Start Up No.2049: AI’s watermarking problem, how Tesla suppressed range complaints, global boiling is here, chatbots ahoy!, and more


Sales of foldable phones (like Samsung’s Galaxy Fold) might hit 100m annually by 2027 – if Apple joins in soon. Don’t hold your breath, though. CC-licensed photo by HS You 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.


It’s Friday, so there’s another post due at the Social Warming Substack at about 0845 UK time.


A selection of 10 links for you. Not origami. I’m @charlesarthur on Twitter. On Mastodon: https://newsie.social/@charlesarthur. Observations and links welcome.


Big AI won’t stop election deepfakes with watermarks • WIRED

Vittoria Elliott:

»

Last week the major AI companies, including OpenAI, Google, Microsoft, and Amazon, promised the US government that they would try to mitigate the harms that could be caused by their technologies. But it’s unlikely to stem the coming tide of AI-generated content and the confusion that it could bring.

The White House says the companies’ “voluntary commitment” includes “developing robust technical mechanisms to ensure that users know when content is AI generated, such as a watermarking system,” as part of the effort to prevent AI from being used for “fraud and deception.”

But experts who spoke to WIRED say the commitments are half measures. “There’s not going to be a really simple yes or no on whether something is AI-generated or not, even with watermarks,” says Sam Gregory, program director at the nonprofit Witness, which helps people use technology to promote human rights.

Watermarking is commonly used by picture agencies and newswires to prevent images from being used without permission—and payment.

But when it comes to the variety of content that AI can generate, and the many models that already exist, things get more complicated. As of yet, there is no standard for watermarking, meaning that each company is using a different method. Dall-E, for instance, uses a visible watermark (and a quick Google search will find you many tutorials on how to remove it), whereas other services might default to metadata, or pixel-level watermarks that are not visible to users. While some of these methods might be hard to undo, others, like visual watermarks, can sometimes become ineffective when an image is resized.

“There’s going to be ways in which you can corrupt the watermarks,” Gregory says.

«

Yes, this is the obvious problem. Does no watermark mean no problem? Of course not. Answers? None so far.
unique link to this extract


Meta’s Threads needs a policy for election disinformation, voting groups say • NPR

Dara Kerr:

»

“If you have that many people, you have a great responsibility to the people that are on the platform,” said Andrea Hailey, CEO of Vote.org. “What we’re asking for here is a real plan, knowing that we’re only a few months out from presidential primaries, and that very soon the presidential election will be on our doorstep.”

The voting rights groups say they have cause for concern. During the past few elections, disinformation involving voter registration, polling places and political candidates was rampant on social media. In 2018, the Cambridge Analytica scandal exposed how that company used Facebook to target and manipulate swing voters. And in 2020, mentions of “stolen election” and “voter fraud” skyrocketed after President Joe Biden won the presidency.

“Misinformation, like social media itself, has gotten considerably more sophisticated,” said Bond Benton, communications associate professor who studies misinformation at Montclair State University. “There are ways that you can manipulate and game the system to get misinformation seen by a lot of people very rapidly. And if you’re not investing to prevent and curtail that, it’s going to find its way through.”

Meta has election disinformation policies for Facebook and Instagram, but it hasn’t published any specifically for Threads. A company spokesman told NPR that Facebook’s rules apply to Threads. So, for example, people can’t post false claims about voter registration. He also said Meta is looking at additional ways to address misinformation in future updates to the Threads app.

The voting rights groups say Threads needs a stand-alone policy. Otherwise, it’s unclear how the rules will be implemented and enforced.

«

I don’t follow how, if Meta says it will use Facebook’s rules on misinformation, that isn’t sufficient for Threads. It seems like that would cover everything – at least as far as the Facebook policy really does cover anything at all.
unique link to this extract


‘Era of global boiling has arrived,’ says UN chief as July set to be hottest month on record • The Guardian

Ajit Niranjan:

»

The era of global warming has ended and “the era of global boiling has arrived”, the UN secretary general, António Guterres, has said after scientists confirmed July was on track to be the world’s hottest month on record.

“Climate change is here. It is terrifying. And it is just the beginning,” Guterres said. “It is still possible to limit global temperature rise to 1.5C [above pre-industrial levels], and avoid the very worst of climate change. But only with dramatic, immediate climate action.”

Guterres’s comments came after scientists confirmed on Thursday that the past three weeks have been the hottest since records began and July is on track to be the hottest month ever recorded.

Global temperatures this month have shattered records, according to the World Meteorological Organization (WMO) and the EU’s Copernicus Earth observation programme, stoked by the burning of fossil fuels and spurring violent weather.

The steady rise in global average temperatures, driven by pollution that traps sunlight and acts like a greenhouse around the Earth, has made weather extremes worse.

“Humanity is in the hot seat,” Guterres told a press conference on Thursday. “For vast parts of North America, Asia, Africa and Europe, it is a cruel summer. For the entire planet, it is a disaster. And for scientists, it is unequivocal – humans are to blame.

«

Scary. Now we need a superconductor so that we can power CO2 extraction machines all over the place. Else we’re just heading rapidly to what William Gibson called “the Jackpot”. It’s not something you want to win.
unique link to this extract


“Diamagnetism is the most likely explanation” • Twitter

Alex Kaplan has been gathering reactions and analysis of LK-99, the allegedly superconducting material made by a Korean team:

»

CMTC [Condensed Matter Theory Center] probably takes the cake at predicting diamagnetism as the most likely explanation. Consider these videos of flux pinning (perfect diamagnetism) vs. moderate diamagnetism. Which one does LK-99 remind you of?

Other susceptibility data points also indicate diamagnetism.

«

Kaplan/CMTC point out too that LK-99 has higher resistance than copper, which is not really what you want from a superconductor.

Still awaiting attempted replication, but the scrutiny so far isn’t encouraging for a dramatic scientific breakthrough. (Everything has diamagnetism.) So.. no easy solution to CO2.
unique link to this extract


Millionaire crypto influencer found dismembered in suitcase • NY Post

Ben Cost:

»

Police have launched a murder investigation after the dismembered remains of missing millionaire Fernando Pérez Algaba, 41, were discovered by a group of children in Argentina over the weekend.

The grisly case came to light after the kids found a red suitcase filled with body parts while playing by a stream in the town of Ingeniero Budge, Buenos Aires Province, on Sunday, Jam Press reported.

The children’s parents notified the Buenos Aires police, who inspected the package and reportedly found the victim’s legs and forearm inside, discovering another whole arm in the stream. On Wednesday, authorities discovered the missing head and torso, El País reported.

The body parts were cleanly amputated, suggesting the work of a professional, local media reported. Meanwhile, a subsequent autopsy revealed that the victim had been shot three times before the dismemberment.

Police identified Algaba by his fingerprints and also by distinctive tattoos on the body parts. The entrepreneur had been declared missing since last Tuesday.

The businessman had accumulated millions renting luxury vehicles and selling cryptocurrency, which he frequently advertised to his 900,000 followers on Instagram. The influencer, who lived in Barcelona, Spain, had reportedly been staying in Argentina for a week prior to his alleged murder.

Police have arrested one suspect in connection with Algaba’s death. And while the motive behind the alleged murder remains unclear, authorities suspect that the influencer was killed over his numerous debts…He had reportedly run afoul of the Barra Bravas, a violent gang heavily involved in Argentina’s soccer scene, which reportedly demanded that he pay them a $40,000 loan.

“If something happens to me, everyone is already warned,” he wrote in a message.

«

The moral, sadly, isn’t too hard to infer. One can only hope that the children weren’t too badly traumatised.
unique link to this extract


Rewarding chatbots for real-world engagement with millions of users • ArXiv

A 14-strong team from Chai Research, Seamless Capital and the University of Cambridge’s Machine Intelligence Laboratory:

»

The emergence of pretrained large language models has led to the deployment of a range of social chatbots for chitchat. Although these chatbots demonstrate language ability and fluency, they are not guaranteed to be engaging and can struggle to retain users. This work investigates the development of social chatbots that prioritize user engagement to enhance retention, specifically examining the use of human feedback to efficiently develop highly engaging chatbots.

The proposed approach uses automatic pseudo-labels collected from user interactions to train a reward model that can be used to reject low-scoring sample responses generated by the chatbot model at inference time. Intuitive evaluation metrics, such as mean conversation length (MCL), are introduced as proxies to measure the level of engagement of deployed chatbots.

A/B testing on groups of 10,000 new daily chatbot users on the Chai Research platform shows that this approach increases the MCL by up to 70%, which translates to a more than 30% increase in user retention for a GPT-J 6B model.

«

It really is the world of Her, where the protagonist suddenly discovers that “his” AI has been talking to thousands, even millions of other people all the time.
unique link to this extract


Intel stock rallies on surprise earnings as PC, data-center sales beat – MarketWatch

Wallace Witkowski:

»

Intel Corp. shares surged in the extended session Thursday after the chip maker posted a surprise profit as PC and data-center sales came in better than expected.

Intel shares surged as much as 7% after hours, following a 0.6% rise to close the regular session at $34.55.

The company reported second-quarter net income of $1.48bn, or 35 cents a share, versus a loss of $454m, or 11 cents a share, in the year-ago period.

After adjusting for restructuring charges and other items, Intel reported 13 cents a share, versus net income of 29 cents a share a year ago.

Revenue fell to $12.95bn from $15.32bn in the year-ago period, and adjusted gross margins came in at 39.8% .

Intel had forecast an adjusted second-quarter loss of 4 cents a share on revenue of about $11.5bn to $12.5bn for the current period, and adjusted gross margins of about 33.2% for the quarter.

«

If you read the actual Intel press release, pretty much every number is down, apart from Foundry Services; the profit seems to have been pulled from a hat. ntel’s not out of the woods yet.
unique link to this extract


Global foldable smartphone shipments to cross 100 million by 2027 • Counterpoint Research

»

Global foldable smartphone shipments are expected to pass the 100-million mark by 2027, according to Counterpoint Research’s latest Global Foldable Smartphone Tracker and Forecast, with Samsung and Apple accounting for the biggest market share.

Commenting on growth expectations, Research Director Tom Kang said, “At the moment, foldables remain niche. But it is an important segment for brands looking to maintain leadership in innovation and a premium market presence.”

Kang added, “Samsung and the Chinese OEMs have been very active, especially in their home markets, with China emerging as the biggest market globally last year. If you want to make it in foldables, you have to make it there.”

Senior analyst Jene Park said, “In the long term, we are waiting to see what Apple does. We are looking at 2025 as the possible year of iPhone’s foldable debut, which could provide another growth spurt for the segment.”

«

Ah, so it’s a forecast that depends on Apple entering that market. Samsung would no doubt be delighted: another customer for its foldable screens. But if that timeline is going to happen, Apple would be designing such a phone now (there’s a two-year timeline), and one might expect to start hearing little noises about Samsung increasing its foldable screen capacity. So keep an eye on that. I’m sceptical: I think Apple has plenty on its plate with the Vision Pro next year, and its followup in 2025, and foldables remain unproven.
unique link to this extract


Facebook’s algorithm is ‘influential’ but doesn’t necessarily change beliefs, researchers say • The New York Times

Mike Isaac and Sheera Frenkel:

»

The algorithms powering Facebook and Instagram, which drive what billions of people see on the social networks, have been in the cross hairs of lawmakers, activists and regulators for years. Many have called for the algorithms to be abolished to stem the spread of viral misinformation and to prevent the inflammation of political divisions.

But four new studies published on Thursday — including one that examined the data of 208 million Americans who used Facebook in the 2020 presidential election — complicate that narrative.

In the papers, researchers from the University of Texas, New York University, Princeton and other institutions found that removing some key functions of the social platforms’ algorithms had “no measurable effects” on people’s political beliefs. In one experiment on Facebook’s algorithm, people’s knowledge of political news declined when their ability to reshare posts was removed, the researchers said.

At the same time, the consumption of political news on Facebook and Instagram was highly segregated by ideology, according to another study: 97% of the links to “untrustworthy” news stories on the apps during the 2020 election were read by users who identified as conservative and largely engaged with right-wing content, the research found.

The studies, which were published in the journals Science and Nature, provide a contradictory and nuanced picture of how Americans have been using — and have been affected by — two of the world’s biggest social platforms.

«

No simple solution in sight. Meanwhile, let’s try Threads! Though a Wired writeup of the studies says we prefer algorithmic feeds, actually.
unique link to this extract


Tesla’s secret team to suppress thousands of driving range complaints • Reuters

Steve Stecklow and Norihiko Shirouzu:

»

In March, Alexandre Ponsin set out on a family road trip from Colorado to California in his newly purchased Tesla, a used 2021 Model 3. He expected to get something close to the electric sport sedan’s advertised driving range: 353 miles on a fully charged battery.

He soon realized he was sometimes getting less than half that much range, particularly in cold weather – such severe underperformance that he was convinced the car had a serious defect.

“We’re looking at the range, and you literally see the number decrease in front of your eyes,” he said of his dashboard range meter.

Ponsin contacted Tesla and booked a service appointment in California. He later received two text messages, telling him that “remote diagnostics” had determined his battery was fine, and then: “We would like to cancel your visit.”

What Ponsin didn’t know was that Tesla employees had been instructed to thwart any customers complaining about poor driving range from bringing their vehicles in for service. Last summer, the company quietly created a “Diversion Team” in Las Vegas to cancel as many range-related appointments as possible.

The Austin, Texas-based electric carmaker deployed the team because its service centers were inundated with appointments from owners who had expected better performance based on the company’s advertised estimates and the projections displayed by the in-dash range meters of the cars themselves, according to several people familiar with the matter.

Inside the Nevada team’s office, some employees celebrated canceling service appointments by putting their phones on mute and striking a metal xylophone, triggering applause from coworkers who sometimes stood on desks. The team often closed hundreds of cases a week and staffers were tracked on their average number of diverted appointments per day.

Managers told the employees that they were saving Tesla about $1,000 for every canceled appointment, the people said. Another goal was to ease the pressure on service centers, some of which had long waits for appointments.

«

unique link to this extract


• Why do social networks drive us a little mad?
• Why does angry content seem to dominate what we see?
• How much of a role do algorithms play in affecting what we see and do online?
• What can we do about it?
• Did Facebook have any inkling of what was coming in Myanmar in 2016?

Read Social Warming, my latest book, and find answers – and more.


Errata, corrigenda and ai no corrida: none notified

Leave a comment

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