The Linux Foundation's director of IT infrastructure says they now spend more CPU cycles "rendering commits for scrapers than we spend on all other kinds of legitimate access." At any one time, across 5 geo-distributed nodes, there are 14 CPU cores doing nothing but rendering git commits as html.... [W]hen a source is guaranteed to be LLM-free, like the entire history of kernel commits, it's worth its weight in gold as a source of training data... At the time of writing, linux.git is about 1.48 million commits. Oh, and we have about 922 forks of it on git.kernel.org — but don't worry, it's actually extremely efficient on the backend, since it's mostly the same objects in every fork. Unless, of course, you're a scraper, in which case you have, oh, several BILLION valid URLs you can scrape, only to get 922 duplicates of the same 1.48 million commits — which is exactly what the scrapers are doing. But wait, it's not just commits itself. You can also ask for patches, plain renders, diffs between arbitrary commits — cgit is happy to let you, which was perfect for the times when the Internet was for humans or crawlers who obeyed robots.txt, and is AWFUL right about now, because we can generate 1.2 METRIC BAJILLION valid URLs just for a single fork of linux.git. Initially, this was the solution — look through the logs, find out which IPs are obvious scraper bots, and fail2ban them. At first, this was easy, because the bots helpfully told you who they were via their user-agent. Then, they wised up and started pretending that they were random vanilla browsers. So, we started banning them by IP — after all, it's easy to figure out that an IP that is trying to grab every possible commit in a 8-year-old abandoned fork of linux is not really some lone Chrome on Windows user who is just furiously clicking every link that comes across their screen. The bots then started fanning out to entire subnets, but this was still meh, because obviously an IP coming from Google Compute is just pretending to be a Firefox user... And... that's when things turned really, really ugly. Suddenly, the crawlers were coming from millions of random residential or mobile IPs, all pretending to be random modern browsers. An IP like that would make 4-5 requests and then never show up in the logs again... They descended like swarms of locust, hit hard and fast until the system fell over and then moved on to the next target until you recovered. Then, they returned. Rinse. Repeat. They still do that — welcome to the wonderful world of "proxy SDK monetization." It's big business, and your TV is probably doing it... Today, git.kernel.org receives about 6M daily requests demanding to see random commits. Of these, 66% are still immediately batted away with the Anubis challenge, but 33% are now solving the math and getting through to the main site — because apparently what we have to offer is worth spending a ton of cycles to calculate the Anubis challenge... With a bunch of generous assumptions, legitimate requests are only about 2% of git.kernel.org traffic — everything else are scrapers... [W]e're turning off features to reduce the number of crawlable URLs and to gate off actions that are expensive for us to run. Expect to lose some functionality, at least when accessing our resources anonymously. Trust me, we hate it just as much as you, but at this point it's a necessity... [W]e promise to still offer all of our data for download to anyone who asks. You just may have to jump through more hoops to get it. Sorry.

Read more of this story at Slashdot.

Six ans après son lancement au salon de Francfort , la Volkswagen ID.3 change une nouvelle fois de nom. Après un premier restylage discret en 2023, la compacte électrique de Wolfsburg revient sous l’appellation ID.3 Neo. Un baptême qui n’a rien d’anodin : Volkswagen a repris sa copie en profondeur, sur le design comme sur l’ergonomie intérieure, deux points qui avaient valu à la ID.3 originelle une volée de critiques à sa sortie. Nous avons pris le volant de la nouvelle venue pour vérifier si le bon sens l’a enfin emporté.

Un style plus mature, moins clivant

La face avant est la première à changer. Elle adopte le nouveau langage stylistique « Pure Positive » signé Andreas Mindt, le directeur du design de la marque. Un bandeau lumineux continu relie désormais les deux blocs optiques sur toute la largeur de la calandre, avec un logo Volkswagen illuminé sur les finitions hautes. Une signature déjà entrevue sur l’ID. Polo, et que l’on retrouvera prochainement sur l’ID. Cross.

À l’arrière, exit le hayon en plastique noir contrasté qui donnait à l’ID.3 ses airs de jouet bicolore. Il est désormais peint dans la couleur de la carrosserie, tout comme le toit. Le résultat est plus homogène, et la silhouette y gagne en sobriété sans devenir banale. On note même un petit air de famille avec la Golf 8 dans le traitement du becquet arrière.

L’ergonomie enfin corrigée

C’est peut-être la meilleure nouvelle de cette ID.3 Neo. Volkswagen a visiblement écouté ses clients, et surtout ses détracteurs. Les commandes tactiles au volant, unanimement détestées, cèdent enfin la place à de vraies touches physiques. Sous l’écran central, plus besoin de deviner où appuyer dans le noir : les réglages essentiels reviennent en dur, avec un sélecteur rotatif dédié au volume sur la console centrale.

À bord, l’écran d’infodivertissement « Innovision » grimpe à 12,9 pouces, épaulé par un combiné d’instrumentation numérique de 10,25 pouces. Clin d’œil appuyé à l’histoire de la marque, l’affichage peut adopter un thème rétro qui rappelle la première Golf de 1974, un principe déjà vu sur le cockpit de l’ID. Polo. Volkswagen ajoute également un assistant vocal intelligent dopé à ChatGPT, une clé numérique via smartphone, et des services connectés offerts pendant dix ans. En option, un affichage tête haute à réalité augmentée vient projeter les informations de conduite directement dans le champ de vision.

Les matériaux progressent aussi : de nouveaux textiles et des surfaces moussées habillent une grande partie de la planche de bord, dans un esprit plus proche d’un T-Roc que de l’ancienne ID.3. La qualité perçue n’atteint pas encore le niveau des meilleures compactes premium, mais l’écart se resserre nettement.

Trois batteries, trois niveaux de puissance

Sous le plancher, l’ID.3 Neo reprend l’architecture MEB et sa transmission propulsion (moteur à l’arrière). L’offre s’articule autour de trois capacités de batterie NMC : 50, 58 et 79 kWh, associées respectivement à 170, 190 et 231 chevaux. Les autonomies WLTP s’échelonnent selon les configurations, la version la plus endurante dépassant les 600 km.

Côté nouveautés électriques, l’ID.3 Neo inaugure le préconditionnement manuel de la batterie et la consultation directe de son état de santé (SOH), deux fonctions réclamées de longue date par les propriétaires d’électriques. La recharge V2L, qui permet de brancher un appareil électrique sur la voiture, est disponible en option.

Sur la route : le sérieux avant le spectacle

Au volant, l’ID.3 Neo ne cherche pas à surprendre, elle cherche à rassurer. La direction, calibrée sans excès de vivacité, et l’amortissement, plutôt souple sans être mou, dessinent le profil d’une compacte pensée pour avaler des kilomètres sans fatiguer son conducteur. La propulsion (moteur à l’arrière) donne une motricité saine à l’accélération, y compris sur chaussée humide, et l’insonorisation a clairement progressé : à vitesse stabilisée sur autoroute, les bruits aérodynamiques et de roulement restent contenus, un point sur lequel l’ancienne ID.3 pêchait un peu.

Sur un parcours mixte, alternant routes secondaires et voies rapides, notre modèle d’essai en finition Style, associée à la batterie intermédiaire de 58 kWh et à des jantes de 19 pouces, s’est stabilisé autour de 14 kWh/100 km. Une consommation cohérente avec la valeur WLTP annoncée par Volkswagen, et qui autorise un rayon d’action réel avoisinant les 380 à 400 km dans des conditions de conduite normales, chauffage ou climatisation inclus. Sur autoroute à vitesse soutenue, la consommation grimpe logiquement, mais l’ID.3 Neo reste dans la moyenne haute de sa catégorie plutôt que de décrocher, comme cela pouvait arriver sur les premières versions du modèle produites à Zwickau et Dresde Volkswagen avait d’ailleurs renoncé à ouvrir une ligne supplémentaire à Wolfsburg faute de demande suffisante. Au chapitre de la recharge rapide, rien ne change fondamentalement par rapport à la devancière, mais le préconditionnement manuel évoqué plus haut permet désormais d’optimiser la vitesse de charge avant une pause sur autoroute, un détail qui change vraiment le quotidien en électrique.

Commandes ouvertes, livraisons à l’été

La Volkswagen ID.3 Neo se décline en trois finitions : Trend, Life et Style. En France, les commandes sont ouvertes depuis le 22 avril 2026, pour des premières livraisons attendues à partir de juillet. Le tarif d’entrée s’établit autour de 40 500 € en finition Life avec la batterie de 58 kWh et les 190 chevaux, contre environ 42 700 € pour la finition Style équivalente. La version Pro S à 79 kWh démarre quant à elle sous la barre des 40 500 €, et reste éligible au bonus écologique sous conditions de prix.

Volkswagen ne réinvente pas l’ID.3, il la corrige. Et c’est bien ce qui manquait depuis 2020 : une compacte électrique qui assume enfin ses ambitions, sans laisser l’ergonomie ou la finition ternir le tableau. La marque tient là sa meilleure ID.3 à ce jour, en attendant la relève annoncée pour 2028 sous la forme d’une Golf électrique.

L’article [VIDÉO] Essai Volkswagen ID3.Neo est apparu en premier sur Le Blog Auto.

Google's AI assistant Gemini crawls your Google apps for "more personalized suggestions," in a feature they call Personal Intelligence. Now when given permission, an Android/iOS app called Dreambeans "uses Personal Intelligence to connect information from Google apps like Gmail, Calendar, Photos, YouTube and Search History, to curate a finite collection of daily stories designed to spark new ideas." In a slick new video Google says the app will "surface what matters to you." The blog 9to5Google notes that after previously requiring a subscription, Dreambeans "is now available for free Google Accounts in the US to test, calling it "very much like a hyper-personalized Google Now or Discover feed." Instead of using stock imagery, each Dreambeans story is accompanied by Nano Banana 2-generated artwork that uses an illustrated style. For example, if Dreambeans recommends a hike, you might see yourself and a group of friends at that location... There's also an "Ask anything" chat experience to get more information. You can leave feedback by tapping the heart or downvoting, while the chat interface allows you to provide more details. "Tune your feed" lets you tell Dreambeans what you "want more stories about" and what it should never show. Or, as TechCrunch senior writer Lucas Ropek puts it, you get a curated list of AI-illustrated "stories" that "will turn your life into a cartoon" [Gozde Oznur, the product lead behind the new app, told TechCrunch the app] has also been built as a doomscrolling antidote, in that it only provides users with a limited number of stories per day — typically 10 to 14. The idea is to get a few inspirational ideas and then go out and live your life, she said. A lot of companies are currently trying to court the user that is sick of phone addiction. I recently reviewed a startup, Bond, which also uses AI to auto-generate lifestyle suggestions for the user. What about privacy protections? According to Oznur, they are pretty solid. The only person with access to the app's stories is the user, she said. Users can also delete their data whenever they want, and can choose which Google services they want to connect to the tool.

Read more of this story at Slashdot.

Axios reports: Some of the world's largest music publishers filed a blockbuster lawsuit against Anthropic late Friday night, alleging "one of the largest and most blatant ongoing thefts of intellectual property in history." The suit is the opening salvo in what is now likely to be a yearslong fight over music, AI, and how intellectual property is protected in a new era of technology... The Sony/Warner lawsuit is notable because it's broad. It alleges Anthropic unlawfully trained its models off "tens of thousands" of music publishers' copyrighted compositions, whereas other lawsuits have focused on a narrower set of works. BMG's lawsuit against Anthropic, for example, claims infringement against 493 compositions. What they're saying: "Defendants Anthropic and its founders Dario Amodei and Benjamin Mann have conducted a brazen campaign of illegally torrenting, scraping, and downloading copyrighted works on a massive scale in order to develop, operate, and reap enormous profits from Anthropic's 'Claude' series of artificial intelligence ('AI') models," the plaintiffs write in the 48-page complaint. They also claim that Anthropic pirated thousands of copyrighted musical works... More details from The Verge: The companies are asking for up to $150,000 per work, plus up to $25,000 for each instance when identifiable copyright data was stripped. In total, the damages could amount to several billion dollars if a court finds in Sony and Warner's favor and awards the maximum amount... The suit alleges that Mann used BitTorrent to download over five million pirated books, and that employees downloaded at least two million more from Pirate Library Mirror. It also claims that Anthropic scraped lyrics from sites like MusixMatch and LyricFind, which paid to license content from the labels.

Read more of this story at Slashdot.

The Washington Post looks "a growing number of Americans who have had their private conversations with AI chatbots made public in court," calling the chatbots "custodians of a fast-growing trove of intimate data that can be sought by law enforcement, or companies or other opponents in civil litigation." A Washington Post review of public records and local news stories found that chatbot logs were cited in 12 court cases over the past two years. It's hard to know how often chatbot material is drawn into investigations and legal proceedings more broadly, because police, and parties in civil cases, don't have to present in court all the evidence they obtain... In some cases, AI companies have reported people on their platforms to the FBI, leading to prosecutions and at least one conviction... OpenAI has said the company uses software to scan conversations for signs of dangerous behavior and flag them to human reviewers. If a reviewer determines a chat indicates "an imminent and credible risk of harm to others," they report the user to law enforcement, the company said in an April blog post. A spokesperson for OpenAI declined to disclose how many times the company had reported users to police.... Data released by OpenAI shows requests by government agencies and law enforcement for user data growing rapidly. In the second half of 2025, the company disclosed data from more than 80 accounts, an increase of more than four times as many over the same period the year before... When police ask someone to open their phone during questioning or after an arrest, they often do so, allowing officers to scroll through the apps they use, [according to Michael Price, the litigation director for the Fourth Amendment Center at the National Association of Criminal Defense Lawyers]... In general, the Fourth Amendment, which protects people against unreasonable searches and seizures of their property, provides the right to refuse a search of the contents of their phone unless police have a warrant. "The short answer is that most people consent," Price said. "They shouldn't, but they do..." OpenAI CEO Sam Altman has argued that conversations with chatbots should receive special legal protections, similar to conversations between clients and their attorneys. Special legal protections aren't recognized by courts today, the article poitns out. A financial executive who became the subject of a federal criminal investigation for fraud asked Claude about potential defense strategies, then asked a judge to disallow the conversations from the prosecution's case, according to the article. "But the judge said that protections for conversations between clients and their attorneys did not apply: Claude is not a lawyer, and the executive's attorneys did not ask him to talk to the chatbot." The article includes quotes from three ChatGPT conversations that later turned up in court proceedings: "I deleted a email a year ago from my yahoo account. Can yahoo still retrieve the email?" "How f----d am i bro. What if i smahsed the s--- oitta mutlipls cars." "My dad Said that I'm will get a settlement worth of 1million dollar." "As AI becomes a bigger part of everyday life, its use as evidence and in trials is likely to become more prevalent," the article points out — especially with the growing popularity of agents that take actions for users.

Read more of this story at Slashdot.

Phoronix reports on Greg Kroah-Hartman's recent slide from his upcoming talk in Paris at Kernel Recipes 2026 (September 21 to 23): With the proliferation of AI/LLM models analyzing the Linux kernel's vast codebase, there has been a surge in the number of CVEs per kernel release. After typically being around 500 CVEs fixed per release, we are now approaching 2,000 CVEs fixed per release and perhaps will break that threshold for Linux 7.3... Fortunately, most often they end up being lower priority vulnerabilities and often within old/obscure driver code, so the impact is often minimal.

Read more of this story at Slashdot.

An anonymous reader quotes a report from TechCrunch: Over a hundred tech companies -- including OpenAI, Anthropic, Google, and Microsoft -- have signed an open letter urging both the private and public sectors to work together to defend themselves from AI-related cyber threats. The letter -- which was also signed by prominent cyber firms like CrowdStrike, Okta, and Fortinet, as well as prominent financial institutions and internet infrastructure firms -- calls for the adoption of new forms of cyber defense, while also encouraging governments at the "local, national, and international levels" to collaborate on security. "In the coming months, AI-enabled cyber attacks will become far more widespread and sophisticated as models around the world become increasingly capable," the letter states. "The companies and public services our communities depend on -- from hospitals to water treatment plants to the infrastructure that powers the internet -- are at risk." [...] The letter further suggests the mobilization of a "collective response," one in which "new partnerships" are formed "to raise security standards and find new solutions to emerging cyber threats." Several of the AI companies that have signed the letter are still actively developing ever more advanced AI models, highlighting their conflicted position. At the same time, they are also offering programs to use frontier AI models for defensive purposes, including OpenAI's Daybreak program, Anthropic's Mythos, and Microsoft's new cyber platform Perception.

Read more of this story at Slashdot.

Plus d'une centaine d'entreprises tech, dont les principaux laboratoires d'IA, viennent de publier une lettre ouverte appelant à une mobilisation mondiale contre les cyberattaques assistées par IA. Une démarche qui tient autant de l'alerte que de l'auto-diagnostic.

An anonymous reader quotes a report fro SFGATE: From "rizz" to "meat proxy," words and phrases have fallen in and out of favor for centuries. Culture, politics and technological development have long influenced these trends, and now, new research shows that ChatGPT is beginning to influence how humans communicate. The findings from a team at the Pew Research Center show how, in a short time since OpenAI released ChatGPT in November 2022 and other chatbots followed, the use of certain words and phrases has spiked on the web. The researchers, led by data scientist Samuel Bestvater, studied nearly half a million English language webpage texts in snapshots over the past five years. Not only has AI-written text become more common, but it could also be influencing how people write, such as trying to avoid using the "tells" that have become associated with AI-generated text. Bestvater and his team detected double the number of em dashes, double the frequency of particular words -- such as "delve," "testament" and "interplay" -- and nearly triple the use of a phrasing known as negative parallelisms ("it's not X, it's Y"), as well as a 63% increase in Oxford commas. But while AI labs continually try to tweak their models to generate more humanlike text, the people they're mimicking aren't staying the same. "Human writers might also be changing the way that they write in response to AI. I've seen some anecdotal evidence of people saying, 'Now I avoid em dashes,'" Bestvater said. "If human writers are adjusting their choices of words and phrases and punctuation to try to differentiate their writing from AI-generated text, then you would see a pattern that looks sort of like that in the chart." "You can see changes in people's speech. People are tending to more commonly use these low-frequency words in their speech because they're spending a lot of time listening to or reading AI-generated content," said Roger Kreuz, a cognitive psychologist at the University of Memphis who recently wrote about how difficult it is for people to identify bot-written text.

Read more of this story at Slashdot.

Google is imposing new memory-use limits on Android apps as the AI data center boom contributes to a broader memory chip shortage that could leave lower-cost phones with less RAM. Developers will have until February 2027 to meet new thresholds for memory and bitmap usage. To aid developers, Google is adding tools to flag apps that exceed the limits and help prevent slowdowns and crashes. TechCrunch reports: The company explains that the mobile industry is now facing "significant hardware supply constraints that are altering device memory availability," which can then, in turn, affect the consumer's experience with their devices. To address this, Google is now establishing new performance thresholds across several areas, like dynamic memory usage and bitmap usage. In addition, Google is adding code optimization requirements designed to prevent things like app slowdowns and crashes related to performance.

Read more of this story at Slashdot.

Wired reports that OpenAI is testing a new "Persistent mode" for Codex that would let the agent keep working until explicitly "put to sleep," proactively creating follow-up tasks for itself across sessions and using prior interactions and knowledge of the user to decide what to do next. Code alluding to the feature was spotted in the product's code base, though it has not been rolled out or officially announced yet. From the report: Persistent mode appears in Codex's "reasoning effort" menu, in which users can select the level of computing power, tokens, and time they want to allow for an AI model to "think" before answering a prompt. It seems to be one of OpenAI's most computationally intensive settings. When users have selected Persistent mode, OpenAI's code base reads that Codex will "continue working until put to sleep." That's a stark contrast to currently available modes, which will stop working on a task after a few minutes or hours, even if it's not complete. In another file in the code base, OpenAI describes a feature within Persistent mode called "proactivity." This appears to be a type of system prompt for agents in Persistent mode, which are told that their work is not done when they finish answering a user's request. Instead, the agent is instructed to proactively create follow-up tasks for itself. The agent is capable of working on those tasks across sessions and using past user interactions and "knowledge of the user" to decide what to work on. It also has a tool to message the user without being asked but is told to send these sparingly. The instructions also set limits for the agent, according to the file. The agent is told that Persistent mode does not expand what it is allowed to do and that altering anything outside the user's own system requires the user's approval first -- seemingly intended to limit how dangerous a persistent AI agent could be. The file sits in the shared core of Codex rather than in the code specific to the terminal, seeming to suggest the proactivity feature is intended for more than the command line tool.

Read more of this story at Slashdot.

An anonymous reader quotes a report from Ars Technica: Documentation files on more than 100 websites are referencing potentially dangerous executable content that gets installed automatically when visited by many AI agents [including Claude, OpenAI's Codex, and Nous Research's Hermes]. A few dozen companies, some of them Fortune 500s, are among those that executed proof-of-concept code. At least one misconfigured site is directing visitors, human or AI, to live malware. The potentially dangerous content is in llms.txt and llms-full.txt files, an emerging convention websites employ to provide machine-readable summaries of the site's content and its high-level structure. These files are the AI equivalent of the robots.txt standard that instructs search engines how to index the site's content. Google Lighthouse, a tool for helping web developers, has more here. Correctly configured llms.txt and llms-full.txt files for Cloudflare are here and here. "The trust model is broken," Alon Hertz, one of the researchers, wrote in an interview. "Agents treat vendor docs as ground truth and don't question them -- and neither do the humans supervising them. Agentic AI usage is exploding, and agents are spreading across every layer -- SaaS, cloud, endpoint. As they multiply, so does the supply-chain surface, and today's guards don't cover it." "An agent doesn't distinguish between a page and a command," the researchers wrote Thursday. "Everything it reads is input, and every input is a potential instruction. Which means the entire corpus of published data that agents are now wired to consume has silently become an execution surface -- and almost none of it carries the integrity guarantees we apply to actual code."

Read more of this story at Slashdot.

Hyundai annonce l'arrivée en France de la Ioniq 6 N, la déclinaison hautes performances de sa berline électrique. Avec 650 chevaux, un 0 à 100 km/h en 3,2 secondes et une panoplie de technologies héritées du sport automobile, la marque coréenne entend transposer l'expérience de sa très remarquée Ioniq 5 N dans une carrosserie plus aérodynamique.

Le 26 août 2026, OpenAI a publié le rapport technique attendu depuis l'incident Hugging Face. L'entreprise américaine y détaille comment des agents IA ont contourné leurs propres restrictions avant de s'attaquer à Hugging Face.

An anonymous reader quotes a report from CNN: Microsoft co-founder Bill Gates argued on Wednesday that artificial intelligence needs significant limits or else the harm to humans will outweigh any potential good. "AI will either be the greatest equalizer ever invented, or the worst source of injustice," he said in a 6,000-word essay, entitled "The turbulent AI era is here. The choices we make now are critical." [...] "Even under the best circumstances, the transition to this new AI era will be one of the most turbulent times in human history," he said, adding that "I don't see evidence that leaders, experts, and communities are confronting the challenges adequately. There is no plan to ease the entry into the AI era." AI can provide great benefits in field like agriculture and medicine, he argued, such as breakthroughs in preventing and treating diseases. But he also noted that the transition carries big risks, like widespread unemployment, harming the educational and social development of children, or making it easier for criminals and bad actors -- or AI itself -- to cause deliberate harm. "As the models become more powerful, they could begin to act against our interests and we could lose control," he warned. These significant risks must be controlled quickly, Gates said. "If someone had a credible plan for slowing down AI advances globally, I would likely support it," he wrote. "However, I don't think that's going to happen. The geopolitical and economic incentives are pushing too hard to go full speed ahead." In an interview with CNN's Anderson Cooper that will air on Wednesday evening, Gates said AI models have gotten dramatically more powerful at a rate faster than he expected. "They're now capable of causing cyberattack risk, bioterrorism risk, psychosocial risk," he said. "I have to say I'm kind of shocked that the exact criteria that we review these models with, and the actions we take to minimize the harms, are really completely missing." In his essay, Gates proposed taxing AI or robots the way you would pay a human employee's payroll tax in order to slow the shift away from using human labor. He also wants to set aside some work to be done by humans only. "My message to leaders is: You have a chance to act now, before unemployment rises sharply, communities are hurting, and public trust has eroded," he said. "You can make sure AI benefits everyone. And you can work with other governments to meet this national and global challenge."

Read more of this story at Slashdot.

Quelques semaines après l'avoir levée, OpenAI réactive sa limite des cinq heures sur Codex et ChatGPT Work pour les abonnés Plus, qui doivent de nouveau composer avec deux plafonds au lieu d'un.

Perplexity and Nvidia have launched "Portable Computer," a local-first version of Perplexity's agent platform that runs AI models, files, tools, and workflows directly on Nvidia-powered Linux hardware. Local tasks incur no token charges and keep data on-device by default, with users asked for permission before the system escalates a step to a cloud model. VentureBeat reports: For Nvidia, which has spent the past two years selling the world on trillion-dollar AI data centers, the announcement signals something subtler but strategically important: the chipmaker believes local AI has crossed a threshold from hobbyist curiosity to practical tool -- and it wants to sell the hardware that runs it. "Local AI reached an inflection point," said Nader, Nvidia's director of developer technology, who focuses on developer tooling and open source. "For the longest time, it was hobbyists and enthusiasts, and they were running these quantized models that were quantized down to be super tiny... And while that's cool, it's not super practical. But all that changed with a lot of these new open source models that have come out that are super useful." [...] Portable Computer arrives today for Pro, Max, Enterprise Pro, and Enterprise Max subscribers on Linux, with Windows support following in September. Any RTX GPU with at least 24GB of VRAM -- roughly a GeForce RTX 3090 or newer -- clears the bar, a threshold Nate called "sort of the floor where we really want to make sure that we can deliver a great experience, but balance that with making it broadly available."

Read more of this story at Slashdot.

An anonymous reader quotes a report from Ars Technica: For years, AI industry watchers of all stripes have been warning of a coming jobs apocalypse driven by ultra-intelligent AI systems that will be able to replicate most human tasks more cheaply. Now, newly updated research from Stanford University economists suggests AI seems to be causing significant entry-level job losses for younger workers in some fields, even as older workers appear largely unaffected so far. The August 2026 edition of "Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence" updates and revises a paper of the same name published last year with fresh data and refined statistics. In that update, the Stanford researchers find the employment trends they identified for entry-level workers last year are persisting and expanding. Specifically, employment levels for workers ages 22 to 25 in the most "AI-exposed" occupations are now 19 percent below those of their peers in fields less exposed to AI disruption. Last year, that gap measured just 13 percent. [...] Digging deeper into the data, the researchers found that this phenomenon is mainly manifesting itself through lower hiring rates for entry-level workers in AI-impacted fields, rather than increased firings or employees quitting. They also found that the labor market effects among this age group were mostly seen in lower overall employment, rather than reduced pay rates. But not all jobs that show potential for AI "disruption" are created equal, the researchers found. In its Economic Index, Anthropic differentiates between queries related to tasks that are "automative" (i.e., fully replacing work previously done by a human) or "augmentative" (i.e., helping human workers be more effective at tasks they are still needed for). By this measure, jobs like "accountants and auditors" and "receptionists and information clerks" were among those judged most susceptible to AI automation, while jobs like "chief executive" and "registered nurse" were among those using AI augmentation most often. Unsurprisingly, jobs where AI automation is prevalent are the ones showing the worst relative employment levels for entry-level workers these days. "The findings are consistent with automation-oriented uses of AI substituting for labor while complementary uses are associated with flat or rising employment," the researchers write. Interestingly, the researchers found that the impact is strongest in entry-level jobs built around "codified" knowledge, which is the formal, documented skills that AI can more easily replicate. Meanwhile, experienced workers in roles relying on tacit, practice-based knowledge have seen stronger employment growth.

Read more of this story at Slashdot.

Quelques semaines après l'avoir levée, OpenAI réactive sa limite des cinq heures sur Codex et ChatGPT Work pour les abonnés Plus, qui doivent de nouveau composer avec deux plafonds au lieu d'un.

The SEC is investigating the near-collapse of AI-focused hedge fund Situational Awareness, sending subpoenas to major Wall Street banks for details about the fund's trades, borrowing, and communications with lenders. The fund, founded by former OpenAI researcher Leopold Aschenbrenner, managed over $30 billion and borrowed tens of billions more before leveraged bets unraveled, forcing it to sell most of its stock portfolio to Citadel at a discount. The New York Times reports: The subpoenas asked for details on the timing of Situational Awareness's trades and for its communications with lenders about the money it was borrowing, also known as "leverage," two of those people said. The subpoenas additionally warned the banks to preserve any information regarding the San Francisco hedge fund. The S.E.C. oversees financial markets with an eye toward protecting small investors, and has brought civil cases regularly against investment firms that produced large losses. Any investigation into Situational Awareness would be at its earliest stages, and it's no guarantee that it would lead to fines or other punishment. The hedge fund has not been accused of wrongdoing. [...] Situational Awareness had a fast rise and an even quicker retreat. Founded just two years ago by Leopold Aschenbrenner, a former researcher at OpenAI, it rode the A.I. boom to soaring investment returns. To achieve those results, however, the fund relied on heavy borrowing, as well as complicated and expensive financial instruments that magnify gains -- and losses. The latter piled up quickly last month when the stock prices of publicly traded, high-flying A.I. companies dipped. At the same time, shares in more traditional technology companies -- which the hedge fund had been betting against -- rose, compounding the problem. Situational Awareness was forced into a fire sale. It wound up selling most of its stock portfolio to a rival, Citadel, at a discount.

Read more of this story at Slashdot.