Join our Folding@Home team:
Main F@H site
Our team page
Support us: Subscribe Here
and buy SoylentNews Swag
We always have a place for talented people, visit the Get Involved section on the wiki to see how you can make SoylentNews better.
Arthur T Knackerbracket has processed the following story:
As numerous Walled Culture posts attest, site blocking is in the vanguard of the actions by copyright companies against sites engaged in the unauthorized sharing of material. Over the past few months, this approach has become even more pervasive, and even more intrusive. For example, in France, the Internet infrastructure company Cloudflare was forced to geoblock more than 400 sports streaming domain names. More worryingly, leading VPN providers were ordered to block similar sites. This represents another attack on basic Internet infrastructure, something this blog has been warning about for years.
In Spain, LaLiga, the country’s top professional football league, has not only continued to block sites, it has even ignored attempts by the Vercel cloud computing service to prevent overblocking, whereby many other unrelated sites are knocked out too. As TorrentFreak reported:
[...] the company [Vercel] set up an inbox which gave LaLiga direct access to its Site Reliability Engineering incident management system. This effectively meant that high priority requests could be processed swiftly, in line with LaLiga's demands while avoiding collateral damage.
Despite Vercel’s attempts to give LaLiga the blocks it wanted without harming other users, the football league ignored the new management system, and continued to demand excessively wide blocks. As Walled Culture has noted, this is not some minor, fringe issue: overblocking could have serious social consequences. That’s something Cloudflare’s CEO underlined in the context of LaLiga’s actions. According to TorrentFreak, he warned:
It's only a matter of time before a Spanish citizen can't access a life-saving emergency resource because the rights holder in a football match refuses to send a limited request to block one resource versus a broad request to block a whole swath of the Internet.
[...] The pioneer of this kind of excessive site blocking is Italy, with its Piracy Shield system. As Walled Culture wrote recently, there are already moves to expand Piracy Shield that will make it worse in a number of ways. The overreach of Piracy Shield has prompted the Computer & Communications Industry Association (CCIA) to write to the European Commission, urging the latter to assess the legality of the Piracy Shield under EU law. And that, finally, is what the European Commission is beginning to do.
A couple of weeks ago, the Commission sent a letter to Antonio Tajani, Italy’s Minister of Foreign Affairs and International Cooperation. In it, the European Commission offered some comments on Italy’s notification of changes in its copyright law. These changes include “amendments in the Anti-Piracy Law that entrusted Agcom [the Italian Authority for Communications Guarantees] to implement the automated platform later called the “Piracy Shield”.” In the letter, the European Commission offers its thoughts on whether Piracy Shield complies with the Digital Services Act (DSA), one of the key pieces of legislation that regulates the online world in the EU.
Arthur T Knackerbracket has processed the following story:
AMD CEO Lia Su said that chips made in TSMC’s Arizona facility are more expensive than those made in a comparable facility in Taiwan. Dr. Su said that U.S.-made chips cost ‘more than 5% but less than 20%’ higher, and she added during an interview with Bloomberg that these are costs that the company must shoulder to have a more resilient supply chain.
“I think the economics of it are we have to consider the resiliency of the supply chain, I think we learned that during the pandemic — the idea that you think about your supply chains not just by the lowest cost, but also about reliability, about resiliency, and all those things. I think that’s how we’re thinking about U.S. manufacturing,” the CEO said to Bloomberg’s Ed Ludlow. “And yes, it will be a little bit more expensive — frankly, some of the work that has been done to encourage semiconductor investment has been helpful. But when you really average it across everything else that you need to build this computing infrastructure, I think it’s a very good investment for us to make to assure that we have American manufacturing and resiliency.”
https://arstechnica.com/health/2025/07/inventor-claims-bleach-injections-will-destroy-cancer-tumors/
Xuewu Liu, a Chinese inventor who has no medical training or credentials of any kind, is charging cancer patients $20,000 for access to an AI-driven but entirely unproven treatment that includes injecting a highly concentrated dose of chlorine dioxide, a toxic bleach solution, directly into cancerous tumors.
One patient tells WIRED her tumor has grown faster since the procedure and that she suspects it may have caused her cancer to spread—a claim Liu disputes—while experts allege his marketing of the treatment has likely put him on the wrong side of US regulations.
[...] Food and Drug Administration recently removed a warning about the substance from its website. The agency says the removal was part of a routine process of archiving old pages on its site, but it has had the effect of emboldening the bleacher community.
"Without the FDA's heavy-handed warnings, it's likely my therapy would have been accepted for trials years earlier, with institutional partnerships and investor support," Liu tells WIRED.
[...] For decades, pseudoscience grifters have peddled chlorine dioxide solutions—sold under a variety of names, such as Miracle Mineral Solution—and despite warnings and prosecutions have continued to claim the toxic substance is a "cure" for everything from HIV to COVID-19 to autism.
[...] Liu claims he has injected himself with the solution more than 50 times and suffered no side effects. "This personal data point encouraged me to continue research," he says.
Liu has been making the solution in his rented apartment in Beijing by mixing citric acid with sodium chlorite
[...] "The blast blacked out my vision," Liu wrote. "Dense clouds of chlorine dioxide burst into my face, filling my eyes, nose, and mouth. I stumbled back into the apartment, rushing to the bathroom to wash out the gas from my eyes and respiratory tract. My lungs were burning. Later, I would find 4–5 cuts on my upper thigh—shards of glass had pierced through my pants." Liu also revealed that his 3-year-old daughter was nearby when the explosion happened.
[...] WIRED spoke to a patient of Liu's, whose descriptions of the treatment appear to undermine his claims of efficacy and raise serious questions about its safety.
[...] "I would welcome the fact that he's not a doctor, that he's not an MD, because he's not clouded, jaded, and biased with all kinds of misguidance that would push them the wrong way,"
[...] When asked about a timeline to have this procedure legally available in the US, Hagerman said he hopes it could be achieved before the end of 2025. Liu, however, thinks it could take slightly longer, saying that he believes clinical trials will begin in 2026.
Obligatory: https://xkcd.com/1217/
TechCrunch has an interesting report on an initiative to rehabilitate inmates in Maine:
If you omit some key details, all Preston Thorpe has to do to become a senior software engineer at a promising tech company is walk through the door.
For about six months, Thorpe was a prolific volunteer contributor to an open source project led by database company Turso. His work was impressive enough that Turso's CEO, Glauber Costa, quickly offered him a job. That was also when Costa realized that Thorpe is anything but an ordinary programmer.
"I checked his GitHub profile, and he mentions the fact that he is incarcerated," Costa told TechCrunch. "It's a story I've never seen before."
It's true: Thorpe is serving his 11th year in prison for drug-related crimes. Still, he has worked full-time from his cell at a venture-funded, San Francisco-based startup since May.
Thorpe is part of an experimental program in the Maine state prison system that allows incarcerated people to work remote jobs from custody. Though unconventional, these opportunities have proven immensely rehabilitative.
[...] The United States criminal justice system is plagued by recidivism, or former prisoners' return to custody after they have been released. Repeat offending creates a financial burden on the state and its taxpayers. But Commissioner Liberty has the data to show it's well worth the effort and investment to expand access to education and addiction treatment.
Is remote education and work a better way to rehabilitate people in prison? Are second chances worth the expense? Is the commissioner's last name a fateful omen?
Google has filed a lawsuit against the Badbox 2.0 botnet operators, after identifying over 10 million infected Android devices:
These devices lack Google's security protections, and the perpetrators pre-installed the Badbox 2.0 malware on them, to create a backdoor and abuse them for large-scale fraud and other illicit schemes.
While updates to Google Play Protect kept the malware away from devices running Google services and automatically blocked associated applications, the fresh lawsuit is meant to help the internet giant dismantle the criminal operation behind the botnet.
Badbox 2.0 "is already the largest known botnet of internet-connected TV devices, and it grows each day. It has harmed millions of victims in the United States and around the world and threatens many more," Google notes in its complaint, a copy of which was shared with SecurityWeek.
[...] According to Google, Badbox 2.0 is operated by multiple cybercrime groups from China, each having a different role in maintaining the botnet, such as establishing infrastructure, developing and pre-installing the malware on devices, and conducting fraud.
Originally spotted on Schneier on Security.
Previously: Thousands of Android Devices Come With Unkillable Backdoor Preinstalled
Arthur T Knackerbracket has processed the following story:
Take a deep breath. A flow of air has rushed into your lungs, where the oxygen moves into your bloodstream, fueling metabolic fires in cells throughout your body. You, being an aerobic organism, use oxygen as the cellular spark that frees molecular energy from the food you eat. But not all organisms on the planet live or breathe this way. Instead of using oxygen to harvest energy, many single-celled life-forms that live in environments far from oxygen’s reach, such as deep-sea hydrothermal vents or stygian crevices in the soil, wield other elements to respire and unlock energy.
This physical separation of the oxygen-rich and oxygen-free worlds is not merely a matter of life utilizing available resources; it’s a biochemical necessity. Oxygen doesn’t play nice with the metabolic pathways that make it possible to respire with the use of other elements, such as sulfur or manganese. It gives aerobes like us life, but for many anaerobes, or creatures that respire without oxygen, oxygen is a toxin that reacts with and damages their specialized molecular machinery.
[...] An ongoing mystery for researchers is how life navigated the shift from anaerobic to aerobic respiration; so much microbial biodiversity had to adapt to a world filled with what was once a biochemical bane. Now researchers have fresh insight into what that transition could have looked like billions of years ago, gleaned from an organism living today. A bacterium that researchers collected from the cauldron of a Yellowstone National Park hot spring does something that life really shouldn’t be able to do: It runs aerobic and anaerobic metabolisms simultaneously. It breathes oxygen and sulfur at the same time.
[...] RSW1 and any other microbes that have dual metabolism make intriguing models for how microbial life may have evolved during the Great Oxygenation Event, Boyd said. “That must have been a quite chaotic time for microbes on the planet,” he said. As a slow drip of oxygen filtered into the atmosphere and sea, any life-form that could handle an occasional brush with the new, poisonous gas — or even use it to its energetic benefit — may have been at an advantage. In that time of transition, two metabolisms may have been better than one.
Arthur T Knackerbracket has processed the following story:
The country also wants to force businesses to inform the government when they plan to make a ransom payment.
Ransomware gangs might have to scratch a few targets off their lists. The UK High Office and National Cyber Security Centre (NCSC) announced proposals to ban ransom payments in an effort to "crack down on cyber criminals and safeguard the public."
According to the announcement, the proposals would prohibit "public sector bodies and operators of critical national infrastructure, including the [National Health Service], local councils and schools," from making ransomware-related payments. They would also require other businesses planning to pay a ransom to notify the UK government so it can "provide those businesses with advice and support" before the payment is made. (Including a heads-up if such a payment would violate sanctions on Russia.)
The proposals wouldn't require companies to inform the UK government of a ransomware attack if they didn't plan to pay the ransom. But the announcement indicated that a mandatory reporting policy is in the works, too, in a bid to "equip law enforcement with essential intelligence to hunt down perpetrators and disrupt their activities" and "better protect British organisations and industry." That should make it more difficult to deploy ransomware in the UK without risking law enforcement's ire.
"The new package of measures will lead the way in tackling ransomware and are designed to strike against cyber criminals’ business model, bolstering our national security and protecting key services and businesses from disruption - delivering on our Plan for Change," the Home Office and NCSC said in the announcement. "They follow an extensive consultation with stakeholders across the UK, which showed strong public backing for tougher action to tackle ransomware and protect vital services."
The UK and Singapore previously said in January 2024 that they "strongly discourage anyone from paying a ransomware demand" because doing so:
- Does not guarantee the end of an incident, or the removal of malicious software from your systems
- Provides incentives for criminals to continue and expand their activities
- Provides funds that criminal actors can use for illicit activity
- Does not guarantee you will get your data back
Now the UK is looking to outright ban those payments rather than merely "strongly discouraging" them. The news follows reports earlier this week that a 158-year-old UK company was forced to shut down following a ransomware attack, at the cost of 700 jobs.
"Cyber criminals have not only cost the nation billions of pounds but in some cases have brought essential services to a standstill," the Home Office and NCSC said. "The devastating consequences are not just financial but can put lives in danger, with an NHS organisation recently identifying a ransomware attack as one of the factors that contributed to a patient’s death. These attacks have brutally exposed the alarming vulnerability at the core of our public and private institutions, from flagship British retailers and essential supermarkets including the Co-op to NHS hospitals."
From the Crystal Palace at London 1851, the telephone at Philadelphia 1876 to the escalator at Paris 1900, the World Expo has always showcased innovations and cutting-edge technologies of the time. At the ongoing Expo 2025 Osaka, host country Japan is promoting its latest technology at an unlikely spot; the bus terminal outside the main venue:
When visitors arrive at the Yumeshima Transportation Terminal 1, they will see more than 250 panels of ultrathin, lightweight "perovskite" solar cells forming the curved roof of the 250-meter-long terminal. These film-like solar panels, Japan hopes, will be the killer technology that not only grants the country more renewable power and reduces its dependency on China, but also gives it the chance to be the leader in the next generation of solar battery technology.
As reported at ZeroHedge:
Perovskite solar cells, discovered in 2009, are made from layers of chemicals just millimeters thick. Though still in early development, they rival traditional silicon-based panels in efficiency while being 20 times thinner and 10 times lighter, allowing installation on walls, rooftops, and even windows—places unsuitable for heavy panels.
"We believe this technology has the potential to beat the conventional silicon-based solar panels in terms of power generation efficiency," said Futoshi Kamiwaki, president of Sekisui Solarfilm, which developed the panels showcased at the Expo.
Japan, with limited flat land, leads major nations in solar capacity per km2 but is running out of space. Installing perovskite cells on buildings could turn cities into vertical solar farms, helping Tokyo meet its 2040 goal: 29% of power from solar, up from under 10% today.
Remember when a hard drive making a repeated grinding sound meant that it was time to backup the data, hope you get it all copied out in time, and prepare to send the hard drive to the great recycling factory in the sky? Fear not, Western Digital is on a mission to help you relive your click-of-death PSTD data loss nightmares with its latest invention: Preventive Wear Leveling aka PWL. The idea of PWL is simple, but the implementation can be a nightmare. It is supposed to move the hard drive mechanics every so often to prevent hard drive failure due to repeated limited motions. The catch is that on some drives the PWL kicks in every five seconds making an audible click that can be felt up to a meter away. This has been asked about on forums and as of July 2025 is still an issue in latest generation Western Digital hard drives that is driving people crazy with the grind and vibration occuring every five seconds. The click of death is now a feature.
Soylentils, do you have one of these drives? What is your experience with the WD five second click of death?
Arthur T Knackerbracket has processed the following story:
One of the not-funny ironies of the 21st century has been that everything we thought was social media is actually just mass media, except it’s terrible and broken. Luckily, journalists and creators are finally figuring out how to leave the old media models behind and enter the future.
The term “mass media” became popular in the 1920s to describe pop culture in the age of industrial production. Mass-produced books, movies and radio shows created a paradigm for audiences where thousands or even millions of people could experience the same exact piece of media at the same time. Before the 20th century, most people experienced their entertainment live, in theatres, bars and concert halls, where the performance was always slightly different. But a movie or radio show was the same for everyone, no matter when or where you experienced it. You could buy standardised media products for the masses, just like shoes or cars.
Social media didn’t change this formula. Platforms such as X, Facebook and TikTok were made for mass consumption. Every post, video and livestream is a product aimed at the broadest possible audience. Yes, you can target your media at certain demographics if you like, or create filter bubbles. But the whole reason why follower counts matter is because we are still in a mass media mindset, looking to see who can deliver content to the largest number of people. That isn’t “social” anything. It’s mass production under a different name.
What if we tried to make media that was truly social, without AI slop and political scapegoating? One possibility is something called cosy media, which refers to apps or other content designed to help you connect with small groups of friends, often in a friendly, calming environment. Imagine the media equivalent of meeting up with friends to knit or play cards and talk beside the fire.
The game Animal Crossing, with its low-stakes missions and cute, natural setting, is an iconic cosy-media experience. App developers are trying to reproduce that aesthetic in social apps too – anything from a group chat to an online book club can be cosy. But it isn’t just about aesthetics. A cosy social app is designed to limit your social interactions with random strangers, steering you towards trusted friends instead.
I have been using the photo-sharing app Retro a lot recently. Unlike Instagram, where Retro’s creators cut their teeth, Retro is primarily intended to be used among small groups of trusted friends. And there are no algorithms pushing videos from strangers into your feed. When I open Retro, I feel like I’m hearing from my pals rather than tuning into a fire hose of nonsense and advertising. Nothing I post there is intended to go beyond a few dozen people. Like a group chat app, Retro lets you choose who you want to talk to in a mindful way, rather than shouting into a giant algorithmic void.
We may need cosy media to soothe ourselves in a frenetic, scary time, but we also need news and analysis. Unfortunately, many of our trusted news sources are falling apart. Journalists in the US, where I live, are leaving media outlets such as The Washington Post, The New York Times and National Public Radio, citing diminishing resources and editorial freedoms.
Some, like economist Paul Krugman and technology researcher Molly White, have created successful, crowdfunded newsletters for their work. But most journalists don’t want to go solo: good reporting and analysis often require a solid team. That is why many are forming worker-owned co-operatives to start new publications, where they get institutional perks like lawyers, editors and helpful colleagues. This model is also good for consumers, who don’t want to search out and subscribe to dozens of individual newsletters just to catch up on current issues.
The worker-owned co-op model has already been a smashing success for several publications that started in the past couple of years. 404 Media [Paywall warning] is one such site, breaking news in the worlds of tech and science. Defector is a worker-owned co-op that covers sports and politics; Aftermath covers games; Hearing Things covers music. Flaming Hydra (to which I contribute) is a collective that publishes political analysis, interviews and cultural criticism. Coyote Media is about to launch in the San Francisco Bay area, to cover local news. And there are many other worker-owned local media co-ops forming.
Like mass media, social media often leads to loneliness and isolation. The point of cosy media and worker-owned publications is to rebuild community and trust. We might be witnessing the birth of a new information ecosystem, designed to help us understand the world again.
Arthur T Knackerbracket has processed the following story:
As expected, last night’s Tesla earnings announcement brought more bad news for the challenged EV-maker with 13.5pc less vehicles delivered in Q2.
Elon Musk’s Tesla woes continue as the electric vehicle (EV) maker again announced disappointing quarter two results, with its biggest decline in revenues in over a decade. And the future does not look bright with the loss in electric vehicle incentives on the way thanks to Donald Trump’s ‘Big Beautiful Bill’.
Revenues at Tesla fell 12pc in Q2 to $22.5bn. On an earnings call, the normally optimistic Musk warned of “rough quarters” ahead, when pressed on the loss of the EV incentives, and markets reacted with a drop of up to 5pc in the share value.
Most analysts believe the launch of a promised new affordable Tesla is the short-term fix, but there was little yesterday to reassure them. Having originally said the long-awaited affordable Tesla would start builds in the first half of the year, it said yesterday that “the first builds” started only in June. Musk mentioned on the call that the new model would be a version of the existing Y model.
“A lightly refreshed product offering, plus increasingly compelling alternatives from competitors in Asia, Europe and North America make it harder to sell Teslas than has been the case until quite recently,” said Forrester principal analyst, Paul Miller.
“The withdrawal of EV incentives in several countries makes the vehicles less attractive and the full impact of tariffs imposed by the US and other countries is not yet clear.”
On the earnings call, Musk continued to promote the idea of his robotaxis as the proverbial white horse that would bring Tesla back to success, but the Austin robotaxi pilot got off to a shaky start and many question Musk’s optimism when it comes to reaching his huge ambitions.
“During the investor call, Elon Musk talked about ‘getting the regulatory approvals’ to expand the Austin pilot even further, and to launch in the Bay Area, Arizona, Nevada and Florida soon,” said Miller. “He went so far as to suggest the company ‘could’ address half the US population by the end of 2025, ‘subject to regulatory approvals’.
“That caveat is an important one, as regulatory approvals take time and there is no evidence that these formal applications to the separate state regulatory processes have begun.”
Arthur T Knackerbracket has processed the following story:
Enterprise CIOs have been mesmerized by GenAI claims of autonomous agents and systems that can figure anything out. But the complexity that such large models deliver is also fueling errors, hallucinations, and spiraling bills.
All of the major model makers – OpenAI, Microsoft, Google, Amazon, Anthropic, Perplexity, etc. - are singing from the same hymnal book, the one that says that the bigger the model, the more magical it is.
But much smaller models might do a better job with controllability and reliability.
Utkarsh Kanwat is an AI engineer with ANZ, a financial institution headquartered in Australia. In a blog post, he broke down the numbers showing that large GenAI models become mathematically unsustainable at scale.
"Here's the uncomfortable truth that every AI agent company is dancing around: error compounding makes autonomous multi-step workflows mathematically impossible at production scale," Kanwat wrote . "Let's do the math. If each step in an agent workflow has 95 percent reliability, which is optimistic for current LLMs," then five steps equal a 77 percent success rate, ten steps is a 59 percent success rate, and 20 steps is a 36 percent success rate.
What does that all mean? "Production systems need 99.9%+ reliability. Even if you magically achieve 99% per-step reliability (which no one has), you still only get 82% success over 20 steps. This isn't a prompt engineering problem. This isn't a model capability problem. This is mathematical reality."
Several analysts and GenAI specialists back Kanwat's view.
Jason Andersen, a principal analyst for Moor Insights & Strategy, said that enterprises often opt for the path of least resistance. If the large model maker is promising to solve all of their problems, they want to believe that. But it is often the much smaller and more-focused strategies that deliver better results. Small, tight and well-scoped is good. Loosey goosey is bad. There is a lot of wisdom in going small
"This points out that the real value of an agent in an enterprise sense is to put boundaries around the model so you can get a certain degree of purpose out of it," Andersen said. "When you have a well-crafted and well-scoped (GenAI) strategy, you are likely to have more success."
The larger the model, "the further away you get the accuracy line, further away from reliability," Andersen said. "Small, tight and well-scoped is good. Loosey goosey is bad. There is a lot of wisdom in going small."
Andersen said that he asks CIOs whether they want the AI model "to be the pilot or the navigator?"
A good example of this, he said, is GenAI-powered vibe coding. Should AI be helping the coder or replacing the coder?
"Both have humans in the loop but what role is the human providing? Is the human running the show or is GenAI running the show?" Andersen asked.
Justin St-Maurice, technical counselor at Info-Tech Research Group, agreed that many enterprises are not doing themselves any favors by focusing overwhelmingly on the largest models.
"We are putting agents into complex sociotechnical systems. Agent systems run the risk of causing feedback loops and going off the rails, and the inherent nature of LLMs is randomness," St-Maurice said. "There is a real balance between taking advantage of the generative nature of GenAI and putting rules around it to make it behave deterministically."
Andersen offered an analogy of hiring a new employee and instead of training that new worker on how the team does things, the executive told the new employee to figure it out on their own. And when that new employee's work is not what the executive wanted, the company blames the employee rather than the executive who didn't want to spend the time or money on training new talent.
Kanwat also argued that the smaller models – even when deployed in massive numbers – can be far more cost-effective and often deliver an outright lower price.
"Context windows create quadratic cost scaling that makes conversational agents economically impossible," Kanwat said, and then he offered what he said was his own financial experience.
"Each new interaction requires processing all previous context. Token costs scale quadratically with conversation length. A 100-turn conversation costs $50-100 in tokens alone," Kanwat said. "Multiply by thousands of users and you're looking at unsustainable economics. I learned this the hard way when prototyping a conversational database agent. The first few interactions were cheap. By the 50th query in a session, each response was costing multiple dollars more than the value it provided. The economics simply don't work for most scenarios."
Kanwat said that many autonomous agent companies are going to have severe economic issues.
"Venture-funded fully autonomous agent startups will hit the economics wall first. Their demos work great with 5-step workflows, but customers will demand 20+ step processes that break down mathematically," Kanwat said. "Burn rates will spike as they try to solve unsolvable reliability problems."
Andersen agreed with the pricing concerns.
"The more context you have to give every step, the more the price goes up. It is a logarithmic pricing model," Andersen said, stressing that the model makers are going to soon be forced to sharply increase what they charge enterprises.
A chorus of AI insiders chimed in. Himanshu Tyagi is the co-founder of AI vendor Sentient and he argued that "there's a trade-off between deep reasoning and streamlined reliability. Both should coexist, not compete. Big Tech isn't going to build this. They'll optimize for lock-in." Robin Brattel, CEO of AI vendor Lab 1, agreed that many enterprises are not sufficiently focusing on the benefits of smaller models.
"AI agents that focus on specific, small-scale applications will have reduced error rates and be far more successful in production," Brattel said. "Multi-step AI agents in production will find data inconsistency and integrations incredibly challenging to resolve, causing costs and error rates to spiral."
Brattel had specific suggestions for what IT should look for when assessing various model and agent options.
Consider the "Low precision requirement. Can the solution be approximately right? Illustrations are easier than code because the illustration can be 20 percent off the ideal and still work," Brattel said. Another factor is "low risk. Generating a poem for a custom birthday card is low risk compared to a self-driving car."
One security executive who also agreed that small can often be better is Chester Wisniewski, director of global field CISO at security vendor Sophos. When Wisniewski read Kanwat's post, he said his first reaction was "Hallelujah!"
"This general LLM experiment that Meta and Google and OpenAI are pushing is all just showoff (that they are offering this) Godlike presence in our lives," Wisniewski said. "If you hypertrain a neural network to do one thing, it will do it better, faster and cheaper. If you train a very small model, it is far more efficient."
The problem, he said, is that creating a large number of smaller models requires more work from IT and it's simply easier to accept a large model that claims to do it all.
Creating those small models "requires a lot of data scientists that know how to do that training," Wisniewski said.
Even Microsoft conceded that small models can often work far better than large models. But one of their AI execs said small only works well for enterprises if the CIO's team has put in the time and thinking to map out a precise AI strategy. For those IT leaders who have yet to figure out exactly what they want AI to do, there is a reason to still embrace the largest of models.
"Large models are still the fastest way to turn an ambiguous business problem into working software. Once you know the shape of the task, smaller custom models can be cheaper and faster," said Asha Sharma, the corporate VP for AI at Microsoft. "Smart companies don't pick a side. They standardize on a common safety and observability stack, then mix and match models to meet quality, cost, and latency goals." (Note: Microsoft declined an interview request from The Register. We reached out to just about every major model maker and they either declined or ignored our request. The Microsoft comment above came from an emailed statement sent after publication.)
Not all enterprises have focused solely on large models. Capital One, for example, has focused on GenAI efforts that limit themselves to their internal data and they also severely limit what can be queried to what the database knows.
Kanwat said most enterprises are not the ideal clean environments for GenAI experiments.
"Enterprise systems aren't clean APIs waiting for AI agents to orchestrate them. They're legacy systems with quirks, partial failure modes, authentication flows that change without notice, rate limits that vary by time of day, and compliance requirements that don't fit neatly into prompt templates," Kanwat said. "Enterprise software companies that bolted AI agents onto existing products will see adoption stagnate. Their agents can't integrate deeply enough to handle real workflows."
The better enterprise approach, Kanwat said, "is not a 'chat with your code' experience. It's a focused tool that solves a specific problem efficiently."
Perhaps not a surprise. But working less, for the same pay, makes workers feel better and more relaxed.
we study how an organization-wide 4-day workweek intervention—with no reduction in pay—affects workers' well-being.
These are the findings from new peer-reviewed research published in Nature Human Behaviour, where researchers monitored the effects of a four-day work week for six months.
About 2,896 employees across 141 organisations in Australia, Canada, New Zealand, the United States, Ireland and the United Kingdom took part.
They answered surveys before and after the trial. Their answers were then compared with 285 employees from 12 companies who worked a normal five-day week.
shows improvements in burnout, job satisfaction, mental health and physical health—a pattern not observed in 12 control companies.
Three key factors mediate the relationship: improved self-reported work ability, reduced sleep problems and decreased fatigue. The results indicate that income-preserving 4-day workweeks are an effective organizational intervention for enhancing workers' well-being.
[...] "We know when people are really stressed and burnt out and not sleeping well, productivity doesn't just continue upwards," Dr Sander said.
Who wouldn't want to work one day less per week for the same pay ...
https://www.abc.net.au/news/2025-07-22/four-day-work-week-health-burnout/105555392
https://www.nature.com/articles/s41562-025-02259-6
https://lists.debian.org/debian-devel-announce/2025/07/msg00003.html
trixie release date
===================We are planning to release trixie on August 9th. There will be release parties, see if you want to join or organize one.
Full freeze
===========The Full Freeze will start on July 27th. Once the Full Freeze is instated every package will need an unblock to be able to migrate to testing. Please see the freeze policy for what qualifies for migration. We are aware that this notice is shorter than we wanted, but we don't want to delay the release.
Note that packages that have not migrated by the full freeze will be frozen, even if they were uploaded before the freeze.
Updates targeting trixie should be uploaded with an unblock request filed before the end of July 30th.
Final days before the release
=============================During the last week before the planned release date, testing will be completely frozen, and no packages will be unblocked except for critical fixes. Please refrain from filing unblock requests for non-urgent issues.
Upgrades to trixie
==================Please help test upgrades from bookworm to trixie. If you encounter any issues, file bugs against the upgrade-reports pseudo-package.
Release notes
=============If there are any noteworthy changes that should be mentioned in the release notes, please submit a bug report against the release-notes package, or a merge request in salsa after checking that it has not been reported already. The current release notes can be viewed at [6][not included].
Our own anonymous AC has found the following story:
MotorTrend reviews the recent Tesla earnings call, where Musk ended by saying that the new lower cost Tesla will be a stripped version of the Model Y SUV -- rear drive only, spartan interior, smaller (and/or lower cost chemistry) battery, etc.
Sales and earnings haven't looked good this year, I wonder if this will be enough to bring people back to Tesla stores after Elon's time in Washington DC?
While not mentioned in the article, I also wonder how current Model Y owners will feel about their neighbors' ability to get a car that looks the same, without paying the Tesla luxury car price?
The story ends with other Tesla news:
The cheap Tesla news is the big headliner of the earnings call, but there were plenty of other interesting tidbits. Per usual, Musk and Tesla championed autonomy and Tesla's nascent robotaxi service. The automaker plans to expand the size and, well, suggestive shape[*] of its robotaxi service's operating area in Austin, Texas, while also eyeing San Francisco as its next location. Robotaxis are apparently already testing in the bay area, as well as Arizona and Florida. There is also the goal of reducing the cost-per-mile of robotaxi service once the Cybercab is out in the wild. Standard Tesla robotaxi vehicles will remain more expensive than Cybercab, but it will also be built differently from regular Teslas, with longer-life tires, a plusher ride, and a much lower top speed. Tesla also expects its robotaxi service to grow much larger in 2026 and "have significant impact" on Tesla's otherwise poor financials in Q2 2025—auto revenue is down by 16 percent, year-over-year, while income from all operations are down by 42 percent, year-over-year. On top of the robotaxi fleet expansion, Musk stated that he's confident that autonomous vehicle deliveries will also expand with Bay Area driverless deliveries starting by the end of this year.
Tesla's woeful financials in part come down to the upcoming revocation of the Federal EV tax credit, tariffs impacting materials cost, and the revocation of different tax credits impacting Tesla's energy storage business. Surely, Musk's polarizing foray into politics and government is having an impact on sales, too, giving some would-be buyers pause. Musk seems to acknowledge the issue, even admitting that he's worried about being ousted as the CEO, mentioning "activist shareholders" pulling a vote to remove him. With Tesla's recent quarter being its worst since 2021 and Q1 2024, there's plenty for Musk and the company to worry about.
[*] https://www.teslaacessories.com/blogs/news/tesla-expands-robotaxi-service-area-in-austin-texas spoiler, on a map it looks like a dick...