat other kinds of algorithms, like spam filters.
Those adversarial examples are also much easier to create than was previously understood, according to research released Wednesday from MIT’s Computer Science and Artificial Intelligence Laboratory. And not just under controlled conditions; the team reliably fooled Google’s Cloud Vision API, a machine learning algorithm used in the real world today.
There’s that need for oversight, except if these things are classifying colossal numbers of objects how will we know when it makes a false negative? (The false positives should stick out a mile.)
link to this extract
Keeper, a password manager software maker, has filed a lawsuit against a news reporter and its publication after a story was posted reporting a vulnerability disclosure.
Dan Goodin, security editor at Ars Technica, was named defendant in a suit filed Tuesday by Chicago-based Keeper Security, which accused Goodin of “false and misleading statements” about the company’s password manager.
Goodin’s story, posted December 15, cited Google security researcher Tavis Ormandy, who said in a vulnerability disclosure report he posted a day earlier that a security flaw in Keeper allowed “any website to steal any password” through the password manager’s browser extension.
Goodin was one of the first to cover news of the vulnerability disclosure. He wrote that the password manager was bundled in some versions of Windows 10. When Ormandy tested the bundled password manager, he found a password stealing bug that was nearly identical to one he previously discovered in 2016.
Wouldn’t expect this to get far given the reality that Goodin’s story was updated in a timely fashion.
link to this extract
Facebook will stop flagging content that’s been declared false by external fact-checkers, and will instead surface fact-checks as related articles in the News Feed, the social media giant announced Wednesday.
The move represents the biggest outward facing change to Facebook’s year-old partnership with fact-checkers. The company said this new approach will be more effective in stopping the spread of misinformation, while also making it easier to scale its effort to other markets and content types.
Tessa Lyons, a News Feed product manager, told BuzzFeed News that surfacing fact-checks as related articles proved more effective in tests than applying a disputed flag to stories in the News Feed.
“Related articles outperformed disputed flags in giving people more information so they could understand what was true or false,” she said. “Hoaxes that had related article fact checks had fewer shares than those with the disputed flag.”
This is much better. Pity it’s about two or more years too late.
link to this extract
Google might want to follow Apple’s lead and force developers to disclose loot box odds • AndroidAuthority
In a move that I think Google should follow with the Play Store, Apple revised its App Store guidelines to force developers to disclose the odds of people receiving each type of item from them.
The updated guidelines are a tad vague, in that they do not say exactly where developers should display those odds, though they state that the odds need to be displayed before folks buy loot boxes:
Apps offering ‘loot boxes’ or other mechanisms that provide randomized virtual items for purchase must disclose the odds of receiving each type of item to customers prior to purchase.
For the uninitiated, loot boxes contain a variety of virtual items that contain everything from common to rare in-game items. Some, if not most, games are designed so that you cannot pick up these items separately — you can only get them in loot boxes.
The problem is that many folks see these loot boxes as a predatory and manipulative business model that get people to spend more money on games. More significant, you can purchase loot boxes either with in-game or real-world currency, which, along with their randomness, have forced people to wonder whether loot boxes constitute gambling.
“Forced” people to wonder? I think they’ve led people to wonder that. Anyway, yes, it would be an excellent move if Google were to follow Apple’s lead here. An even better one just to ban the damn things, but let’s win the small victories first.
link to this extract
Smart speakers to outsell wearables during U.S. holidays, as demand for wearables slows • TechCrunch
Smart speakers will likely outsell wearable devices this holiday season. That’s the latest prediction from analysts at eMarketer, which forecasts a slowing growth rate for devices like fitness trackers and smartwatches here in the US. The wearable market is continuing to grow, to be clear, but it’s struggling to reach the mainstream. Next year, only 20% of the U.S. adult population will use a wearable devices at least once a month, the firm says.
Note that eMarketer is looking at wearable usage and market penetration here, not sales.
That being said, the firm is estimating that usage of wearable will grow just 11.9% in 2018, rising from 44.7m adult wearable users in 2017 to 50.1m in 2018. As a percentage of the population, that’s a climb from 17.7% to 19.6%.
Things won’t improved much in the next few years, either, if the forecast holds out. The growth rate will slow to single digits in 2019. By 2021, eMarketer is estimating 59.5m adult wearable users, representing 22.6% of the population.
The firm attributes the majority of the growth in the sector – a market today that’s dominated by fitness trackers – to new users of smartwatches, like the Apple Watch…
…“Other than early adopters, consumers have yet to find a reason to justify the cost of a smartwatch, which can sometimes cost as much as a smartphone,” eMarketer forecasting analyst Cindy Liu said. “Instead, for this holiday season, we expect smart speakers to be the gift of choice for many tech enthusiasts, because of their lower price points.”
[Neil] Cavuto, who anchors one hour each weekday on Fox News Channel and two on the Fox Business Network, revealed in an on-air commentary that he won’t ask for an interview. He said he spoke publicly after some viewers and administration officials remarked that things he had done weren’t helping his chances of speaking to the president. The Trump campaign had not appreciated a Cavuto interview with Mitt Romney attacking Trump. Cavuto has criticized Trump’s use of Twitter and suggested he needs to show loyalty in order to receive it.
He said he’s been called an “Obama toady” for saying that former President Barack Obama improved the economy.
“I’m a numbers nerd,” Cavuto said in an interview. “He came into a meltdown and a mess, and the numbers when he got out were a lot better. You can credit him, or you can say he got lucky. But did it happen under his watch? Yeah. These are the numbers we use as business journalists to judge the success or failure of a presidency.”
Presidential interviews are often unproductive because they have a limited amount of time and are skilled at filibustering when there are subjects they want to avoid, Cavuto said.
Trump adds other complications. A study in The New York Times on Sunday said Trump had made 103 “demonstrably and substantially false statements” during his first 10 months in office, compared with 18 by Obama during his eight-year presidency.
“Any interview would require me to get clarifications on many of the president’s own statements,” Cavuto said. “I could conceivably be spending half the allotted time just trying to have him explain his saying this is the largest tax cut in history when it isn’t or that he inherited the biggest economic mess ever when he didn’t. Just trying to set the record straight, I’d run straight into a wall and the interview would be over.”
I’d agree with Cavuto: unless you can get the time to demonstrate that a lying liar is lying, or challenge them substantially, there isn’t much point in the exercise. Multiple interviews with Trump show that normal discourse just doesn’t work.
link to this extract
There’s a new leader in the sweepstakes for the zaniest name change in the crypto craze.
Long Island Iced Tea Corp. shares rose 238% after the company rebranded itself Long Blockchain Corp. It’s the latest in a near-daily phenomenon sweeping the stock market, where obscure microcap companies reorient to focus on some aspect of the mania sparked by bitcoin’s 1,600% rally this year.
Long Blockchain, whose business has been selling non-alcoholic beverages, says it will now seek to partner with or invest in companies that develop the decentralized ledgers known as blockchain, the technology that underpins bitcoin.
As with many of the recently christened crypto companies — a list that includes former makers of juice, sports bras and sofas — Long Blockchain so far has little to show for its aspirations. It has no agreements with any blockchain firms, and says “there is no assurance that a definitive agreement with these, or any other entity, will be entered into or ultimately consummated.”
I think they mean the shares tripled and that bitcoin has gone up 15-fold this year, but with percentages over 199, who knows?
Also this is completely redolent of the dot-com bubble when if you didn’t dot-com, then don’t-come to the stock market.
link to this extract
Maybe you’re one of those people who care enough about the security and privacy of your computer that you enable the facial recognition feature built into versions of Windows 10, but find it too much of a pain to set up a password.
If so, you’re potentially at risk of having your computer unlocked by an attacker holding a modified low resolution laser-printed photograph of you in front of your webcam.
As described on the Full Disclosure mailing list, a team of German penetration testers discovered it was all too easy to trick a locked Windows 10 system into letting them login using a “modified printed photo of an authorised user.”
Windows Hello is a feature currently only shipping in Windows 10, allowing PCs with the necessary hardware to use special imaging techniques to let you sign in with just a look.
The researchers tested the spoofing attack against a Dell Latitude E7470 laptop running Windows 10 Pro (Version 1703) with a Windows Hello compatible webcam, and against a Microsoft Surface Pro 4 device running Windows 10 Pro (Version 1607) with a built-in camera.
Microsoft has included updates to try to stop this in its October update, but you’d ideally set up your facial authentication all over again.
Presently, this does leave Apple’s iPhone X as the only one where I haven’t seen video of two different non-twin adults unlocking someone else’s phone using facial recognition.
link to this extract
I wanted to understand how many clicks per search are happening each month, so I made a new chart that illustrates that trend:
The metric of “clicks / ten search queries” helps us avoid seasonality biases and look instead at the rate of queries that lead to traffic opportunity. Here, the reality is sobering.
• The high point was the first month of the graph, November 2015
• Since then, there have been two significant declines in organic clicks/query (12/2015 and 11/2016) and one significant decline in paid clicks/query (01/2016, though it’s now nearly recovered)
• The 5.41 clicks/10 queries in October, 2017 is 23% lower than the 6.97 clicks/10 queries we had back in November, 2015. That’s a lot of lost SEO opportunity
• I haven’t yet tried to tie the drops back to noted changes in the SERPs, but I suspect the growth in featured snippets, instant answers, and knowledge panels in the results are at least partially responsible
• The growth of search volume has made up for much of the lost click opportunities, but this is a tough trend chart to see as an SEO
• That said, SEO still gets ~20X more traffic than PPC, and it doesn’t cost anything close to as much, so there’s still a massive advantage to ranking organically.
My conclusion from this — we’re living in a world with slightly less SEO opportunity and a trendline over the last couple years that worries and frustrates me.
There’s also a really interesting graph of “no-click seaches” for mobile v desktop (ie, someone does a query but then doesn’t visit a result) which indicates that Google changed something in November 2016 to dramatically increase those numbers on mobile, but not desktop.
link to this extract
Errata, corrigenda and ai no corrida: none notified