Author name: Kris Guyer

modder-packs-an-entire-nintendo-wii-into-a-box-the-size-of-a-pack-of-cards

Modder packs an entire Nintendo Wii into a box the size of a pack of cards

gaming, modding, Nintendo, Wii / /

wii micro —

There’s no disc drive, but there are still ports for GameCube controllers.

–

Its creator calls the

Enlarge / Its creator calls the “Short Stack” the world’s smallest scale model replica of the Nintendo Wii (bottom).

The miniaturization of retro tech has always been a major obsession for modders, from the person who fit an original NES into a Game Boy-sized portable to the person who made a mini-er version of Apple’s Mac mini.

One mod in this storied genre that caught our eye this week is the “Short Stack,” a scale model of the Nintendo Wii that packs the 2006 console’s internal hardware into a 3D-printed enclosure roughly the size of a deck of playing cards.

“You could fit 13.5 of these inside an original Wii,” writes James Smith (aka loopj), the person behind the project. All the design details, custom boards, and other information about recreating the mod are available on GitHub.

Like many space-saving console mods, the Short Stack requires a cut-down version of the original Wii’s PCB, retaining (and occasionally relocating) the original console’s CPU, GPU, RAM, and NAND flash chip. Power delivery, USB, the Wi-Fi and Bluetooth chips, and GameCube controller ports were all relocated to separate custom PCBs, which also allowed Smith to add HDMI output and a microSD card slot (the original Wii used a full-size SD card and didn’t support digital video output).

  • The Short Stack cleverly hides its GameCube controller ports behind a sliding panel on the side.

  • A headphone jack-to-GameCube dongle preserves GameCube controller compatibility while saving space in the console itself.

  • The cut-down Wii PCB, featuring the original console’s CPU, GPU, RAM, and NAND flash.

  • The Short Stack next to a deck of cards.

Some sacrifices were made in the name of miniaturization. The console’s disc drive is gone, so any games will need to be loaded from a microSD card instead. And the four GameCube controller ports are actually headphone jacks that work over a special adapter. Smith made these headphone-to-GameCube dongles pin-compatible with an earlier mod called the GC Nano, a project that did for the GameCube what the Short Stack does for the Wii.

Smith also designed custom front and rear PCBs for the console to handle things like the power button and the glowing blue light around the Short Stack’s (aesthetic, non-functional) DVD slot. A custom heatsink (Smith uses aluminum, though it can also be made with copper to improve heat transfer at the expense of weight) and a tiny fan keep the console cool.

Nintendo released its own Wii Mini toward the end of the Wii’s life in 2012, but it came with significant compromises: no online connectivity, no GameCube controller ports or game compatibility, and no SD card slot. The Short Stack loses the optical drive for space-saving reasons but otherwise retains all of the features of the original Wii.

Smith says that the Short Stack could probably be as much as 20 to 30 percent smaller without giving up features. But one of the goals of the Short Stack project was to make a scale model of the original Wii, and further shrinkage would make the project “tricky to assemble.”

Modder packs an entire Nintendo Wii into a box the size of a pack of cards Read More »

roku-forcing-2-factor-authentication-after-2-breaches-of-600k-accounts

Roku forcing 2-factor authentication after 2 breaches of 600K accounts

passwords, Roku, Security, smart tvs, streaming, Tech, two-factor authentication / /

Roku account breach —

Accounts with stored payment information went for as little as $0.50 each.

–

Roku logo on TV with remote in foreground

Getty Images

Everyone with a Roku TV or streaming device will eventually be forced to enable two-factor authentication after the company disclosed two separate incidents in which roughly 600,000 customers had their accounts accessed through credential stuffing.

Credential stuffing is an attack in which usernames and passwords exposed in one leak are tried out against other accounts, typically using automated scripts. When people reuse usernames and passwords across services or make small, easily intuited changes between them, actors can gain access to accounts with even more identifying information and access.

In the case of the Roku attacks, that meant access to stored payment methods, which could then be used to buy streaming subscriptions and Roku hardware. Roku wrote on its blog, and in a mandated data breach report, that purchases occurred in “less than 400 cases” and that full credit card numbers and other “sensitive information” was not revealed.

The first incident, “earlier this year,” involved roughly 15,000 user accounts, Roku stated. By monitoring these accounts, Roku identified a second incident, one that touched 576,000 accounts. These were collectively “a small fraction of Roku’s more than 80M active accounts,” the post states, but the streaming giant will work to prevent future such stuffing attacks.

The affected accounts will have their passwords reset and will be notified, along with having charges reversed. Every Roku account, when next requiring a login, will now need to verify their account through a link sent to their email address. Alternatively, one can use the device ID of any linked Roku device, according to Roku’s support page. (Forcing this upgrade yourself is probably a good idea for past or present Roku owners.)

Security blog BleepingComputer reported around the time of the incident that breached Roku accounts were sold for as little as 50 cents each and likely obtained using commonly available stuffing tools that bypass brute-force protections through proxies and other means. BleepingComputer reported that “a source” tied Roku’s recent updates to its Dispute Resolution Terms, which all but locked Roku devices until a customer agreed, to the fraudulent activity. Roku told BleepingComputer that the two were not related.

Roku forcing 2-factor authentication after 2 breaches of 600K accounts Read More »

sleeping-more-flushes-junk-out-of-the-brain

Sleeping more flushes junk out of the brain

Biology, brain, Neuroscience, Science, sleep / /

Better sleep on it —

Rhythmic activity during sleep may get fluids in the brain moving.

–

Abstract image of a pink brain against a blue background.

As if we didn’t have enough reasons to get at least eight hours of sleep, there is now one more. Neurons are still active during sleep. We may not realize it, but the brain takes advantage of this recharging period to get rid of junk that was accumulating during waking hours.

Sleep is something like a soft reboot. We knew that slow brainwaves had something to do with restful sleep; researchers at the Washington University School of Medicine in St. Louis have now found out why. When we are awake, our neurons require energy to fuel complex tasks such as problem-solving and committing things to memory. The problem is that debris gets left behind after they consume these nutrients. As we sleep, neurons use these rhythmic waves to help move cerebrospinal fluid through brain tissue, carrying out metabolic waste in the process.

In other words, neurons need to take out the trash so it doesn’t accumulate and potentially contribute to neurodegenerative diseases. “Neurons serve as master organizers for brain clearance,” the WUSTL research team said in a study recently published in Nature.

Built-in garbage disposal

Human brains (and those of other higher organisms) evolved to have billions of neurons in the functional tissue, or parenchyma, of the brain, which is protected by the blood-brain barrier.

Everything these neurons do creates metabolic waste, often in the form of protein fragments. Other studies have found that these fragments may contribute to neurodegenerative diseases such as Alzheimer’s.

The brain has to dispose of its garbage somehow, and it does this through what’s called the glymphatic system (no, that’s not a typo), which carries cerebrospinal fluid that moves debris out of the parenchyma through channels located near blood vessels. However, that still left the questions: What actually powers the glymphatic system to do this—and how? The WUSTL team wanted to find out.

To see what told the glymphatic system to dump the trash, scientists performed experiments on mice, inserting probes into their brains and planting electrodes in the spaces between neurons. They then anesthetized the mice with ketamine to induce sleep.

Neurons fired strong, charged currents after the animals fell asleep. While brain waves under anesthesia were mostly long and slow, they induced corresponding waves of current in the cerebrospinal fluid. The fluid would then flow through the dura mater, the outer layer of tissue between the brain and the skull, taking the junk with it.

Just flush it

The scientists wanted to be sure that neurons really were the force that pushed the glymphatic system into action. To do that, they needed to genetically engineer the brains of some mice to nearly eliminate neuronal activity while they were asleep (though not to the point of brain death) while leaving the rest of the mice untouched for comparison.

In these engineered mice, the long, slow brain waves seen before were undetectable. As a result, the fluid was no longer pushed to carry metabolic waste out of the brain. This could only mean that neurons had to be active in order for the brain’s self-cleaning cycle to work.

Furthermore, the research team found that there were fluctuations in the brain waves of the un-engineered mice, with slightly faster waves thought to be targeted at the debris that was harder to remove (at least, this is what the researchers hypothesized). It is not unlike washing a plate and then needing to scrub slightly harder in places where there is especially stubborn residue.

The researchers also found out why previous experiments produced different results. Because the flushing out of cerebrospinal fluid that carries waste relies so heavily on neural activity, the type of anesthetic used mattered—anesthetics that inhibit neural activity can interfere with the results. Other earlier experiments worked poorly because of injuries caused by older and more invasive methods of implanting the monitoring hardware into brain tissues. This also disrupted neurons.

“The experimental methodologies we used here largely avoid acute damage to the brain parenchyma, thereby providing valuable strategies for further investigations into neural dynamics and brain clearance,” the team said in the same study.

Now that neurons are known to set the glymphatic system into motion, more attention can be directed towards the intricacies of that process. Finding out more about the buildup and cleaning of metabolic waste may contribute to our understanding of neurodegenerative diseases. It’s definitely something to think about before falling asleep.

Nature, 2024.  DOI: 10.1038/s41586-024-07108-6

Sleeping more flushes junk out of the brain Read More »

why-do-some-people-always-get-lost?

Why do some people always get lost?

navigation, Science, syndication / /

I don’t know where I’m going —

Experience may matter more than innate ability when it comes to sense of direction.

–

Scientists are homing in on how navigation skills develop.

Enlarge / Scientists are homing in on how navigation skills develop.

Knowable Magazine (CC BY-ND)

Like many of the researchers who study how people find their way from place to place, David Uttal is a poor navigator. “When I was 13 years old, I got lost on a Boy Scout hike, and I was lost for two and a half days,” recalls the Northwestern University cognitive scientist. And he’s still bad at finding his way around.

The world is full of people like Uttal—and their opposites, the folks who always seem to know exactly where they are and how to get where they want to go. Scientists sometimes measure navigational ability by asking someone to point toward an out-of-sight location—or, more challenging, to imagine they are someplace else and point in the direction of a third location—and it’s immediately obvious that some people are better at it than others.

“People are never perfect, but they can be as accurate as single-digit degrees off, which is incredibly accurate,” says Nora Newcombe, a cognitive psychologist at Temple University who coauthored a look at how navigational ability develops in the 2022 Annual Review of Developmental Psychology. But others, when asked to indicate the target’s direction, seem to point at random. “They have literally no idea where it is.”

While it’s easy to show that people differ in navigational ability, it has proved much harder for scientists to explain why. There’s new excitement brewing in the navigation research world, though. By leveraging technologies such as virtual reality and GPS tracking, scientists have been able to watch hundreds, sometimes even millions, of people trying to find their way through complex spaces, and to measure how well they do. Though there’s still much to learn, the research suggests that to some extent, navigation skills are shaped by upbringing.

Nurturing navigation skills

The importance of a person’s environment is underscored by a recent look at the role of genetics in navigation. In 2020, Margherita Malanchini, a developmental psychologist at Queen Mary University of London, and her colleagues compared the performance of more than 2,600 identical and nonidentical twins as they navigated through a virtual environment to test whether navigational ability runs in families. It does, they found—but only modestly. Instead, the biggest contributor to people’s performance was what geneticists call the “nonshared environment”—that is, the unique experiences each person accumulates as their life unfolds. Good navigators, it appears, are mostly made, not born.

A remarkable, large-scale experiment led by Hugo Spiers, a cognitive neuroscientist at University College London, gave researchers a glimpse at how experience and other cultural factors might influence wayfinding skills. Spiers and his colleagues, in collaboration with the telecom company T-Mobile, developed a game for cellphones and tablets, Sea Hero Quest, in which players navigate by boat through a virtual environment to locate a series of checkpoints. The game app asked participants to provide basic demographic data, and nearly 4 million worldwide did so. (The app is no longer accepting new participants except by invitation of researchers.)

Through the app, the researchers were able to measure wayfinding ability by the total distance each player traveled to reach all the checkpoints. After completing some levels of the game, players also had to shoot a flare back toward their point of origin—a dead-reckoning test analogous to the pointing-to-out-of-sight-locations task. Then Spiers and his colleagues could compare players’ performance to the demographic data.

Several cultural factors were associated with wayfinding skills, they found. People from Nordic countries tended to be slightly better navigators, perhaps because the sport of orienteering, which combines cross-country running and navigation, is popular in those countries. Country folk did better, on average, than people from cities. And among city-dwellers, those from cities with more chaotic street networks such as those in the older parts of European cities did better than those from cities like Chicago, where the streets form a regular grid, perhaps because residents of grid cities don’t need to build such complex mental maps.

Why do some people always get lost? Read More »

us-drug-shortages-reach-record-high-with-323-meds-now-in-short-supply

US drug shortages reach record high with 323 meds now in short supply

adderall, ASHP, cancer drugs, DEA, drug shortages, fda, generic drugs, health, pharmacists, pharmacy, Science / /

Terrible —

The shortages affect everything from generic cancer drugs to ADHD medication.

–

Takeda Pharmaceutical Co. Adderall XR brand medication arranged at a pharmacy in Provo, Utah, in November 2023.

Enlarge / Takeda Pharmaceutical Co. Adderall XR brand medication arranged at a pharmacy in Provo, Utah, in November 2023.

Drug shortages in the US have reached an all-time high, with 323 active and ongoing shortages already tallied this year, according to data collected by the American Society of Health-System Pharmacists (ASHP).

The current drug shortage total surpasses the previous record of 320, set in 2014, and is the highest recorded since ASHP began tracking shortages in 2001.

“All drug classes are vulnerable to shortages,” ASHP CEO Paul Abramowitz said in a statement Thursday. “Some of the most worrying shortages involve generic sterile injectable medications, including cancer chemotherapy drugs and emergency medications stored in hospital crash carts and procedural areas. Ongoing national shortages of therapies for attention-deficit/hyperactivity disorder [ADHD] also remain a serious challenge for clinicians and patients.”

Erin Fox, associate chief pharmacy officer of University of Utah Health, told CBS MoneyWatch, that most of the drugs in short supply are generic, older products, and around half are injectable drugs that require more stringent manufacturing processes.

There are myriad reasons for the hundreds of drug shortages now facing doctors and patients, many of which remain unclear. But, as Ars has reported before, the root cause of shortages of low-cost, off-patent generic drugs is well established. These drugs have razor-thin to non-existent profit margins, driven by middle managers who have, in recent years, pushed down wholesale prices to rock-bottom levels. In some cases, generic manufacturers lose money on the drugs, disincentivizing other players in the pharmaceutical industry from stepping in to bolster fragile supply chains. Several generic manufacturers have filed for bankruptcy recently.

For other drugs, the situation is more complicated. The ADHD drug Adderall, for instance, has been in critical shortage since October 2022, causing millions of patients around the country to struggle to fill their prescriptions. It began when a manufacturing delay for one manufacturer kicked off a shortfall. Although that problem has since been resolved, it came amid a significant increase in Adderall prescriptions, which spiked further during the pandemic when telehealth prescribing became more common. Additionally, because Adderall—made of amphetamine-mixed salts—is a controlled substance, the Drug Enforcement Administration sets limits or “quotas” on how much of it manufacturers can make. Such quotas can exacerbate shortages, ASHP said.

But, in a joint letter with the Food and Drug Administration last August, the DEA said that, by its data, manufacturers of amphetamine products (including Adderall) only sold approximately 70 percent of their allotted quotas in 2022. That meant that there were approximately 1 billion more doses that they could have produced but did not make or ship. At the time, the agencies said its 2023 data was trending in the same direction.

In 60 percent of cases, manufacturers do not know or do not provide reasons why their drugs fall into short supply, ASHP found.

The organization has put forth policy recommendations to prevent shortages and improve supply chains, advocating for federal and regulatory changes. “Much work remains to be done at the federal level to fix the root causes of drug shortages,” Abramowitz said. “ASHP will continue to engage with policymakers regularly as we guide efforts to draft and pass new legislation to address drug shortages and continue to strongly advocate on behalf of our members for solutions that work.”

US drug shortages reach record high with 323 meds now in short supply Read More »

sd-cards-finally-expected-to-hit-4tb-in-2025

SD cards finally expected to hit 4TB in 2025

microSD, SanDisk, SD cards, Tech, Western Digital / /

Not quite at 128TB yet —

For media pros’ cameras and laptops.

–

Two SD cards on a wood surface

Enlarge / Generic, non-Western Digital SD cards.

Western Digital plans to release the first 4TB SD card next year. On Thursday, the storage firm announced plans to demo the product in person next week.

Western Digital will launch the SD card, which follows the SD Association’s Secure Digital Ultra Capacity (SDUC) standard, under its SanDisk brand and market it toward “complex media and entertainment workflows,” such as high-resolution video with high framerates, using cameras and laptops, the announcement said.

The spacious card will use the Ultra High Speed-1 (UHS-1) bus interface, supporting max theoretical transfer rates of up to 104 MB per second. It will support minimum write speeds of 10 MB/s, AnandTech reported. Minimum sequential write speeds are expected to reach 30 MB/s, the publication said.

With those specs, the upcoming SD card should be able to fit storage-hungry video formats, including 8K, although the card won’t be speedy enough to support raw 8K video recording. That helps explain why Western Digital is initially previewing the offering at The National Association of Broadcasters’ annual event for broadcast and media professionals, which starts tomorrow in Las Vegas.

“Attendees will get a preview of the 4TB SD card’s full capacity and learn more about how it will expand the creative possibilities for cameras and laptops,” Western Digital said.

Hopefully, Western Digital will provide more information about the SD card at the event, as this week’s announcement didn’t go into further details, such as the type of NAND the card will use or whether it will support the SanDisk-created DDR200/DDR208 mode, which could enable higher data transfer speeds of up to 170 MB/s (but only with a host that supports that mode).

Western Digital didn’t say what the SD card would cost, but with its advanced capabilities and targeted audience of professional creators, the offering will likely have premium pricing. The 1TB SanDisk Extreme Pro SDXC UHS-I Card currently has a $140 MSRP.

WD’s announcement comes six years after the SD Association, which writes SD standards, announced the SDUC standard that bumped the max possible capacity for SD cards from 2TB to 128TB. As with many releases of new standards, some were eager to believe that “SD cards could soon hold 128TB of storage.” But with 4TB not expected to arrive until 2025 (assuming there are no delays), “soon,” even today, is unrealistic. The 2018 standard only made such capacities a possibility. At the time, the max theoretical capacities for an SD card had been 2TB for about nine years, but the most spacious SD cards actually available for purchase at the time were 512GB.

Today, the highest capacity SD cards readily available are 1TB (Western Digital also this week announced plans to release a 2TB SanDisk Extreme Pro SDXC UHS-I memory card at an unspecified date), while 2TB microSD cards only became available this year.

With Western Digital’s announcement, the industry inches closer to SD cards’ maximum potential. For professionals with high-storage needs, that’s exciting news, as having more capable technology opens more possibiltiies, like working with high-res media.

However, Western Digital’s announcement also comes as SanDisk’s reputation for reliable storage is in serious question by professional and long-time customers. There are multiple lawsuits about SanDisk Extreme portable SSDs reportedly failing unexpectedly. These alleged failures, combined with frustration around Western Digital’s limited response to reported data losses, could have professionals with work-critical storage needs consider waiting for another brand to make the leap to 4TB.

SD cards finally expected to hit 4TB in 2025 Read More »

“highly-capable”-hackers-root-corporate-networks-by-exploiting-firewall-0-day

“Highly capable” hackers root corporate networks by exploiting firewall 0-day

Biz & IT, exploits, Palo Alto Networks, pan-os, Security, vulnerabilities, zerodays / /
The word ZERO-DAY is hidden amidst a screen filled with ones and zeroes.

Highly capable hackers are rooting multiple corporate networks by exploiting a maximum-severity zero-day vulnerability in a firewall product from Palo Alto Networks, researchers said Friday.

The vulnerability, which has been under active exploitation for at least two weeks now, allows the hackers with no authentication to execute malicious code with root privileges, the highest possible level of system access, researchers said. The extent of the compromise, along with the ease of exploitation, has earned the CVE-2024-3400 vulnerability the maximum severity rating of 10.0. The ongoing attacks are the latest in a rash of attacks aimed at firewalls, VPNs, and file-transfer appliances, which are popular targets because of their wealth of vulnerabilities and direct pipeline into the most sensitive parts of a network.

“Highly capable” UTA0218 likely to be joined by others

The zero-day is present in PAN-OS 10.2, PAN-OS 11.0, and/or PAN-OS 11.1 firewalls when they are configured to use both the GlobalProtect gateway and device telemetry. Palo Alto Networks has yet to patch the vulnerability but is urging affected customers to follow the workaround and mitigation guidance provided here. The advice includes enabling Threat ID 95187 for those with subscriptions to the company’s Threat Prevention service and ensuring vulnerability protection has been applied to their GlobalProtect interface. When that’s not possible, customers should temporarily disable telemetry until a patch is available.

Volexity, the security firm that discovered the zero-day attacks, said that it’s currently unable to tie the attackers to any previously known groups. However, based on the resources required and the organizations targeted, they are “highly capable” and likely backed by a nation-state. So far, only a single threat group—which Volexity tracks as UTA0218—is known to be leveraging the vulnerability in limited attacks. The company warned that as new groups learn of the vulnerability, CVE-2024-3400, is likely to come under mass exploitation, just as recent zero-days affecting products from the likes of Ivanti, Atlassian, Citrix, and Progress have in recent months.

“As with previous public disclosures of vulnerabilities in these kinds of devices, Volexity assesses that it is likely a spike in exploitation will be observed over the next few days by UTA0218 and potentially other threat actors who may develop exploits for this vulnerability,” company researchers wrote Friday. “This spike in activity will be driven by the urgency of this window of access closing due to mitigations and patches being deployed. It is therefore imperative that organizations act quickly to deploy recommended mitigations and perform compromise reviews of their devices to check whether further internal investigation of their networks is required.”

The earliest attacks Volexity has seen took place on March 26 in what company researchers suspect was UTA0218 testing the vulnerability by placing zero-byte files on firewall devices to validate exploitability. On April 7, the researchers observed the group trying unsuccessfully to install a backdoor on a customer’s firewall. Three days later, the group’s attacks were successfully deploying malicious payloads. Since then, the threat group has deployed custom, never-before-seen post-exploitation malware. The backdoor, which is written in the Python language, allows the attackers to use specially crafted network requests to execute additional commands on hacked devices.

“Highly capable” hackers root corporate networks by exploiting firewall 0-day Read More »

words-are-flowing-out-like-endless-rain:-recapping-a-busy-week-of-llm-news

Words are flowing out like endless rain: Recapping a busy week of LLM news

AI, Anthropic, Biz & IT, Chatbot Arena, chatgpt, chatgtp, Claude 3, Claude 3 Opus, Claude 3 Sonnet, Gemini, Gemini 1.5, Google Gemini, GPT-4, GPT-4-turbo, large language models, machine learning, Meta, Mistral, Mixtral, Mixtral 8x22B, openai, RECAP, Simon Willison, whirlwind / /

many things frequently —

Gemini 1.5 Pro launch, new version of GPT-4 Turbo, new Mistral model, and more.

–

An image of a boy amazed by flying letters.

Enlarge / An image of a boy amazed by flying letters.

Some weeks in AI news are eerily quiet, but during others, getting a grip on the week’s events feels like trying to hold back the tide. This week has seen three notable large language model (LLM) releases: Google Gemini Pro 1.5 hit general availability with a free tier, OpenAI shipped a new version of GPT-4 Turbo, and Mistral released a new openly licensed LLM, Mixtral 8x22B. All three of those launches happened within 24 hours starting on Tuesday.

With the help of software engineer and independent AI researcher Simon Willison (who also wrote about this week’s hectic LLM launches on his own blog), we’ll briefly cover each of the three major events in roughly chronological order, then dig into some additional AI happenings this week.

Gemini Pro 1.5 general release

On Tuesday morning Pacific time, Google announced that its Gemini 1.5 Pro model (which we first covered in February) is now available in 180-plus countries, excluding Europe, via the Gemini API in a public preview. This is Google’s most powerful public LLM so far, and it’s available in a free tier that permits up to 50 requests a day.

It supports up to 1 million tokens of input context. As Willison notes in his blog, Gemini 1.5 Pro’s API price at $7/million input tokens and $21/million output tokens costs a little less than GPT-4 Turbo (priced at $10/million in and $30/million out) and more than Claude 3 Sonnet (Anthropic’s mid-tier LLM, priced at $3/million in and $15/million out).

Notably, Gemini 1.5 Pro includes native audio (speech) input processing that allows users to upload audio or video prompts, a new File API for handling files, the ability to add custom system instructions (system prompts) for guiding model responses, and a JSON mode for structured data extraction.

“Majorly Improved” GPT-4 Turbo launch

A GPT-4 Turbo performance chart provided by OpenAI.

Enlarge / A GPT-4 Turbo performance chart provided by OpenAI.

Just a bit later than Google’s 1.5 Pro launch on Tuesday, OpenAI announced that it was rolling out a “majorly improved” version of GPT-4 Turbo (a model family originally launched in November) called “gpt-4-turbo-2024-04-09.” It integrates multimodal GPT-4 Vision processing (recognizing the contents of images) directly into the model, and it initially launched through API access only.

Then on Thursday, OpenAI announced that the new GPT-4 Turbo model had just become available for paid ChatGPT users. OpenAI said that the new model improves “capabilities in writing, math, logical reasoning, and coding” and shared a chart that is not particularly useful in judging capabilities (that they later updated). The company also provided an example of an alleged improvement, saying that when writing with ChatGPT, the AI assistant will use “more direct, less verbose, and use more conversational language.”

The vague nature of OpenAI’s GPT-4 Turbo announcements attracted some confusion and criticism online. On X, Willison wrote, “Who will be the first LLM provider to publish genuinely useful release notes?” In some ways, this is a case of “AI vibes” again, as we discussed in our lament about the poor state of LLM benchmarks during the debut of Claude 3. “I’ve not actually spotted any definite differences in quality [related to GPT-4 Turbo],” Willison told us directly in an interview.

The update also expanded GPT-4’s knowledge cutoff to April 2024, although some people are reporting it achieves this through stealth web searches in the background, and others on social media have reported issues with date-related confabulations.

Mistral’s mysterious Mixtral 8x22B release

An illustration of a robot holding a French flag, figuratively reflecting the rise of AI in France due to Mistral. It's hard to draw a picture of an LLM, so a robot will have to do.

Enlarge / An illustration of a robot holding a French flag, figuratively reflecting the rise of AI in France due to Mistral. It’s hard to draw a picture of an LLM, so a robot will have to do.

Not to be outdone, on Tuesday night, French AI company Mistral launched its latest openly licensed model, Mixtral 8x22B, by tweeting a torrent link devoid of any documentation or commentary, much like it has done with previous releases.

The new mixture-of-experts (MoE) release weighs in with a larger parameter count than its previously most-capable open model, Mixtral 8x7B, which we covered in December. It’s rumored to potentially be as capable as GPT-4 (In what way, you ask? Vibes). But that has yet to be seen.

“The evals are still rolling in, but the biggest open question right now is how well Mixtral 8x22B shapes up,” Willison told Ars. “If it’s in the same quality class as GPT-4 and Claude 3 Opus, then we will finally have an openly licensed model that’s not significantly behind the best proprietary ones.”

This release has Willison most excited, saying, “If that thing really is GPT-4 class, it’s wild, because you can run that on a (very expensive) laptop. I think you need 128GB of MacBook RAM for it, twice what I have.”

The new Mixtral is not listed on Chatbot Arena yet, Willison noted, because Mistral has not released a fine-tuned model for chatting yet. It’s still a raw, predict-the-next token LLM. “There’s at least one community instruction tuned version floating around now though,” says Willison.

Chatbot Arena Leaderboard shake-ups

A Chatbot Arena Leaderboard screenshot taken on April 12, 2024.

Enlarge / A Chatbot Arena Leaderboard screenshot taken on April 12, 2024.

Benj Edwards

This week’s LLM news isn’t limited to just the big names in the field. There have also been rumblings on social media about the rising performance of open source models like Cohere’s Command R+, which reached position 6 on the LMSYS Chatbot Arena Leaderboard—the highest-ever ranking for an open-weights model.

And for even more Chatbot Arena action, apparently the new version of GPT-4 Turbo is proving competitive with Claude 3 Opus. The two are still in a statistical tie, but GPT-4 Turbo recently pulled ahead numerically. (In March, we reported when Claude 3 first numerically pulled ahead of GPT-4 Turbo, which was then the first time another AI model had surpassed a GPT-4 family model member on the leaderboard.)

Regarding this fierce competition among LLMs—of which most of the muggle world is unaware and will likely never be—Willison told Ars, “The past two months have been a whirlwind—we finally have not just one but several models that are competitive with GPT-4.” We’ll see if OpenAI’s rumored release of GPT-5 later this year will restore the company’s technological lead, we note, which once seemed insurmountable. But for now, Willison says, “OpenAI are no longer the undisputed leaders in LLMs.”

Words are flowing out like endless rain: Recapping a busy week of LLM news Read More »

texas-surgeon-accused-of-secretly-blocking-patients-from-getting-transplants

Texas surgeon accused of secretly blocking patients from getting transplants

data manipulation, kidney, liver, Science, texas, transplant / /

Bizarre —

The motive for the alleged data manipulation is unknown.

–

Texas surgeon accused of secretly blocking patients from getting transplants

An accomplished and prominent transplant surgeon in Texas allegedly falsified patient data in a government transplant waiting list, which may have prevented his own patients from receiving lifesaving liver transplants, according to media reports and hospital statements.

Memorial Hermann-Texas Medical Center halted its liver transplant program on April 3 after finding “irregularities” with donor acceptance criteria, the Houston Chronicle reported based on a statement from the hospital. At the time there were 38 patients on the hospital’s wait list for a liver. Earlier this week, the hospital also halted its kidney transplant program, telling the Chronicle that it was pausing operations to “evaluate a new physician leadership structure.”

Memorial Hermann has not named the surgeon behind the “inappropriate changes,” but The New York Times identified him as Dr. Steve Bynon, a surgeon who has received numerous accolades and, at one point, appears to have been featured on a billboard. Bynon oversaw both the liver and kidney transplant programs at Memorial Hermann.

According to the Times, the hospital said that a doctor in its liver transplant program had admitted to changing patient records, and those changes led to patients being denied transplants. Officials who spoke with the Times identified said physician as Bynon.

Bynon is a surgical transplant director at the University of Texas Health Science Center at Houston (UTHealth Houston), as well as the leader of Memorial Hermann’s abdominal transplant program. Ars was not able to reach Bynon for comment. When reporters for the Times reached Bynon by phone Thursday, he did not confirm that he had admitted to manipulating the data and referred further questions to UTHealth Houston.

On Friday, UTHealth Houston released a statement defending Bynon, saying that he is “an exceptionally talented and caring physician, and a pioneer in abdominal organ transplantation.” The statement continued by saying that the survival rates and outcomes for Dr. Bynon’s transplant patients are among the best in the country. “Our faculty and staff members, including Dr. Bynon, are assisting with the inquiry into Memorial Hermann’s liver transplant program and are committed to addressing and resolving any findings identified by this process,” it said.

Officials at Memorial Hermann told the Times that the data allegedly manipulated involved criteria for organ donors. When doctors place patients on a waiting list for an organ donation, they specify criteria for acceptable donors. This includes factors like a potential donor’s age and weight. But, according to the officials, entries for some of the patients in Bynon’s program had criteria that were set to impossible conditions, such as a patient only accepting organs from a toddler who weighed 300 pounds. This effectively made Bynon’s patients ineligible for a transplant.

The Chronicle notes that Memorial Hermann has seen increasing numbers of liver transplant candidates die or become too sick for a transplant while on the waitlist. According to federal data, in 2021, only four patients died or got too sick for a transplant while on the hospital’s liver transplant waiting list. In 2022, the number increased to 11, and in 2023, it was 14. So far this year, there are five patients who have died or gotten too sick while waiting.

If the allegations are true, it’s unclear what motivated Bynon to alter the data.

It’s also unclear when the hospital’s transplant programs will resume. In a statement Friday, Memorial Hermann said it is working to “make the necessary changes that will allow for the quick reactivation of the kidney transplant program under a different physician leadership structure.” It did not address when the liver transplant program may resume.

Texas surgeon accused of secretly blocking patients from getting transplants Read More »

google-mocks-epic’s-proposed-reforms-to-end-android-app-market-monopoly

Google mocks Epic’s proposed reforms to end Android app market monopoly

android, android app marketplace, Android Apps, anticompetitive behavior, Antitrust law, App Store, epic games, epic games store, gaming, Google, Google Play Store, monopoly, Policy / /
Google mocks Epic’s proposed reforms to end Android app market monopoly

Epic Games has filed a proposed injunction that would stop Google from restricting third-party app distribution outside Google Play Store on Android devices after proving that Google had an illegal monopoly in markets for Android app distribution.

Epic is suggesting that competition on the Android mobile platform would be opened up if the court orders Google to allow third-party app stores to be distributed for six years in the Google Play Store and blocks Google from entering any agreements with device makers that would stop them from pre-loading third-party app stores. This would benefit both mobile developers and users, Epic argued in a wide-sweeping proposal that would greatly limit Google’s control over the Android app ecosystem.

US District Court Judge James Donato will ultimately decide the terms of the injunction. Google has until May 3 to respond to Epic’s filing.

A Google spokesperson confirmed to Ars that Google still plans to appeal the verdict—even though Google already agreed to a $700 million settlement with consumers and states following Epic’s win.

“Epic’s filing to the US Federal Court shows again that it simply wants the benefits of Google Play without having to pay for it,” Google’s spokesperson said. “We’ll continue to challenge the verdict, as Android is an open mobile platform that faces fierce competition from the Apple App Store, as well as app stores on Android devices, PCs, and gaming consoles.”

If Donato accepts Epic’s proposal, Google would be required to grant equal access to the Android operating system and platform features to all developers, not just developers distributing apps through Google Play. This would allow third-party app stores to become the app update owner, updating any apps downloaded from their stores as seamlessly as Google Play updates apps.

Under Epic’s terms, any app downloaded from anywhere would operate identically to apps downloaded from Google Play, without Google imposing any unnecessary distribution fees. Similarly, developers would be able to provide their own in-app purchasing options and inform users of out-of-app purchasing options, without having to use Google’s APIs or paying Google additional fees.

Notably, Epic filed its lawsuit after Google removed the Epic game Fortnite from the Google Play Store because Epic tried to offer an “Epic Direct Payment” option for in-game purchases.

“Google must also allow developers to communicate directly with their consumers, including linking from their app to a website to make purchases and get deals,” Epic said in a blog post. “Google would be blocked from using sham compliance programs like User Choice Billing to prevent competing payment options inside an app or on a developer’s website.”

Unsurprisingly, Epic’s proposed injunction includes an “anti-retaliation” section specifically aimed at protecting Epic from any further retaliation. If Donato accepts the terms, Google would be violating the injunction order if the tech giant fails to prove that it is not “treating Epic differently than other developers” by making it “disproportionately difficult or costly” for Epic to develop, update, and market its apps on Android.

That part of the injunction would seem important since, last month, Epic announced that an Epic Games Store was “coming to iOS and Android” later this year. According to Inc, Epic told Game Developers Conference attendees that its app-distribution platform will be the “first ever game-focused, multiplatform store,” working across “Android, iOS, PC and macOS.”

Google mocks Epic’s proposed reforms to end Android app market monopoly Read More »

the-diskmantler-violently-shakes-hard-drives-for-better-rare-earth-recovery

The DiskMantler violently shakes hard drives for better rare-earth recovery

e-waste, electronics recycling, hard drives, hdd, hdds, rare earth, rare earth elements, rare earth magnets, rare earth metals, Tech / /

Rare-earth magnet recycling —

A nifty HDD disassembly tool, sure, but we also have a larger global need for magnets.

–

From magnets we came, to magnets we return.

Enlarge / From magnets we came, to magnets we return.

Garner Products

There is the mental image that most people have of electronics recycling, and then there is the reality, which is shredding.

Less than 20 percent of e-waste even makes it to recycling. That which does is, if not acquired through IT asset disposition (ITAD) or spotted by a worker who sees some value, heads into the shredder for raw metals extraction. If you’ve ever toured an electronics recycling facility, you can see for yourself how much of your stuff eventually gets chewed into little bits, whether due to design, to unprofitable reuse markets, or sheer volume concerns.

Traditional hard drives have some valuable things inside them—case, cover, circuit boards, drive assemblies, actuators, and rare-earth magnets—but only if they avoid the gnashing teeth. That’s where the DiskMantler comes in. Garner Products, a data elimination firm, has a machine that it claims can process 500 hard drives (the HDD kind) per day in a way that leaves a drive separated into those useful components. And the DiskMantler does this by shaking the thing to death (video).

  • Insert the drive into the DiskMantler like you’re nostalgic for the VCR days.

    Garner Products

  • The DiskMantler shakes the drive until the screws fly out of the thing.

    Garner Products

  • The disassembled drive pops out or lands on a conveyor belt.

    Garner Products

  • The component parts that the DiskMantler breaks down to.

    Garner Products

  • The DiskMantler itself, which needs an air supply and power.

    Garner Products

The DiskMantler, using “shock, harmonics, and vibration,” vibrates most drives into pieces in between 8–90 seconds, depending on how much separation you want. Welded helium drives take about two minutes. The basic science for how this works came from Gerhard Junker, the perfectly named German scientist who fully explored the power of vibrations, or “shear loading perpendicular to the fastener axis,” to loosen screws and other fasteners.

As Garner’s chief global development officer, Michael Harstrick, told E-Scrap News, the device came about when a client needed a way to extract circuit boards from drives fastened with proprietary screw heads. Prying or other destruction would have been too disruptive and potentially damaging. After testing different power levels and durations, Garner arrived at a harmonic vibration device that can take apart pretty much any drive, even those with more welding than screws. “They still come apart,” Harstrick told E-Scrap News. “It just takes a little bit.”

Improving the recovery and sorting ease of hard drives is itself a useful thing, but the potential for rare-earth magnet recycling is particularly attractive. Most rare-earth magnet recycling involves “long-loop” recycling, or breaking them down into rare earth elements and then putting those back into the magnet production stream, which is energy-intensive and not very cost-effective. Electric vehicles and wind turbines have huge amounts of rare-earth magnets in them but rarely see recycling. Hard drives, while individually small, are massive in scale, with roughly 259 million shipped in 2021.

One Canadian firm, based on a University of Birmingham-patented process, wants to reuse drive magnets more directly, creating new sources that don’t require extraction and aren’t quite so globally concentrated. That Canadian firm, HyProMag, uses robotics to find and extract drives’ permanent magnets, then sends the rest of the disk off for recycling.

The technology is not all there yet, but soon enough, it looks like something interesting will shake out.

The DiskMantler violently shakes hard drives for better rare-earth recovery Read More »

spacex’s-most-flown-reusable-rocket-will-go-for-its-20th-launch-tonight

SpaceX’s most-flown reusable rocket will go for its 20th launch tonight

b1062, cape canaveral, Commercial space, falcon 9, launch, reusable rocket, Science, Space, spacex, starlink / /
File photo of a Falcon 9 rocket rolling out of its hangar at Cape Canaveral Space Force Station, Florida.

Enlarge / File photo of a Falcon 9 rocket rolling out of its hangar at Cape Canaveral Space Force Station, Florida.

For the first time, SpaceX will launch one of its reusable Falcon 9 boosters for a 20th time Friday night on a flight to deliver 23 more Starlink Internet satellites to orbit.

This milestone mission is scheduled to lift off at 9: 22 pm EDT Friday (01: 22 UTC Saturday) from Space Launch Complex 40 (SLC-40) at Cape Canaveral Space Force Station, Florida. Forecasters from the US Space Force predict “excellent” weather for the primetime launch.

Falcon 9 will blaze a familiar trail into space, following the same profile as dozens of past Starlink missions.

The rocket’s first-stage booster will shut off its nine kerosene-fueled Merlin engines about two-and-a-half minutes into the flight, reaching a top speed of more than 5,000 mph (8,000 km per hour). The first stage will detach from the Falcon 9’s upper stage, which will continue firing into orbit. The 15-story-tall Falcon 9 booster, meanwhile, will follow an arcing trajectory before braking for a vertical landing on a drone ship floating in the Atlantic Ocean near the Bahamas.

The 23 flat-packed Starlink spacecraft will deploy from the upper stage a little more than an hour after liftoff, bringing the total number of Starlinks in low-Earth orbit to more than 5,800 spacecraft.

A hunger for launch

Pretty much every day, SpaceX is either launching a rocket or rolling one out of the hangar to the launch pad. At this pace, SpaceX is redefining what is routine in the space industry, but the rapid-fire launch rate also means the company is continually breaking records, mostly its own.

Friday night’s launch will break another one of those records. This first-stage booster, designated by the tail number B1062, has flown 19 times since its first flight in November 2020. The booster will now be the first in SpaceX’s inventory to go for a 20th flight, breaking a tie with three other rockets as the company’s fleet leader.

When SpaceX debuted the latest version of its Falcon 9 rocket, the Falcon 9 Block 5, officials said the reusable first stage could fly 10 times with minimal refurbishment and perhaps additional flights with a more extensive overhaul. Now, SpaceX is certifying Falcon 9 boosters for 40 flights.

This particular rocket has not undergone any extended maintenance or long-term grounding. It has flown an average of once every two months since debuting three-and-a-half years ago. So the 20-flight milestone SpaceX will achieve Friday night means this rocket has doubled its original design life and, at the same time, has reached the halfway point of its extended service life.

In its career, this booster has launched eight people and 530 spacecraft, mostly Starlinks. The rocket’s first two flights launched GPS navigation satellites for the US military, then it launched two commercial human spaceflight missions with Dragon crew capsules. These were the all-private Inspiration4 mission and Axiom Mission 1, the first fully commercial crew flight to the International Space Station.

A SpaceX Falcon 9 rocket lifts off Sunday, April 7, on the Bandwagon 1 rideshare mission.

Enlarge / A SpaceX Falcon 9 rocket lifts off Sunday, April 7, on the Bandwagon 1 rideshare mission.

Remarkably, this will be the sixth Falcon 9 launch in less than eight days, more flights than SpaceX’s main US rival, United Launch Alliance, has launched in 17 months.

It will be the 38th Falcon 9 launch of the year and the 111th flight of a Falcon 9 or Falcon Heavy rocket—the 114th launch by SpaceX overall—in the last 365 days. More than a third of SpaceX’s Falcon 9 or Falcon Heavy missions, a number that will stand at 332 after Friday night’s flight, have launched in the past year.

This month, for the first time, SpaceX demonstrated it could launch two Falcon 9 rockets in less than five days from the company’s launch pad at Vandenberg Space Force Base, California. SpaceX has also cut the turnaround time between Falcon 9 rockets at Launch Complex 39A at NASA’s Kennedy Space Center. The company’s most-used launch pad, SLC-40, can handle two Falcon 9 flights in less than four days.

It’s not just launch pad turnaround. SpaceX uses its drone ships—two based in Florida and one in California—for most Falcon 9 landings. In order to meet the appetite for Falcon 9 launches, SpaceX is getting rockets back to port and re-deploying drone ships back to sea at a faster rate.

SpaceX’s most-flown reusable rocket will go for its 20th launch tonight Read More »