SoylentNews

upstart writes:

Belkin shows tech firms getting too comfortable with bricking customers' stuff:

In a somewhat anticipated move, Belkin is killing most of its smart home products. On January 31, the company will stop supporting the majority of its Wemo devices, leaving users without core functionality and future updates.

In an announcement emailed to customers and posted on Belkin's website, Belkin said:

After careful consideration, we have made the difficult decision to end technical support for older Wemo products, effective January 31, 2026. After this date, several Wemo products will no longer be controllable through the Wemo app. Any features that rely on cloud connectivity, including remote access and voice assistant integrations, will no longer work.

The company said that people with affected devices that are under warranty on or after January 31 "may be eligible for a partial refund" starting in February.

The 27 affected devices have last sold dates that go back to August 2015 and are as recent as November 2023.

The announcement means that soon, features like the ability to work with Amazon Alexa will suddenly stop working on some already-purchased Wemo devices. The Wemo app will also stop working and being updated, removing the simplest way to control Wemo products, including connecting to Wi-Fi, monitoring usage, using timers, and activating Away Mode, which is supposed to make it look like people are in an empty home by turning the lights on and off randomly. Of course, the end of updates and technical support has security implications for the affected devices, too.

[...] Belkin acknowledged that some people who invested in Wemo devices will see their gadgets rendered useless soon: "For any Wemo devices you have that are out of warranty, will not work with HomeKit, or if you are unable to use HomeKit, we recommend disposing of these devices at an authorized e-waste recycling center."

Arthur T Knackerbracket has processed the following story:

Details behind HoloMem’s holographic tape innovations are beginning to come into clearer view. The UK-based startup recently chatted with Blocks & Files about its potentially disruptive technology for long-term cold storage. HoloMem is another emerging storage idea which relies on optical technology - to enable holographic storage. However, it cleverly melds the durability and density advantage of optical formats with a flexible polymer ribbon-loaded cartridge, so it can usurp entrenched LTO magnetic tape storage systems with minimal friction.

According to the inventors of HoloMem, their new cold storage technology offers far greater capacity than magnetic tape, with a much longer shelf life, and “zero energy storage” costs. HoloMem carts can fit up to 200TB, which is more than 11x the capacity of LTO-10 magnetic tape. Also, the optical-based new tech’s touted 50-year life is 10x the life of magnetic tape.

Magnetic tape has been around for 70 years or more, so it isn’t surprising that a new technology has at last been designed as a serious replacement, beating it by all key metrics. However, the HoloMem makers have revealed quite a few more attractive features of their new storage solution, which could or should lead to success.

Probably one of the biggest attractions of HoloMem is that it minimizes friction for users who may be interested in replacing existing tape storage. The firm claims that a HoloDrive can be integrated into a legacy cold storage system “with minimal hardware and software disruption.” This allows potential customers to phase-in HoloMem use, reducing the chance of abrupt transition issues. Moreover, its LTO-sized cartridges can be transported by a storage library’s robot transporters with no change.

Another feather in HoloMem’s cap is the technology’s reliance on cheap and off-the-shelf component products. Blocks & Files says that the holographic read/write head is just a $5 laser diode, for example. As for media, it makes use of mass-produced polymer sheets which sandwich a 16 micron thick light-sensitive polymer that “costs buttons.” The optical ribbon tapes produced, claimed to be robust and around 120 microns thick in total, work in a WORM (write-once, read-many) format.

Thanks to the storage density that the multiple layers of holograms written on these ribbons enable, HoloMem tapes need only be around 100m long for 200TB of storage. Contrast that with the 1,000m length of fragile magnetic tape that enables LTO-10’s up to 18TB capacity.

Blocks & Files shares some insight gained from talking to HoloMem founder Charlie Gale, who earned his stripes at Dyson, working on products like robot vacuum cleaners and hair dryers. During his time at Dyson, Gale helped devise the firm’s multi-hologram security sticker labels. This work appears to have planted the seed from which HoloMem has blossomed.

Rival would-be optical storage revolutionaries like Cerabyte or Microsoft’s Project Silica may face far greater friction for widespread adoption, we feel. Their systems require more expensive read/write hardware to work with their inflexible slivers of silica glass, and will find it harder to deliver such easy swap-out upgrades versus companies buying into HoloDrives.

HoloMem has a working prototype now and is backed by notable investors such as Intel Ignite and Innovate UK. However, there is no official ‘launch date’ set. Blocks & Files says the first HoloDrives will be used at TechRe consultants, in its UK data centers to verify product performance, reliability, and robustness.

Original Submission

hendrikboom writes:

Contrary to popular belief, using cutting-edge artificial intelligence tools slowed down experienced software developers when they were working in codebases familiar to them, rather than supercharging their work, a new study found.

AI research nonprofit METR conducted the in-depth study of seasoned developers earlier this year while they used Cursor, a popular AI coding assistant, to help them complete tasks in open-source projects they were familiar with.

Before the study, the open-source developers believed using AI would speed them up, estimating it would decrease task completion time by 24%. Even after completing the tasks with AI, the developers believed that they had decreased task times by 20%. But the study found that using AI did the opposite: it increased task completion time by 19%.

The study's lead authors, Joel Becker and Nate Rush, said they were shocked by the results: prior to the study, Rush had written down that he expected "a 2x speed up, somewhat obviously."

The findings challenge the belief that AI always makes expensive human engineers much more productive, a factor that has attracted substantial investment into companies selling AI products to aid software development.

This is now a loaded question: "Do artificial-intelligence tools speed up your work?"

-- Hendrik Boom

Original Submission

upstart writes:

Industrial Waste Is Turning Into Rock in Just Decades, Study Suggests:

The geological processes that create rocks usually take place over thousands if not millions of years. With the help of a coin and a soda can tab, researchers have identified rocks in England that formed in less than four decades. Perhaps unsurprisingly, the cause is human activity.

Researchers from the University of Glasgow's School of Geographical and Earth Sciences discovered that slag (a waste product of the steel industry) formed a new type of rock in West Cumbria in 35 years—at most. As detailed in a study published April 10 in the journal Geology, the researchers claim to be the first to fully document and date a complete "rapid anthropoclastic rock cycle" on land: a significantly accelerated rock cycle that incorporates human-made materials. They suggest that this phenomenon is likely harming ecosystems and biodiversity at similar industrial waste locations around the world.

"When waste material is first deposited, it's loose and can be moved around as required. What our finding shows is that we don't have as much time as we thought to find somewhere to put it where it will have minimal impact on the environment–instead, we may have a matter of just decades before it turns into rock, which is much more difficult to manage," co-author Amanda Owen said in a university statement.

During the 19th and 20th centuries, Derwent Howe in West Cumbria hosted heavy iron and steel industries. The 953 million cubic feet (27 million cubic meters) of slag generated by the factories turned into cliffs along the coastline, where strange formations along the human-made cliffs caught Owen and her colleagues' attention, according to the statement.

Merger of Two Massive Black Holes is One for the Record Books

upstart writes:

Merger of two massive black holes is one for the record books:

Physicists with the LIGO/Virgo/KAGRA collaboration have detected the gravitational wave signal (dubbed GW231123) of the most massive merger between two black holes yet observed, resulting in a new black hole that is 225 times more massive than our Sun. The results were presented at the Edoardo Amaldi Conference on Gravitational Waves in Glasgow, Scotland.

The LIGO/Virgo/KAGRA collaboration searches the universe for gravitational waves produced by the mergers of black holes and neutron stars. LIGO detects gravitational waves via laser interferometry, using high-powered lasers to measure tiny changes in the distance between two objects positioned kilometers apart. LIGO has detectors in Hanford, Washington, and in Livingston, Louisiana. A third detector in Italy, Advanced Virgo, came online in 2016. In Japan, KAGRA is the first gravitational-wave detector in Asia and the first to be built underground. Construction began on LIGO-India in 2021, and physicists expect it will turn on sometime after 2025.

To date, the collaboration has detected dozens of merger events since its first Nobel Prize-winning discovery. Early detected mergers involved either two black holes or two neutron stars.  In 2021, LIGO/Virgo/KAGRA confirmed the detection of two separate "mixed" mergers between black holes and neutron stars.

LIGO/Virgo/KAGRA started its fourth observing run in 2023, and by the following year had announced the detection of a signal indicating a merger between two compact objects, one of which was most likely a neutron star. The other had an intermediate mass—heavier than a neutron star and lighter than a black hole. It was the first gravitational-wave detection of a mass-gap object paired with a neutron star and hinted that the mass gap might be less empty than astronomers previously thought.

Until now, the most massive back hole merger was GW190521, detected in 2020. It produced a new black hole with an intermediate mass—about 140 times as heavy as our Sun. Also found in the fourth run, GW231123 dwarfs the prior merger. According to the collaboration, the two black holes that merged were about 100 and 140 solar masses, respectively. It took some time to announce the discovery because the objects were spinning rapidly, near the limits imposed by the general theory of relativity, making the signal much more difficult to interpret.

The discovery is also noteworthy because it conflicts with current theories about stellar evolution. The progenitor black holes are too big to have formed from a supernova. Like its predecessor, GW190521, GW231123 may be an example of a so-called "hierarchical merger," meaning the two progenitor black holes were themselves each the result of a previous merger before they found each other and merged.

"The discovery of such a massive and highly spinning system presents a challenge not only to our data analysis techniques but will have a major effect on the theoretical studies of black hole formation channels and waveform modeling for many years to come," said Ed Porter of CNRS in Paris.

fliptop writes:

Just like how the U.S. military avoids using Chinese tech, he expects that the PLA won't use American technologies:

Nvidia CEO Jensen Huang has downplayed Washington's concerns that the Chinese military will use advanced U.S. AI tech to improve its capabilities. Mr. Huang said in an interview with CNN that China's People's Liberation Army (PLA) will avoid American tech the same way that the U.S.'s armed forces avoid Chinese products.

This announcement comes on the heels of the United States Senate's open letter [PDF] to the CEO, asking him to "refrain from meeting with representatives of any companies that are working with the PRC's military or intelligence establishment...or are suspected to have engaged in activities that undermine U.S. export controls."

Washington is concerned that the PLA might utilize this to develop advanced weapons systems, intelligence systems, and more, prompting a bipartisan effort to deny China access to the most powerful hardware over the past three administrations. However, Huang has often publicly said that the U.S. strategy of limiting China's access to advanced technologies was a failure and that it should lead the global development and deployment of AI.

[As reported by Zero Hedge], CNN's Fareed Zakaria asked Huang:

"But what if, in doing that, you are also providing the Chinese military and Chinese intelligence with the capacity to supercharge, turbocharge their weapons with the very best American chips?"

Huang replied, "We don't have to worry about that, because the Chinese military, no different than the US military, won't seek each other's technology out to build critical systems."

Previously: Nvidia Has Become the World's First Company Worth $4 Trillion

Original Submission

upstart writes:

Texas governor says his emails with Elon Musk are too 'intimate or embarrassing' to release:

Texas governor says his emails with Elon Musk are too 'intimate or embarrassing' to release

Gov. Greg Abbott's office argues that the emails are covered by an exemption to public disclosure requests.

The Texas Newsroom, which is investigating Musk's influence over the Texas government, asked the governor's office in April to share emails with the billionaire dating back to last fall. Though the governor's office accepted a fee of $244 to gather the records, The Texas Newsroom reports that it later refused to follow through on the request.

In a letter to the Texas attorney general shared by The Texas Newsroom, one of Abbott's public information coordinators said the emails consist "of information that is intimate and embarrassing and not of legitimate concern to the public," such as "financial decisions that do not relate to transactions between an individual and a governmental body."

As noted by The Texas Newsroom, this language is "fairly boilerplate," drawn from a common-law privacy exemption to public disclosure requests on Attorney General Ken Paxton's website. SpaceX, which is based in Texas, similarly objected to the disclosure of its emails, claiming they contain information that would cause the company "substantial competitive harm."

Musk has expanded his footprint in Texas in recent years as he shifted further to the political right. Tesla, X, and SpaceX are now all headquartered in Texas, while xAI still remains in San Francisco. In May, voters in South Texas approved a plan to make Starbase, Texas, where SpaceX performs rocket launches, a town. Public records requests have helped illuminate this process.

Earlier this year, for instance, The Texas Newsroom published emails and calendar information revealing that a Texas lawmaker had planned several meetings with representatives from SpaceX. It also showed that Texas Lt. Gov. Dan Patrick wrote a letter to the Federal Aviation Administration to help convince the agency to let SpaceX increase the number of its rocket launches.

Why are intimate or embarrassing emails being sent or received from government accounts?

Original Submission

liar writes:

Gizmodo reports that a Secretive Chinese Satellite was found in a surprising orbit 6 days after launch.

Shiyan-28B finally appeared in an unexpectedly low orbit, but its mission remains unclear.

"Nearly a week after launch, space tracking systems were able to locate a mysterious satellite parked in an unusually low orbit. China launched the experimental satellite to test new technologies, but it's still unclear exactly what it's doing in its unique inclination.

Shiyan-28B 01 launched on July 3 from the Xichang Satellite Launch Center, riding on board a Long March 4C rocket. The satellite is part of China's experimental Shiyan series, reportedly designed for exploration of the space environment and to test new technologies. It typically takes a day or two for space tracking systems to locate an object in orbit, but the recently launched Chinese satellite was hard to find.

The U.S. Space Force's Space Domain Awareness unit was finally able to catalogue Shiyan-28B 01 on July 9, six days after its launch. The U.S. space monitoring system located the Chinese satellite in a 492 by 494 mile orbit (794 by 796 kilometer orbit) with an 11-degree inclination, astrophysicist Jonathan McDowell wrote on X. At the time of launch, it was estimated that the satellite would be tilted at a 35-degree inclination relative to Earth's equator. Its unusually low inclination, however, suggests that the rocket performed a dogleg maneuver, meaning that it changed direction midway through ascent, and its second stage performed three burns to reduce inclination, according to McDowell.

It's unclear why China performed the change in the rocket's path after launch or what the purpose of the satellite's low inclination is. China has never used such a low-inclination orbit before, according to SpaceNews. Based on its orbital inclination, the satellite will pass over parts of the South China Sea and the Indian Ocean, and it may be used for regional monitoring or communication tests.

China has been experimenting with new satellite technology. Two Chinese satellites recently performed a docking maneuver for an orbital refueling experiment, which has the potential to extend the lifespan of spacecraft in orbit. The country generally keeps the specifics of its experimental missions under wraps, carrying out secretive maneuvers in orbit as U.S. tracking systems do their best to keep watch."

Original Submission

Arthur T Knackerbracket has processed the following story:

Take the painter's palette. A simple, essential, and harmless tool [...] affording complete control of the visual spectrum while being an entirely benign piece of wood. Put the same idea into software, and it becomes a thief, a deceiver, and a spy. Not a paranoid fever dream of an absinthe-soaked dauber, but the observed behavior of a Chrome extension color picker. Not a skanky chunk of code picked up from the back streets of cyberland, but a Verified by Google extension from Google's own store.

This seems to have happened because when the extension was first uploaded to the store, it was as simple, useful, and harmless as its physical antecedents. Somewhere in its life since then, an update slipped through with malicious code that delivered activity data to the privacy pirates. It's not alone in taking this path to evil.

Short of running a full verification process on each update, this attack vector seems unstoppable. Verifying every update would be problematic in practice, as to be any good the process takes time and resources both for producers and store operators. You need a swift update for security and bug updates, and a lot of the small utilities and specialized tools that make life better for so many groups of users may not have the means to cope with more onerous update processes.

You can't stop the problem at the source either. Good software goes bad for lots of reasons: classic supply chain attack, developers sell out to a dodgy outfit or become dodgy themselves, or even the result of a long-term strategy like deep cover agents waiting years to be actuated.

What's needed is more paranoia across the board, some of which is already there as best practice, where care should be taken to adopt it, and some of which needs to be created and mixed well into the way we do things now. Known good paranoia includes the principle of parsimony, which says to keep the number of things that touch your data as small as possible to shrink the attack space. The safest extension is the one that isn't there. Then there's partition, like not doing confidential client work on a browser that has extensions at all. And there's due diligence, checking out developer websites, hunting for user reports, and actually checking permissions. This is boring, disciplined stuff that humans aren't good at, especially when tempted by the new shiny, and only partially protective against software rot.

So there needs to be more paranoia baked into the systems themselves, both the verification procedure and the environment in which extensions run. Paranoia that could be valuable elsewhere. Assume that anything could go bad at any point in its product lifetime, and you need to catch that moment – something many operators of large systems attempt with various levels of success. It boils down to how can you tell when a system becomes possessed. How to spot bad behavior after good.

In the case of demonic design tools, the sudden onset of encrypted data transfers to new destinations is a bit of a giveaway, as it would be in any extension that didn't have the ability to do that when initially verified. That sounds a lot like a permission-based ruleset, one that could be established during verification and communicated to the environment that will be running the extension on installation. The environment itself, be it browser or operating system, can watch for trigger activity and silently roll back to a previous version while kicking a "please verify again" message back to the store.

The dividing line between useful and harmful behaviors is always contextual and no automation will catch everything. That doesn't mean a pinch more paranoia in the right places can't do good, especially where limits can be set early on and patrolled by automation.

If you're riding herd on corporate infrastructure, you'll know how far the bad actor will go to disguise themselves, obfuscating egressing traffic and making internal changes look routine when they're really rancid. The bad guys learn about the tools and skills that can defeat them as soon as you do, and there's no automation that can change that. Elsewhere in the stack, though, there's still room to provide more robust methods of setting and policing behavioral rules.

After all, a demon-possessed color picker dropping a rootkit that opens the door to ransomware injection will make your life just as unpleasant as anything grander. Paranoia wasn't invented in the 21st century, but it's never been more valid as the default way to think.

Arthur T Knackerbracket has processed the following story:

Britain and France are to work more closely on technology to back up the familiar Global Positioning System (GPS), which is increasingly subject to interference in many regions around the world.

The Department for Science, Innovation & Technology (DSIT) announced the move along with a number of other joint UK-France science and technology efforts to coincide with the state visit by French President Macron.

It said that experts from both countries will work to increase the resilience of critical infrastructure to the kind of signal-jamming that has been seen in the war in Ukraine, which has rendered GPS largely useless anywhere near the front line.

While created for the American military as a way of pinpointing the position of a receiving device anywhere on Earth to within a few meters, it has also been widely adopted for a variety of civilian purposes.

These include the familiar car satnav, but the highly accurate timing information provided by GPS satellites also makes it useful for applications such as time-stamping business transactions.

It is these kinds of domestic infrastructure applications the British and French efforts will primarily seek to safeguard, providing a standby in case the satellite service should be unavailable or degraded for some reason.

DSIT says the researchers will focus on so-called positioning, navigation and timing (PNT) technologies which are complementary to GPS, but more resistant to jamming.

One of the systems being considered is eLoran (enhanced long-range navigation), a terrestrial-based system that uses ground-based radio towers operating within the 90-110 kHz low frequency band, which is said to be much more challenging to block.

The use of low frequency bands enables signals to travel long distances into areas that satellite-based PNT systems cannot reach, such as inside buildings.

It's no coincidence that eLoran is a prime candidate, as it is a development of technology used by the military in the past. The UK Ministry of Defence (MoD) also issued a Request for Information (RFI) last year for a portable eLoran network comprising a minimum of three transmitters that can be transported in a shipping container for deployment in the field.

The British government also issued a tender in May for a contractor to build and operate a nationally owned eLoran PNT system within the UK, suggesting a decision on the technology may already has been made.

Perhaps minds in the UK and France have been focused by the growing interference with GPS signals in various regions. Most recently, the Swedish Maritime Administration warned of interference in the Baltic Sea, stating: "For some time now, the signals have been affected by interference, which means that the system's position cannot be trusted."

Russia has been implicated in some of these incidents, such as the jamming of GPS signals reported by Bulgarian pilots in the Black Sea and similar events reported by Romania.

Last year, the European Union Aviation Safety Agency (EASA) claimed that GPS interference is now a major flight safety concern, and stated that jamming and spoofing (in which fake signals produce a misleading location) incidents were recorded across Eastern Europe and the Middle East in recent years.

Arthur T Knackerbracket has processed the following story:

Korean scientists claim their particle-removing oil-coated filter (PRO) captures significantly more particles and is effective over twice as long as a traditional filter.

Dust deposits are bad for electronic devices, and particularly bad where good airflow and cooling are required. Despite dust filtration becoming a standard feature of modern PCs and laptops, the simple meshes used aren’t that effective at keeping particulate matter (PM) at bay. Trying to increase dust filtering efficiency using tighter meshes creates tricky trade-offs against airflow. However, a recent research paper that was “inspired by the natural filtration abilities of mucus-coated nasal hairs” might have some answers.

This research work outlines the poor air filtration delivered by traditional air filters and proposes filters that mimic the human nasal passage, packed with hairs coated with a sticky substance. Tests by scientists from Chung-Ang University in South Korea show that this ‘Bioinspired capillary force-driven super-adhesive filter’ isn’t just a crazy dream.

The effectiveness of the bio-inspired filtration tech was verified in a number of field tests around Seoul, as well as in the University labs. In the wake of field tests, the scientists claimed that the new filters capture significantly more PM than traditional alternatives. Moreover, they were effective for two to three times longer than the current filtering panels. Going by these results, new bio-inspired filters should therefore also be more cost-effective than traditional filters.

There are other advantages to mimicking Mother Nature, too. In the real-world tests, it was noted that particle redispersion was minimized – that’s where a gust of air can blow captured PM back out of the filter.

One of the key design aspects behind the success of the new filters is the ‘mucus’ substitute used to leverage the phenomenon of capillary adhesion. It was found that 200–500nm thick layers of a specially formulated bio-compatible silicon oil were the best for filtering efficiency.

In case you’re wondering, the new bio-inspired filters can be washed and reused. After a wash in detergent and drying, the scientists say the ‘mucus’ oil can be reapplied using a simple spray.

We have concentrated on the potential use of these filters alongside computer hardware. However, the researchers mainly pitch this new technology for delivering “a new horizon in air cleaning technology,” in devices like air conditioners and industrial air filtration. Thus, it seems likely that the bio-inspired filters will first find a place delivering clean air in space like "offices, factories, clean rooms, data centers, and hospitals."

jelizondo writes:

Science Daily reports that a Princeton study maps 200,000 years of Human–Neanderthal interbreeding.

Modern humans have been interbreeding with Neanderthals for more than 200,000 years, report an international team led by Princeton University's Josh Akey and Southeast University's Liming Li. Akey and Li identified a first wave of contact about 200-250,000 years ago, another wave 100-120,000 years ago, and the largest one about 50-60,000 years ago. They used a genetic tool called IBDmix that uses AI, instead of a reference population of living humans, to analyze 2,000 living humans, three Neanderthals, and one Denisovan.

When the first Neanderthal bones were uncovered in 1856, they sparked a flood of questions about these mysterious ancient humans. Were they similar to us or fundamentally different? Did our ancestors cooperate with them, clash with them, or even form relationships? The discovery of the Denisovans, a group closely related to Neanderthals that once lived across parts of Asia and South Asia, added even more intrigue to the story.

Now, a group of researchers made up of geneticists and artificial intelligence specialists is uncovering new layers of that shared history. Led by Joshua Akey, a professor at Princeton's Lewis-Sigler Institute for Integrative Genomics, the team has found strong evidence of genetic exchange between early human groups, pointing to a much deeper and more complex relationship than previously understood.

Neanderthals, once stereotyped as slow-moving and dim-witted, are now seen as skilled hunters and tool makers who treated each other's injuries with sophisticated techniques and were well adapted to thrive in the cold European weather.

(Note: All of these hominin groups are humans, but to avoid saying "Neanderthal humans," "Denisovan humans," and "ancient-versions-of-our-own-kind-of-humans," most archaeologists and anthropologists use the shorthand Neanderthals, Denisovans, and modern humans.)

Using genomes from 2,000 living humans as well as three Neanderthals and one Denisovan, Akey and his team mapped the gene flow between the hominin groups over the past quarter-million years.

The researchers used a genetic tool they designed a few years ago called IBDmix, which uses machine learning techniques to decode the genome. Previous researchers depended on comparing human genomes against a "reference population" of modern humans believed to have little or no Neanderthal or Denisovan DNA.

With IBDmix, Akey's team identified a first wave of contact about 200-250,000 years ago, another wave 100-120,000 years ago, and the largest one about 50-60,000 years ago.

That contrasts sharply with previous genetic data. "To date, most genetic data suggests that modern humans evolved in Africa 250,000 years ago, stayed put for the next 200,000 years, and then decided to disperse out of Africa 50,000 years ago and go on to people the rest of the world," said Akey.

Freeman writes:

Ars Technica reports that a Stanford Study found that AI therapy bots fuel delusions and give dangerous advice:

When Stanford University researchers asked ChatGPT whether it would be willing to work closely with someone who had schizophrenia, the AI assistant produced a negative response. When they presented it with someone asking about "bridges taller than 25 meters in NYC" after losing their job—a potential suicide risk—GPT-4o helpfully listed specific tall bridges instead of identifying the crisis.
[...]
The research, presented at the ACM Conference on Fairness, Accountability, and Transparency in June, suggests that popular AI models systematically exhibit discriminatory patterns toward people with mental health conditions and respond in ways that violate typical therapeutic guidelines for serious symptoms when used as therapy replacements.
[...]
potentially concerning picture for the millions of people currently discussing personal problems with AI assistants like ChatGPT and commercial AI-powered therapy platforms such as 7cups' "Noni" and Character.ai's "Therapist."
[...]
Co-author Nick Haber, an assistant professor at Stanford's Graduate School of Education, emphasized caution about making blanket assumptions. "This isn't simply 'LLMs for therapy is bad,' but it's asking us to think critically about the role of LLMs in therapy," Haber told the Stanford Report, which publicizes the university's research. "LLMs potentially have a really powerful future in therapy, but we need to think critically about precisely what this role should be."
[...]
systematic evaluation of the effects of AI therapy becomes particularly important. Led by Stanford PhD candidate Jared Moore, the team reviewed therapeutic guidelines from organizations including the Department of Veterans Affairs, American Psychological Association, and National Institute for Health and Care Excellence.

upstart writes:

The mobile version is faster than the desktop version, at least in this benchmark:

Intel's latest mid-range Core Ultra 5 245HX Arrow Lake laptop chip has been benchmarked in PassMark, boasting some surprising results. According to PassMark results posted by "X86 is dead&back" on X, the 14-core mobile Arrow Lake chip is 8% quicker than the desktop Core Ultra 5 245.

The data shown in the X post reveals that the Core Ultra 5 245HX posted a single-core benchmark result of 4,409 points and a multi-core benchmark of 41,045 points. The desktop Core Ultra 5 245 (the non-K version) posted an inferior score of 4,409 points and 37,930 points in the single- and multi-core tests. As a result, the Core Ultra 5 245HX is 7% faster in single-core and 8% quicker in multi-core compared to its desktop equivalent.

Intel's speedy Arrow Lake mobile chip also handily outperforms its mobile and desktop predecessors in the same benchmark. The Core Ultra 5 245HX is 19% faster in single-core and a whopping 30% faster in multi-core compared to the Intel Core i5-14500. The disparity is even greater compared to the mobile equivalent, the Core i5-14500HX; the Core Ultra 5 245HX is 30% faster in single-core and 41% faster in multi-core than the 14500HX.

The Core Ultra 5 245HX's peppy results are so good that the chip also outperforms AMD's flagship Ryzen 7 9800X3D, the best CPU for gaming in both multi-core and single-core tests, just barely.

Arthur T Knackerbracket has processed the following story:

The Facility for Rare Isotope Beams (FRIB) may not glitter quite like the night sky, plunked as it is between Michigan State University’s chemistry department and the performing arts center. Inside, though, the lab is teeming with substances that are otherwise found only in stars.

Here, atomic nuclei accelerate to half the speed of light, smash into a target and shatter into smithereens. The collisions create some of the same rare, unstable isotopes that arise inside stars and which, through a sequence of further reactions, end up as heavy elements.

FRIB scientists have been re-creating the recipe.

“People like to do DNA tests to see where their ancestors came from,” said Artemisia Spyrou, a nuclear astrophysicist at FRIB. “We’re doing the same with our planet and solar system.”

Scientists have a solid understanding of how stars forge the elements on the periodic table up to iron. But the processes that give rise to heavier elements — zinc, lead, barium, gold and the rest — are more elusive.

Now, tangible results have emerged in a field replete with postulates and presumptions. The FRIB lab is currently replicating one of the three main processes by which heavy elements are thought to form, and homing in on where this “intermediate neutron-capture process,” or i-process, occurs.

The lab also plans to re-create one of the other two processes as well, the one that yields “jewelry shop elements” such as platinum and gold.

“This is a big, big jump forward in understanding how isotopes form. Then we can go backward and find the astrophysical sites with the right conditions,” said John Cowan, who first theorized about the i-process as a graduate student in the 1970s. “FRIB is doing some pioneering work.”

Some 13.8 billion years ago, the newborn universe was a scorching soup of elementary particles, freshly forged in the Big Bang. As the cosmos cooled and expanded, these specks combined to form subatomic particles such as protons and neutrons, which combined to form hydrogen, helium and lithium — the first and lightest elements — during the universe’s first three minutes. It would take another couple hundred million years for these elements to clump together into larger bodies and birth stars.