Threads is testing a new feature that lets users hide spoiler content by blurring images or text, which can then be revealed with a tap. The Verge reports: Meta spokesperson Alec Booker told The Verge that this is a "global test," though it's not clear how many people will gain access to it. Spoilers will also look a bit different depending on which device you're using. On desktop, spoilers are hidden by a gray block, but they appear behind a bunch of floating dots on mobile (which you can see in the GIF embedded [here]). "This feature is currently optimized for mobile, but we're working to improve the experience for desktop," Booker said.

A new Salesforce-led study found that LLM-based AI agents struggle with real-world CRM tasks, achieving only 58% success on simple tasks and dropping to 35% on multi-step ones. They also demonstrated poor confidentiality awareness. "Agents demonstrate low confidentiality awareness, which, while improvable through targeted prompting, often negatively impacts task performance," a paper published at the end of last month said. The Register reports: The Salesforce AI Research team argued that existing benchmarks failed to rigorously measure the capabilities or limitations of AI agents, and largely ignored an assessment of their ability to recognize sensitive information and adhere to appropriate data handling protocols.

The research unit's CRMArena-Pro tool is fed a data pipeline of realistic synthetic data to populate a Salesforce organization, which serves as the sandbox environment. The agent takes user queries and decides between an API call or a response to the users to get more clarification or provide answers.

"These findings suggest a significant gap between current LLM capabilities and the multifaceted demands of real-world enterprise scenarios," the paper said. [...] AI agents might well be useful, however, organizations should be wary of banking on any benefits before they are proven.

An anonymous reader quotes a report from TechCrunch: While Silicon Valley executives like those from Palantir, Meta, and OpenAI are grabbing headlines for trading their Brunello Cucinelli vests for Army Reserve uniforms, a quieter transformation has been underway in the U.S. Navy. How so? Well, the Navy's chief technology officer, Justin Fanelli, says he has spent the last two and a half years cutting through the red tape and shrinking the protracted procurement cycles that once made working with the military a nightmare for startups. The efforts represent a less visible but potentially more meaningful remaking that aims to see the government move faster and be smarter about where it's committing dollars.

"We're more open for business and partnerships than we've ever been before," Fanelli told TechCrunch in a recent episode of StrictlyVC Download. "We're humble and listening more than before, and we recognize that if an organization shows us how we can do business differently, we want that to be a partnership." Right now, many of these partnerships are being facilitated through what Fanelli calls the Navy's innovation adoption kit, a series of frameworks and tools that aim to bridge the so-called Valley of Death, where promising tech dies on its path from prototype to production. "Your granddaddy's government had a spaghetti chart for how to get in," Fanelli said. "Now it's a funnel, and we are saying, if you can show that you have outsized outcomes, then we want to designate you as an enterprise service."

In one recent case, the Navy went from a Request for Proposal (RFP) to pilot deployment in under six months with Via, an eight-year-old, Somerville, Massachusetts-based cybersecurity startup that helps big organizations protect sensitive data and digital identities through, in part, decentralization, meaning the data isn't stored in one central spot that can be hacked. (Another of Via's clients is the U.S. Air Force.) The Navy's new approach operates on what Fanelli calls a "horizon" model, borrowed and adapted from McKinsey's innovation framework. Companies move through three phases: evaluation, structured piloting, and scaling to enterprise services. The key difference from traditional government contracting, Fanelli says, is that the Navy now leads with problems rather than predetermined solutions. "Instead of specifying, 'Hey, we'd like this problem solved in a way that we've always had it,' we just say, 'We have a problem, who wants to solve this, and how will you solve it?'" Fanelli said.

Thousands of American investors have lost millions of dollars to sophisticated pump-and-dump schemes involving small Chinese companies listed on Nasdaq, prompting the Justice Department to declare the fraud a priority under the Trump administration's white-collar enforcement program.

The scams recruit victims through social media ads and WhatsApp messages, directing them to purchase shares in obscure Chinese firms whose stock prices are artificially inflated before collapsing. Since 2020, nearly 60 China-based companies have conducted initial public offerings on Nasdaq raising $15 million or less each, with more than one-third experiencing sudden single-day price drops exceeding 50%. In one recent case, seven traders earned over $480 million by defrauding 600 victims who purchased shares in China Liberal Education Holdings.

Tensions between OpenAI and Microsoft over the future of their famed AI partnership are flaring up. WSJ, minutes ago: OpenAI wants to loosen Microsoft's grip on its AI products and computing resources, and secure the tech giant's blessing for its conversion into a for-profit company. Microsoft's approval of the conversion is key to OpenAI's ability to raise more money and go public.

But the negotiations have been so difficult that in recent weeks, OpenAI's executives have discussed what they view as a nuclear option: accusing Microsoft of anticompetitive behavior during their partnership, people familiar with the matter said. That effort could involve seeking federal regulatory review of the terms of the contract for potential violations of antitrust law, as well as a public campaign, the people said.

Researchers are cautioning users against clicking unsubscribe links embedded in email bodies, citing new data showing such actions can expose recipients to malicious websites and confirm active email addresses to attackers. DNSFilter found that one in every 644 clicks on unsubscribe links leads users to potentially malicious websites.

"You've left the safe, structured environment of your email client and entered the open web," TK Keanini, DNSFilter's chief technology officer, told WSJ. The risks range from confirming to bad actors that an email address belongs to an active user to redirecting victims to fake websites designed to steal login credentials or install malware. Clicking such links "can make you a bigger target in the future," said Michael Bargury, CTO of security company Zenity.

A Dutch court on Monday confirmed a 2021 consumer watchdog's ruling saying that Apple had abused its dominant position by imposing unfair conditions on providers of dating apps in the App Store. From a report: The Rotterdam District Court ruled that the Dutch Authority for Consumers and Markets (ACM) was therefore right to impose an order subject to a penalty for non-compliance. The court ruled that ACM was right in finding that dating app providers had to use Apple's own payment system, were not allowed to refer to payment options outside the App Store, and had to pay a 30% commission (15% for small providers) to Apple.

Microsoft has disabled Windows Hello's ability to authenticate users in low-light environments through a recent security update that now requires both infrared sensors and color cameras to verify faces. The change forces the system to see a visible face through the webcam before completing authentication with IR sensors.

Windows Hello earlier relied solely on infrared sensors to create 3D facial scans, allowing the feature to work in complete darkness similar to iPhone's Face ID. Microsoft pushed the dual-camera requirement to address a spoofing vulnerability in the biometric system.

An anonymous reader shares a report: Japan is the latest nation hoping to tempt disgruntled US researchers alarmed by the Trump administration's hostile attitude to academia to relocate to the Land of the Rising Sun. The Japanese government aims to create an elite research environment, and has detailed a $693 million package to attract researchers from abroad, including those from America who may have seen their budgets slashed or who fear a clampdown on their academic freedom.

Researchers have built a breakthrough random number generator that solves a critical problem: for the first time, every step of creating random numbers can be independently verified and audited, with quantum physics guaranteeing the numbers were truly unpredictable.

Random numbers are essential for everything from online banking encryption to fair lottery drawings, but current systems have serious limitations. Computer-based generators follow predictable algorithms -- if someone discovers the starting conditions, they can predict all future outputs. Hardware generators that measure physical processes like electronic noise can't prove their randomness wasn't somehow predetermined or tampered with.

The new system, developed by teams at the University of Colorado Boulder and the National Institute of Standards and Technology, uses quantum entanglement -- Einstein's "spooky action at a distance" -- to guarantee unpredictability. The setup creates pairs of photons that share quantum properties, then sends them to measurement stations 110 meters apart. When researchers measure each photon's properties, quantum mechanics ensures the results are fundamentally random and cannot be influenced by any classical communication between the stations.

The team created a system called "Twine" that distributes the random number generation process across multiple independent parties, with each step recorded in tamper-proof digital ledgers called hash chains. This means no single organization controls the entire process, and anyone can verify that proper procedures were followed. During a 40-day demonstration, the system successfully generated random numbers in 7,434 of 7,454 attempts -- a 99.7% success rate. Each successful run produced 512 random bits with mathematical certainty of randomness bounded by an error rate of 2^-64, an extraordinarily high level of confidence.

The Trump Organization on Monday unveiled a mobile phone plan and a $499 smartphone that is set to launch in September. CNBC: The new service, Trump Mobile, will offer a $47.45-per-month plan that includes "unlimited" talk, text and data, as well as roadside assistance and a "Telehealth and Pharmacy Benefit," according to its website. The company, owned by President Donald Trump, also announced it will sell a "T1" smartphone, which appears to feature a gold-colored metal case etched with an American flag. Further reading: I Tried Pre-Ordering the Trump Phone. The Page Failed and It Charged My Credit Card the Wrong Amount.

Microsoft's latest Windows 11 preview builds contain a bug that replaces the operating system's startup sound with Windows Vista's iconic boot chime from 2006. Microsoft acknowledged the bug in its release notes -- describing it as a "delightful blast from the past" -- and said it was working on a fix.

Pharmaceutical giant Novo Nordisk forfeited patent protection for semaglutide -- the active ingredient in blockbuster diabetes and weight loss drugs Ozempic and Wegovy -- in Canada after failing to pay a $450 maintenance fee in 2019. The company had paid maintenance fees through 2018 but requested a refund for the 2017 fee, apparently seeking more time to decide whether to continue protecting the patent.

When the 2019 fee came due at $450 with late penalties, Novo never paid despite having a one-year grace period. Canadian patent authorities confirmed the patent "cannot be revived" once lapsed. The oversight is particularly costly given Canada represents the world's second-largest semaglutide market, worth billions annually. Generic drugmaker Sandoz plans to launch a competing version in early 2026, while Novo's U.S. patent protection extends until at least 2032.

An anonymous reader shares a report: When Facebook bought WhatsApp for $19 billion in 2014, the messaging app had a clear focus. No ads, no games and no gimmicks. For years, that is what WhatsApp's two billion users -- many of them in Brazil, India and other countries around the world -- got. They chatted with friends and family unencumbered by advertising and other features found on social media. Now that is set to change.

On Monday, WhatsApp said it would start showing ads inside its app for the first time. The promotions will appear only in an area of the app called Updates, which is used by around 1.5 billion people a day. WhatsApp will collect some data on users to target the ads, such as location and the device's default language, but it will not touch the contents of messages or whom users speak with. The company added that it had no plans to place ads in chats and personal messages.

[...] In-app ads are a significant change from WhatsApp's original philosophy. Jan Koum and Brian Acton, who founded WhatsApp in 2009, were committed to building a simple and quick way for friends and family to communicate with end-to-end encryption, a method of keeping texts, photos, videos and phone calls inaccessible by third parties. Both left the company seven years ago. Since then, Mark Zuckerberg, the chief executive of Facebook, now Meta, has focused on WhatsApp's growth and user privacy while also melding the app into the company's other products, including Instagram and Messenger.

"Walmart is bringing drone deliveries to three more states," reports CNBC: On Thursday, the big-box retailer said it plans to launch the speedier delivery option at 100 stores in Atlanta, Charlotte, Houston, Orlando and Tampa within the coming year. With the expansion, Walmart's drone deliveries will be available in a total of five states: [parts of northwest] Arkansas, Florida, Georgia, North Carolina and [the Dallas-Fort Worth area of] Texas... The drone operator will have an up to a six-mile range from stores.
Walmart tells CNBC the most frequently delivered items include ice cream, fresh fruit, and pet food, as well as "urgent items, such as hamburger buns for a cookout, eggs to make brownies or Tylenol or cold medicine needed when sick." It's all part of Walmart's effort to compete with Amazon: With more than 4,600 Walmart stores across the U.S., the retailer has used its large footprint to get online orders to customers faster. It has an Express Delivery service that drops purchases at customers' doors in as fast as 30 minutes, along with InHome, a subscription-based service, that puts items directly into people's fridges. The company began same-day prescription deliveries last fall and has expanded the service across the country.... Walmart stores have an assortment of over 150,000 items in a location. Over 50% of those can be delivered by drone, said Greg Cathey [Walmart's senior VP for U.S. transformation and innovation]...

Walmart's drone delivery count so far is modest. The company did not share the specific count, but said it has racked up a total of more than 150,000 drone deliveries since 2021.

KDE isn't the only organization reaching out to " as Microsoft prepares to end support for Windows 10.

"Now, The Document Foundation, maker of LibreOffice, has also joined in to support the Endof10 initiative," reports the tech blog Neowin: The foundation writes: "You don't have to follow Microsoft's upgrade path. There is a better option that puts control back in the hands of users, institutions, and public bodies: Linux and LibreOffice. Together, these two programmes offer a powerful, privacy-friendly and future-proof alternative to the Windows + Microsoft 365 ecosystem."

It further adds the "real costs" of upgrading to Windows 11 as it writes:

"The move to Windows 11 isn't just about security updates. It increases dependence on Microsoft through aggressive cloud integration, forcing users to adopt Microsoft accounts and services. It also leads to higher costs due to subscription and licensing models, and reduces control over how your computer works and how your data is managed. Furthermore, new hardware requirements will render millions of perfectly good PCs obsolete.... The end of Windows 10 does not mark the end of choice, but the beginning of a new era. If you are tired of mandatory updates, invasive changes, and being bound by the commercial choices of a single supplier, it is time for a change. Linux and LibreOffice are ready — 2025 is the right year to choose digital freedom!"
The first words on LibreOffice's announcement? "The countdown has begun...."

Spain's newspaper El Pais found an entire fake album on YouTube titled Rumba Congo (1973). And they cite a study from France's International Confederation of Societies of Authors and Composers that estimated revenue from AI-generated music will rise to $4 billion in 2028, generating 20% of all streaming platforms' revenue: One of the major problems with this trend is the lack of transparency. María Teresa Llano, an associate professor at the University of Sussex who studies the intersection of creativity, art and AI, emphasizes this aspect: "There's no way for people to know if something is AI or not...." On Spotify Community — a forum for the service's users — a petition is circulating that calls for clear labeling of AI-generated music, as well as an option for users to block these types of songs from appearing on their feeds. In some of these forums, the rejection of AI-generated music is palpable.

Llano mentions the feelings of deception or betrayal that listeners may experience, but asserts that this is a personal matter. There will be those who feel this way, as well as those who admire what the technology is capable of... One of the keys to tackling the problem is to include a warning on AI-generated songs. YouTube states that content creators must "disclose to viewers when realistic content [...] is made with altered or synthetic media, including generative AI." Users will see this if they glance at the description. But this is only when using the app, because on a computer, they will have to scroll down to the very end of the description to get the warning....

The professor from the University of Sussex explains one of the intangibles that justifies the labeling of content: "In the arts, we can establish a connection with the artist; we can learn about their life and what influenced them to better understand their career. With artificial intelligence, that connection no longer exists."
YouTube says they may label AI-generated content if they become aware of it, and may also remove it altogether, according to the article. But Spotify "hasn't shared any policy for labeling AI-powered content..." In an interview with Gustav Söderström, Spotify's co-president and chief product & technology officer, he emphasized that AI "increases people's creativity" because more people can be creative, thanks to the fact that "you don't need to have fine motor skills on the piano." He also made a distinction between music generated entirely with AI and music in which the technology is only partially used. But the only limit he mentioned for moderating artificial music was copyright infringement... something that has been a red line for any streaming service for many years now. And such a violation is very difficult to legally prove when artificial intelligence is involved.

An anonymous reader shared this report from Bloomberg: Amazon's hard-line stance on getting disabled employees to return to the office has sparked a backlash, with workers alleging the company is violating the Americans with Disabilities Act as well as their rights to collectively bargain. At least two Amazon employees have filed complaints with the Equal Employment Opportunity Commission (EEOC) and the National Labor Relations Board, federal agencies that regulate working conditions. One of the workers said they provided the EEOC with a list of 18 "similarly situated" employees to emphasize that their experience isn't isolated and to help federal regulators with a possible investigation.

Disabled workers frustrated with how Amazon is handling their requests for accommodations — including exemptions to a mandate that they report to the office five days a week — are also venting their displeasure on internal chat rooms and have encouraged colleagues to answer surveys about the policies. Amazon has been deleting such posts and warning that they violate rules governing internal communications. One employee said they were terminated and another said they were told to find a different position after advocating for disabled workers on employee message boards. Both filed complaints with the EEOC and NLRB.
Amazon has told employees with disabilities they must now submit to a "multilevel leader review," Bloomberg reported in October, "and could be required to return to the office for monthlong trials to determine if accommodations meet their needs." (They received calls from "accommodation consultants" who also reviewed medical documentation, after which "another Amazon manager must sign off. If they don't, the request goes to a third manager...")

Bloomberg's new article remembers how several employees told them in November. "that they believed the system was designed to deny work-from-home accommodations and prompt employees with disabilities to quit, which some have done. Amazon denied the system was designed to encourage people to resign." Since then, workers have mobilized against the policy. One employee repeatedly posted an online survey seeking colleagues' reactions, defying the company's demands to stop. The survey ultimately generated feedback from more than 200 workers even though Amazon kept deleting it, and the results reflected strong opposition to Amazon's treatment of disabled workers. More than 71% of disabled Amazon employees surveyed said the company had denied or failed to meet most of their accommodation requests, while half indicated they faced "hostile" work environments after disclosing their disabilities and requesting accommodations.

One respondent said they sought permission to work from home after suffering multiple strokes that prevented them from driving. Amazon suggested moving closer to the office and taking mass transit, the person said in the survey. Another respondent said they couldn't drive for longer than 15-minute intervals due to chronic pain. Amazon's recommendation was to pull over and stretch during their commute, which the employee said was unsafe since they drive on a busy freeway... Amazon didn't dispute the accounts and said it considered a range of solutions to disability accommodations, including changes to an employee's commute.
Amazon is also "using AI to parse accommodation requests, read doctors' notes and make recommendations based on keywords," according to the article — another policy that's also generated internal opposition (and formed a "key element" of the complaint to the Equal Employment Opportunity Commission).

"The dispute could affect thousands of Amazon workers. An internal Slack channel for employees with disabilities has 13,000 members, one of the people said..."

In Tokyo Mitsubishi is deploying "an innovative new battery swap network for electric cars" in a multi-year test program reports the EV news site Electrek.

But it's not just for electric cars. Along with the 14 modular battery swapping stations, Mitsubishi is also deploying "more than 150 battery-swappable commercial electric vehicles" from truck maker Fuso: A truck like the Mitsubishi eCanter typically requires a full night of AC charging to top off its batteries, and at least an hour or two on DC charging in Japan, according to Fuso. This joint pilot by Mitsubishi, Mitsubishi Fuso Trucks, and [EV battery swap specialist] Ample aims to circumvent this issue of forced downtime with its swappable batteries, supporting vehicle uptime by delivering a full charge within minutes.

The move is meant to encourage the transport industry's EV shift while creating a depository of stored energy that can be deployed to the grid in the event of a natural disaster — something Mitsubishi in Japan has been working on for years.
The article's author also adds their own opinion about battery-swapping technology. "When you see how simple it is to add hundreds of miles of driving in just 100 seconds — quicker, in many cases, than pumping a tank of liquid fuel into an ICE-powered car — you might come around, yourself."

Timothy B. Lee has written for the Washington Post, Vox.com, and Ars Technica — and now writes a Substack blog called "Understanding AI."

This week he visits recent research by computer scientists and legal scholars from Stanford, Cornell, and West Virginia University that found that Llama 3.1 70BÂ(released in July 2024) has memorized 42% of the first Harry Potter book well enough to reproduce 50-token excerpts at least half the time... The paper was published last month by a team of computer scientists and legal scholars from Stanford, Cornell, and West Virginia University. They studied whether five popular open-weight models — three from Meta and one each from Microsoft and EleutherAI — were able to reproduce text from Books3, a collection of books that is widely used to train LLMs. Many of the books are still under copyright... Llama 3.1 70B — a mid-sized model Meta released in July 2024 — is far more likely to reproduce Harry Potter text than any of the other four models....

Interestingly, Llama 1 65B, a similar-sized model released in February 2023, had memorized only 4.4 percent of Harry Potter and the Sorcerer's Stone. This suggests that despite the potential legal liability, Meta did not do much to prevent memorization as it trained Llama 3. At least for this book, the problem got much worse between Llama 1 and Llama 3. Harry Potter and the Sorcerer's Stone was one of dozens of books tested by the researchers. They found that Llama 3.1 70B was far more likely to reproduce popular books — such as The Hobbit and George Orwell's 1984 — than obscure ones. And for most books, Llama 3.1 70B memorized more than any of the other models...

For AI industry critics, the big takeaway is that — at least for some models and some books — memorization is not a fringe phenomenon. On the other hand, the study only found significant memorization of a few popular books. For example, the researchers found that Llama 3.1 70B only memorized 0.13 percent of Sandman Slim, a 2009 novel by author Richard Kadrey. That's a tiny fraction of the 42 percent figure for Harry Potter... To certify a class of plaintiffs, a court must find that the plaintiffs are in largely similar legal and factual situations. Divergent results like these could cast doubt on whether it makes sense to lump J.K. Rowling, Richard Kadrey, and thousands of other authors together in a single mass lawsuit. And that could work in Meta's favor, since most authors lack the resources to file individual lawsuits.
Why is it happening? "Maybe Meta had trouble finding 15 trillion distinct tokens, so it trained on the Books3 dataset multiple times. Or maybe Meta added third-party sources — such as online Harry Potter fan forums, consumer book reviews, or student book reports — that included quotes from Harry Potter and other popular books..."

"Or there could be another explanation entirely. Maybe Meta made subtle changes in its training recipe that accidentally worsened the memorization problem."