Last week the Guardian reported on "thousands of AI workers contracted for Google through Japanese conglomerate Hitachi's GlobalLogic to rate and moderate the output of Google's AI products, including its flagship chatbot Gemini... and its summaries of search results, AI Overviews." "AI isn't magic; it's a pyramid scheme of human labor," said Adio Dinika, a researcher at the Distributed AI Research Institute based in Bremen, Germany. "These raters are the middle rung: invisible, essential and expendable...." Ten of Google's AI trainers the Guardian spoke to said they have grown disillusioned with their jobs because they work in siloes, face tighter and tighter deadlines, and feel they are putting out a product that's not safe for users... In May 2023, a contract worker for Appen submitted a letter to the US Congress that the pace imposed on him and others would make Google Bard, Gemini's predecessor, a "faulty" and "dangerous" product
This week Google laid off 200 of those moderating contractors, reports Wired. "These workers, who often are hired because of their specialist knowledge, had to have either a master's or a PhD to join the super rater program, and typically include writers, teachers, and people from creative fields." Workers still at the company claim they are increasingly concerned that they are being set up to replace themselves. According to internal documents viewed by WIRED, GlobalLogic seems to be using these human raters to train the Google AI system that could automatically rate the responses, with the aim of replacing them with AI. At the same time, the company is also finding ways to get rid of current employees as it continues to hire new workers. In July, GlobalLogic made it mandatory for its workers in Austin, Texas, to return to office, according to a notice seen by WIRED...

Some contractors attempted to unionize earlier this year but claim those efforts were quashed. Now they allege that the company has retaliated against them. Two workers have filed a complaint with the National Labor Relations Board, alleging they were unfairly fired, one due to bringing up wage transparency issues, and the other for advocating for himself and his coworkers. "These individuals are employees of GlobalLogic or their subcontractors, not Alphabet," Courtenay Mencini, a Google spokesperson, said in a statement...
"Globally, other AI contract workers are fighting back and organizing for better treatment and pay," the article points out, noting that content moderators from around the world facing similar issues formed the Global Trade Union Alliance of Content Moderators which includes workers from Kenya, Turkey, and Colombia.

Thanks to long-time Slashdot reader mspohr for sharing the news.

When Meta finally unveiled its newest smart glasses, CEO Mark Zuckerberg "drew more snickers than applause," wrote the New York Times. (Mashable points out a video call failing onstage followed by an unsuccessful recipe demonstration.)

Meta chief technology officer Andrew Bosworth later explained the funny reason their demo didn't work, reports TechCrunch, while answering questions on Instagram: "When the chef said, 'Hey, Meta, start Live AI,' it started every single Ray-Ban Meta's Live AI in the building. And there were a lot of people in that building," Bosworth explained. "That obviously didn't happen in rehearsal; we didn't have as many things," he said, referring to the number of glasses that were triggered... The second part of the failure had to do with how Meta had chosen to route the Live AI traffic to its development server to isolate it during the demo. But when it did so, it did this for everyone in the building on the access points, which included all the headsets. "So we DDoS'd ourselves, basically, with that demo," Bosworth added... Meta's dev server wasn't set up to handle the flood of traffic from the other glasses in the building — Meta was only planning for it to handle the demos alone.

The issue with the failed WhatsApp call, on the other hand, was the result of a new bug. The smart glasses' display had gone to sleep at the exact moment the call came in, Bosworth said. When Zuckerberg woke the display back up, it didn't show the answer notification to him. The CTO said this was a "race condition" bug... "We've never run into that bug before," Bosworth noted. "That's the first time we'd ever seen it. It's fixed now, and that's a terrible, terrible place for that bug to show up." He stressed that, of course, Meta knows how to handle video calls, and the company was "bummed" about the bug showing up here... "It really was just a demo fail and not, like, a product failure," he said.
Thanks to Slashdot reader fjo3 for sharing the news.

The consumer advocacy nonprofit PIRG (Public Interest Research Group) is now petitioning Microsoft to reconsider pulling support for Windows 10 in 2025, since "as many as 400 million perfectly good computers that can't upgrade to Windows 11 will be thrown out." In a petition addressed to Microsoft CEO Satya Nadella, the group warned the October 14 end of free support could cause "the single biggest jump in junked computers ever, and make it impossible for Microsoft to hit their sustainability goals." About 40% of PCs currently in use can't upgrade to Windows 11, even if users want to... Less than a quarter of electronic waste is recycled, so most of those computers will end up in landfills.
Consumer Reports recently also urged Microsoft to not to "strand millions of customers.". And now more groups are also pushing back, according to a post from the blog Windows: Central The Restart Project co-developed the "End of 10" toolkit, which is designed to support Windows 10 users who can't upgrade to Windows 11 after the operating system hits its end-of-support date.
They also note that a Paris-based company called Back Market plans to sell Windows 10 laptops refurbished with Ubuntu Linux or ChromeOS Flex. ("We refuse to watch hundreds of millions of perfectly good computers end up in the trash as e-waste," explains their web site.) Back Market's ad promises an "up-to-date, secure operating system — so instead of paying for a new computer you don't need, you can help us give this one a brand new life."

Right now Windows 10 holds 71.9% of Microsoft's market share, with Windows 11 at 22.95%, according to figures from StatCounter cited by the blog Windows Central. And HP and Dell "recently indicated that half of the global PCs are still running Windows 10," according to another Windows Central post...

A large U.S.-made bomb left over from World War II was discovered at a construction site, reports the Associated Press: Police said the bomb was 1.5 meters (nearly 5 feet) in length and weighed about 1,000 pounds (450 kilograms). It was discovered by construction workers in Quarry Bay, a bustling residential and business district on the west side of Hong Kong island... [A police official] said that because of "the exceptionally high risks associated with its disposal," approximately 1,900 households involving 6,000 individuals were "urged to evacuate swiftly." The operation to deactivate the bomb began late Friday and lasted until around 11:30 a.m. Saturday. No one was injured in the operation.

Bombs left over from World War II are discovered from time to time in Hong Kong. The city was occupied by Japanese forces during the war, when it became a base for the Japanese military and shipping. The United States, along with other Allied forces, targeted Hong Kong in air raids to disrupt Japanese supply lines and infrastructure.
"Bombs from the war have triggered evacuations and emergency measures around the globe in recent months," reports CBS News: Earlier this month, a 500-pound bomb was discovered in Slovakia's capital during construction work, prompting evacuations. In August, large parts of Dresden, Germany, were evacuated so experts could defuse an unexploded World War II bomb found during clearance work for a collapsed bridge. In June, over 20,000 people were evacuated from Cologne after three unexploded U.S. bombs from the war were found... In March, a World War II bomb was found near the tracks of Paris' Gare du Nord station. In February, more than 170 bombs were found near a children's playground in northern England. And in October 2024, a World War II bomb exploded at a Japanese airport.

"A cyberattack targeting check-in and boarding systems disrupted air traffic and caused delays at several of Europe's major airports on Saturday," reports the Associated Press.

"While the impact on travelers appeared to be limited, experts said the intrusion exposed vulnerabilities in security systems." The disruptions to electronic systems initially reported at Brussels, Berlin's Brandenburg and London's Heathrow airports meant that only manual check-in and boarding was possible. Many other European airports said their operations were unaffected... Airports said the issue centered around a provider of check-in and boarding systems — not airlines or the airports themselves. Collins Aerospace, whose systems help passengers check themselves in, print boarding passes and bag tags and dispatch their luggage from a kiosk, cited a "cyber-related disruption" to its MUSE (Multi-User System Environment) software at "select airports."
Brussels Airport initially reported a "large impact" on flight schedules," according to the article, with a spokesperson telling broadcaster VTM that by mid-morning nine flights had been canceled, with four more redirected to another airport and 15 delayed an hour or more. The airport later told Reuters there were "delays on most of the departing flights."

Reuters notes it's "the latest in a string of hacks targeting governments and companies across the world, hitting sectors from healthcare and defence to retail and autos.: A recent breach at luxury carmaker Jaguar Land Rover brought its production to a halt...

At Heathrow, Berlin and Brussels, 29 flight departures and arrivals had been cancelled as of 1130 GMT, aviation data provider Cirium said. In total, 651 departures were scheduled from Heathrow, 228 from Brussels and 226 from Berlin on Saturday... Brussels Airport said it had asked airlines to cancel half of their scheduled departing flights on Sunday to avoid long queues and late cancellations, signalling that the disruption would continue through the weekend.

A European Commission spokesperson said there were currently no indications of a "widespread or severe attack" and that the origin of the incident was still under investigation.

The Leon Thevenin, Africa's only permanently stationed cable repair ship, maintains over 60,000 kilometers of undersea internet infrastructure from Madagascar to Ghana. The 43-year-old vessel employs a 60-person crew who perform precision repairs on fiber-optic cables that carry data for Alphabet, Meta, and Amazon -- companies that consumed 3.6 billion megabits per second of bandwidth in 2023.

Operating costs range from $70,000 to $120,000 daily, according to owner Orange Marine. The ship has experienced increased demand due to unusual underwater landslides in the Congo Canyon causing frequent cable breaks. Cable jointer Shuru Arendse and his team spend up to 48 hours on repairs that require fusing hair-thin glass fibers in conditions where a speck of dust can ruin the joint. The vessel gained Starlink connectivity last year after decades of relying on satellite phones and shared computers for crew communication. Sixty-two cable repair ships operate globally to maintain the infrastructure supporting streaming media and AI applications.

Meta is moving to break into the wholesale power-trading business to better manage the massive electricity needs of its data centers. Bloomberg: The company, which owns Facebook, filed an application with US regulators this week seeking authorization to do so. A Meta representative said it was a natural next step to participate in energy markets as it looks to power operations with clean energy.

Buying electricity has become an increasingly urgent challenge for technology companies including Meta, Microsoft and Alphabet's Google. They're all racing to develop more advanced artificial intelligence systems and tools that are notoriously resource-intensive. Amazon, Google and Microsoft are already active power traders, according to filings with US regulators.

Chinese consumer tech giant Xiaomi will remotely fix a flaw in the assisted driving system on over 110,000 of its popular SU7 electric cars, the firm and regulators said Friday, months after a deadly crash involving the model. From a report: China's tech companies and automakers have poured billions of dollars into smart-driving technology, a new battleground in the country's cutthroat domestic car market. But Beijing has moved to tighten safety rules after a Xiaomi SU7 in assisted driving mode crashed and killed three college students this year. It also raised concerns over the advertising of cars as being capable of autonomous driving. On Friday, the State Administration for Market Regulation said Xiaomi's highway assisted driving system showed insufficient recognition, warning and handling ability in some extreme driving conditions.

Google has added Gemini features to Chrome for all desktop users in the US browsing in English following a limited release to paying subscribers in May. The update introduces a Gemini button in the browser that launches a chatbot capable of answering questions about page content and synthesizing information from multiple tabs. Users can remove the Gemini sparkle icon from Chrome's interface.

Google will add its AI Mode search feature to Chrome's address bar before September ends. The feature will suggest prompts based on webpage content but won't replace standard search functionality. Chrome on Android already includes Gemini features. The company plans to add agentic capabilities in coming months that would allow Gemini to perform tasks like adding items to online shopping carts by controlling the browser cursor.

An anonymous reader quotes a report from Reuters: On the outskirts of this city of 21 million, a showroom in a shopping mall offers extraordinary deals on new cars. Visitors can choose from some 5,000 vehicles. Locally made Audis are 50% off. A seven-seater SUV from China's FAW is about $22,300, more than 60% below its sticker price. These deals -- offered by a company called Zcar, which says it buys in bulk from automakers and dealerships -- are only possible because China has too many cars. Years of subsidies and other government policies have aimed to make China a global automotive power and the world's electric-vehicle leader. Domestic automakers have achieved those goals and more -- and that's the problem.

China has more domestic brands making more cars than the world's biggest car market can absorb because the industry is striving to hit production targets influenced by government policy, instead of consumer demand, a Reuters examination has found. That makes turning a profit nearly impossible for almost all automakers here, industry executives say. Chinese electric vehicles start at less than $10,000; in the U.S., automakers offer just a few under $35,000. Most Chinese dealers can't make money, either, according to an industry survey published last month, because their lots are jammed with excess inventory. Dealers have responded by slashing prices. Some retailers register and insure unsold cars in bulk, a maneuver that allows automakers to record them as sold while helping dealers to qualify for factory rebates and bonuses from manufacturers.

Unwanted vehicles get dumped onto gray-market traders like Zcar. Some surface on TikTok-style social-media sites in fire sales. Others are rebranded as "used" -- even though their odometers show no mileage -- and shipped overseas. Some wind up abandoned in weedy car graveyards. These unusual practices are symptoms of a vastly oversupplied market -- and point to a potential shakeout mirroring turmoil in China's property market and solar industry, according to many industry figures and analysts. They stem from government policies that prioritize boosting sales and market share -- in service of larger goals for employment and economic growth -- over profitability and sustainable competition. Local governments offer cheap land and subsidies to automakers in exchange for production and tax-revenue commitments, multiplying overcapacity across the country.

prisoninmate shares a report from 9to5Linux: The GNOME Project released today GNOME 49 "Brescia" as the latest stable version of this widely used desktop environment for GNU/Linux distributions, a major release that introduces exciting new features. Highlights of GNOME 49 include a new "Do Not Disturb" toggle in Quick Settings, a dedicated Accessibility menu in the login screen, support for handling unknown power profiles in the Quick Settings menu, support for YUV422 and YUV444 (HDR) color spaces, support for passive screen casts, and support for async keyboard map settings.

GNOME 49 also introduces support for media controls, restart and shutdown actions on the lock screen, support for dynamic users for greeter sessions in the GNOME Display Manager (GDM), and support for per-monitor brightness sliders in Quick Settings on multi-monitor setups. For a full list of changes, check out the release notes.

Two Xpeng AeroHT flying cars collided during a rehearsal for the Changchun Air Show in China, with one vehicle catching fire upon landing. While the company reported no serious injuries, CNN reported one person was injured in the crash. The BBC reports: Footage on Chinese social media site Weibo appeared to show a flaming vehicle on the ground which was being attended to by fire engines. One vehicle "sustained fuselage damage and caught fire upon landing," Xpeng AeroHT said in a statement to CNN. "All personnel at the scene are safe, and local authorities have completed on-site emergency measures in an orderly manner," it added.

The electric flying cars take off and land vertically, and the company is hoping to sell them for around $300,000 each. In January, Xpeng claimed to have around 3,000 orders for the vehicle. [...] It has said it wants to lead the world in the "low-altitude economy."

China's internet regulator has told the country's biggest technology companies to stop buying all of Nvidia's artificial intelligence chips and terminate their existing orders, as Beijing steps up efforts to boost its homegrown semiconductor industry and compete with the US. From a report: The Cyberspace Administration of China (CAC) informed companies including ByteDance and Alibaba this week to terminate their testing and orders of the RTX Pro 6000D, Nvidia's tailor-made product for the country introduced two months ago, according to three people with knowledge of the matter.

Several companies had indicated they would order tens of thousands of the RTX Pro 6000D, and had started testing and verification work with Nvidia's server suppliers before telling them to stop the work after receiving the CAC order, said the people. Nvidia CEO responds: In response to a question on the FT report, Huang said Wednesday that "we can only be in service of a market if the country wants us to be."

"We probably contributed more to the China market than most countries have. And I'm disappointed with what I see," Huang said. "But they have larger agendas to work out between China and the United States, and I'm understanding of that."

It comes after a tumultuous few years for Nvidia's business in China, which Huang described as "a bit of a rollercoaster."

"We've guided all financial analysts not to include China" in financial forecasts, Huang told reporters Wednesday at a press briefing in London. "The reason for that is because that's largely going to be within the discussions of the United States government and Chinese government."

BrianFagioli shares a report from NERDS.xyz: The Fedora Project has announced Fedora Linux 43 Beta, giving users and developers the opportunity to test the distribution ahead of its final release. This beta introduces improvements across installation, system tools, and programming languages while continuing Fedora's pattern of cleaning out older components. The beta can be downloaded in Workstation, KDE Plasma, Server, IoT, and Cloud editions. Spins and Labs are also available, though Mate and i3 are not provided in some builds. Existing systems can be upgraded with DNF system-upgrade. Fedora CoreOS will follow one week later through its "next" stream. The beta brings enhancements to its Anaconda WebUI, moves to Python 3.14, and supports Wayland-only GNOME, among many other changes. A full list of improvements and system enhancements can be found here.

The official release should be available in late October or early November.

Consumer Reports has urged Microsoft to keep supporting Windows 10 beyond its October 2025 cutoff, saying the move will "strand millions of consumers" who have machines incompatible with Windows 11. The Verge reports: As noted by Consumer Reports, data suggests that around 46.2 percent of people around the world still use Windows 10 as of August 2025, while around 200 to 400 million PCs can't be upgraded to Windows 11 due to missing hardware requirements.

In the letter, Consumer Reports calls Microsoft "hypocritical" for urging customers to upgrade to Windows 11 to bolster cybersecurity, but then leaving Windows 10 devices susceptible to cyberattacks. It also calls out the $30 fee Microsoft charges customers for "a mere one-year extension to preserve their machine's security," as well as the free support options that force people to use Microsoft products, allowing the company to "eke out a bit of market share over competitors."

Consumer Reports asks that Microsoft continue providing support for Windows 10 computers for free until more people have upgraded to Windows 11.

An anonymous reader quotes a report from Ars Technica: Verizon agreed to offer $20-per-month broadband service to people with low incomes in California in exchange for a merger approval. In a bid to complete its $9.6 billion purchase of Frontier Communications, Verizon committed to offering $20 fiber-to-the-home service with symmetrical speeds of 300Mbps. Verizon also committed to offering a $20 fixed wireless service with download speeds of 100Mbps and upload speeds of 20Mbps. Verizon would be required to offer the plans for at least 10 years, according to a joint motion (PDF) to approve the settlement agreement. After three years, Verizon would need to "make commercially reasonable efforts" to increase the speeds "while retaining the $20 price point."

The joint motion filed by Verizon and the California Public Advocates Office seeks approval from the California Public Utilities Commission (CPUC). The $20 plans would be available to people who meet income eligibility guidelines and can be paired with Lifeline discounts. "My team required those options to be California Lifeline eligible, which effectively makes it free for low-income Californians throughout the state," wrote Ernesto Falcon, a program manager at the Public Advocates Office. California's Lifeline program provides $19 discounts. Falcon also wrote that the settlement would expand fiber deployment beyond what Frontier would have offered on its own. "If the merger is approved, Verizon will deliver 75,000 new fiber-to-the-home connections in California beyond Frontier's entire buildout plan with a priority for low-income households," he wrote. The deal also requires 250 new cell sites for Verizon's 5G network.

Waymo is now permitted to test its robotaxi service at San Francisco International Airport (SFO), a big win for the company as it seeks to expand its service area and tackle more popular, revenue-generating destinations. From a report: After years of back-and-forth negotiations, Waymo signed "Testing and Operations Pilot Permit" with SFO, San Francisco Mayor Daniel Lurie said in a release. Under the agreement, Waymo will roll out its service to SFO in three phases, including testing vehicles with a human driver, testing without a driver, and eventually beginning commercial service.

Waymo will start its tests with employees before eventually inviting members of the public to take trips to and from the airport. Pickups and dropoffs will initially take place at SFO's Kiss & Fly lot, which is accessible to the terminals via the AirTrain.

Nature's editorial board urged governments on Tuesday to reverse decades of rail disinvestment as railways mark their 200th anniversary September 27, citing transport sector emissions that grew 1.7% annually from 1990-2022 and now generate one-quarter of global CO2. Rail produces one-fifth the emissions of cars per passenger kilometer yet carries just 8.4% of EU passenger traffic versus 73% for automobiles.

The journal called for broader investment criteria beyond narrow profitability metrics and noted only one-third of countries have incorporated transport into their Paris Agreement commitments. Global rail freight fell from 38% to 24% between 1980-2017 while US networks shrank from 400,000 to 200,000 kilometers since 1914. Africa operates 87,000 rail kilometers continent-wide compared to India's 65,000 kilometers in one-tenth the area. Transport emissions must decline 3% yearly to meet net-zero targets.

An anonymous reader quotes a report from Ars Technica: The companies seeking to build larger AI models have been increasingly stymied by a lack of high-quality training data. As tech firms scour the web for more data to feed their models, they could increasingly rely on potentially sensitive user data. A team at Google Research is exploring new techniques to make the resulting large language models (LLMs) less likely to 'memorize' any of that content. LLMs have non-deterministic outputs, meaning you can't exactly predict what they'll say. While the output varies even for identical inputs, models do sometimes regurgitate something from their training data -- if trained with personal data, the output could be a violation of user privacy. In the event copyrighted data makes it into training data (either accidentally or on purpose), its appearance in outputs can cause a different kind of headache for devs. Differential privacy can prevent such memorization by introducing calibrated noise during the training phase.

Adding differential privacy to a model comes with drawbacks in terms of accuracy and compute requirements. No one has bothered to figure out the degree to which that alters the scaling laws of AI models until now. The team worked from the assumption that model performance would be primarily affected by the noise-batch ratio, which compares the volume of randomized noise to the size of the original training data. By running experiments with varying model sizes and noise-batch ratios, the team established a basic understanding of differential privacy scaling laws, which is a balance between the compute budget, privacy budget, and data budget. In short, more noise leads to lower-quality outputs unless offset with a higher compute budget (FLOPs) or data budget (tokens). The paper details the scaling laws for private LLMs, which could help developers find an ideal noise-batch ratio to make a model more private. The work the team has done here has led to a new Google model called VaultGemma, its first open-weight model trained with differential privacy to minimize memorization risks. It's built on the older Gemma 2 foundation and sized at 1 billion parameters, which the company says performs comparably to non-private models of similar size.

It's available now from Hugging Face and Kaggle.

Bruce66423 shares a report from The Guardian: MI5 has conceded it "unlawfully" obtained the communications data of a former BBC journalist, in what was claimed to be an unprecedented admission from the security services. The BBC said it was a "matter of grave concern" that the agency had obtained communications data from the mobile phone of Vincent Kearney, a former BBC Northern Ireland home affairs correspondent. The admission came in a letter to the BBC and to Kearney, in relation to a tribunal examining claims that several reporters in Northern Ireland were subjected to unlawful scrutiny by the police. It related to work carried out by Kearney for a documentary into the independence of the Office of the Police Ombudsman for Northern Ireland (PONI). Kearney is now the northern editor at Irish broadcaster RTE.

In documents submitted to the Investigatory Powers Tribunal (IPT), MI5 conceded it obtained phone data from Kearney on two occasions in 2006 and 2009. Jude Bunting KC, representing Kearney and the BBC, told a hearing on Monday: "The MI5 now confirms publicly that in 2006 and 2009 MI5 obtained communications data in relation to Vincent Kearney." He said the security service accepted it had breached Kearney's rights under article 8 and article 10 of the European convention on human rights. They relate to the right to private correspondence and the right to impart information without interference from public authorities. "This appears to be the first time in any tribunal proceedings in which MI5 publicly accept interference with a journalist's communications data, and also publicly accept that they acted unlawfully in doing so," Bunting said. He claimed the concessions that it accessed the journalist's data represented "serious and sustained illegality on the part of MI5." Bruce66423 comments: "The good news is that it's come out. The bad news is that it has taken 16 years to do so. The interesting question is whether there will be any meaningful consequences for individuals within MI5; there's a nice charge of 'malfeasance in public office' that can be used to get such individuals into a criminal court. Or will the outcome be like that of when the CIA hacked the US Senate's computers, lied about it, and nothing happened?"