Anthropic is releasing Claude Fable 5, a Mythos-class AI model for enterprise customers and paid subscribers. The company says broader access is possible thanks to new safeguards that block high-risk requests in areas like cybersecurity and biology. "For us, it's really around what we call 'race to the top,' being able to provide this technology in a valuable fashion, and at the same time providing the right safety guardrails so that it can do asymmetrically more benefits than harm," Dianne Penn, Anthropic's head of product management for research, told CNBC in an interview. CNBC reports: [W]ith the launch of Claude Fable 5, Anthropic is honoring its stated "eventual goal" to deploy Mythos-class models at scale. It's also capitalizing on growing momentum and investor interest in its technology ahead of a potentially massive IPO, which is expected to take place as soon as this year. Anthropic said Claude Fable 5 shows "exceptional performance" across software engineering and knowledge work tasks. On some benchmarks, it scored more than 10% higher than Claude Opus 4.8, another model the company announced late last month, according to a blog post.
Claude Fable 5 represents a "significant jump" in capability, which is why Anthropic had to implement additional guardrails to prevent misuse, Penn said. If a user asks a high-risk question, like how to make ricin, a toxin, for instance, the model will block its response and fall back to Claude Opus 4.8 to deliver a safe answer. "What we wanted to do was to be very intentional about building new types of classifiers and new types of safety guardrails in place for this launch," Penn said. Anthropic also released an updated Mythos model called Claude Mythos 5. "It's the same underlying model as Claude Fable 5, but with the safeguards lifted in some areas," reports CNBC.
The European Commission says Apple's decision not to launch Siri AI in the EU is Apple's alone, arguing that the company sought an exemption from Digital Markets Act interoperability rules instead of building a compliant privacy- and security-preserving solution. Apple, meanwhile, says regulators rejected its proposals and claims the DMA would require giving third-party AI systems overly broad access to users' devices. MacRumors reports: Commission spokesperson Thomas Regnier told reporters in Brussels: "The decision not to roll out Siri AI in the EU is Apple's and Apple's only. Apple was simply unable to develop interoperability solutions that meet essential EU privacy and security standards. Instead of trying to find a suitable compliance solution, Apple simply made a request to the European Commission to be exempted from their interoperability obligations. That's not an option."
Craig Federighi, Apple's senior vice president of Software Engineering, said the company was "deeply disappointed" and cited what it described as regulators' refusal to accept any of Apple's proposals, including a system called Trusted System Agent that would have allowed third-party virtual assistants to safely access the same device capabilities as Siri AI.
The Commission's account tells a different story. Rather than negotiating over Apple's proposed solutions, regulators say Apple simply requested a blanket exemption from its interoperability obligations under the Digital Markets Act, something the Commission says is not an available option. Apple's statement framed the DMA's requirements as demanding that any AI system be given "nearly unlimited access" to a user's device.
NHS England plans to roll out Microsoft Copilot to 505,000 clinicians and support staff after a 30,000-person pilot claimed the AI assistant saved users an average of 43 minutes a day on administrative work. The Register reports: The rollout won't happen overnight. NHS England said that each trust will receive a central allocation of licenses based on headcount, typically starting with around 2,000 Copilot seats, and that more than half a million staff are expected to have access by October 2026. The NHS has no shortage of administrative work to throw at the software. The rollout envisions Copilot helping with discharge paperwork, bed management, rota planning, meeting minutes, board papers, briefings, data analysis, and assorted HR, finance, and procurement tasks.
NHS organizations will also receive access to Copilot Studio, Microsoft's toolkit for building custom AI agents. NHS England said trusts will be able to develop agents for tasks such as handling Freedom of Information requests, processing complaints, reducing helpdesk workloads, and assisting with financial analysis. A governance framework called Agent 365 will oversee the deployment of those systems.
At WWDC 2026, Apple announced a new "Siri AI," describing it as a more conversational, personalized, and systemwide assistant that can understand on-screen context and interact with apps while relying on on-device processing or Private Cloud Compute. The relaunch comes two years after Apple's original Apple Intelligence promises stumbled and "never fully materialized," reports The Verge. MacRumors reports: Siri is now embedded directly in the Dynamic Island, accessible by swiping down from it, pressing the side button, or saying "Hey Siri." A revamped voice engine makes the assistant sound more expressive, with micro-adjustable voice settings available during initial setup.
During Apple's keynote demo, presenters showed Siri handling chained, multi-step requests with apparent ease. In one sequence, a presenter asked about a Suki Waterhouse concert, was told tickets require a lottery entry, and asked Siri to set a reminder when the lottery opens, which it did. In another, the assistant identified a photo's landmark, pulled up navigation to that location, and surfaced photos from a recent family trip, adding a specific image to a shared family album on request.
Another demo showcased Siri's ability to synthesize information across apps. A presenter asked about a dessert he had heard about at an event, and Siri located the relevant details from his Messages history. It then compiled the information into a watch-party menu, drafted a message to his contacts with the menu included, and presented send and edit options. In a further demo, a presenter asked about something his son had shared in a message and followed it up by asking Siri to compose an email on the subject.
A new dedicated Siri app lets users scroll back through prior conversations and kick off new ones, with conversation history synced via iCloud so sessions carry seamlessly between devices. The app is also coming to watchOS. On the Mac, Siri is now also integrated into Spotlight and available via right-click context menus on any file or window. On visionOS, Siri AI gains a 3D visualization that users can place anywhere in their space.
OpenAI a déployé deux nouvelles options de sécurité pour les comptes ChatGPT : un mode de verrouillage pour limiter les risques d'exfiltration liés aux injections de prompts, et une fonctionnalité de gestion des sessions actives ouverte à tous.
Elle n'a jamais vendu la moindre action, mais le 12 juin, elle pourrait pourtant devenir la 8ᵉ entreprise la plus cotée de la planète : le cas SpaceX est vertigineux et inquiétant pour certains analystes, voici pourquoi.
Dans un article paru le 7 juin 2026, le quotidien britannique The Guardian décrit comment des escrocs sont parvenus à faire remonter des boutiques clonées dans les réponses de ChatGPT. Une technique qui a exploité à la fois les failles des LLM et la disparition d'une marque emblématique.
La 5e édition des eTrophées de l’AMAM (Association des Médias Auto et Moto) s’est tenue en région parisienne les 3 et 4 juin 2026. Il s’agissait de distinguer les meilleures voitures électrifiées lancées en France en 2025. Leblogauto.com était représenté au sein du jury comme l’an passé. Retour sur cet événement qui offre une vision intéressante de notre marché.
28 voitures à l’essai, 35 journalistes au volant
Pendant deux jours, l’association et ses équipes ont rassemblé en un seul lieu 28 voitures, un nouveau record battant les 26 modèles de l’édition précédente, hybrides, PHEV ou 100 % électriques, réparties dans plusieurs catégories. Pour les départager, 35 journalistes issus de médias français, ainsi que, pour la première fois, une consœur et un confrère exerçant respectivement en Suisse et en Belgique. Nous avons pris le volant des véhicules sur un parcours mêlant tous types de routes : nationales, départementales, voies rapides, avec des incursions en milieu urbain et périurbain.
En passant d’un modèle à l’autre, il est ainsi possible de comparer immédiatement les différences entre chacun d’eux, ce dont rêverait n’importe quel acheteur hésitant entre deux voitures. À la fin de chaque journée, les consommations moyennes des véhicules sont relevées et transmises au jury. À l’issue de l’événement, chaque voiture aura parcouru entre 400 et 500 km, ce qui permet d’obtenir des chiffres très proches d’une utilisation réelle.
Autre nouveauté notable de cette édition : l’AMAM a proposé aux constructeurs de devenir leur propre jury. À l’issue d’une session d’essai qui leur était réservée, chaque représentant de marque a pu évaluer plusieurs véhicules concurrents et voter pour celui qui l’avait le plus séduit.
De la mini citadine au grand SUV, en passant par des berlines et même un mini-bus électrique, les modèles engagés dans la compétition étaient les suivants : Alpine A390, Audi Q3 e-hybrid, BMW iX3, BYD Atto 2 DM-i, Cadillac Optiq, Citroën ë-C3, Citroën C5 Aircross PHEV, Cupra Raval VZ, Dacia Bigster Hybrid GPL 4×4, Denza Z9 GT, Honda Prelude, Jaecoo 5 HEV, Lexus RZ, Mazda 6e, MG 4 Urban, Nissan Micra, Nissan Qashqai, Omoda 9 PHEV, Peugeot e-408, Polestar 3, Polestar 4, Porsche Cayenne electric, Renault Twingo, Suzuki eVitara, Volkswagen ID. 3 GTX, Volkswagen ID. Buzz GTX, Volvo ES90 et Zeekr 7X. Sur les 24 marques représentées, cinq sont nouvelles sur le marché français : Denza, Jaecoo, Omoda, Polestar et Zeekr, signe d’un marché en plein renouvellement.
Le Palmarès
Onze prix ont été décernés lors de la cérémonie organisée à l’issue des deux journées d’essai, animée par Nicolas Heidet et Alexandre Lenoir, respectivement Président et Trésorier de l’association.
Meilleure Électrique : 1ère BMW iX3, 2e Porsche Cayenne electric, 3e MG 4 Urban
Meilleure PHEV : 1ère Citroën C5 Aircross, 2e Audi Q3 e-hybrid, 3e BYD Atto 2 DM-i
Meilleure HEV : 1ère Honda Prelude, 2e Jaecoo 5, 3e Nissan Qashqai e-Power
Meilleur Grand SUV : 1er Porsche Cayenne electric, 2e BMW iX3, 3e Polestar 3
Meilleur SUV compact : 1er Jaecoo 5, 2e Dacia Bigster Hybrid GPL 4×4, 3e Nissan Qashqai
Meilleure Familiale : 1ère Polestar 3, 2e BMW iX3, 3e Mazda 6e
Meilleure Compacte : 1ère MG 4 Urban, 2e Volkswagen ID. 3 GTX, 3e Honda Prelude
Meilleure Citadine : 1ère Renault Twingo, 2e Cupra Raval, 3e Nissan Micra
Meilleur Intérieur : 1er Volvo ES90, 2e BMW iX3, 3e Renault Twingo
Meilleure Ergonomie : 1ère BMW iX3, 2e Honda Prelude, 3e Cupra Raval
Prix des constructeurs : Renault Twingo
Le prix « Coup de Foudre » de l’AMAM
Le prix Coup de Foudre, décerné conjointement par le jury et les membres de l’AMAM, a été attribué cette année à la Honda Prelude, saluant son caractère et ses qualités globales qui ont su séduire aussi bien les journalistes que les représentants des constructeurs.
Nicolas Heidet, Président de l’AMAM, a résumé l’esprit de ces cinquièmes eTrophées : « L’AMAM se réjouit que l’engouement autour des eTrophées continue de grandir malgré un marché automobile toujours dans la grisaille en France. C’est la preuve d’un réel dynamisme et d’une forte volonté de la part des constructeurs de proposer de nouvelles solutions aux automobilistes. Depuis maintenant 5 ans, les eTrophées de l’AMAM sont devenus un événement incontournable, tant pour les constructeurs auto qui se prêtent au jeu de la confrontation, que pour les médias français. »
The University of California at Berkeley discovered the percentage of failing grades in multiple CS classes this spring "is significantly higher than past semesters," reports the campus's student newspaper.
"Instructors point to students' increased reliance on AI, lack of mathematical preparedness and understaffing as potential contributing factors."
According to [coursework platform] Berkeleytime, 35.3% of CS 10 students and 10.6% of CS 61A students received F's in spring 2026. In spring 2025 and spring 2024, the percentage of F's did not exceed 10% for either class. The electrical engineering and computer sciences department's grading guidelines state that 7% of students in lower division courses, including CS 10 and CS 61A, should receive D's and F's...
[UC Berkeley teaching professor Dan Garcia, who taught both classes] believes the "primary driver" of these abnormally high failing rates is due to a "vast increase in academic dishonesty" due to students' usage of large language models, such as Claude, ChatGPT and Google Gemini. "Some of the numbers that you saw from the number of students who receive failing grades were because we caught them (cheating) and prosecuted them and are sending their cases to the Center for Student Conduct," Garcia said. "But in other cases, it's students who are leaning a little too hard on LLMs to do their work for them, and then at exam time just really aren't ready." According to Garcia, nearly 30 students in CS 10 were "caught cheating on take-home exams" in spring 2026...
In addition to overreliance on AI, Garcia also pointed out that many students are underprepared mathematically, a concern echoed by campus associate teaching professor Gireeja Ranade. Ranade noticed a similar lack of prerequisite mathematical skills in her spring 2026 EECS 127 class, "Optimization Models in Engineering," which she described as "differently challenging" to teach this semester. The class saw a 16.8% F rate, far higher than the 5% of D's and F's that the EECS department describes as "typical" for an upper division course...
Both Garcia and Ranade have joined more than 1,300 UC faculty in signing a petition calling for the reinstatement of ACT and SAT standardized testing scores for STEM admissions in the UC system.
Thanks to long-time Slashdot reader theodp for sharing the article.
The company has evolved from an AI culling-focused tool into a full all-in-one platform for culling (Select), AI + manual editing (Edit), retouching (Retouch), and client delivery via Aftershoot Galleries. This shift eliminates the need to switch between multiple apps (e.g., Lightroom for editing/delivery).
Use coupon code RUMORS for a 15% discount at Aftershoot.
Here are the details:
Key Elements of the May 2026 Update
Built-in RAW Editing: New dedicated RAW editor with manual tonal adjustments, batch processing, sliders, and fine-tuning of AI edits. You can now edit RAW files directly in Aftershoot and export JPEGs or TIFFs without leaving the app. This includes self-improving AI profiles that learn from your adjustments, plus pipeline tools such as advanced masking, denoising, object removal, HSL/color grading, and more precise cropping/straightening (including aggressive, subject-focused crops).
Aftershoot Galleries (new, highlighted with “New” on the homepage): Integrated client delivery platform with 100 GB free storage. Features include built-in proofing, face detection/filtering by person, favorites, branded galleries, and direct export from Aftershoot (no re-uploads or downloads needed). It integrates print sales via labs like WHCC, Bay Photo, and Atkins Pro Labs. The goal is a better client experience than basic shares or Google Drive links.
Culling (Select) Enhancements: Smarter duplicate/variation detection (distinguishes intentional variations from true duplicates, ~20% tighter culls), improved blur/closed-eye detection, key subjects prioritization, binary selection states, and “cull to target” (specify how many keepers you want). UI now clearly distinguishes AI Automated Cull vs. AI Assisted Cull.
Editing & Retouching Improvements: More consistent white balance/exposure across albums, better before/after previews, enhanced acne/blemish/hair/glare detection with texture preservation, Background Replacement (beta with custom uploads), Cloth Dewrinkle, object removal, and subtle body reshaping. Retouching works on RAW files.
The update positions Aftershoot as a privacy-focused (local processing), non-destructive workflow tool that learns your style via training on your edits/presets or imported Lightroom profiles. It includes 30+ pro styles in the marketplace.
2026 Roadmap
Aftershoot also published its 2026 Roadmap (full AI Retouching rollout with one-click skin/hair/glare tools, advanced culling, smarter editing, and expanded creator styles/marketplace):
Further AI refinements in culling (smarter technical detection, key subjects, cull-to-target).
Enhanced editing (better profile accuracy, advanced masking/cropping, more marketplace styles with previews).
Expanded RAW Editing (histogram, tone curves, deeper masking, etc.).
More AI Retouching tools (one-click object removal, body reshaping, background removal, natural skin work).
Tethered shooting (sneak peek): Planned rock-solid tethering with fast/stable connections, broad camera support, auto-import/organization, on-set proofing, and instant gallery pushes. Not yet released.
The company claims users save significant time (e.g., 60+ days per year in repetitive work across their user base) while maintaining creative control and style consistency.
Pricing Updates
Aftershoot introduced more flexible modular pricing aligned with the complete workflow:
Aftershoot outlines three firm commitments: they will never create tools to replace you, they will always ask explicit permission before using your images (with easy opt-out or deletion), and they will build the product together with their Founders Community of photographers who help shape the roadmap. A personal note from founder Harshit explains how the company started in 2019 with a simple request to reduce culling time and has stayed true to its mission of returning time to creators ever since.
An official trailer dropped this week for Tomb Raider: Legacy of Atlantis. It's "a full-blown remake of the original 1996 Tomb Raider game," reports Kotaku, "rebuilt from the ground up using Unreal Engine 5." Developed by Flying Wild Hog (with assistance/guidance from longtime Tomb Raider studio Crystal Dynamics), "it will also make some changes to puzzles, combat, platforming..." The game's Steam page acknowledges that AI-assisted tools were used during development "to support some early exploration and temporary development content," but that any AI-assisted assets were "either replaced or refined by humans in order to maintain the creative and artistic vision of the development team." In a statement to Eurogamer, Crystal Dynamics clarifies that they "leverage" AI tools "to help our teams iterate on ideas faster and more efficiently, while ensuring that all finished content in the final product is human-crafted." (But are they considering AI-assisted assets "refined" by humans as "human-crafted"?)
Polygon reports that "The early response to the news has been mixed to negative on the Tomb Raider subreddit, ranging from vague hopes that the generative-AI craze will simply go away to grim resignation that this is the future of game development." Beyond labor concerns, art theft worries, and environmental issues, the most straightforward reason AI art has been unpopular is that many players find it hideous. We'll find out for sure whether Tomb Raider: Legacy of Atlantis' use of AI is particularly blatant when it comes out in February 2027. Its release date is February 12, 2027 on PS5, Xbox Series X/S, Switch 2, and PC.
Anthropic is urging leading AI labs to consider slowing development, warning that frontier models are advancing fast enough that they may soon be able to improve themselves without direct human intervention. The company says a global ability to pause or slow AI development would "likely be a good thing," citing internal data about accelerating model capabilities. From a blog post: Using public benchmarks and previously unreported data from within Anthropic, The Anthropic Institute is showing that AI is already accelerating the development of AI systems. To take just one example: today, Anthropic engineers on average ship 8x as much code per quarter as they did from 2021-2025.
The technical trends discussed in this piece suggest that AI systems are going to become much more capable in coming years. These trends have huge implications. AI that can build itself would be a major development in the history of technology -- one that could bring enormous good for the world in science, healthcare, and beyond. But full recursive self-improvement also might increase the risks of humans losing control over AI systems. If systems are capable of fully building their own successors, the ways we secure them, monitor them, and shape their behavior all grow much more important. [...]
If it were possible to effectively slow the development of this technology to give ourselves more time to deal with its immense implications, we think that would likely be a good thing. But if a slowdown simply lets the least cautious actors catch up technologically, it could leave everyone less safe. Without a global coordination mechanism, companies and governments will have to make difficult decisions about safety while under competitive and geopolitical pressures.
We believe it would be good for the world to have the option to slow or temporarily pause frontier AI development to enable societal structures and alignment research to keep up with the advance of the technology. The Anthropic Institute will conduct research -- in collaboration with many others -- and take actions to help build the systems that a credible slowdown or pause would require. These systems would enable frontier AI developers to verify that others globally have actually stopped or slowed, and that a bad actor could not use the auspices of a coordinated slowdown to jump ahead in secret. If such systems existed, we expect that we would slow down or temporarily pause, if other developers at or near the frontier also did so in a verifiable manner...
Meta has reportedly delayed the developer release of its Muse Spark AI model API multiple times, and as of Tuesday, had no scheduled launch date, according to the Wall Street Journal (paywalled). Reuters reports: A Meta spokesperson told Reuters on Wednesday that the company is already testing the Application Programming Interface (API) with some early partners and is looking forward to releasing it this month. "The muse spark API will be coming soon," Meta AI Chief Alexandr Wang announced in a post on X in April.
Meta unveiled Muse Spark in April as the first model built to close the gap with rivals. Muse Spark is the first in a new series of models created by the company's Superintelligence Labs. Earlier on Wednesday, Meta unveiled an AI agent aimed at helping businesses carry out day-to-day operations, hinting at the company's ambitions to compete with rivals such as OpenAI, Anthropic and Alphabet's Google.
Dans un article paru le 3 juin 2026, le site américain 404 Media révèle comment des entreprises exploitent Reddit pour manipuler les réponses de ChatGPT et de la recherche IA de Google.
Meta is scaling back parts of its employee tracking initiative after staff objected to software that collected mouse movements, clicks, keystrokes, and other actions for AI training data. According to Reuters, the company will now let workers pause collection for up to 30 minutes and request exemptions. Reuters reports: [Stephane Kasriel, a vice president in Meta's AI model-building Superintelligence Labs unit] said the team behind the software had also introduced "several optimizations" to reduce its impact on computer battery life, after employees complained it was consuming so much data it was causing their home internet usage to spike. "While we remain confident in the privacy protections we put in place at launch, which went through several layers of risk review, we have heard your concerns about personal data on work devices, battery life, and wanting more control over when capturing happens," Kasriel said in the memo.
An anonymous reader quotes a report from GeekWire: A team inside Microsoft has been quietly building a platform for devices that run AI agents instead of apps, based on Android instead of Windows, with two working hardware designs so far, and an initial set of big-name companies lined up to run pilots. The platform, dubbed "Project Solara," is Microsoft's bet that AI will open up entirely new scenarios for computing -- using agents to avoid the constraints of traditional software, and off-the-shelf components to develop new devices quickly and inexpensively. [...] The company unveiled Solara on Tuesday at its Build conference in San Francisco, describing it as a new platform that spans from chip to cloud. GeekWire got a behind-the-scenes look at the project during a briefing last week in Redmond, including demos of the first two concept devices based on the platform:
- A desktop hub that sits beside a PC and responds to voice commands, signs users in using facial recognition, and surfaces the day's most pressing items. With a monitor attached, it becomes a full Windows machine running in the cloud.
- A wearable badge that reimagines the standard employee ID card. A fingerprint button wakes an agent in one press; a single tap records and transcribes a conversation; and a built-in camera lets the agent act on what the user sees.
Microsoft says it won't ship these devices itself. Instead, it envisions hardware makers and other industry partners turning the reference designs into implementations of their own, each intended for a specific industry, company, or scenario. For example, in one demo shown by the company, the high-tech badge ran on agents designed for use by a health-care worker, including the ability to scan a patient's QR code, record and transcribe the visit, log vitals, and start a prescription. In another application of the same badge, the built-in camera scanned a brainstorm board with ideas for an office revamp, and made a suggestion: add some plants.
The two devices are a starting point. The bigger opportunity, the company says, is all the tasks and workflows where a PC or phone gets in the way or isn't practical to use. [...] In the coming months, companies including AccuWeather, Best Buy, CVS Health, Levi's, and Target are expected to begin pilots of devices based on the reference designs. The operating system is the Microsoft Device Ecosystem Platform, or MDEP, an enterprise version of Android that Microsoft developed for devices including Teams meeting-room hardware. The company says it chose MDEP over Windows deliberately, to run on smaller, lower-power devices while keeping the management and security features IT departments expect: patch and over-the-air updates, device integrity, Microsoft Defender, Intune, and Entra ID sign-in. While the project is still in the early stages, Microsoft CEO Satya Nadella encouraged the team to show it at Build sooner than the company would normally show its work in public. "That underscores just how competitive and fast-moving the AI world is right now, but it also illustrates the pace that the new technologies are enabling," reports GeekWire.
The report notes that the business model for the platform still needs to be worked out. The devices run on Microsoft's Azure cloud, but beyond that, "the economics are still taking shape."
Qualcomm and MediaTek have been chosen as the first chip partners. "The badge runs on a new Qualcomm wearable chip; the desk hub runs on MediaTek IoT silicon," reports GeekWire. "Both are off-the-shelf, not custom, which is central to how Microsoft plans to keep devices cheap and fast to build."
Longtime Slashdot reader Matt_Bennett shares a blog post from Adafruit: Adafruit received at 10:38 p.m. ET on May 22, 2026 a letter from former FBI chief of staff, Jonathan F. Lenzner, and partner at Fenwick & West LLP, counsel for Flux, demanding, among other things, that Adafruit refrain from publishing an article addressing what the letter characterizes as false and potentially defamatory claims about Flux, including statements about Flux's intellectual property, commercial traction and user base.
The letter further asserts claims under the Computer Fraud and Abuse Act. Adafruit accessed only information that Flux's own systems made publicly available through a server misconfiguration. Adafruit's reporting concerns a matter of public security interest and was conducted in the ordinary course of responsible disclosure.
Although Adafruit vigorously rejects the assertions made in Flux's May 22, 2026 demand letter, we have temporarily stopped publishing on the Adafruit blog while we consider our response and next steps. We will update the community as appropriate. For context, Adafruit is a major open-source hardware company and electronics retailer known for its maker-focused boards, components, tutorials, and community publishing. Flux.ai is relevant because it is building an AI-assisted circuit-board design platform aimed at changing how engineers create and collaborate on PCB designs.
"Adafruit probably did a review of AI PCB tools," writes HN user karmicthreat. "I've used Flux.ai before; it was a pretty bad experience. After about 50-100$ in tokens a couple of times, I couldn't get more than a couple of simple components on the schematic. And not in sensible positions..."
Redditor AlexTaradox adds: "Nothing was published as far as I know. I assume they did review of AI tools and likely contacted flux with some preliminary results, but flux saw where it is going and decided to block them from publishing any results. Flux is garbage and they obviously know it, but they need to hold for some time until some other scam acquires them. Doing anything with them is just asking to be screwed..."
Further discussions are taking place on Reddit and Hacker News.
An anonymous reader quotes a report from Ars Technica: In April, GitHub announced that it was moving subscribers from request-based billing to a usage-based model for its AI-powered Copilot service. As that new pricing model goes into effect today, many GitHub Copilot users are reporting some extreme sticker shock as they realize just how quickly their previous "normal" usage is burning through their newly limited monthly allotment of AI credits. Across social media and forums, many Copilot users are sharing personal statistics showing how just a few hours of AI usage can now account for a large chunk of their new monthly subscription caps. For some users, it reportedly took less than a day to use up a month's usage quota.
That's a big change from previous months, when GitHub Copilot subscribers were allocated a certain number of "requests" and "premium requests" based on their payment tier. GitHub said that the old system meant that "a quick chat question and a multi-hour autonomous coding session [could] cost the user the same amount," forcing Copilot itself to "absorb much of the escalating inference cost behind that usage." [...] Indeed, some Copilot users have been sharing estimates from GitHub's own tool showing that their previous monthly usage would rack up bills in the thousands of dollars under the new pricing plan. Under GitHub's new usage-based pricing system, paid Copilot subscriptions instead grant users a certain number of AI "credits" each month, with one credit corresponding to $0.01 of usage. Subscribers also get bonus credits depending on their subscription level: the $10/month Pro plan includes 1,500 credits ($15 worth); the $39 Pro+ plan includes 7,000 credits ($70 worth); and the $100/month Copilot Max plan includes 20,000 credits ($200 worth).
The precise number of Copilot credits used by a given prompt is determined by the number of input and output tokens used and the rates charged by the underlying large language model. That means pricing is highly dependent not just on the type of request but on the specific model that a user chooses. One million output tokens from OpenAI's GPT-5.4 nano would run just $1.25 on GitHub Copilot, but that same level of output would run $30 on the frontier GPT-5.5 model (Copilot users who rely on "Auto" mode to pick the most appropriate available model for any request should be extremely careful, as some users report it can switch to expensive models for extremely simple queries).
D'après le site américain Media 404, des hackers seraient parvenus à détourner le chatbot d’assistance de Meta pour prendre le contrôle de comptes Instagram. Une faille simple, basée sur la fragilité des gardes-fous de l'IA.
Premier État américain à poursuivre OpenAI, la Floride accuse l’entreprise d’avoir bâti le succès de ChatGPT sur des pratiques trompeuses, une collecte massive de données et des garde-fous insuffisants, notamment pour les mineurs et les utilisateurs vulnérables.
Connexion
