RSS 생중계

Google today unveiled significant upgrades to its Gemini 2.5 AI models, introducing an experimental "Deep Think" reasoning mode for 2.5 Pro that allows the model to consider multiple hypotheses before responding. The new capability has achieved impressive results on complex benchmarks, scoring highly on the 2025 USA Mathematical Olympiad and leading on LiveCodeBench, a competition-level coding benchmark. Gemini 2.5 Pro also tops the WebDev Arena leaderboard with an ELO score of 1420. "Based on Google's experience with AlphaGo, AI model responses improve when they're given more time to think," said Demis Hassabis, CEO of Google DeepMind. The enhanced Gemini 2.5 Flash, Google's efficiency-focused model, has improved across reasoning, multimodality, and code benchmarks while using 20-30% fewer tokens. Both models now feature native audio capabilities with support for 24+ languages, thought summaries, and "thinking budgets" that let developers control token usage. Gemini 2.5 Flash is currently available in preview with general availability expected in early June, while Deep Think remains limited to trusted testers during safety evaluations.

Read more of this story at Slashdot.

Google is adding AI-powered live translation to Meet, enabling participants to converse in their native languages while the system automatically translates in real time with the speaker's original vocal characteristics intact. Initially launching with English-Spanish translation this week, the technology processes speech with minimal delay, preserving tone, cadence, and expressions -- creating an effect similar to professional dubbing but with the speaker's own voice, the company announced at its developer conference Tuesday. In some testings, WSJ found occasional limitations: initial sentences sometimes appear garbled before smoothing out, context-dependent words like "match" might translate imperfectly (rendered as "fight" in Spanish), and the slight delay can create confusing crosstalk with multiple participants. Google plans to extend support to Italian, German, and Portuguese in the coming weeks. The feature is rolling out to Google AI Pro and Ultra subscribers now, with enterprise availability planned later this year. The company says that no meeting data is stored when translation is active, and conversation audio isn't used to train AI models.

Read more of this story at Slashdot.

Ihor Solodrai has been working on the BPF subsystem's continuous-integration (CI) testing for the last six months. At the 2025 Linux Storage, Filesystem, Memory-Management, and BPF Summit, he remotely shared an update on his work, and solicited feedback on how the tests could be further improved. Much of the work he's done has been specific to the BPF subsystem, but some is more generic and could potentially be of use to other subsystems. He also shared some general lessons learned from working on the BPF CI tests.

Despite careful planning and months of warning, Debian developer Mo Zhou has acknowledged that the project needs more time to grapple with the questions around AI models and the Debian Free Software Guidelines (DFSG). For now, he has withdrawn his proposed General Resolution (GR) that would have required the original training data for AI models to be released in order to be considered DFSG-compliant—though the debates on the topic continue.

Adobe will rebrand its Creative Cloud All Apps subscription to "Creative Cloud Pro" on June 17 for North American users, making significant price increases while bundling AI features. Individual annual subscribers will see monthly rates jump from $59.99 to $69.99, while monthly non-contracted subscribers face a $15 hike to $104.99. The revamped plan includes unlimited generative AI image credits, 4,000 monthly "premium" AI video and audio credits, access to third-party models like OpenAI's GPT, and the beta Firefly Boards collaborative whiteboard. Adobe will also offer a cheaper "Creative Cloud Standard" option at $54.99 monthly with severely reduced AI capabilities, but this plan remains exclusive to existing subscribers -- forcing new customers into the pricier AI-focused tier.

Read more of this story at Slashdot.

Microsoft is starting to integrate AI shortcuts, or what it calls AI actions, into the File Explorer in Windows 11. From a report: These shortcuts let you right-click on a file and quickly get to Windows AI features like blurring the background of a photo, erasing objects, or even summarizing content from Office files. Four image actions are currently being tested in the latest Dev Channel builds of Windows 11, including Bing visual search to find similar images on the web, the blur background and erase objects features found in the Photos app, and the remove background option in Paint. Similar AI actions will soon be tested with Office files, The Verge added.

Read more of this story at Slashdot.

America's regional state universities are experiencing steep enrollment declines, triggering economic crises in the towns that depend on them, while flagship universities continue to thrive. At Western Illinois University in Macomb, enrollment has plummeted 47% since 2010, driving the city's population down 23% to 14,765. Empty dorms have been repurposed or demolished, while local businesses struggle to survive. "It's almost like you're watching the town die," Kalib McGruder, a 28-year veteran of the campus police department, told WSJ. An analysis of 748 public four-year institutions reveals enrollment at prestigious state universities increased 9% between 2015 and 2023, while regional state schools saw a 2% decline. The University of Tennessee Knoxville's enrollment jumped 30% as the state's regional colleges collectively fell 3%. With high school graduate numbers expected to decline starting next year after reaching a record high in 2024, the outlook for struggling college towns appears bleak.

Read more of this story at Slashdot.

French authorities have denied Telegram founder Pavel Durov's request to travel to the U.S. for "negotiations with investment funds." From a report: The Paris prosecutor's office told POLITICO that it rendered its decision on May 12 "on the grounds that such a trip abroad did not appear imperative or justified." Durov was arrested in August 2024 at a French airport and has been under strict legal control since last September, when he was indicted on six charges related to illicit activity on the messaging app he operates. He is forbidden to leave France without authorization -- which he obtained to travel to Dubai from March 15 to April 7, the prosecutor's office said. Russian-born Durov is a citizen, among other countries, of France and the United Arab Emirates.

Read more of this story at Slashdot.

Red Hat has announced the release of Red Hat Enterprise Linux (RHEL) 10. A blog post accompanying the release provides details on some of the more notable features, such as encrypted DNS, a developer preview of RHEL 10 for RISC-V, and image mode for RHEL using bootc.

Image mode for RHEL lets you deploy your OS as a bootc image to your hardware, virtual machine or cloud, and then layer your app on top of it. That's a far less complex operation than traditional packaged deployments, and it gives developers and image maintainers a common experience and total control over their environment.

RHEL 10 includes the 6.12.0 kernel, GCC 14.2, GNU Binutils 2.41, GNU C Library (glibc) 2.39, Python 3.12, Perl 5.40, and more. See the release notes for a full list of changes. LWN covered CentOS Stream 10 in December, which provided an early look at what would be in the RHEL 10 release.

An anonymous reader shares a report: Intel has considered divesting its network and edge businesses as the chipmaker looks to shave off parts of the company its new chief executive does not see as crucial, three sources familiar with the matter said. Talks about the potential sale of the group, once called NEX in Intel's financial results, are a part of CEO Lip-Bu Tan's strategy to focus its tens of thousands of employees on areas in which it has historically thrived: PC and data center chips.

Read more of this story at Slashdot.

Survey response rates have collapsed from 30-50% in the 1970s to as low as 5% today, while AI agents now account for an estimated 20% of survey responses, according to a new analysis. The UK's Office for National Statistics has seen response rates drop from 40% to 13%, with some labor market questions receiving only five human responses. The U.S. Current Population Survey hit a record low 12.7% response rate, down from 50% historically.

Read more of this story at Slashdot.

The US tech sector shed 214,000 jobs in April amid continuing economic uncertainty, according to CompTIA analysis of Bureau of Labor Statistics data. Companies are extending hiring timelines to two or three times longer than last year while significantly raising skill requirements, particularly for AI competencies. "It's the great hesitation," said George Denlinger of Robert Half, noting employers now demand 10-12 skills instead of 6-7 previously. Entry-level programming positions are disappearing as AI assumes those functions, with Janco Associates CEO Victor Janulaitis observing that "a job that has been eliminated from almost all IT departments is an entry-level IT programmer."

Read more of this story at Slashdot.

Security updates have been issued by Debian (firefox-esr, openjdk-11, openjdk-17, and wireless-regdb), Fedora (iputils, open-vm-tools, sfnt2woff-zopfli, and woff), Red Hat (postgresql:12), SUSE (apache2-mod_auth_openidc, brltty, helm, python-maturin, and rubygem-rack), and Ubuntu (linux-azure-fips).

Klarna's net loss more than doubled in the first quarter [non-paywalled link] as more consumers failed to repay loans from the Swedish "buy now, pay later" lender as concerns rose about the financial health of US consumers. Financial Times: The fintech, which offers interest-free consumer loans to allow customers to make retail purchases, on Monday reported a net loss of $99 million for the three months to March, up from $47 million a year earlier. The company, which makes money by charging fees to merchants and to consumers who fail to repay on time, said its customer credit losses had risen to $136 million, a 17% year-on-year increase. The increased failure to repay comes on the back of gloomy economic sentiment in the US, where a closely watched measure of consumers' confidence last week fell to its second-lowest level on record. US President Donald Trump's trade war has driven expectations of higher inflation. Further reading: The Klarna Hype Machine.

Read more of this story at Slashdot.

SAG-AFTRA has filed a labor complaint against Fortnite developer Epic Games, alleging the game improperly used AI to replicate James Earl Jones' Darth Vader voice without bargaining with the union, despite the estate's approval. Gizmodo reports: The union has now filed an unfair labor practice charge (link to the PDF is on the SAG-AFTRA website) that calls out "Fortnite's signatory company, Llama Productions" for "[replacing] the work of human performers with AI technology" without "providing any notice of their intent to do this and without bargaining with us over appropriate terms." The union notes that it's not against the general idea here: "We celebrate the right of our members and their estates to control the use of their digital replicas and welcome the use of new technologies to allow new generations to share in the enjoyment of those legacies and renowned roles." The problem is that the AI being used here makes human voice actors obsolete, and "we must protect our right to bargain terms and conditions around uses of voice that replace the work of our members, including those who previously did the work of matching Darth Vader's iconic rhythm and tone in video games." So far there's been no response from Epic Games on the filing. The Hollywood Reporter notes that despite the SAG-AFTRA's still-ongoing Interactive Media Agreement strike, which has been stuck for months on negotiating "AI protections for voice actors in video games," actors can actually work on Fortnite without violating the strike, since the game falls under an exception for titles that were in production before August 2023.

Read more of this story at Slashdot.

An anonymous reader quotes a report from Ars Technica: There's a lot of matter around, which ensures that any antimatter produced experiences a very short lifespan. Studying antimatter, therefore, has been extremely difficult. But that's changed a bit in recent years, as CERN has set up a facility that produces and traps antimatter, allowing for extensive studies of its properties, including entire anti-atoms. Unfortunately, the hardware used to capture antiprotons also produces interference that limits the precision with which measurements can be made. So CERN decided that it might be good to determine how to move the antimatter away from where it's produced. Since it was tackling that problem anyway, CERN decided to make a shipping container for antimatter, allowing it to be put on a truck and potentially taken to labs throughout Europe. [...] Overall, the hardware stayed cold, generally at a bit over 5 Kelvin. The exception was when the system was reconnected to the antimatter source hardware and the system reconnected to the electrical system at CERN. While those actions show up as temperature spikes, the superconducting magnets remained well under 7 Kelvin. An accelerometer was in place to track the forces experienced by the hardware while the truck was moving. This showed that changes in the truck's speed produced turbulence in the liquid helium, making measurements of its presence unreliable. Levels had dropped from about 75 percent of maximum to 30 percent by the time the system was reconnected, suggesting that liquid helium presents the key limiting factor in shipping. Measurements made while the system was in transit suggest that the whole process occurred losslessly, meaning that not a single proton escaped during the entire transport. The findings have been published in the journal Nature.

Read more of this story at Slashdot.

According to Bloomberg's Mark Gurman (paywalled), Apple has big plans to turn Siri into a true ChatGPT competitor. "A next-generation, chatbot version of Siri has reportedly made significant progress during testing over the past six months; some executives allegedly now see it as 'on par' with recent versions of ChatGPT," reports MacRumors. "Apple is also apparently discussing giving Siri the ability to access the internet to gather and synthesize data from multiple sources, just like ChatGPT." From the report: The report added that Apple now has artificial intelligence offices in Zurich, where employees are working on an all-new software architecture for Siri. This "monolithic model" is entirely built on an LLM engine that will eventually replace Siri's current "hybrid" architecture that has been incoherently layered up with different functionality over many years. The new model will make Siri more conversational and better at synthesizing information. Google's Gemini is expected to be added to iOS 19 as an alternative to ChatGPT in Siri, but Apple is also apparently in talks with Perplexity to add their AI service as another option in the future, for both Siri and Safari search.

Read more of this story at Slashdot.

Sony has begun mass suspensions of PlayStation accounts linked to Russian users, including those registered in other countries like Turkiye and Kazakhstan, in an effort to crack down on circumvention of regional restrictions like PS Plus access. This move follows Sony's 2022 exit from the Russian market due to the war in Ukraine. Ukrainska Pravda reports: Russians are trying to bypass the blocks and are contacting PlayStation support, but these attempts are to no avail. Users with purchased accounts encounter the most problems -- Sony checks IP addresses for verification. There are already many similar reports. In some cases, accounts with thousands of roubles worth of games purchased have been blocked. Russian media outlets like RT are reporting the news.

Read more of this story at Slashdot.

Qualcomm announced plans to re-enter the data center market with custom CPUs designed to integrate with Nvidia GPUs and software. As CNBC reports, the move supports Qualcomm's broader strategy to diversify beyond smartphones and into high-growth areas like data centers, PCs, and automotive chips. From the report: "I think we see a lot of growth happening in this space for decades to come, and we have some technology that can add real value added," Cristiano Amon, CEO of Qualcomm, told CNBC in an interview on Monday. "So I think we have a very disruptive CPU." Amon said the company will make an announcement about the CPU roadmap and the timing of its release "very soon," without offering specifics. The data center CPU market remains highly competitive. Big cloud computing players like Amazon and Microsoft already design and deploy their own custom CPUs. AMD and Intel also have a strong presence. Addressing the competition, Amon said that there will be a place for Qualcomm in the data center CPU space. "As long as ... we can build a great product, we can bring innovation, and we can add value with some disruptive technology, there's going to be room for Qualcomm, especially in the data center," Amon said. "[It] is a very large addressable market that will that will see a lot of investment for decades to come." Last week, Qualcomm signed a memorandum of understanding with Saudi-based AI frim Humain to develop data centers, joining a slew of U.S. tech companies making deals in the region. Humain will operate under Saudi Arabia's Public Investment Fund.

Read more of this story at Slashdot.

An anonymous reader quotes a report from Bloomberg: While using website data to build a Google Search topped with artificial intelligence-generated answers, an Alphabet executive acknowledged in an internal document that there was an alternative way to do things: They could ask web publishers for permission, or let them directly opt out of being included. But giving publishers a choice would make training AI models in search too complicated, the company concludes in the document, which was unearthed in the company's search antitrust trial. It said Google had a "hard red line" and would require all publishers who wanted their content to show up in the search page to also be used to feed AI features. Instead of giving options, Google decided to "silently update," with "no public announcement" about how they were using publishers' data, according to the document, written by Chetna Bindra, a product management executive at Google Search. "Do what we say, say what we do, but carefully." "It's a little bit damning," said Paul Bannister, the chief strategy officer at Raptive, which represents online creators. "It pretty clearly shows that they knew there was a range of options and they pretty much chose the most conservative, most protective of them -- the option that didn't give publishers any controls at all." For its part, Google said in a statement to Bloomberg: "Publishers have always controlled how their content is made available to Google as AI models have been built into Search for many years, helping surface relevant sites and driving traffic to them. This document is an early-stage list of options in an evolving space and doesn't reflect feasibility or actual decisions." They added that Google continually updates its product documentation for search online.

Read more of this story at Slashdot.