The Register

Biting the hand that feeds IT — Enterprise Technology News and Analysis

OpenAI throws Astra into the top-tier model ring

After a brief training pause to reflect on the inadequacy of its AI security measures, OpenAI has released GPT-6 Astra, initially for those in its Trusted Access Program. Assuming all goes well and there are no more incidents like the Hugging Face hack, the AI biz in a few days will allow Plus, Pro, Business, and Enterprise subscribers to try out what the company describes as "our most intelligent model yet, with state-of-the-art performance in computer use, browsing, software engineering, science, and professional work." GPT-6 Astra arrives just two days after rival Anthropic debuted Fable 5.1, at the same eye-watering price: $10 per million input tokens and $50 per million output tokens. Astra, OpenAI says, is the first of its models to reach the "Critical" level for cybersecurity capabilities under its Preparedness Framework. The 22-page policy document has been panned [PDF] by researchers, who argue that it "allows OpenAI’s CEO to deploy even more dangerous capabilities, especially if other AI developers do so." And here we are. OpenAI's cautionary but upbeat assessment of Astra describes it as "a significant step up in cyber capabilities." "This means that, with the right tools and access, GPT‑6 Astra can find previously unknown security flaws and develop new ways to exploit them across many well-protected systems without a person guiding each step," the company said. OpenAI goes on to argue not to worry since Astra is more robust than prior models, better aligned, more comprehensively monitored, and safer in high-risk scenarios. "[W]e built a new evaluation informed by the Hugging Face incident that evaluates whether a model facing a difficult or impossible task will go beyond its intended scope," OpenAI said. "Compared to GPT‑5.6 Sol, which without production safeguards went beyond the authorized target 48 percent of the time, GPT‑6 Astra did this in 0 percent of cases." Other longstanding AI model problems are also said to have been reduced. On OpenAI's internal hallucination benchmark, Astra scored 2 percent, down from 9.4 percent for GPT-5.6 Sol. And Astra is said to be three times less likely than GPT-5.6 Sol to misrepresent what it can do. "GPT-6 Astra surpasses the human baseline in action efficiency on ARC-AGI-3," said Greg Kamradt, president of ARC Prize Foundation, in a blog post. "It used fewer actions than the median tested human on 96 percent of levels." But it does so at far greater cost – about $360 per game in GPU-forged tokens compared to $0.00067 per game if one had to purchase electricity from the grid to run a human brain. Fable 5.1 hasn't been publicly evaluated on ARC-AGI-3, but for version two of the test it scored 90 percent, compared to 95 percent for Astra. According to Artificial Analysis, GPT-6 Astra matches GPT-5.6 Sol on its Intelligent Index with a score of 61, five points less than Claude Fable 5.1 and also trailing Meta Muse Spark 1.3. In terms of its competency as a coding agent, Astra scores 67, on par with Opus 5, Fable 5, and Muse Spark 1.3. But Fable 5.1 in Claude Code tops current coding agent scores at 70. Fable 5.1, however, appears to be significantly more expensive on a cost per task basis. Artificial Analysis' numbers put Fable 5.1 at $9.18 per task, compared to $4.72 per task with GPT-6 Astra. Astra brings with it improvements in the Codex harness, which OpenAI claims result in a 1.9x faster task completion rate. Codex has also incorporated a new approach to context maintenance – retaining prior session text to inform future work. "In Codex, Astra can keep notes across context windows, preserving accumulated details without repeatedly compressing them into a single summary," the company explains. "Earlier context windows remain searchable, so Astra can find requirements or test results from previous messages and tool outputs – even if that information wasn’t captured in its notes." This feature is currently experimental and must be enabled through the Codex config.toml file until it becomes the default in a few weeks.®

Hugging Face is too important to fall into Nvidia's hands

OPINION Nvidia on Thursday announced an agreement to acquire open source AI evangelist and model depot Hugging Face for $12.9 billion in a deal that's expected to close next year, assuming regulators don't get in the way. They should get in the way. As antitrust magnets go, Nvidia buying Hugging Face takes the cake. You wouldn't let an automaker acquire the primary means of fuel distribution. Nor would you let it buy the primary means by which the mechanics are trained. Yet, Nvidia's acquisition of Hugging Face is tantamount to both. Unfortunately, the Trump administration likely lacks both the competence and teeth necessary to litigate the case, something that probably factored into CEO Jensen Huang's decision to buy Hugging Face now. To understand why Nvidia's latest acquisition is so troublesome, we need to talk about just how important Hugging Face has become to the broader AI ecosystem. Have you seen the XKCD comic in which a piece of open source software maintained by a single developer is holding up all of modern infrastructure? It could be argued that Hugging Face is that piece for the AI industry. Founded in 2016, the AI model repository has become the beating heart of the machine learning community. When a new open weights model drops, Hugging Face is the first place people look. A key element of its success has been its unrelenting commitment to open source AI, which has made it a sort of Switzerland for researchers, enthusiasts, and software devs to collaborate on neutral ground. If you've ever downloaded an open weights model, it almost certainly came from Hugging Face, whether you knew it or not. No matter how fervently Nvidia promises not to pollute Hugging Face's hallowed ground, the temptation to use the platform to advance its hardware and software interests will inevitably prove too great for Huang and his lieutenants to resist for long. In spite of this, Hugging Face CEO Clem Delangue sang Nvidia's praises on Thursday, casting the acquisition and the capital it will bring as an opportunity to grow the company's user base from around 18 million today to more than 100 million in the years to come. Yet, just a year ago Hugging Face rebuffed a $500 million investment by Nvidia, suggesting Delangue knows he's just made a deal with the devil. That said, with a cool $1 billion earmarked for Hugging Face employees joining Nvidia, Delangue and his cohort are going to have plenty of Benjamins with which to wipe away any tears of regret. What does buying Hugging Face actually get Nvidia? Nvidia would probably prefer to paint the acquisition as an altruistic move to provide financial security to one of the most important AI resources on the internet today. The case certainly can be made that Hugging Face is that one brick holding up the rest of the AI ecosystem. If it ever did exhaust its runway, the consequences would be catastrophic. And Hugging Face's business is capital-intensive. Storing petabytes' worth of AI models and datasets isn't cheap, nor is the bandwidth required to continuously serve those models at scale. Nvidia's involvement all but ensures Hugging Face never has to worry about infrastructure demand going forward. But while the benefits of Nvidia's patronage are obvious, so is the potential harm to the GPU giant's competitors. Despite frequent comparisons to GitHub, Hugging Face is more than a place for storing and sharing model weights and training data. For example, Hugging Face is home to arguably the most comprehensive documentation on AI development on the internet. It wouldn't be hard for Nvidia to use its newfound position to ensure its products are always better documented than its competitors'. You can claim until your face is red that the platform is still "open," but just because something is open doesn't mean it's unbiased. Then there are Hugging Face's considerable software contributions. If you weren't aware, the popular local AI inference engine llama.cpp became part of Hugging Face earlier this year. Meanwhile, Hugging Face's Transformers Python library — not to be confused with the model architecture — is a key piece of major inference platforms like vLLM and SGLang, which compete directly with Nvidia's own TRT-LLM offering. The acquisition puts Nvidia in a prime position to prioritize its own hardware and software products. The AI arms dealer wouldn't necessarily need to do something as brash as ending support for competing platforms. It could simply ensure that the models and frameworks it favors are always better documented, and run first on its kit. One method to tip the scales would be to flood Hugging Face with cheap Nvidia-based compute. Hugging Face has offered compute resources in the form of its inference endpoints, providers, and Spaces for several years now. To supply this compute, Hugging Face worked with numerous providers across a range of hardware, including major cloud providers, inference-as-a-service vendors, and chip designers such as Nvidia, AMD, Cerebras, SambaNova, and Groq. As Hugging Face's parent company, Nvidia would be in the position to subsidize compute through its various partners and economically incentivize developers to build for its hardware first. The GPU slinger would have to be subtle. Tying a certain percentage of a neocloud's compute allocations to low-cost educational and development resources would be one way to do it without raising too many suspicions. No matter how you slice it, nor how much Nvidia insists it's totally not going to thumb the scales, its acquisition of Hugging Face has a high likelihood of harming competition. Hugging Face works better as AI Switzerland than it does as part of the most powerful company in the industry. ®

The Guardian

Latest news, sport, business, comment, analysis and reviews from the Guardian, the world's leading liberal voice

Daniel Kretinsky to become biggest West Ham shareholder in blow to Amanda Staveley

  • Royal Mail owner made agreement with David Sullivan

  • Staveley was pushing to buy stake in the club

Daniel Kretinsky is set to become the biggest shareholder at West Ham after striking an agreement with David Sullivan, dealing a blow to Amanda Staveley’s hopes of buying a stake in the Championship club.

Staveley was a guest at the London Stadium for West Ham’s 4-2 win over Wolves earlier this week and was pushing to buy Vanessa Gold’s 25.1% stake in the club. The British businesswoman, previously a director at Newcastle, had struck a deal with the Gold family in July and was fronting a consortium looking for a way on to the east London club’s board.

Continue reading...

Genetically modified pig kidney keeps man alive for record nine months

Organ kept Tim Andrews off dialysis and acted as a ‘bridge’ while he waited for a human transplant in the US

A genetically modified kidney from a pig called Wilma has kept a 66-year-old man with kidney failure alive for a record nine months and successfully acted as a “bridge” to a human transplant in a world first.

Tim Andrews was seriously ill with end-stage kidney disease caused by type 2 diabetes and had only a 9% chance of receiving a human kidney within five years, against a more than 40% risk of dying or being removed from the waiting list.

Continue reading...

Sadiq Khan agrees to have his texts and emails searched in Palantir legal battle

US tech company has sued London mayor after he prevented it from working with Metropolitan police

Sadiq Khan has agreed for his text messages and emails to be searched as part of a legal battle with Palantir, the high court has been told, after he prevented the US technology company from working with the Metropolitan police.

The London mayor stepped in to block a £50m deal between Palantir and the Met in May, with the artificial intelligence developer suing Khan’s office over the decision.

Continue reading...

Volkswagen confirms it will cut 100,000 jobs by 2030

Under-pressure carmaker will shed 15% of workforce and halve its product line in sector’s biggest ever restructure

The car company Volkswagen has confirmed that it will shed 100,000 jobs by the end of the decade after being hit by US tariffs and fierce competition from Chinese rivals.

The German manufacturer said management and unions had agreed as part of a sweeping cost-cutting plan to cut a further 50,000 positions by 2030, bringing total job losses in the pipeline to 100,000.

Continue reading...

Federal judge blocks Trump’s latest attempt to restrict birthright citizenship

Supreme court ruled in June against president’s earlier order to change citizenship rules for babies born in US

A federal judge has temporarily blocked Donald Trump’s latest attempt to restrict birthright citizenship – the longstanding principle that all babies born in the US are guaranteed American citizenship.

The decision came nearly a month after the president tried for a second time to limit via executive order which people born in the US can become American citizens. The supreme court ruled in late June that Trump’s previous sweeping order denying birthright citizenship to the children of undocumented immigrants or temporary visitors was unconstitutional.

Continue reading...

‘It’s huge’: Nike deal breaks new ground in WSL for London City Lionesses

  • Club owner says deal is largest in women’s game

  • Claimed to exceed some men’s Premier League shirt deals

The Women’s Super League club London City Lionesses’ new front-of-shirt sponsorship deal with Nike is the largest in the women’s game, according to their American owner, Michele Kang.

The length and value of the deal has not been disclosed but sources close to the WSL club are hailing it as a gamechanging moment and Kang claims that it even exceeds some men’s Premier League clubs’ shirt deals.

Continue reading...

Naomi Osaka grinds it out to see off Siniakova’s stubborn US Open challenge

  • Former champion into third round with 6-2, 5-7, 6-1 win

  • ‘I hope to continue making good memories here’

For a second consecutive match at the US Open, Naomi Osaka appeared to be cruising. Everything was in hand as the two-time champion established a set and a break lead in her round two match against Katerina Siniakova, an uneventful return to the third round beckoning.

Instead of carrying that great momentum forward until the end, though, Osaka was overcome by tension in the final stages of the second set and she soon found herself in a decider. It took a significant amount of resilience and restraint for her to reset and find a way through to the third round with a comprehensive final set to win 6-2, 5-7, 6-1.

Continue reading...

JD Vance reacts to airstrike that killed four at Iran wedding: ‘Things happen’

US vice-president says he’s ‘extremely skeptical’ of Iran reports when asked whether US was responsible for deaths

The US vice-president, JD Vance, said “sometimes things happen” when asked on Thursday whether an American strike was responsible for killing four people, including a child, at a wedding celebration in southern Iran this week.

The strike hit Kuhestak, a town in Sirik county on the strait of Hormuz. The toll was reported not by the Iranian government but by the Iranian Red Crescent, a member of the International Red Cross and Red Crescent Movement that operates independently of Tehran and is generally treated by international monitors as a credible source on casualty figures.

Continue reading...

The Blooming Dahlia

BertvB posted a photo:

The Blooming Dahlia

Tsuruhashi, Osaka, Japan 鶴橋、大阪

Mr Mikage (ミスター御影) posted a photo:

Tsuruhashi, Osaka, Japan 鶴橋、大阪

Slashdot

News for nerds, stuff that matters

'Welcome to the AGI Era,' OpenAI Says As GPT-6 Astra Debuts

OpenAI has released GPT-6 Astra, a new model that president Greg Brockman called a "generational leap" and suggested may represent the arrival of AGI. "I think it might be about this model," Brockman said in a briefing with reporters. He ended the briefing by saying: "Welcome to the AGI era." Axios reports: Astra pushes AI agents closer to doing complex professional work on their own -- while also raising questions about how safely they can be deployed. OpenAI said that Astra was built on its largest-ever training run, using more than 100,000 GPUs at its Stargate site in Texas.

The company also said this is its first model to use other models in a significant role in supervising Astra's training.
GPT-6 Astra will first be available to a limited set of organizations in OpenAI's Daybreak Access program and will be available "in the coming days" for ChatGPT Plus, Pro, Business and Enterprise customers and API developers.

OpenAI said earlier this week that Astra would be released soon, but that its most powerful cybersecurity capabilities would remain limited to a small group of trusted testers. Astra is the first model OpenAI has designated as reaching its "critical" cybersecurity threshold under its preparedness framework -- meaning it can potentially find and exploit previously unknown vulnerabilities across well-protected systems without step-by-step human guidance. OpenAI previously slowed Astra's release to add safety testing after determining that its cyber capabilities could reach the critical threshold. Further reading: OpenAI's New Reasoning Technique Alarms AI Safety Experts

Read more of this story at Slashdot.

Nvidia Launches Free Tool That Links Idle Computers Into a Personal AI Data Center

Nvidia has launched PAIR, a free open-source tool that links compatible computers on a home network so they can pool idle processing power for local AI inference and agentic workloads. "While the compatible devices are mostly Nvidia GeForce GPUs (PAIR works with RTX 20-series cards and newer, as well as RTX Pro GPUs and DGX Spark systems), Apple's M4 chips or newer will also work," reports The Verge. From the report: The key thing here is that PAIR uses your in-home systems when they're idle to avoid interfering with other tasks. And this disaggregated system of computers can work in parallel to chew through lots of processing requests -- which should be helpful for an agentic workflow that breaks complex tasks into smaller jobs. This should prevent large bottlenecks on a single GPU, and Nvidia says PAIR can adapt as devices join or leave the network -- including if a user does something like start playing a game on their desktop PC.

[...] Nvidia says PAIR is secured by pairing all devices through a six digit code and then securing the channel via mTLS (Mutual Transport Layer Security), to create an encrypted communication line that's trusted in both directions between computers. The Nvidia PAIR beta is available today, with support for Windows, Linux, and macOS.

Read more of this story at Slashdot.

Wel.nl

Minder lezen, Meer weten.

Trump vraagt hooggerechtshof om beperken poststemmen toe te staan

WASHINGTON (ANP/RTR) - De Amerikaanse president Donald Trump heeft het hooggerechtshof om toestemming gevraagd voor het beperken van het poststemmen. Een lagere rechter blokkeerde zijn plannen, maar Trump wil dat het hoogste hof die beperking toch toestaat, voor de tussentijdse verkiezingen van november.

Trump stapte al eerder naar het hooggerechtshof over het stemmen per post en kreeg toen goedkeuring om zijn nieuwe regels door te zetten. Maar omdat het hof nog de mogelijkheid openliet voor nieuwe rechtszaken, kon een lagere rechter de plannen opnieuw blokkeren. Nu moet het hooggerechtshof zich daarom opnieuw uitspreken.

De president probeert al jaren om het stemmen per post te stoppen. Volgens hem zijn die briefstemmen fraudegevoelig. Trump beweert dat hij daardoor de verkiezingen van 2020 heeft verloren van Joe Biden, maar heeft daar nooit bewijs voor geleverd. Amerikaanse media hebben zijn claims over die fraude weerlegd.


Behance Featured Projects

The latest projects featured on the Behance

Klar Reward Agency brand illustrations


Nagasaki Peace Bell

stan.jernigan has added a photo to the pool:

Nagasaki Peace Bell

I took this photo of the “Nagasaki Peace Bell” with my iPhone 17 Pro Max during a visit to Nagasaki’s Peace Memorial Park in Japan. The bell is topped with four childlike figures that appear to float or fly in midair. The figures represent the women and mobilized schoolchildren who were forced to work in the city’s munition factories during World War II. Every month on the 9th at exactly 11:02 AM (the precise time the bomb detonated), survivors and visitors ring the bell to pray for permanent world peace...

Part of the Urakami Cathedral

stan.jernigan has added a photo to the pool:

Part of the Urakami Cathedral

I took this photo of the “Urakami Cathedral Post” with my iPhone 17 Pro Max during a visit to Nagasaki’s Peace Memorial Park in Japan. The original red-brick Urakami Cathedral—once the largest Christian church in East Asia—stood just about 500 meters from the hypocenter...

MetaFilter

The past 24 hours of MetaFilter

Decade of work helps remote island's native wildlife "return to 1616."

Decade of work helps remote island's native wildlife "return to 1616." Researchers were unafraid to get their hands dirty in an ambitious bid to restore Western Australia's largest island to its pre-colonial state.

The molecular study of animal faeces has helped return native bettongs, or "rat kangaroos", to Dirk Hartog Island. Researchers had to determine if the island could support the marsupials' fungal diet, after years of damage by invasive species. Boodies, another variety of bettong, are set to be reintroduced in spring and are among the final species to participate in the major ecological project.

VK: Voorpagina

Volkskrant.nl biedt het laatste nieuws, opinie en achtergronden

Het stikstofbeleid gaat de goede kant op, maar er zal nog een schepje bovenop moeten