Top Eurocops are appealing for help from lawmakers to undermine a privacy-enhancing technology (PET) they say is hampering criminal investigations -- and it's not end-to-end encryption this time. Not exactly. From a report: Europol published a position paper today highlighting its concerns around SMS home routing -- the technology that allows telcos to continue offering their services when customers visit another country. Most modern mobile phone users are tied to a network with roaming arrangements in other countries. EE customers in the UK will connect to either Telefonica or Xfera when they land in Spain, or T-Mobile in Croatia, for example.

While this usually provides a fairly smooth service for most roamers, Europol is now saying something needs to be done about the PETs that are often enabled in these home routing setups. According to the cops, they pointed out that when roaming, a suspect in a criminal case who's using a SIM from another country will have all of their mobile communications processed through their home network. If a crime is committed by a Brit in Germany, for example, then German police couldn't issue a request for unencrypted data as they could with a domestic operator such as Deutsche Telekom.

An anonymous reader shares a report: Apple has removed several apps offering virtual private network services from the Russian App Store, following a request from Roskomnadzor, Russia's media regulator, independent news outlet Mediazona reported on Thursday.

The VPN services removed by Apple include leading services such as ProtonVPN, Red Shield VPN, NordVPN and Le VPN. Those living in Russia will no longer be able to download the services, while users who already have them on their phones can continue using them, but will be unable to update them. Red Shield VPN posted a notice from Apple on X, which said that their app would be removed following a request from Roskomnadzor, "because it includes content that is illegal in Russia."

Cloudflare today unveiled a new feature part of its content delivery network (CDN) that prevents AI developers from scraping content on the web. According to Cloudflare, the feature is available for both the free and paid tiers of its service. SiliconANGLE reports: The feature uses AI to detect automated content extraction attempts. According to Cloudflare, its software can spot bots that scrape content for LLM training projects even when they attempt to avoid detection. "Sadly, we've observed bot operators attempt to appear as though they are a real browser by using a spoofed user agent," Cloudflare engineers wrote in a blog post today. "We've monitored this activity over time, and we're proud to say that our global machine learning model has always recognized this activity as a bot."

One of the crawlers that Cloudflare managed to detect is a bot that collects content for Perplexity AI Inc., a well-funded search engine startup. Last month, Wired reported that the manner in which the bot scrapes websites makes its requests appear as regular user traffic. As a result, website operators have struggled to block Perplexity AI from using their content. Cloudflare assigns every website visit that its platform processes a score of 1 to 99. The lower the number, the greater the likelihood that the request was generated by a bot. According to the company, requests made by the bot that collects content for Perplexity AI consistently receive a score under 30.

"When bad actors attempt to crawl websites at scale, they generally use tools and frameworks that we are able to fingerprint," Cloudflare's engineers detailed. "For every fingerprint we see, we use Cloudflare's network, which sees over 57 million requests per second on average, to understand how much we should trust this fingerprint." Cloudflare will update the feature over time to address changes in AI scraping bots' technical fingerprints and the emergence of new crawlers. As part of the initiative, the company is rolling out a tool that will enable website operators to report any new bots they may encounter.

Ahead of its one-year anniversary, Meta CEO Mark Zuckerberg announced that Threads has reached more than 175 million monthly active users. The Verge reports: Back when it arrived in the App Store on July 5th, 2023, Musk was taking a wrecking ball to the service formerly called Twitter and goading Zuckerberg into a literal cage match that never happened. A year later, Threads is still growing at a steady clip -- albeit not as quickly as its huge launch -- while Musk hasn't shared comparable metrics for X since he took over.

As with any social network, and especially for Threads, monthly users only tell part of the growth story. It's telling that, unlike Facebook, WhatsApp, and Instagram, Meta hasn't shared daily user numbers yet. That omission suggests Threads is still getting a lot of flyby traffic from people who have yet to become regular users. I've heard from Meta employees in recent months that much of the app's growth is still coming from it being promoted inside Instagram. Both apps share the same account system, which isn't expected to change.

Matthew Connatser reports via The Register: Utility firms American Electric Power (AEP) and Exelon have filed an official objection with the Federal Energy Regulatory Commission (FERC) over Talen Energy's nuclear power deal with Amazon. Back in March, Amazon bought a nuclear-powered datacenter from Talen Energy -- an operator of electricity generation and transmission facilities in the US. As part of the deal, Amazon would get 480 MW straight from the 2.7 GW Talen nuclear power plant in Susquehanna, Pennsylvania, and may even be able to upgrade to 960 MW down the line. However, that Susquehanna atomic plant also provides power to PJM Interconnection, the regional power grid operator for much of the eastern US. The two companies -- Talen and PJM -- have an interconnection service agreement (ISA) that sets the rules for how Talen should deliver power to PJM's transmission system.

To better accommodate the nuclear datacenter, Talen and PJM agreed to a new ISA, which has caught the ire of AEP and Exelon. The duo claim Talen and Amazon are basically getting a free ride that other PJM ratepayers will have to pay for, saying that even though Amazon's datacenter isn't directly connected to PJM, it still benefits from the power grid, meaning the other ratepayers are left holding the short end of the stick. Each of the station's two reactors has 1,350 MW available, and Amazon is already able to use 480 MW, and up to 960 MW in the future. If one of the reactors experiences an outage, the ISA says the datacenter is first in line for power from the other reactor, which leaves PJM with far less electricity than normal. That in turn would mean a lower energy supply for PJM's customers, who would have to pay more, at least according to the complaint's reasoning. The Talen-PJM ISA states that in this event, the nuclear datacenter will separate from the plant and get its power elsewhere, but AEP and Exelon are skeptical and want to know how exactly that would work.

The complaint argues Amazon's DC is essentially using the grid, saying the "premise" of the tweaked ISA "is that this datacenter co-located load is like load on a remote island -- one that simply has no impact on the PJM grid and would thus be properly excluded from economic and other responsibility for maintaining the PJM grid. But that storyline does not stand up to scrutiny." "They present their filing as no more than a replacement of older agreements with updated terms and 'clarifications' regarding the parties' roles and obligations," the two utility companies told [PDF] FERC, requesting a hearing over the matter. "The filing [new ISA] casts the submission as a mere housekeeping exercise, as if there is nothing to see here."

The protest adds: "The co-located load should not be allowed to operate as a free rider, making use of, and receiving the benefits of, a transmission system paid for by transmission ratepayers. We have no objection to co-location per se, but such load should pay its fair share of system use and other charges, just like other loads and customers." AEP and Exelon claim the new terms of the ISA contains a key loophole that hinges on the datacenter's co-location with the nuclear power plant, which allows its power usage to not be considered "in-network," even though the power load is synced to PJM's grid and could theoretically get power from it. [...] The end result, or so AEP and Exelon allege, is that Talen would be able to benefit from PJM's services without the associated cost. That would cost other customers between $58 million and $140 million per year overall, according to an analysis from Concentric Energy Advisors CEO Danielle Powers and chairman John Reed included with the filing. AEP and Exelon asked FERC to either hold a hearing to answer questions it feels are unresolved or to reject the new ISA outright. For its part, Talen claims the complaint's narrative is "demonstrably false" and that "transmission is not implicated."

An anonymous reader quotes a report from ZDNet, written by Steven Vaughan-Nichols: Hold onto your SSH keys, folks! A critical vulnerability has just rocked OpenSSH, Linux's secure remote access foundation, causing seasoned sysadmins to break out in a cold sweat. Dubbed "regreSSHion" and tagged as CVE-2024-6387, this nasty bug allows unauthenticated remote code execution (RCE) on OpenSSH servers running on glibc-based Linux systems. We're not talking about some minor privilege escalation here -- this flaw hands over full root access on a silver platter. For those who've been around the Linux block a few times, this feels like deja vu. The vulnerability is a regression of CVE-2006-5051, a bug patched back in 2006. This old foe somehow snuck back into the code in October 2020 with OpenSSH 8.5p1. Thankfully, the Qualys Threat Research Unit uncovered this digital skeleton in OpenSSH's closet. Unfortunately, this vulnerability affects the default configuration and doesn't need any user interaction to exploit. In other words, it's a vulnerability that keeps security professionals up at night.

It's hard to overstate the potential impact of this flaw. OpenSSH is the de facto standard for secure remote access and file transfer in Unix-like systems, including Linux and macOS. It's the Swiss Army knife of secure communication for sysadmins and developers worldwide. The good news is that not all Linux distributions have the vulnerable code. Old OpenSSH versions earlier than 4.4p1 are vulnerable to this signal handler race condition unless they are patched for CVE-2006-5051 and CVE-2008-4109. Versions from 4.4p1 up to, but not including, 8.5p1 are not vulnerable. The bad news is that the vulnerability resurfaced in OpenSSH 8.5p1 up to, but not including, 9.8p1 due to the accidental removal of a critical component. Qualys has found over 14 million potentially vulnerable OpenSSH server internet instances. The company believes that approximately 700,000 of these external internet-facing instances are definitely vulnerable. A patch, OpenSSH 9.8/9.8p1 is now available. Many, but not all, Linux distributions have made it available. If you can get it, install it as soon as possible. If for whatever reason you're not able to install a patch, Vaughan-Nichols recommends you set LoginGraceTime to 0 in the sshd configuration file and use network-based controls to restrict SSH access, while also configuring firewalls and monitoring tools to detect and block exploit attempts.

Digital audio broadcasting (DAB+) and internet radio have largely replaced traditional FM radio in Switzerland, with digital radio holding an 81% share of use in spring 2023. Due to the high costs of maintaining FM transmitters and declining financial resources, Switzerland plans to fully transition to digital radio by the end of 2026, phasing out FM broadcasting completely. From a report: DAB+ and the internet offer better quality and a larger program selection, are more energy and cost efficient, and can provide additional information in text and images, it said. To receive DAB+ requires a corresponding device or adapter, and new cars have been equipped with digital technology as standard for several years. In addition, the Federal Roads Office (FEDRO) will upgrade all tunnels on the national road network for digital radio reception by the end of the year and switch off FM transmitters.

FM was originally expected to be switched off throughout Switzerland by the end of 2024. The government extended FM licenses for the radio industry for the last time in October 2023 to the end of 2026, after which radio stations in Switzerland will no longer be able to broadcast via FM, only digitally. OFCOM announced at the time that the final extension would give the radio industry the flexibility to complete the transition process from analogue to digital radio.

An anonymous reader shares a report: Amazon built a $2 trillion company through years of aggressive spending on its retail and logistics businesses. Its future gains will likely be determined by the billions designated to fund its artificial-intelligence push. Amazon is planning to spend more than $100 billion over the next decade on data centers, an impressive level of investment even for a company known for its spending ways. The Seattle company is now devoting more investment money to its cloud computing and AI infrastructure than to its sprawling network of e-commerce warehouses.

Amazon Web Services, the arm that manages Amazon's cloud business, has opened data centers for years, but executives said there is a surge in investment now to meet demand triggered by the excitement around AI. "We have to dive in. We have to figure it out," said John Felton, who took over as AWS's chief financial officer this year after spending most of his career in Amazon's retail fulfillment operations. The company's financial commitment reflects the importance and high costs of AI. Felton said building for AI today feels like building that massive delivery network in years past. "It's a little uncertain," he said. AWS is expanding in Virginia, Ohio and elsewhere.

The Los Angeles Times reports that "multiple projects seeking to remove carbon dioxide from the air have been launched across Los Angeles County: When completed, Project Monarch and its wastewater component, Pure Water Antelope Valley, will purify up to 4.5 million gallons of water each day and capture 25,000 tons of atmospheric CO2 each year. (The typical gasoline-powered automobile spews 4.6 tons of carbon each year, according to the Environmental Protection Agency).... But the Palmdale project isn't the only new carbon-capture development in L.A. County. On Friday, officials from CarbonCapture Inc. gathered in Long Beach to introduce the first commercial-scale U.S. direct air capture, or DAC, system designed for mass production. The unit, which resembles a shipping container, can remove more than 500 tons of atmospheric CO2 per year... The L.A.-based company also announced that it will mass-produce up to 4,000 of its DAC modules annually at a new facility in Mesa, Arizona. It joins similar efforts from L.A.-based Captura, which is working to remove CO2 from the upper ocean; L.A.-based Avnos, which produces water while capturing carbon; and L.A.-based Equatic, which is working to remove atmospheric CO2 using the ocean...

[Equatic's] San Pedro facility pumps seawater through a series of electric plates that separate the water into hydrogen and oxygen as well as acidic and alkaline streams of liquid. The alkaline, or base, stream is exposed to the atmosphere, where it mineralizes CO2 into carbonates that are then dissolved and discharged back into the ocean for permanent storage, operators say Additionally, the hydrogen produced by the process is carbon-negative, making it a source of renewable energy that can be used to fuel the CO2 removal process or sold to other users, said Edward Sanders, chief operating officer at Equatic.

Equatic announced this month that it will partner with a Canadian carbon removal project developer, Deep Sky, to build North America's first commercial-scale ocean-based CO2 removal plant in Quebec, following the success of its effort in Los Angeles as well as another facility in Singapore. While the San Pedro facility can capture about 40 tons of CO2 per year, the Quebec facility will capture about 100,000 tons per year, Sanders said.
Meanwhile, two new projects by direct air capture company Heirloom were announced this week in Louisiana. Those projects are "expected to remove hundreds of thousands of tons of carbon dioxide from the air per year," according to the Associated Press, "and store it deep underground... part of "a slew of carbon removal and storage projects that have been announced in Louisiana." Heirloom estimates that they will eventually remove 320,000 tons of carbon dioxide each year... The company uses limestone, a natural absorbent, to extract carbon dioxide from the air. Heirloom's technology reduces the time it takes to absorb carbon dioxide in nature from years to just three days, according to the company's press release. The carbon dioxide is then removed from the limestone material and stored permanently underground.
In May America's Energy department also announced $3.5 billion in funding for its carbon-capture program — four large-scale, regional direct air capture hubs "that each comprise a network of carbon dioxide removal projects..." The hubs will have the capacity to capture and then permanently store at least one million metric tons of CO2 from the atmosphere annually, either from a single unit or from multiple interconnected units.
And Shell Canada has a pair of carbon capture projects in Alberta it expects to have operational toward the end of 2028, according to the CBC: The Polaris project is designed to capture about 650,000 tonnes of carbon dioxide annually from the Scotford complex. That works out to approximately 40 per cent of Scotford's direct CO2 emissions from the refinery and 22 per cent of its emissions from the chemicals complex.

Researchers in South Africa have injected radioactive material into the horns of 20 rhinos to deter poaching, aiming to leverage existing radiation detectors at borders for early detection and interception of trafficked horns. The Associated Press reports: The research, which has included the participation of veterinarians and nuclear experts, begins with the animal being tranquilized before a hole is drilled into its horn and the nuclear material carefully inserted. This week, researchers at the University of the Witwatersrand's Radiation and Health Physics Unit in South Africa injected 20 live rhinos with these isotopes. They hope the process can be replicated to save other wild species vulnerable to poaching -- like elephants and pangolins. "We are doing this because it makes it significantly easier to intercept these horns as they are being trafficked over international borders, because there is a global network of radiation monitors that have been designed to prevent nuclear terrorism," said Professor James Larkin, who heads the project. "And we're piggybacking on the back of that."

According to figures by the International Union for Conservation of Nature, an international conservation body, the global rhino population stood at around 500,000 at the beginning of the 20th century. It now stands at around 27,000 due to continued demand for rhino horns on the black market. South Africa has the largest population of rhinos with an estimated 16,000, making it a hotspot with over 500 rhinos killed yearly. [...] While the idea has received support from some in the industry, the researchers have had to jump many ethical hurdles posed by critics of their methodology.

Pelham Jones, chairperson of the Private Rhino Owners Association, is among the critics of the proposed method and doubts that it would effectively deter poachers and traffickers. "(Poachers) have worked out other ways of moving rhino horn out of the country, out of the continent or off the continent, not through traditional border crossings," he said. "They bypass the border crossings because they know that is the area of the highest risk of confiscation or interception." Professor Nithaya Chetty, dean of the science faculty at Witwatersrand, said the dosage of the radioactivity is very low and its potential negative impact on the animal was tested extensively.

According to Interpol, nearly 4,000 people around the world have been arrested for a variety of online crimes, with $257 million in assets seized. The Record reports: The operation, dubbed First Light, was conducted by police officers from 61 countries and targeted phishing, investment fraud, fake online shopping sites, romance scams, and impersonation scams, according to a statement by Interpol. In addition to arresting thousands of potential cybercriminals, the police also identified over 14,600 other possible suspects across all continents.

During the searches, law enforcement seized suspects' real estate, high-end vehicles, expensive jewelry, and many other high-value items and collections. They also froze 6,745 bank accounts used for transferring money obtained through illegal operations. In one case, the police intercepted $331,000 gleaned from a business email compromise fraud involving a Spanish victim who unknowingly transferred money to someone in Hong Kong. In another case, authorities in Australia successfully recovered $3.7 million on behalf of an impersonation scam victim after the funds were fraudulently transferred to bank accounts in Malaysia and Hong Kong.

The criminal networks identified during the operation were spread around the globe. In Namibia, for example, the police rescued 88 local youths who were forced into conducting scams as part of a sophisticated international crime network, according to Interpol. Law enforcement from Singapore, Hong Kong, and China prevented an attempted tech support scam, saving a 70-year-old victim from losing $281,200 worth of savings.

TeamViewer, the company that makes widely used remote access tools for companies, has confirmed an ongoing cyberattack on its corporate network. TechCrunch: In a statement Friday, the company attributed the compromise to government-backed hackers working for Russian intelligence, known as APT29 (and Midnight Blizzard). The Germany-based company said its investigation so far points to an initial intrusion on June 26 "tied to credentials of a standard employee account within our corporate IT environment."

TeamViewer said that the cyberattack "was contained" to its corporate network and that the company keeps its internal network and customer systems separate. The company added that it has "no evidence that the threat actor gained access to our product environment or customer data." Martina Dier, a spokesperson for TeamViewer, declined to answer a series of questions from TechCrunch, including whether the company has the technical ability, such as logs, to determine what, if any, data was accessed or exfiltrated from its network.

A study from the National Institutes of Health (NIH) analyzed data from nearly 400,000 healthy adults over 20 years and determined that "multivitamin use to improve longevity is not supported." The findings were published in JAMA Network Open. ABC News reports: The study found no evidence that daily multivitamin consumption reduced the risk of death from conditions such as heart disease or cancer. Rather than living longer, otherwise healthy people who took daily multivitamins were slightly more likely (4%) than non-users to die in the study period, according to researchers. Researchers reported nearly 165,000 deaths occurring during the follow-up period of the study, out of the initial group of 390,000 participants. The study, however, did not analyze data from people with pre-existing vitamin deficiencies. "What this study shows is that, generally, multivitamins aren't going to help you live longer," Dr. Jade A Cobern, MD, MPH, board-certified physician in pediatrics and general preventive medicine, told ABC News. "Even though the cost of many multivitamins isn't high, this is still an expense that many people can be spared from."

Al Michaels, the 79-year-old American broadcaster, who first covered the Olympics decades ago, is returning to broadcasting via an AI clone. NBCUniversal and Peacock will use AI-generated narration by Al Michaels for daily customized highlight reels of the Summer Olympics. Officials say they anticipate seven million different variations of the customized highlights throughout the games. The New York Times reports: Al Michaels, the 79-year-old American broadcaster, who first covered the Olympics decades ago, is coming back to primetime. It does raise a key question, one that recalls Mr. Michaels's most famous Olympic call: Do NBCUniversal executives believe in miracles? NBC has been exclusively broadcasting the Olympics in the United States since 1996, and the network frequently finds itself subject to intense public scrutiny for its coverage of the Games. [...]

Subscribers who want the daily Peacock highlight reel will be able choose the Olympic events that interest them most, and the types of highlights they want to see, such as viral clips, gold medalists or elimination events. From there, Peacock's A.I. machines will get to work each evening cranking out the most notable moments and putting them together in a tidy customized package. Mr. Michaels's recreated voice will be piped over the reels. (Humans will make quality control checks on the A.I. highlight reels.)

Many American subscribers are unable to use their phones overseas because all three major U.S. carriers are experiencing outages. According to The Register, the outages have been ongoing for several hours and stem from third-party communications technology company Syniverse. From the report: "Since the onset of these issues, Syniverse has been working closely with our network partners to restore full service," Syniverse, a US-based comms provider that focuses on roaming services, said in a statement confirming the breakdown. "We understand the inconvenience this has caused and appreciate your patience as we navigate this challenge."

"We're one of several providers impacted by a third-party vendor's issue that is intermittently affecting some international roaming service," T-Mo told us. "We're working with them to resolve it." Similarly, AT&T stated: "The AT&T network is operating normally. Some customers traveling internationally may be experiencing service disruptions due to an issue outside the AT&T network. We're working with one of our roaming connectivity providers to resolve the issue." Likewise, Verizon said, "An international third party communications provider is having issues with making voice and data connections with US based customers traveling overseas."

The international roaming outage has hit users' ability to do calls and texts, and reach the internet. According to Verizon, it's not a complete blackout. "70 percent of calls and data connections are going through at this time," the carrier firm told The Register in the past hour or so. Developing...

An anonymous reader quotes a report from Wired: In recent weeks, as so-called cheap fake video clips suggesting President Joe Biden is unfit for office have gone viral on social media, a Kremlin-affiliated disinformation network has been promoting a parody music video featuring Biden wearing a diaper and being pushed around in a wheelchair. The video is called "Bye, Bye Biden" and has been viewed more than 5 million times on X since it was first promoted in the middle of May. It depicts Biden as senile, wearing a hearing aid, and taking a lot of medication. It also shows him giving money to a character who seems to represent illegal migrants while denying money to US citizens until they change their costume to mimic the Ukrainian flag. Another scene shows Biden opening the front door of a family home that features a Confederate flag on the wall and allowing migrants to come in and take over. Finally, the video contains references to stolen election conspiracies pushed by former president Donald Trump.

The video was created by Little Bug, a group that mimics the style of Little Big, a real Russian band that fled the country in 2022 following Russia's invasion of Ukraine. The video features several Moscow-based actors -- who spoke with Russian media outlet Agency.Media -- but also appears to use artificial intelligence technology to make the actors resemble Biden and Trump, as well as Ilya Prusikin, the lead singer of Little Big. "Biden and Trump appear to be the same actor, with deepfake video-editing changing his facial features until he resembles Biden in one case and Trump in the other case," says Alex Fink, an AI and machine-vision expert who analyzed the video for WIRED. "The editing is inconsistent, so you can see that in some cases he resembles Biden more and in others less. The facial features keep changing." An analysis by True Media, a nonprofit that was founded to tackle the spread of election-related deepfakes, found with 100 percent confidence that there was AI-generated audio used in the video. It also assessed with 78 percent confidence that some AI technology was used to manipulate the faces of the actors.

Fink says the obvious nature of the deepfake technology on display here suggests that the video was created in a rush, using a small number of iterations of a generative adversarial network in order to create the characters of Biden and Trump. It is unclear who is behind the video, but "Bye, Bye Biden" has been promoted by the Kremlin-aligned network known as Doppelganger. The campaign posted tens of thousands of times on X and was uncovered by Antibot4Navalny, an anonymous collective of Russian researchers who have been tracking Doppelganger's activity for the past six months. The campaign first began on May 21, and there have been almost 4,000 posts on X promoting the video in 13 languages that were promoted by a network of almost 25,000 accounts. The Antibot4Navalny researchers concluded that the posts were written with the help of generative AI technology. The video has been shared 6.5 million times on X and has been viewed almost 5 million times.

Chinese undersea cable manufacturers are experiencing growth despite U.S. trade restrictions, as Beijing pushes for self-sufficiency in critical communications infrastructure, Nikkei is reporting. FiberHome International Technologies, blacklisted by the U.S. in 2020, reports increased business due to China's drive for domestic production. "We don't need foreign technologies," a FiberHome executive told Nikkei.

China is challenging U.S. dominance in the global undersea cable network, which spans 1.4 million kilometers and carries over 95% of the world's data. Chinese companies are expected to contribute 45% of new cable installations from 2023 to 2028, according to industry estimates. The Asia-Pacific region leads in subsea cable investment, with China spearheading multiple projects connecting to Southeast Asian nations. This aligns with President Xi Jinping's "Digital Silk Road" initiative, part of the broader Belt and Road strategy. However, geopolitical tensions have led to project delays and increased costs. Some cables are being rerouted to avoid contentious areas like the South China Sea, potentially creating parallel networks divided along geopolitical lines.

Researchers from UC Santa Cruz, UC Davis, LuxiTech, and Soochow University have developed a new method to run AI language models more efficiently by eliminating matrix multiplication, potentially reducing the environmental impact and operational costs of AI systems. Ars Technica's Benj Edwards reports: Matrix multiplication (often abbreviated to "MatMul") is at the center of most neural network computational tasks today, and GPUs are particularly good at executing the math quickly because they can perform large numbers of multiplication operations in parallel. [...] In the new paper, titled "Scalable MatMul-free Language Modeling," the researchers describe creating a custom 2.7 billion parameter model without using MatMul that features similar performance to conventional large language models (LLMs). They also demonstrate running a 1.3 billion parameter model at 23.8 tokens per second on a GPU that was accelerated by a custom-programmed FPGA chip that uses about 13 watts of power (not counting the GPU's power draw). The implication is that a more efficient FPGA "paves the way for the development of more efficient and hardware-friendly architectures," they write.

The paper doesn't provide power estimates for conventional LLMs, but this post from UC Santa Cruz estimates about 700 watts for a conventional model. However, in our experience, you can run a 2.7B parameter version of Llama 2 competently on a home PC with an RTX 3060 (that uses about 200 watts peak) powered by a 500-watt power supply. So, if you could theoretically completely run an LLM in only 13 watts on an FPGA (without a GPU), that would be a 38-fold decrease in power usage. The technique has not yet been peer-reviewed, but the researchers -- Rui-Jie Zhu, Yu Zhang, Ethan Sifferman, Tyler Sheaves, Yiqiao Wang, Dustin Richmond, Peng Zhou, and Jason Eshraghian -- claim that their work challenges the prevailing paradigm that matrix multiplication operations are indispensable for building high-performing language models. They argue that their approach could make large language models more accessible, efficient, and sustainable, particularly for deployment on resource-constrained hardware like smartphones. [...]

The researchers say that scaling laws observed in their experiments suggest that the MatMul-free LM may also outperform traditional LLMs at very large scales. The researchers project that their approach could theoretically intersect with and surpass the performance of standard LLMs at scales around 10^23 FLOPS, which is roughly equivalent to the training compute required for models like Meta's Llama-3 8B or Llama-2 70B. However, the authors note that their work has limitations. The MatMul-free LM has not been tested on extremely large-scale models (e.g., 100 billion-plus parameters) due to computational constraints. They call for institutions with larger resources to invest in scaling up and further developing this lightweight approach to language modeling.

An anonymous reader quotes a report from TorrentFreak: Last week, an in-depth investigative report from JBTC revealed that Korean Internet provider KT, formerly known as Korea Telecom, distributed malware onto subscribers' computers to interfere with and block torrent traffic. File-sharing continues to be very popular in South Korea, but operates differently than in most other countries. "Webhard" services, short for Web Hard Drive, are particularly popular. These are paid BitTorrent-assisted services, which also offer dedicated web seeds, to ensure that files remain available.

Webhard services rely on the BitTorrent-enabled 'Grid System', which became so popular in Korea that ISPs started to notice it. Since these torrent transfers use a lot of bandwidth, which is very costly in the country, providers would rather not have this file-sharing activity on their networks. KT, one of South Korea's largest ISPs with over 16 million subscribers, was previously caught meddling with the Grid System. In 2020, their throttling activities resulted in a court case, where the ISP cited 'network management' costs as the prime reason to interfere. The Court eventually sided with KT, ending the case in its favor, but that wasn't the end of the matter. An investigation launched by the police at the time remains ongoing. New reports now show that the raid on KT's datacenter found that dozens of devices were used in the 'throttling process' and they were doing more than just limiting bandwidth.

When Webhard users started reporting problems four years ago, they didn't simply complain about slow downloads. In fact, the main concern was that several Grid-based Webhard services went offline or reported seemingly unexplainable errors. Since all complaining users were KT subscribers, fingers were pointed in that direction. According to an investigation by Korean news outlet JBTC, the Internet provider actively installed malware on computers of Webhard services. This activity was widespread and effected an estimated 600,000 KT subscribers. The Gyeonggi Southern Police Agency, which carried out the raid and investigation, believes this was an organized hacking attempt. A dedicated KT team allegedly planted malware to eavesdrop on subscribers and interfere with their private file transfers. [...] Why KT allegedly distributed the malware and what it precisely intended to do is unclear. The police believe there were internal KT discussions about network-related costs, suggesting that financial reasons played a role.

An anonymous reader shares a report: We're at a transitional moment in streaming -- user growth is slowing and major players are looking to consolidate, but the long-promised dream of profitability finally seems within reach (especially if you're Netflix). The perfect time, then, for The New York Times to interview many of the industry's big names -- including Netflix co-CEO Ted Sarandos, Amazon's Prime Video head Mike Hopkins, and IAC chairman Barry Diller -- about what they think comes next.

There seemed to be broad agreement on most of the big themes: More ads, higher prices, and fewer big swings on prestige TV. These changes are all united by the shift towards profitability, rather than growth-at-all-costs. If the initial prices of many streaming services seemed unsustainably low at launch, it turns out they were -- prices have been steadily rising, while the streamers have also introduced more affordable subscription tiers for viewers who are willing to watch ads. In fact, some execs told The Times that streamers will keep raising prices for the ad-free tiers with the aim of pushing more customers to sign up for ad-supported subscriptions instead. The growth of ad-supported streaming could also affect the kinds of movies and shows that get produced, since advertisers generally want to reach a mass audience -- think of the heyday of ad-supported network TV, with its endless shows about doctors and cops, compared to the more ambitious fare on subscription-supported HBO.