Actualités — Page13

Florida Attorney General James Uthmeier filed a landmark state lawsuit Monday against OpenAI and its chief executive Sam Altman, accusing the company of ignoring safety warnings and enabling harmful behavior. The 83‑page complaint links the chatbot to mass shootings, suicides and other violent incidents, alleging that OpenAI put children at risk and allowed a dangerous product to reach millions of Floridians. OpenAI has denied responsibility, pointing to ongoing legal battles and previous dismissals of similar claims.Lire la suite

Google let a reporter test Gemini Spark, its new 24/7 AI assistant, over a week. The tool drafted emails, pulled data from Drive, created calendar events and spreadsheets without direct prompts. While the results impressed, the service costs $99.99 a month, is limited to U.S. users, and requires constant oversight. Privacy advocates note the AI accesses personal data, and the reporter found the need to monitor every step. Spark’s performance suggests a powerful addition to Google’s ecosystem, but its price tag and data‑handling raise doubts about broad adoption.Lire la suite

OpenAI rolled out a new feature for its Codex coding assistant that lets users call up AI‑generated animated pets. The optional companions appear as a floating overlay, signaling what Codex is working on, when it finishes a task, or when it needs user input. Developers can type "/pet" to summon or dismiss a companion and "/hatch" to create a custom creature. Eight preset pets ship with the tool, and early users have already shared dozens of variations, including a nod to Microsoft’s classic Clippy. The feature is live on Windows and macOS, and OpenAI is bundling a limited‑time 30‑day ChatGPT Pro trial with ten favorite pets.Lire la suite

Anthropic announced on Monday that it has submitted a confidential draft registration statement to the U.S. Securities and Exchange Commission, marking the first formal step toward an initial public offering. Valued at a post‑money $965 billion after its latest fundraising round, the AI startup now eclipses rival OpenAI’s $852 billion valuation. The filing comes as SpaceX prepares for a June 12 IPO that could raise $80 billion, and follows a recent court ruling that dismissed Elon Musk’s claims against OpenAI. Anthropic’s move intensifies the competition among the sector’s most valuable private firms.Lire la suite

Anthropic, the AI firm behind the Claude chatbot, submitted a confidential registration statement to the U.S. Securities and Exchange Commission on Monday, signalling its intent to go public. The filing follows a $65 billion Series H financing that lifted the company’s valuation to roughly $965 billion. Anthropic did not disclose the number of shares, price range or timing, saying the offering will depend on market conditions. The move places the startup in a crowded IPO season that also includes SpaceX and adds pressure to its rival OpenAI, which is preparing its own public listing.Lire la suite

Florida Attorney General James Uthmeier filed a civil lawsuit on June 1, 2026, accusing OpenAI and its chief executive Sam Altman of deceptive trade practices, negligence and product liability. The complaint alleges the company knowingly released ChatGPT despite risks of addiction, cognitive decline, suicide and violence. Citing several shootings and suicide cases linked to the chatbot, the suit seeks penalties and court orders to hold Altman personally liable for alleged harms to Floridians.Lire la suite

OpenAI announced that its latest artificial‑intelligence system has produced a counterexample to the unit distance conjecture, a problem that has resisted proof for eight decades. By tweaking the spacing of a mathematical grid and applying the Pythagorean theorem, the model identified a configuration where a unit‑distance circle intersects far more grid points than previously thought possible. Researchers say the breakthrough demonstrates AI’s growing capacity to tackle abstract mathematical challenges that have stumped human experts for generations.Lire la suite

Nvidia announced that AI labs Anthropic and OpenAI, along with SpaceX and Oracle, are among the first major users of its new Vera data‑center processor. Unveiled at Computex in Taipei, Vera replaces Nvidia's Grace CPU with a custom design built around 88 Olympus cores and up to 1.2 TB/s of memory bandwidth. Independent benchmarks show the chip outpacing Intel Xeon and AMD EPYC on several AI‑agent workloads. Oracle Cloud Infrastructure will be the first hyperscaler to run Vera at scale, with broader cloud availability slated for the second half of 2026.Lire la suite

Inherent, a London‑based artificial‑intelligence lab founded by former DeepMind researchers, announced a $50 million seed round led by Index Ventures and Radical Ventures. The funding will support development of Faraday, an AI‑native science platform designed to identify the most promising research questions rather than simply answer them. Backed by Nvidia’s venture arm NVentures and a roster of other investors, Inherent operates as a public‑benefit corporation, signaling a blend of cutting‑edge technology and societal responsibility.Lire la suite

Meta will begin testing an artificial‑intelligence‑enabled pendant within the next year, according to an internal memo seen by The Information. The same document outlines a new business‑focused subscription called Wearables for Work, aimed at turning the company’s hardware—particularly its smart glasses—into productivity tools. The move follows a string of costly setbacks for Reality Labs, which posted a $4 billion loss in the first quarter of 2026, and builds on Meta’s 2025 acquisition of Limitless, the startup behind an early AI‑pendant prototype.Lire la suite

Meta is set to begin testing an AI-enabled pendant within the next twelve months, according to an internal memo obtained by The Information. The device builds on technology from Limitless, a startup the company bought at the end of 2025, and is part of a broader push that includes new AI glasses and a business‑focused subscription called Wearables for Work. The move is intended to turn around Reality Labs, which posted a $4 billion loss in the first quarter of 2026.Lire la suite

Anthropic announced it is cutting the list of secondary‑market platforms it deems unauthorized from eight firms to four after a wave of investor confusion and criticism. The revised warning now names Open Door Partners, Unicorns Exchange, Pachamama and Upmarket, removing platforms such as Hiive that had protested the original notice. The move comes as the AI company finalizes a $65 billion funding round that values it at $965 billion, the largest private raise in history. Anthropic has not commented on the revision, and the remaining platforms have yet to respond.Lire la suite

Microsoft’s GitHub Copilot will move from a flat‑rate subscription to a token‑usage model on June 1, charging users for every token the AI consumes. The shift has sparked a wave of criticism on Reddit and X, with developers warning that monthly costs could surge from the current $29 to upwards of $750. Some users blame the new pricing on excessive “vibe‑coding” while others question the sustainability of the former subsidized model. Microsoft has not responded to requests for comment.Lire la suite

A comprehensive glossary of artificial‑intelligence terms has been released, offering clear definitions for buzzwords such as large language models, artificial general intelligence, AI agents, and chain‑of‑thought reasoning. Updated regularly to keep pace with the field, the resource seeks to help developers, researchers, and the broader public navigate an expanding lexicon that often leaves even seasoned technologists unsure of meaning.Lire la suite

Google unveiled Gemini Spark, an always‑on AI assistant that taps into users' Gmail, Docs, and Calendar to automate tasks like event planning. The beta, limited to AI Ultra subscribers at $100 a month, demonstrated its ability to generate a full birthday itinerary from a single prompt, but also sparked privacy concerns after the agent labeled the author's long‑term boyfriend merely as a "close friend and frequent companion."Lire la suite

A METR research team discovered that most developers now balk at performing any coding task without artificial‑intelligence assistance. The lab’s follow‑up survey found programmers believe AI doubles their value, yet recent reports from Amazon, Uber and independent researchers suggest the promised productivity boost may be illusory, with token‑tracking systems shut down and maintenance problems on the rise. The clash between developer reliance on AI and mounting expenses is reshaping how firms evaluate coding tools.Lire la suite

Groq is raising $650 million from its existing backers to fund its inference‑cloud business, six months after Nvidia struck a $20 billion not‑acqui‑hire that paid out Groq’s investors, hired several senior engineers and licensed the company’s chip technology. The round is effectively guaranteed, with Disruptive and Infinitium stepping in to fill any gaps. Led on an interim basis by CEO Adam Winter and CFO Matt Eng, Groq aims to leverage its purpose‑built Language Processing Unit hardware to compete in a market increasingly dominated by GPUs and aggressive pricing from AI model providers.Lire la suite

A new study by METR could not replicate a 2025 experiment that claimed AI made developers more productive, as participants refused to work without AI assistance. Subsequent surveys and corporate reports reveal that AI coding tools often slow development, generate more bugs, and inflate costs. Amazon halted a token‑tracking leaderboard after employees gamed it, while Uber burned through its entire AI budget in four months without measurable gains. Experts warn that the industry’s reliance on AI-driven token consumption as a productivity metric is misguided and call for stronger code‑review practices.Lire la suite

OpenAI announced the retirement of GPT‑4.5, the final GPT‑4‑based model still available in ChatGPT, marking the end of an era that propelled generative AI into mainstream use. The company posted a brief note in ChatGPT’s release notes, confirming that GPT‑4.5 will be removed on June 27, 2026 after a 30‑day sunset period, while the older o3 model will disappear on August 26, 2026. Users, especially paid subscribers, expressed disappointment on X, recalling the models’ writing and reasoning strengths. The move signals OpenAI’s shift toward newer, more capable models such as GPT‑5.5.Lire la suite

A hands‑on test revealed that a straightforward prompt can make ChatGPT answer in a style that mirrors Google’s Gemini model. By asking the chatbot to be more structured, analytical and less conversational, the researcher noted a shift from ChatGPT’s usual warm, supportive tone to a more restrained, academic voice. The change affected how the AI framed advice, used qualifiers and organized its responses, underscoring how personality cues shape user experience even when the underlying model stays the same.Lire la suite