Kalakshar US

Scams using AI deepfakes of celebrities have become an increasingly prominent issue for Meta over the last couple of years. Now, the Oversight Board has weighed in and has seemingly confirmed what other critics have said: Meta isn’t doing enough to enforce its own rules, and makes it far too easy for scammers to get away with these schemes.

“Meta is likely allowing significant amounts of scam content on its platforms to avoid potentially overenforcing a small subset of genuine celebrity endorsements,” the board wrote in its latest decision. “At-scale reviewers are not empowered to enforce this prohibition on content that establishes a fake persona or pretends to be a famous person in order to scam or defraud.”

That conclusion came as the result of a case involving an ad for an online casino-style game called Plinko that used an AI-manipulated video of Ronaldo Nazário, a retired Brazilian soccer player. The ad, which according to the board showed obvious signs of being fake, was not removed by Meta even after it was reported as a scam more than 50 times. Meta later removed the ad, but not the underlying Facebook post behind it until the Oversight Board agreed to review the case. It was viewed more than 600,000 times.

The board says that the case highlights fundamental flaws in how Meta approaches content moderation for reported scams involving celebrities and public figures. The board says that Meta told its members that “it enforces the policy only on escalation to ensure the person depicted in the content did not actually endorse the product” and that individual reviewers’ “interpretation of what constitutes a ‘fake persona’ could vary across regions and introduce inconsistencies in enforcement.” The result, according to the Oversight Board, is that a “significant” amount of scam content is likely slipping through the cracks.

In its sole recommendation to Meta, the board urged the company should update its internal guidelines, empower content reviewers to identify such scams and train them on “indicators” of AI-manipulated content. In a statement, a spokesperson for Meta said that “many of the Board’s claims are simply inaccurate” and pointed to a test it began last year that uses facial recognition technology to fight “celeb-bait” scams.

“Scams have grown in scale and complexity in recent years, driven by ruthless cross-border criminal networks,” the spokesperson said. “As this activity has become more persistent and sophisticated, so have our efforts to combat it. We’re testing the use of facial recognition technology, enforcing aggressively against scams, and empowering people to protect themselves through many different on platform safety tools and warnings. While we appreciate the Oversight Board’s views in this case, many of the Board’s claims are simply inaccurate and we will respond to the full recommendation in 60 days in accordance with the bylaws.”

Scams using AI deepfakes of celebrities has become a major problem for Meta as AI tech gets cheaper and more easily accessible. Earlier this year, I reported that dozens of pages were running ads featuring deepfakes of Elon Musk and Fox News personalities promoting supplements that claimed to cure diabetes. Some of these pages repeatedly ran hundreds of versions of these ads with seemingly few repercussions. Meta disabled some of the pages after my reporting, but similar scam ads persist on Facebook to this day. Actress Jamie Lee Curtis also recently publicly slammed Mark Zuckerberg for not removing a deepfaked Facebook ad that featured her (Meta removed the ad after her public posts).

The Oversight Board similarly highlighted the scale of the problem in this case, noting that it found thousands of video ads promoting the Plinko app in Meta’s Ad Library. It said that several of these featured AI deepfakes, including ads featuring another Brazilian soccer star, Cristiano Ronaldo, and Meta’s own CEO Mark Zuckerberg.

The Oversight Board isn’t the only group that’s raised the alarm about scams on Meta’s platforms. The Wall Street Journal recently reported that Meta “accounted for nearly half of all reported scams on Zelle for JPMorgan Chase between the summers of 2023 and 2024” and that “British and Australian regulators have found similar levels of fraud originating on Meta’s platforms.” The paper noted that Meta is “reluctant” to add friction to its ad-buying process and that the company “balks” at banning advertisers, even those with a history of conducting scams.

This article originally appeared on Engadget at https://www.engadget.com/social-media/the-oversight-board-says-meta-isnt-doing-enough-to-fight-celeb-deepfake-scams-194636203.html?src=rss 

In space, no one can hear you, uh, stream. We finally have a real trailer for the upcoming Alien: Earth TV series and it looks extremely cool. FX dropped a teaser a few months back, but that was just vibes. This is an actual two-minute trailer that’s absolutely packed with footage.

For the uninitiated, Alien: Earth is the first TV series in the franchise’s long history and is being helmed by Noah Hawley. That’s the guy who made the Fargo TV series and the criminally underrated Marvel show Legion. Franchise creator Ridley Scott is also on board as an executive producer.

This is a prequel that’s set just two years before the original film. As the name suggests, it takes place on Earth. The trailer indicates the story will involve a ship crashing into a large corporate-controlled city. This vessel may or may not have a fearsome Xenomorph aboard, among other notable alien species.

A character in the clip says that the ship “collected five different life forms from the darkest corners of the universe,” going on to suggest that one is “predatory.” That’s right. We could be getting a dang Predator in our Alien TV show.

Alien: Earth premieres on August 12 on FX and will be available to stream on Hulu. The show stars Sydney Chandler as a hybrid robot/human lifeform tasked with investigating the crash. The rest of the cast includes Timothy Olyphant, Alex Lawther and Samuel Blenkin, among others.

This article originally appeared on Engadget at https://www.engadget.com/entertainment/tv-movies/the-full-alien-earth-trailer-is-finally-here-and-its-a-doozy-184921361.html?src=rss 

Nintendo Switch 2 Welcome Tour is a $10 museum-like experience dedicated to playfully explaining features of the Nintendo Switch 2, and technically, it costs a lot more than $10 to truly finish. As Eurogamer writes and multiple Nintendo webpages note, you can’t actually finish the game without using all of the Switch 2’s many accessories.

All 12 of the different areas in Welcome Tour are accessible whether you’re plugging in a camera or not, but to play and beat every minigame, “additional accessories [are] required,” according to Nintendo. Those accessories include a USB camera, a controller with GL and GR buttons (the kind in the Switch 2 Pro Controller’s grips) and a 4K-compatible TV. If you wanted to go the official route and use Nintendo accessories, that means paying $55 for the Nintendo Switch 2 Camera and $85 for the Nintendo Switch 2 Pro Controller, assuming you already own a 4K TV.

No one needs to 100-percent complete Nintendo Switch 2 Welcome Tour, of course, but it is funny that price was one of the biggest complaints about the Switch 2’s answer to Wii Sports, and now to totally complete it you might have to pay even more.

The patient among us (who didn’t pre-order) can wait and see if the hidden secrets of Welcome Tour are truly worth spending the extra cash on accessories. For the rest, you might be better off spending your time unlocking characters in Mario Kart World.

This article originally appeared on Engadget at https://www.engadget.com/gaming/nintendo/truly-completing-nintendo-switch-2-welcome-tour-will-cost-you-192255868.html?src=rss 

AI companies claim their tools couldn’t exist without training on copyrighted material. It turns out, they could — it’s just really hard. To prove it, AI researchers trained a new model that’s less powerful but much more ethical. That’s because the LLM’s dataset uses only public domain and openly licensed material.

The paper (via The Washington Post) was a collaboration between 14 different institutions. The authors represent universities like MIT, Carnegie Mellon and the University of Toronto. Nonprofits like Vector Institute and the Allen Institute for AI also contributed.

The group built an 8 TB ethically-sourced dataset. Among the data was a set of 130,000 books in the Library of Congress. After inputting the material, they trained a seven-billion-parameter large language model (LLM) on that data. The result? It performed about as well as Meta’s similarly sized Llama 2-7B from 2023. The team didn’t publish benchmarks comparing its results to today’s top models.

Performance comparable to a two-year-old model wasn’t the only downside. The process of putting it all together was also a grind. Much of the data couldn’t be read by machines, so humans had to sift through it. “We use automated tools, but all of our stuff was manually annotated at the end of the day and checked by people,” co-author Stella Biderman told WaPo. “And that’s just really hard.” Figuring out the legal details also made the process hard. The team had to determine which license applied to each website they scanned.

So, what do you do with a less powerful LLM that’s much harder to train? If nothing else, it can serve as a counterpoint.

In 2024, OpenAI told a British parliamentary committee that such a model essentially couldn’t exist. The company claimed it would be “impossible to train today’s leading AI models without using copyrighted materials.” Last year, an Anthropic expert witness added, “LLMs would likely not exist if AI firms were required to license the works in their training datasets.”

Of course, this study won’t change the trajectory of AI companies. After all, more work to create less powerful tools doesn’t jive with their interests. But at least it punctures one of the industry’s common arguments. Don’t be surprised if you hear about this study again in legal cases and regulation arguments.

This article originally appeared on Engadget at https://www.engadget.com/ai/it-turns-out-you-can-train-ai-models-without-copyrighted-material-174016619.html?src=rss