When a startup announced plans last fall to recreate lost footage from Orson Welles’ classic film “The Magnificent Ambersons” using generative AI, I was skeptical. More than that, I was baffled why anyone would spend time and money on something that seemed guaranteed to outrage cinephiles while offering negligible commercial value.

This week, an in-depth profile by the New Yorker’s Michael Schulman provides more details about the project. If nothing else, it helps explain why the startup Fable and its founder Edward Saatchi are pursuing it: It seems to come from a genuine love of Welles and his work.

Saatchi (whose father was a founder of advertising firm Saatchi & Saatchi) recalled a childhood of watching films in a private screening room with his “movie mad” parents. He said he first saw “Ambersons” when he was twelve.

The profile also explains why “Ambersons,” while much less famous than Welles’ first film “Citizen Kane,” remains so tantalizing — Welles himself claimed it was a “much better picture” than “Kane,” but after a disastrous preview screening, the studio cut 43 minutes from the film, added an abrupt and unconvincing happy ending, and eventually destroyed the excised footage to make space in its vaults.

“To me, this is the holy grail of lost cinema,” Saatchi said. “It just seemed intuitively that there would be some way to undo what had happened.”

Saatchi is only the latest Welles devotee to dream of recreating the lost footage. In fact, Fable is working with filmmaker Brian Rose, who already spent years trying to achieve the same thing with animated scenes based on the movie’s script and photographs, and on Welles’ notes. (Rose said that after he screened the results for friends and family, “a lot of them were scratching their heads.”)

So while Fable is using more advanced technology — filming scenes in live action, then eventually overlaying them with digital recreations of the original actors and their voices — this project is best understood as a slicker, better-funded version of Rose’s work. It’s a fan’s attempt to glimpse Welles’ vision.

Notably, while the New Yorker article includes a few clips of Rose’s animations, as well as images of Fable’s AI actors, there’s no footage showing the results of Fable’s live action-AI hybrid.

By the company’s own admission, there are significant challenges, whether that’s fixing obvious blunders like a two-headed version of the actor Joseph Cotten, or the more subjective task of recreating the complex beauty of the film’s cinematography. (Saatchi even described a “happiness” problem, with the AI tending to make the film’s women look inappropriately happy.)

As for whether this footage will ever be released to the public, Saatchi admitted it was “a total mistake” not to speak to Welles’ estate before his announcement. Since then, he has reportedly been working to win over both the estate and Warner Bros., which owns the rights to the film. Welles’ daughter Beatrice told Schulman that while she remains “skeptical,” she now believes “they are going into this project with enormous respect toward my father and this beautiful movie.”

The actor and biographer Simon Callow — who’s currently writing the fourth book in his multi-volume Welles biography — has also agreed to advise the project, which he described as a “great idea.” (Callow is a family friend of the Saatchis.)

But not everyone has been convinced. Melissa Galt said her mother, the actress Anne Baxter, would “not have agreed with that at all.”

“It’s not the truth,” Galt said. “It’s a creation of someone else’s truth. But it’s not the original, and she was a purist.”

And while I’ve become more sympathetic to Saatchi’s aims, I still agree with Galt: At its best, this project will only result in a novelty, a dream of what the movie might have been.

In fact, Galt’s description of her mother’s position that “once the movie was done, it was done,” reminded me of a recent essay in which the writer Aaron Bady compared AI to the vampires in “Sinners.” Bady argued that when it comes to art, both vampires and AI will always come up short, because “what makes art possible” is a knowledge of mortality and limitations.

“There is no work of art without an ending, without the point at which the work ends (even if the world continues),” he wrote, adding, “Without death, without loss, and without the space between my body and yours, separating my memories from yours, we cannot make art or desire or feeling.”

In that light, Saatchi’s insistence that there must be “some way to undo what had happened” feels, if not outright vampiric, then at least a little childish in its unwillingness to accept that some losses are permanent. It may not, perhaps, be all that different from a startup founder claiming they can make grief obsolete — or a studio executive insisting that “The Magnificent Ambersons” needed a happy ending.

Just in time to create a new Super Bowl ad, Crypto.com founder Kris Marszalek has made the priciest domain purchase in history, buying AI.com for $70 million, according to the Financial Times. The deal, paid entirely in cryptocurrency to an unknown seller, shatters previous records. (Broker Larry Fischer, who facilitated the sale, is presumably celebrating his good fortune.)

Marszalek plans to debut the site during Sunday’s big game, offering consumers a personal AI agent for messaging, app usage, and stock trading. “If you take a long-term view — 10 to 20 years – [AI] is going to be one of the greatest technological waves of our lifetime,” he told the FT.

The purchase rewrites the domain record books — not that crypto industry itself is known for its restraint when it comes to spending. Previously, CarInsurance.com held the crown at $49.7 million (2010), followed by VacationRentals.com ($35 million in 2007) and Voice.com ($30 million in 2019). Other eye-popping sales include PrivateJet.com ($30 million), 360.com ($17 million), and Sex.com, which has sold twice for over $13 million each time, though its second owner went bankrupt trying to monetize it.

“With assets like AI.com, there are no substitutes,” Fischer told the FT. “When one becomes available, the opportunity may never present itself again.”

Whether these mega-dollar domains actually deliver returns remains an open question. But for Marszalek, who already owns Crypto.com and dropped $700 million on stadium naming rights, owning two category-defining domains is apparently worth the outlay.

Comparing Our Favorite Vacuum Cleaners

Honorable Mentions

Bissell PowerClean FurFinder for $210: This was our previous top pick for cordless vacuums, and it’s still a great affordable stick vacuum. But if you’re looking for a cheap option, the regular Bissell PowerClean ($200) is a touch cheaper since it doesn’t come with the FurFinder upholstery attachment, while Dyson and Ryobi’s vacuums have more powerful suction for pet hair.

Black and Decker Dustbuster Flex for $95: This is another cool handheld vacuum that’s great for cars, or even indoor areas like staircases. It has a 4-foot hose, longer than most compact vacuums, and has a handy little charging mount that the accessories clip into. And thank goodness for the charging mount, since the battery only lasts 15 minutes.

Dyson Car+Boat Handheld Vacuum for $290: Dyson’s latest handheld vacuum was designed specifically for cars (and boats!) in mind, so if you have a vehicle to clean, this is made for it. It’s got a fantastic battery life compared to other handheld vacuums.

Dyson Gen5 Detect Cordless Vacuum for $800: This is Dyson’s current top-of-the-line stick vacuum, and it’s fantastic, especially for homes with a lot of pet hair. But it’s expensive and rarely on sale, making it a hard upgrade when you’ll still get a great experience with the Dyson V15 Detect. But this vacuum does have a HEPA filter, while the V15 Detect does not.

Ecovacs Deebot X11 Omnicyclone for $999: If you want a newer robot vacuum, the Ecovacs Deebot X11 Omnicyclone has a unique design with no dust bag. Instead, it has a rounded canister like a Dyson or stick vacuum, circling the debris to keep it from tangling. It’s a good design and this is overall a great robot vacuum, especially if you don’t want to buy dust bags over and over.

Roborock Qrevo S for $350 (used): This was our previous top robot vacuum pick, but Roborock updated the Qrevo line with new models that aren’t shipping to the United States, and the model we liked best is only available secondhand.

Tineco Pure One Station 5 for $459: If you want a cordless stick vacuum but don’t want to deal with emptying it all the time, this Tineco vacuum comes with a self-emptying docking station. You’ll eventually have to empty the station, but it’s a great bonus feature and keeps the vacuum from falling over around your home since it’s stored safely in the docking station.

Power up with unlimited access to WIRED. Get best-in-class reporting and exclusive subscriber content that’s too important to ignore. Subscribe Today.

Emily Musil reviews her 11- and 13-year-old children’s report cards, scanning through the typical categories: language arts, math, social studies.

But one day she is hopeful there will be a new metric for assessing all children: artificial intelligence literacy.

“Yes, I do think we’ll get there,” Musil says, pointing toward the recent rapid progression in what elementary schools offer, from typing class to computer literacy instruction to teaching coding. But currently “as a parent, I’m not seeing how my children are doing in understanding deep computing and AI tools — and that needs to shift.”

Musil is a managing director of social innovation at the nonprofit think tank Milken Institute. She served as the head researcher for a report released in November focused on building the nation’s talent engine in the age of AI.

“If you care about economic mobility, what choices do you need to make?” she asks. “We are behind, because technology is so rapidly advancing and is so tied to all jobs.”

The report pushed for K-12 institutions to emphasize AI literacy in conjunction with critical thinking and decision-making skills. But getting to that point of an expanded curriculum — let alone one that tackles the nuances of AI technology — may prove difficult. That’s because it will take concerted efforts from institutions, schools and leaders to make it happen.

Lack of Standards and Expertise

Federal standards for AI education began in the Obama administration and were most recently pushed again by the Trump administration, with the presidential action “Advancing Artificial Intelligence Education for American Youth.” Yet when it comes to local implementation, it is largely up to the school or administrator — and more than half of U.S. schools or districts, many of them rural or classified as Title 1, have no standards at all.

According to the report, 60 percent of U.S. schools or districts have no guidance for generative AI usage. Many schools have previously told EdSurge that because the technology is so rapidly changing, decisions are often left up to the teacher’s discretion.

The lack of standards could also be a lack of expertise on AI and technology as a whole in the classroom. For example, according to the Milken report, only 17 percent of current computer science teachers have computer science degrees. While the report did not delve into what those teachers may have majored in instead, Musil added that some teachers are asked to cover the subject as the scope of their workload increases.

The same phenomenon could happen with AI literacy curricula.

“If you’ve been a teacher for 20 years, suddenly, you may not be an expert in medieval history, but you had to do something on it,” she says. “So, they’re teaching something they’re not necessarily deeply skilled in.”

Collective Action

The report had four specific focus points for K-12 schools: developmentally appropriate instruction in AI; ethical and critical use of AI tools; pairing human cognition with AI use; and learning through human interaction rather than screens alone.

For students, “K–12 education is often the first place they encounter STEM and computing topics,” the report says. “As an AI-driven workforce demands specialized skills earlier and earlier, K–12 has become an even more crucial intervention point. By building future-ready curricula and support systems, we can address gaps early and support student flourishing.”

They are lofty goals. A related challenge is the dearth of girls pursuing STEM fields. The report found roughly half (49 percent) of elementary school computer science students are girls. That drops to 44 percent by middle school, 33 percent by high school, and to about 20 percent by college graduation.

The Milken report acknowledges there is no easy, silver bullet solution to achieving these goals. There are the necessary federal efforts at play. And Musil suggested that employers or individual philanthropists could help support schools in funding, advocating and collaborating in curricula changes, to benefit both students and hiring organizations.

“This report makes it clear that the challenge is national in scope and the solutions must be collective,” says Michael Ellison, co-founder and CEO of CodePath, a nonprofit focused on diversifying the technology industry. The organization supported the Milken Institute in producing its report. “Philanthropists, industry leaders, policymakers, and educators all must act to rewire education and workforce systems for an AI-driven world.”

Risks of AI Integration

But there are also considerations to take when integrating the rapidly changing technology. A report released last month by The Center for Democracy and Technology found that schools’ embrace of AI was connected to increased risk of worse outcomes for students: half of the surveyed students said using AI in class makes them feel less connected with their teacher.

“As many hype up the possibilities for AI to transform education, we cannot let the negative impact on students get lost in the shuffle,” Elizabeth Laird, director of the Equity in Civic Technology Project at CDT, said in a statement. “Our research shows AI use in schools comes with real risks … Acknowledging those risks enables education leaders, policymakers, and communities to mount prevention and response efforts so that the positive uses of AI are not overshadowed by harm to students.”

And in a 2023 report titled “Artificial Intelligence and the Future of Teaching and Learning,” the Department of Education warns of unchecked usage.

“We especially call upon leaders to avoid romancing the magic of AI or only focusing on promising applications or outcomes, but instead to interrogate with a critical eye how AI-enabled systems and tools function in the educational environment,” the report says.

But Musil points out that whether schools have specific rules or not when it comes to AI integration, the students will be using it in their free time — and it is best to teach them the best way to avoid those negative outcomes.

“My daughter is told AI is cheating, but there’s lots of things to do with pedagogy with AI; that piece of it is going to be their future.” she says. “When I’m hiring, I want someone to use AI and know when it is cheating, when it isn’t, and when it supports human thinking and when it supplants it.”

After a better-than-expected opening weekend in theaters, box office for Amazon’s “Melania” fell 67%, to an estimated $2.37 million, in its second weekend.

The documentary about First Melania Trump has grossed a total of $13.5 million so far (almost all of that in the United States), which means it’s extremely unlikely the film — which Amazon spent $40 million to acquire and $35 million to market — will break even in theaters.

Before “Melania”’s release, a former Amazon film executive asked how the price tag could be motivated by anything other than “currying favor” with the Trump administration or “an outright bribe.” 

Moviegoing typically slows during Super Bowl weekend, but the weekend’s top film, “Send Help,” only declined 47%. And after placing third on the charts last weekend, “Melania” is ninth this time.

Perhaps anticipating discussion about the film’s decline, Amazon released a statement from its head of domestic theatrical distribution Kevin Wilson, who said, “Together, theatrical and streaming represent two distinct value creating moments that amplify the film’s overall impact.”

“Melania” received universally negative reviews from critics, but its 99% audience rating on Rotten Tomatoes was so good that the site put out a statement insisting that the score was real.

Photo credit: Jose-Luis Olivares, MIT
MIT engineers have come up with some ingenious work: they’ve created tiny silicon structures that can crunch numbers using heat that would otherwise be wasted rather than energy. It’s a game changer for dealing with heat in electronics, and a team of MIT researchers lead by undergraduate physics wiz Caio Silva has proved that it truly works. In a study published in Physical Review Applied, they show off the results of their simulation work, demonstrating that these small devices can do some rather important math operations with surprising accuracy.

The truth is, most of the time, heat from computers and chips is just a nuisance, a byproduct of processors doing their tasks that must be removed as soon as possible via fans and cooling, but the researchers at this lab did something completely different. They decided to utilize heat to represent the actual calculations themselves, and guess what? It works.

Anker Prime Charging Station, 250W 6-Port GaN USB C Charger, 2.26″ LCD Smart Display Control, Compatible…

• Super Fast 250W Output: Experience the convenience of high-capacity charging with this 250W desktop charger, featuring four USB-C and two USB-A ports…
• High-Efficiency Charger: Power up your 16-inch MacBook Pro to 50% in just 25 minutes using the first USB-C port, which delivers 140W to significantly…
• Instant Power Feedback: Monitor your charging in real time with the 2.26-inch LCD display, providing visual feedback on charging status and speed.

Each of these tiny structures is roughly the size of a particle of dust and is constructed of silicon with carefully arranged holes. This allows you to precisely control how heat travels through the material, which happens quite naturally. Naturally, heat goes from hot to cold. It was the researchers that came up with the idea of using this natural flow to perform actual computations. They’ve devised a method to depict the data entering a calculation as temperature patterns derived from the heat that ordinarily accumulates beneath the feet of a device. As the heat passes through the silicon, the unique shape of the structure directs the flow, resulting in your response as the quantity of power collected at a specified cooler end.

This is accomplished entirely using analog computing, which employs continuous physical processes rather than the discrete electrical impulses seen in most computers. They were able to get matrix-vector multiplication to operate, which is a calculation used in many machine learning models. The team’s simulations revealed that these teeny-tiny structures can correctly predict the answer more than 99% of the time for small matrices.

Getting all of this set up in silicon eliminates the need for traditional guesswork regarding shape. These researchers automated all of their work with an optimization technique. They give the computer what math problem they want to answer, such as a matrix of numbers indicating how the input should be converted into the output. The computer then gets to work, determining where to place the pores and how thick each component of the structure should be. All the time, it is verifying and rechecking to ensure that the heat flow is doing precisely what it should be doing in order to obtain the correct answer.

Heat conduction only travels one way, from hot to cold, thus positive values are easy to obtain; however, negative values require a method to make them work. The researchers work around this by dividing any matrix with negative integers into positive and negative ones. Each portion is ran through its own customized configuration before the results are pooled using a simple subtraction method. The ability to modify the structure’s thickness allows you to fine-tune how easy heat may pass through at certain spots.

These structures offer a wide range of possible applications beyond simple calculation, such as monitoring how heat flows through electronics without using extra power or separate temperature sensors. In many circumstances, isolated hot spots lead to the discovery of other concerns, such as when some piece of electronics is prone to failure owing to overheating or an uneven build-up of stress. The beauty of this design is that you can just embed the devices directly into the electronics, and they will automatically detect these issues as they occur, using the heat to report back to you on the status of things. Temperature gradients are basically a goldmine of information, according to Giuseppe Romano, senior researcher at MIT’s Institute for Soldier Nanotechnologies. However, if you get a heat source in a place it shouldn’t be, you know you’ve got a problem, and these structures can pick on it without having to go digital.
[Source]

• Broadcom launches new Wi-Fi 8 chips for enterprise access points and switches
• Access point chips combine processing, networking, and wireless functions in a single device
• Wi-Fi 8 radio chips handle wireless transmission across enterprise environments efficiently

Broadcom has announced a new set of enterprise Wi-Fi 8 chipsets covering access points and campus switching hardware.

The launch builds on the company’s earlier disclosure of Wi-Fi 8 radios and extends into full wired and wireless infrastructure.