RSS 생중계

Leading AI systems are solving less than 2% of problems in a new advanced mathematics benchmark, revealing significant limitations in their reasoning capabilities, research group Epoch AI reported this week. The benchmark, called FrontierMath, consists of hundreds of original research-level mathematics problems developed in collaboration with over 60 mathematicians, including Fields Medalists Terence Tao and Timothy Gowers. While top AI models like GPT-4 and Gemini 1.5 Pro achieve over 90% accuracy on traditional math tests, they struggle with FrontierMath's problems, which span computational number theory to algebraic geometry and require complex reasoning. "These are extremely challenging. [...] The only way to solve them is by a combination of a semi-expert like a graduate student in a related field, maybe paired with some combination of a modern AI and lots of other algebra packages," Tao said. The problems are designed to be "guessproof," with large numerical answers or complex mathematical objects as solutions, making it nearly impossible to solve without proper mathematical reasoning. Further reading: New secret math benchmark stumps AI models and PhDs alike.

Read more of this story at Slashdot.

Dutch publisher Veen Bosch & Keuning has announced plans to use AI for translating commercial fiction, drawing sharp criticism from literary professionals despite promises of human oversight and author consent. Award-winning translator Michele Hutchison, who won the 2020 International Booker Prize, argues that translation extends beyond word conversion. "We build bridges between cultures, taking into account the target readership every step of the way," she said, noting that translators convey rhythm, poetry, and cultural nuances while conducting precise terminology research.

Read more of this story at Slashdot.

Microsoft seems set to rebrand the AI-powered features in Windows to "Windows Intelligence" even if some of the more controversial elements, such as Recall, are to remain as they are. The Register: Word of Windows Intelligence has circulated for a while, although Microsoft has yet to issue any official confirmation. In October, Tero Alhonen posted what appeared to be options for apps that use AI services. Over the weekend, X user Albacore turned up a placeholder page in a Windows 24H2 build for Windows Intelligence settings. Although Microsoft has made substantial investments in artificial intelligence, AI as part of a brand is a little generic. Apple's approach, to define AI as being "Apple Intelligence," manages to keep the familiar "AI" initialism while ensuring its own brand is kept front and center. With Windows Intelligence, Microsoft is attempting something similar, although "Apple Intelligence" can be handily shortened to "AI". The recently overhauled Copilot and delayed Recall have sparked debate in the Windows community, yet neither seems likely to be rebranded to Windows Intelligence at this stage. However, Windows Intelligence could represent an umbrella for AI technologies on the Microsoft platform and provide users with a quick and easy way of controlling the access AI apps have to user data and how that data is used.

Read more of this story at Slashdot.

AmiMoJo writes: The climate-damaging vapors left behind by jet planes could be easily tackled, aviation experts say, with a new study suggesting they could be eliminated for a few pounds per flight. Jet condensation trails, or contrails, have spawned wild conspiracy theories alleging mind control and the spreading of disease, but scientists say the real problem is their warming effect. "They create an artificial layer of clouds, which traps the heat from the Earth that's trying to escape to outer space," said Carlos Lopez de la Osa, from the Transport & Environment campaign group, which has carried out a new study on the solutions to contrails. "The scale of the warming that's associated with them is roughly having a similar impact to that of aviation carbon emissions." Tweaking the flight paths of a handful of aircraft could reduce contrail warming by more than half by 2040, at a cost of less than $5.1 per flight. Geography and a flight's latitude have a strong influence on whether a contrail is warming. Time of day also influences the climate effects of contrails. Those formed by evening and night flights have the largest warming contribution. Seasonality is also important -- the most warming contrails tend to occur in winter. "Planes are already flying around thunderstorms and turbulence areas," Mr Lopez de la Osa said. "We will need to add one more constraint to flight planning, which is avoiding areas of contrail formation."

Read more of this story at Slashdot.

Mark Rainey: Engineers are currently debugging why the Voyager 1 spacecraft, which is 15 billions miles away, turned off its main radio and switched to a backup radio that hasn't been used in over forty years! I've had some tricky debugging issues in the past, including finding compiler bugs and debugging code with no debugger that had been burnt into prom packs for terminals, however I have huge admiration for the engineers maintaining the operation of Voyager 1. Recently they sent a command to the craft that caused it to shut off its main radio transmitter, seemingly in an effort to preserve power and protect from faults. This prompted it to switch over to the backup radio transmitter, that is lower power. Now they have regained communication they are trying to determine the cause on hardware that is nearly 50 years old. Any communication takes days. When you think you have a difficult issue to debug, spare a thought for this team.

Read more of this story at Slashdot.

An anonymous reader shares a report: Officials inside the Secret Service clashed over whether they needed a warrant to use location data harvested from ordinary apps installed on smartphones, with some arguing that citizens have agreed to be tracked with such data by accepting app terms of service, despite those apps often not saying their data may end up with the authorities, according to hundreds of pages of internal Secret Service emails obtained by 404 Media. The emails provide deeper insight into the agency's use of Locate X, a powerful surveillance capability that allows law enforcement officials to follow a phone, and person's, precise movements over time at the click of a mouse. In 2023, a government oversight body found that the Secret Service, Customs and Border Protection, and Immigration and Customs Enforcement all used their access to such location data illegally. The Secret Service told 404 Media in an email last week it is no longer using the tool. "If USSS [U.S. Secret Service] is using Locate X, that is most concerning to us," one of the internal emails said. 404 Media obtained them and other documents through a Freedom of Information Act (FOIA) request with the Secret Service.

Read more of this story at Slashdot.

Paul Kinlan, who leads the Chrome and the Open Web Developer Relations team at Google, asks and answers the question (with a no.): Frameworks are abstractions over a platform designed for people and teams to accelerate their teams new work and maintenance while improving the consistency and quality of the projects. They also frequently force a certain type of structure and architecture to your code base. This isn't a bad thing, team productivity is an important aspect of any software. I'm of the belief that software development is entering a radical shift that is currently driven by agents like Replit's and there is a world where a person never actually has to manipulate code directly anymore. As I was making broad and sweeping changes to the functionality of the applications by throwing the Agent a couple of prompts here and there, the software didn't seem to care that there was repetition in the code across multiple views, it didn't care about shared logic, extensibility or inheritability of components... it just implemented what it needed to do and it did it as vanilla as it could. I was just left wondering if there will be a need for frameworks in the future? Do the architecture patterns we've learnt over the years matter? Will new patterns for software architecture appear that favour LLM management?

Read more of this story at Slashdot.

Programming language polyglots are files that are valid programs in multiple languages, and do different things in each. While polyglots are normally nothing more than a curiosity, the Cosmopolitan Libc project has been trying to put them to a novel use: producing native, multi-platform binaries that run directly on several operating systems and architectures. There are still some rough edges with the project's approach, but it is generally possible to build C programs into a polyglot format with with minimal tweaking.

Apple executives have defended the relocation of the power button to the bottom of its new M4 Mac mini, citing the computer's significantly reduced size as the driving factor behind the design change. In a Bilibili video interview, Apple's Greg Joswiak and John Ternus explained that the Mac mini's form factor, now half the size of its predecessor, necessitated finding a new position for the power button. The executives said that the bottom placement allows for convenient access despite initial user criticism.

Read more of this story at Slashdot.

Security updates have been issued by AlmaLinux (expat), Fedora (chromium and golang-github-nvidia-container-toolkit), Mageia (curl, expat, mpg123, networkmanager-libreswan, openssl, php-tcpdf, qbittorrent, and x11-server, x11-server-xwayland, and tigervnc), Red Hat (kernel and libsoup), Slackware (mozilla), SUSE (firefox, kernel, python-PyPDF2, and xen), and Ubuntu (dotnet9, ghostscript, linux-aws, linux-oem-6.8, and pydantic).

OpenAI's latest large language model, known internally as Orion, has fallen short of performance targets, marking a broader slowdown in AI advancement across the industry's leading companies, according to Bloomberg, corroborating similar media stories in recent days. The model, which completed initial training in September, showed particular weakness in novel coding tasks and failed to demonstrate the same magnitude of improvement over its predecessor as GPT-4 achieved over GPT-3.5, the publication reported Wednesday. Google's upcoming Gemini software and Anthropic's Claude 3.5 Opus are facing similar challenges. Google's project is not meeting internal benchmarks, while Anthropic has delayed its model's release, Bloomberg said. Industry insiders cited by the publication pointed to growing scarcity of high-quality training data and mounting operational costs as key obstacles. OpenAI's Orion specifically struggled due to insufficient coding data for training, the report said. OpenAI has moved Orion into post-training refinement but is unlikely to release the system before early 2024. The report adds: [...] AI companies continue to pursue a more-is-better playbook. In their quest to build products that approach the level of human intelligence, tech firms are increasing the amount of computing power, data and time they use to train new models -- and driving up costs in the process. Amodei has said companies will spend $100 million to train a bleeding-edge model this year and that amount will hit $100 billion in the coming years. As costs rise, so do the stakes and expectations for each new model under development. Noah Giansiracusa, an associate professor of mathematics at Bentley University in Waltham, Massachusetts, said AI models will keep improving, but the rate at which that will happen is questionable. "We got very excited for a brief period of very fast progress," he said. "That just wasn't sustainable." Further reading: OpenAI and Others Seek New Path To Smarter AI as Current Methods Hit Limitations.

Read more of this story at Slashdot.

An anonymous reader quotes a report from The Register: Royal assent was granted to two right to repair bills last week that amend Canada's Copyright Act to allow the circumvention of technological protection measures (TPMs) if this is done for the purposes of "maintaining or repairing a product, including any related diagnosing," and "to make the program or a device in which it is embedded interoperable with any other computer program, device or component." The pair of bills allow device owners to not only repair their own stuff regardless of how a program is written to prevent such non-OEM measures, but said owners can also make their devices work with third-party components without needing to go through the manufacturer to do so. Bills C-244 (repairability) and C-294 (interoperability) go a long way toward advancing the right to repair in Canada and, as iFixit pointed out, are the first federal laws anywhere that address how TPMs restrict the right to repair -- but they're hardly final. TPMs can take a number of forms, from simple administrative passwords to encryption, registration keys, or even the need for a physical object like a USB dongle to unlock access to copyrighted components of a device's software. Most commercially manufactured devices with proprietary embedded software include some form of TPM, and neither C-244 nor C-294 place any restrictions on the use of such measures by manufacturers. As iFixit points out, neither Copyright Act amendments do anything to expand access to the tools needed to circumvent TPMs. That puts Canadians in a similar position to US repair advocates, who in 2021 saw the US Copyright Office loosen DMCA restrictions to allow limited repairs of some devices despite TPMs, but without allowing access to the tools needed to do so. [...] Canadian Repair Coalition co-founder Anthony Rosborough said last week that the new repairability and interoperability rules represent considerable progress, but like similar changes in the US, don't actually amount to much without the right to distribute tools. "New regulations are needed that require manufacturers and vendors to ensure that products and devices are designed with accessibility of repairs in mind," Rosborough wrote in an op-ed last week. "Businesses need to be able to carry out their work without the fear of infringing various intellectual property rights."

Read more of this story at Slashdot.

A new analysis of data from NASA's Voyager 2 mission reveals that the planet Uranus and its five largest moons might harbor subsurface oceans and potential conditions for life. The BBC reports: Much of what we know about them was gathered by Nasa's Voyager 2 spacecraft which visited nearly 40 years ago. But a new analysis shows that Voyager's visit coincided with a powerful solar storm, which led to a misleading idea of what the Uranian system is really like. [...] So, for 40 years we have had an incorrect view of what Uranus and its five largest moons are normally like, according to Dr William Dunn of University College London. "These results suggest that the Uranian system could be much more exciting than previously thought. There could be moons there that could have the conditions that are necessary for life, they might have oceans below the surface that could be teeming with fish!". It has been nearly 40 years since Voyager 2 last flew past the icy world and its moons. Nasa has plans to launch a new mission, the Uranus Orbiter and Probe, to go back for a closer look in 10 years' time. According to Nasa's Dr Jamie Jasinski, whose idea it was to re-examine the Voyager 2 data, the mission will need to take his results into account when designing its instruments and planning the scientific survey. "Some of the instruments for the future spacecraft are very much being designed with ideas from what we learned from Voyager 2 when it flew past the system when it was experiencing an abnormal event. So we need to rethink how exactly we are going to design the instruments on the new mission so that we can best capture the science we need to make discoveries." Nasa's Uranus probe is expected to arrive by 2045, which is when scientists hope to find out whether these far-flung icy moons, once thought of as being dead worlds, might have the possibility of being home to life. The findings have been published in the journal Nature Astronomy.

Read more of this story at Slashdot.

Longtime Slashdot reader thephydes writes: The hearing will go ahead on November 13 at 11:30 ET (16:30 GMT). Apparently, it will "further pull back the curtain on secret UAP research programs conducted by the U.S. government, and undisclosed findings they have yielded," according to a House statement. It's driven by two republicans, Nancy Mace (R-S.C.) and Glenn Grothman (R-Wis.), who say: "Americans deserve to understand what the government has learned about UAP sightings, and the nature of any potential threats these phenomena pose. We can only ensure that understanding by providing consistent, systemic transparency. We look forward to hearing from expert witnesses on ways to shed more light and bring greater accountability to this issue." "Expert witnesses in the hearing will include Luis Elizondo, a decorated former counterintelligence officer who has claimed for years that the U.S. government is hiding knowledge of UAP, including materials recovered from crashed flying saucers," reports Space.com. "The House hearing will also include Tim Gallaudet, a retired U.S. Navy Rear Admiral who unidentified submersible objects, arguing that 'these underwater anomalies jeopardize US maritime security.'" "Other speakers at the hearing include journalist Michael Shellenberger, who has also claimed the U.S. government is hiding UFO crash retrieval programs, and former NASA Associate Administrator of Space Policy and Partnerships Michael Gold, who is a member of NASA's independent UAP study team."

Read more of this story at Slashdot.

An anonymous reader quotes a report from NBC News: Former Massachusetts Air National Guard member Jack Teixeira was sentenced Tuesday to 15 years for stealing classified information from the Pentagon and sharing it online, the U.S. Attorney for Massachusetts announced. Teixeira received the sentence before Judge Indira Talwani in U.S. District Court for the District of Massachusetts. In March, the national guardsman pleaded guilty to six counts of willful retention and transmission of national defense information under the Espionage Act. He was arrested by the FBI in North Dighton, Massachusetts, in April 2023 and has been in federal custody since mid-May 2023. According to court documents, Teixeira transcribed classified documents that he then shared on Discord, a social media platform mostly used by online gamers. He began sharing the documents in or around 2022. A document he was accused of leaking included information about providing equipment to Ukraine, while another included discussions about a foreign adversary's plot to target American forces abroad, prosecutors said. [...] While the documents were discovered online in March 2023, Teixeira had been sharing them online since January of that year, according to prosecutors.

Read more of this story at Slashdot.

"Researchers at the University of Texas have unveiled a new thermal interface material that could revolutionize cooling, outperforming top liquid metal solutions by up to 72% in heat dissipation," writes Slashdot reader jjslash. "This breakthrough not only improves energy efficiency but also enables higher-density data center setups, cutting cooling costs and energy usage significantly." TechSpot reports: Thanks to a mechanochemically engineered combination of the liquid metal alloy Galinstan and ceramic aluminum nitride, this thermal interface material, or TIM, outperformed the best commercial liquid metal cooling products by a staggering 56-72% in lab tests. It allowed dissipation of up to 2,760 watts of heat from just a 16 square centimeter area. The material pulls this off by bridging the gap between the theoretical heat transfer limits of these materials and what's achieved in real products. Through mechanochemistry, the liquid metal and ceramic ingredients are mixed in an extremely controlled way, creating gradient interfaces that heat can flow across much more easily. Beyond just being better at cooling, the researchers claim that the higher performance reduces the energy needed to run cooling pumps and fans by up to 65%. It also unlocks the ability to cram more heat-generating processors into the same space without overheating issues. [...] As for how you can get your hands on the material: it's yet to make it out of the labs. The UT team has so far only tested it successfully at small scales but is now working on producing larger batches to put through real-world trials with data center partners. The material has been detailed in a paper published in the journal Nature Nanotechnology.

Read more of this story at Slashdot.

If you have Spotify's soon-to-be-bricked Car Thing, there are a few ways you can give it a new lease on life. YouTuber Dammit Jeff has showcased modifications to Car Thing that makes the device useful as a desktop music controller, customizable shortcut tool, or a simple digital clock. Ars Technica's Kevin Purdy reports: Spotify had previously posted the code for its uboot and kernel to GitHub, under the very unassuming name "spsgsb" and with no announcement (as discovered by Josh Hendrickson). Jeff has one idea why the streaming giant might not have made much noise about it: "The truth is, this thing isn't really great at running anything." It has half a gigabyte of memory, 4GB of internal storage, and a "really crappy processor" (Amlogic S905D2 SoC) and is mostly good for controlling music. How do you get in? The SoC has a built-in USB "burning mode," allowing for a connected computer, running the right toolkit, to open up root access and overwrite its firmware. Jeff has quite a few issues getting connected (check his video description for some guidance), but it's "drag and drop" once you're in. Jeff runs through a few of the most popular options for a repurposed Car Thing: - DeskThing, which largely makes Spotify desk-friendly, but adds a tiny app store for weather (including Jeff's own WeatherWave), clocks, and alternate music controls - GlanceThing, which keeps the music controls but also provides some Stream-Deck-like app-launching shortcuts for your main computer. - Nocturne, currently invite-only, is a wholly redesigned Spotify interface that restores all its Spotify functionality.

Read more of this story at Slashdot.

An anonymous reader quotes a report from BleepingComputer: VMware has announced that its VMware Fusion and VMware Workstation desktop hypervisors are now free to everyone for commercial, educational, and personal use. In May, the company also made VMware Workstation Pro and Fusion Pro free for personal use, allowing students and home users to set up virtualized test labs and experiment with other OSs by running virtual machines and Kubernetes clusters on Windows, Linux, and macOS devices. Starting this week, the Pro versions and the two products will no longer be available under a paid subscription model. "Effective immediately, both VMware Fusion and VMware Workstation will transition away from the paid subscription model, meaning you can now utilize these tools without any cost. The paid versions of these offerings -- Workstation Pro and Fusion Pro -- are no longer available for purchase," said Broadcom product marketing director Himanshu Singh. "If you're currently under a commercial contract, you can rest easy knowing that your agreement will remain in effect until the end of your term. You will continue to receive the full level of service and enterprise-grade support as per your contract." While the free versions will include all the features available in the paid products, Broadcom will no longer provide users with support ticketing for troubleshooting. Broadcom plans to continue developing new features and improvements and ensure that updates are rolled out promptly. "We're actively investing in new features, usability improvements, and other valuable enhancements," Singh added. "Our engineering teams are committed to maintaining our high standards for stability, with timely updates and reliable performance." You can download VMware Fusion and VMware Workstation here (sign-in required).

Read more of this story at Slashdot.

A new update is rolling out that automatically starts Microsoft's Edge browser and prompts users to import their Chrome tabs -- a move that has sparked criticism over its invasive tactics to encourage Edge adoption. The Verge's Tom Warren reports: My colleague Richard Lawler noticed that Edge started automatically on his PC last week at boot and offered up a new prompt to "enhance your browsing experience." The pop-up has a "bring over your data from other browsers regularly" option ticked by default, and encourages people to confirm and continue with a big blue button. If you want to dismiss this prompt there's a tiny white X button that looks similar to the sparkles Microsoft is using in the background of the prompt. If you simply hit confirm and continue then Microsoft Edge will import your Chrome data and continually import your tabs if you have Chrome set as default. The prompt seems to mainly appear on PCs with Chrome installed, suggesting that Microsoft is once again targeting Chrome users. Microsoft confirmed the new "feature" to The Verge. "This is a notification giving people the choice to import data from other browsers," explains Microsoft spokesperson Caitlin Roulston. "There is an option to turn it off."

Read more of this story at Slashdot.

The genetic testing company 23andMe announced it will cut 40% of its workforce, or 200 jobs, and halt the work on therapies it was developing. As the BBC notes, the company is fighting for survival after hackers gained access to personal information of millions of its users, causing the stock to crater by more than 70%. All seven of its independent directors also resigned in September, following a protracted negotiation with founder and Chief Executive Anne Wojcicki over her plan to take the company private. The BBC reports: On Tuesday, the company warned investors of "substantial doubt" about its ability to continue operating, as it reported that revenue had fallen to $44 million between July and September compared to $50 million in the same period last year. Losses fell to $59 million from $75 million. The job cuts are expected to lead to one-off costs of $12 million, including severance pay, for the plan that will result in savings of $35 million. "We are taking these difficult but necessary actions as we restructure 23andMe and focus on the long-term success of our core consumer business and research partnerships," Ms Wojcicki said. The company also said it is considering what to do with the therapies it had in development, including licensing or selling them. 23andMe is a giant of the growing ancestor-tracing industry. It offers genetic testing from DNA, with ancestry breakdown and personalised health insights. Its customers include famous names, from rapper Snoop Dogg to multi-billionaire investor Warren Buffett. The company was valued at roughly $3.5 billion when it listed on the Nasdaq stock exchange in 2021 and its share price peaked at $17.65. But they have since tumbled and are currently trading at less than $5.

Read more of this story at Slashdot.