RSS 생중계

An anonymous reader quotes a report from The Verge: Google DeepMind is releasing a new version of its AI "world" model, called Genie 3, capable of generating 3D environments that users and AI agents can interact with in real time. The company is also promising that users will be able to interact with the worlds for much longer than before and that the model will actually remember where things are when you look away from them. [...] Genie 3 seems like it could be a notable step forward. Users will be able to generate worlds with a prompt that supports a "few" minutes of continuous interaction, which is up from the 10-20 seconds of interaction possible with Genie 2, according to a blog post. Google says that Genie 3 can keep spaces in visual memory for about a minute, meaning that if you turn away from something in a world and then turn back to it, things like paint on a wall or writing on a chalkboard will be in the same place. The worlds will also have a 720p resolution and run at 24fps. DeepMind is adding what it calls "promptable world events" into Genie 3, too. Using a prompt, you'll be able to do things like change weather conditions in a world or add new characters. The model is launching as "a limited research preview" available to "a small cohort of academics and creators," according to Google. It's "exploring" how to bring Genie 3 to "additional testers."

Read more of this story at Slashdot.

Standard DDR4 DRAM prices doubled between May and June 2025, with 8-gigabit units reaching $4.12 and 4-gigabit units hitting $3.14 -- the latter's highest level since July 2021, according to electronics trading companies cited by Nikkei Asia. The unprecedented single-month doubling follows speculation that Chinese manufacturer ChangXin Memory Technologies has halted DDR4 production to shift factories toward DDR5 memory for AI applications. DDR4 currently comprises 60% of desktop PC memory while DDR5 accounts for 40%, per Tokyo-based BCN research. Samsung Electronics, SK Hynix, and Micron Technology controlled 90% of the global DRAM market in Q2 2025.

Read more of this story at Slashdot.

The U.S. Transportation Department is proposing new rules to speed deployment of drones beyond the visual line of sight of operators, a key change needed to advance commercial uses like package deliveries. From a report: "We are going to unleash American drone dominance," Transportation Secretary Sean Duffy said at a press conference on Tuesday. Under current rules, operators need to get individual waivers or exemptions to use drones without visual line of sight. The department said eliminating those requirements "will significantly expand the use-case for drone technologies in areas like: manufacturing, farming, energy production, filmmaking, and the movement of products including lifesaving medications." The proposal includes new requirements for manufacturers, operators, and drone traffic-management services to keep drones safely separated from other drones and airplanes. "It's going to change the way that people and products move throughout our airspace... so you may change the way you get your Amazon package, you may get a Starbucks cup of coffee from a drone," Duffy said.

Read more of this story at Slashdot.

Nearly one-third of all retracted papers at PLoS ONE can be traced back to just 45 researchers who served as editors at the journal, an analysis of its publication records has found. Nature: The study, published in Proceedings of the National Academy of Sciences (PNAS), found that 45 editors handled only 1.3% of all articles published by PLoS ONE from 2006 to 2023, but that the papers they accepted accounted for more than 30% of the 702 retractions that the journal issued by early 2024. Twenty-five of these editors also authored papers in PLoS ONE that were later retracted. The PNAS authors did not disclose the names of any of the 45 editors. But, by independently analysing publicly available data from PLoS ONE and the Retraction Watch database, Nature's news team has identified five of the editors who handled the highest number of papers that were subsequently retracted by the journal. Together, those editors accepted about 15% of PLoS ONE's retracted papers up to 14 July.

Read more of this story at Slashdot.

OpenAI has released two open-weight language models, marking the startup's first such release since GPT-2 in 2019. The models, gpt-oss-120b and gpt-oss-20b, can run locally on consumer devices and be fine-tuned for specific purposes. Both models use chain-of-thought reasoning approaches first deployed in OpenAI's o1 model and can browse the web, execute code, and function as AI agents. The smaller 20-billion-parameter model runs on consumer devices with 16 GB of memory. Gpt-oss-120B model will require about 80 GB of memory. OpenAI said the 120-billion-parameter model performs similarly to the company's proprietary o3 and o4-mini models. The models are available free on Hugging Face under the Apache 2.0 license after safety testing that delayed their March announcement.

Read more of this story at Slashdot.

Proxmox Virtual Environment 9.0, based on Debian 13 ("trixie"), has been released. Notable new features include snapshots for thick-provisioned LVM shared storage, affinity rules for high availability (HA) clusters, and a modernized mobile web interface for managing Proxmox systems. See the release notes and known issues for more details about the release.

The Government Accountability Office has issued reports criticizing the Department of Homeland Security, Environmental Protection Agency, and General Services Administration for failing to implement critical IT and cybersecurity recommendations. DHS leads with 43 unresolved recommendations dating to 2018, including seven priority matters. The EPA has 11 outstanding items, including failures to submit FedRAMP documentation and conduct organization-wide cybersecurity risk assessments. GSA has four pending recommendations. All three agencies failed to properly log cybersecurity events and conduct required annual IT portfolio reviews. The DHS' HART biometric program remains behind schedule without proper cost accounting or privacy controls, with all nine 2023 recommendations still open.

Read more of this story at Slashdot.

Wikipedia editors have adopted a policy enabling administrators to delete AI-generated articles without the standard week-long discussion period. Articles containing telltale LLM responses like "Here is your Wikipedia article on" or "Up to my last training update" now qualify for immediate removal. Articles with fabricated citations -- nonexistent papers or unrelated sources such as beetle research cited in computer science articles -- also meet deletion criteria.

Read more of this story at Slashdot.

Palantir chief executive Alex Karp has told analysts and investors that the company treats Harvard, Princeton and Yale graduates the same as those without college degrees, calling employment at the data analytics firm "a new credential independent of class and background." During the earnings call Monday where Palantir reported its first billion-dollar revenue quarter, Karp said university graduates come to the company after being "engaged in platitudes" and claimed workers without college degrees sometimes create more value than degree holders using Palantir products. The company launched its Meritocracy Fellowship this spring to recruit talent outside traditional university pathways.

Read more of this story at Slashdot.

An anonymous reader shares a report: Microsoft has published a new video that appears to be the first in an upcoming series of videos dubbed "Windows 2030 Vision," where the company outlines its vision for the future of Windows over the next five years. It curiously makes references to some potentially major changes on the horizon, in the wake of AI. This first episode features David Weston, Microsoft's Corporate Vice President of Enterprise & Security, who opens the video by saying "the world of mousing and keyboarding around will feel as alien as it does to Gen Z [using] MS-DOS." Right out of the gate, it sounds like he's teasing the potential for a radical new desktop UX made possible by agentic AI. Weston later continues, "I truly believe the future version of Windows and other Microsoft operating systems will interact in a multimodal way. The computer will be able to see what we see, hear what we hear, and we can talk to it and ask it to do much more sophisticated things."

Read more of this story at Slashdot.

The use of huge pages can significantly increase the performance of many workloads by reducing both memory-management overhead in the kernel and pressure on the system's translation lookaside buffer (TLB). The addition of transparent huge pages (THP) for the 2.6.38 kernel release in 2011 caused the kernel to allocate huge pages automatically to make their benefits available to all workloads without any effort needed on the user-space side. But it turns out that use of huge pages can make some workloads slower as the result of internal memory fragmentation, so the THP feature is often disabled. Two patch sets aimed at better targeting the use of transparent huge pages are currently working their way through the review process.

AI meeting transcription software is inadvertently sharing private conversations with all meeting participants through automated summaries. WSJ found a series of mishaps that people confirmed on-record. Digital marketing agency owner Tiffany Lewis discovered her "Nigerian prince" joke about a potential client was included in the summary sent to that same client. Nashville branding firm Studio Delger received meeting notes documenting their discussion about "getting sandwich ingredients from Publix" and not liking soup when their client failed to appear. Communications agency coordinator Andrea Serra found her personal frustrations about a neighborhood Whole Foods and a kitchen mishap while making sweet potato recipes included in official meeting recaps distributed to colleagues.

Read more of this story at Slashdot.

An anonymous reader shares a report: A researcher has scraped nearly 100,000 conversations from ChatGPT that users had set to share publicly and Google then indexed, creating a snapshot of all the sorts of things people are using OpenAI's chatbot for, and inadvertently exposing. 404 Media's testing has found the dataset includes everything from the sensitive to the benign: alleged texts of non-disclosure agreements, discussions of confidential contracts, people trying to use ChatGPT to understand their relationship issues, and lots of people asking ChatGPT to write LinkedIn posts. The news follows a July 30 Fast Company article which reported "thousands" of shared ChatGPT chats were appearing in Google search results. People have since dug through some of the chats indexed by Google. The around 100,000 conversation dataset provides a better sense of the scale of the problem, and highlights some of the potential privacy risks in using any sharing features of AI tools. OpenAI did not dispute the figure of around 100,000 indexed chats when contacted for comment.

Read more of this story at Slashdot.

The call for topics for the 2025 Maintainers Summit has been posted. The Summit, to be held in Tokyo on December 10, will involve around 30 developers gathered to discuss development-process issues for the kernel. Anybody who is interested in attending is encouraged to post a nomination along with the topic they would like to discuss. Nominations and topics are best sent before September 10.

The call for topics for the Kernel Summit, which runs as a Linux Plumbers Conference track, is also out.

The U.S. Coast Guard determined the implosion of the Titan submersible that killed five people while traveling to the wreckage of the Titanic was a preventable disaster caused by OceanGate Expeditions's inability to meet safety and engineering standards. WSJ: A 335-page report [PDF] detailing a two-year inquiry from the U.S. Coast Guard's Marine Board of Investigation found the company that owned and operated the Titan failed to follow maintenance and inspection protocols for the deep-sea submersible. OceanGate avoided regulatory review and managed the submersible outside of standard protocols "by strategically creating and exploiting regulatory confusion and oversight challenges," the report said. The Coast Guard opened its highest-level investigation into the event in June 2023, shortly after the implosion occurred. "There is a need for stronger oversight and clear options for operators who are exploring new concepts outside of the existing regulatory framework," Jason Neubauer, the chair of the Coast Guard Marine Board of Investigation for the Titan submersible, said in a statement.

Read more of this story at Slashdot.

An anonymous reader shares a report: In a landmark move, Illinois state lawmakers have passed a bill banning AI from acting as a standalone therapist and placing firm guardrails on how mental health professionals can use AI to support care. Governor JB Pritzker signed the bill into law on Aug. 1. The legislation, dubbed the Wellness and Oversight for Psychological Resources Act, was introduced by Rep. Bob Morgan and makes one thing clear: only licensed professionals can deliver therapeutic or psychotherapeutic services to another human being. [...] Under the new state law, mental health providers are barred from using AI to independently make therapeutic decisions, interact directly with clients, or create treatment plans -- unless a licensed professional has reviewed and approved it. The law also closes a loophole that allows unlicensed persons to advertise themselves as "therapists."

Read more of this story at Slashdot.

Antonio Cuni, who is a longtime Python performance engineer and PyPy developer, gave a presentation at EuroPython 2025 about "Myths and fairy tales around Python performance" on the first day of the conference in Prague. As might be guessed from the title, he thinks that much of the conventional wisdom about Python performance is misleading at best. With lots of examples, he showed where the real problems that he sees lie. He has come to the conclusion that memory management will ultimately limit what can be done about Python performance, but he has an early-stage project called SPy that might be a way toward a super-fast Python.

Security updates have been issued by AlmaLinux (python-requests), Fedora (mingw-libxslt), Red Hat (gdk-pixbuf2, jq, kernel, mod_security, ncurses, nodejs:22, opentelemetry-collector, python-setuptools, python3-setuptools, python3.12-setuptools, qt5-qt3d, redis, redis:6, redis:7, sqlite, and unbound), SUSE (apache2, cairo, chromium, djvulibre, govulncheck-vulndb, grub2, java-11-openjdk, java-17-openjdk, liblua5_5-5, nvidia-open-driver-G06-signed, python, python310, python314, python39, redis, sqlite3, and systemd), and Ubuntu (apport, linux, linux-aws, linux-aws-hwe, linux-azure, linux-azure-4.15, linux-gcp, linux-gcp-4.15, linux-hwe, linux-kvm, linux-aws-fips, linux-azure-fips, linux-fips, linux-gcp-fips, linux-azure, and linux-oracle).

For years, whistle-blowers have warned that fake results are sneaking into the scientific literature at an increasing pace. A new statistical analysis backs up the concern. From a report: A team of researchers found evidence of shady organizations churning out fake or low-quality studies on an industrial scale. And their output is rising fast, threatening the integrity of many fields. "If these trends are not stopped, science is going to be destroyed," said LuÃs A. Nunes Amaral, a data scientist at Northwestern University and an author of the study, which was published in the Proceedings of the National Academy of Sciences on Monday. Science has made huge advances over the past few centuries only because new generations of scientists could read about the accomplishments of previous ones. Each time a new paper is published, other scientists can explore the findings and think about how to make their own discoveries. Fake scientific papers produced by commercial "paper mills" are doubling every year and a half, according to the report. Northwestern University researchers examined over one million papers and identified networks of fraudulent studies sold to scientists seeking to pad their publication records. The team estimates the actual scope of fraud may be 100 times greater than currently detected cases. Paper mills charge hundreds to thousands of dollars for fake authorship and often target specific research fields like microRNA cancer studies.

Read more of this story at Slashdot.

BrianFagioli shares a report from NERDS.xyz: Synchron has just released a public demo showing something that used to feel impossible. A man with ALS is now using his iPad with nothing but his brain. No hands. No voice. No eye-tracking. Just thought. The man in the video is named Mark. He's part of Synchron's COMMAND clinical study and has an implant called the Stentrode. It sits inside his brain's blood vessels and picks up his motor intention. Those signals get sent wirelessly to an external decoder, which then tells the iPad what to do. It's all made possible by Apple's new Brain-Computer Interface Human Interface Device protocol, which lets iPadOS treat brain activity like an actual input method. Apple's built-in Switch Control feature makes the whole thing work on the software side. The iPad even sends back screen context to the BCI decoder to make everything run more smoothly and accurately. [...] Synchron was the first company to start clinical trials with a permanently implanted BCI. The big difference here is that it doesn't require open brain surgery. The device is implanted through the blood vessels, which makes it way more practical for real-world use.

Read more of this story at Slashdot.