Vue normale

Reçu aujourd’hui — 15 juillet 2025Actualités numériques

☕️ X visé par une enquête du parquet de Paris pour ingérence étrangère

15 juillet 2025 à 06:39

L’AFP a appris vendredi 11 juillet que le parquet de Paris a ouvert une enquête contre le réseau social d’Elon Musk, qu’il suspecte de « biaiser le débat démocratique en France ».

La procureure de Paris, Laure Beccuau, a confié cette enquête à la gendarmerie nationale. Elle vise aussi bien le réseau social que « les personnes physiques » qui la dirigent.

L’enquête a été déclenchée suite à deux signalements en janvier dernier, dont l’un provient du député Eric Bothorel et l’autre d’un directeur de cybersécurité dans la fonction publique, selon le Canard enchaîné.

Cette annonce a été faite alors que le bot de xAI a récemment publié sur le réseau social X plusieurs fois des messages antisémites, généré des avis positifs sur l’extrême droite française et fait l’éloge d’Hitler.

Le communiqué du parquet affirme que la procédure a été ouverte « sur le fondement de vérifications, de contributions de chercheurs français et d’éléments apportés par différentes institutions publiques ». L’enquête porte « notamment » sur l’ « altération du fonctionnement d’un système de traitement automatisé de données en bande organisée, ainsi qu’extraction frauduleuse de données d’un système de traitement automatisé de données en bande organisée », explique l’AFP.

☕️ X visé par une enquête du parquet de Paris pour ingérence étrangère

15 juillet 2025 à 06:39

L’AFP a appris vendredi 11 juillet que le parquet de Paris a ouvert une enquête contre le réseau social d’Elon Musk, qu’il suspecte de « biaiser le débat démocratique en France ».

La procureure de Paris, Laure Beccuau, a confié cette enquête à la gendarmerie nationale. Elle vise aussi bien le réseau social que « les personnes physiques » qui la dirigent.

L’enquête a été déclenchée suite à deux signalements en janvier dernier, dont l’un provient du député Eric Bothorel et l’autre d’un directeur de cybersécurité dans la fonction publique, selon le Canard enchaîné.

Cette annonce a été faite alors que le bot de xAI a récemment publié sur le réseau social X plusieurs fois des messages antisémites, généré des avis positifs sur l’extrême droite française et fait l’éloge d’Hitler.

Le communiqué du parquet affirme que la procédure a été ouverte « sur le fondement de vérifications, de contributions de chercheurs français et d’éléments apportés par différentes institutions publiques ». L’enquête porte « notamment » sur l’ « altération du fonctionnement d’un système de traitement automatisé de données en bande organisée, ainsi qu’extraction frauduleuse de données d’un système de traitement automatisé de données en bande organisée », explique l’AFP.

Souveraineté : défaillances et contradictions de l’État sur la commande publique

15 juillet 2025 à 06:27
Même une faute politique ? La question est posée
Souveraineté : défaillances et contradictions de l’État sur la commande publique

« Choose France », « buy european »… des termes qui ne trouvent pas suffisamment d’écho sur la commande publique, notamment dans le numérique. C’est la conclusion d’un rapport d’une commission d’enquête pour qui « l’État est pris en faute ».

Après des dizaines d’auditions et quatre mois de travaux, le rapport de la commission d’enquête sur la commande publique est disponible (les comptes rendus des auditions arriveront dans un second temps). Il est assorti de pas moins de 67 recommandations pour « mieux exploiter le potentiel de la commande publique au service de la souveraineté économique et numérique européenne ».

L’enjeu est de taille puisque la commande publique française représentait, selon le rapport, 170 milliards d’euros en 2023, soit deux fois plus qu’en 2014 (83 milliards d’euros). Le rapport tire à boulet rouge contre l’État qui n’a « pas su se hisser à la hauteur des enjeux ces dernières années », notamment pour ce qui est « d’assurer la protection des données publiques face aux géants du numérique ». Dans cet article, nous allons rester sur le volet numérique du rapport.

Inertie, défaillances, contradictions, faute… L’État en prend pour son grade


Il reste 85% de l'article à découvrir.
Vous devez être abonné•e pour lire la suite de cet article.
Déjà abonné•e ? Générez une clé RSS dans votre profil.

Zotac aurait du matériel MXM en réserve : le format modulaire de GPU encore en vie ?

Sur PC portable, les slots PCIe sont quasiment inexistants. En effet, si les M.2 NVMe sont techniquement bien câblés en utilisant le protocole PCI Express (ce qui permet l’existence d’adaptateur M2 vers GPU plus ou moins douteux !), il s’agit bien de la seule interface interne capable de transporter...

New Effort To Upstream LTTng In The Linux Kernel Draws Criticism From Torvalds

15 juillet 2025 à 00:37
The LTTng tracing toolkit is twenty years old this year and it's seen significant adoption by different hyperscalers and other notable organizations like IBM and Sony and Siemens beyond basic end-users and administrators for system tracing/debugging. While having many successes over the past two decades, the kernel modules remain outside of the kernel tree. Even with around four different upstreaming attempts to get the LTTng code into the mainline kernel, it still has not happened. A fifth attempt began today but still looks like it could be an uphill battle...

Meta's Superintelligence Lab Considers Shift To Closed AI Model

Par :BeauHD
15 juillet 2025 à 00:10
An anonymous reader quotes a report from Investing.com: Meta's newly formed superintelligence lab is discussing potential changes to the company's artificial intelligence strategy that could represent a major shift for the social media giant. A small group of top members of the lab, including 28-year-old Alexandr Wang, Meta's new chief A.I. officer, talked last week about abandoning the company's most powerful open source A.I. model, called Behemoth, in favor of developing a closed model, according to a report in the New York Times, citing people familiar with the matter. Meta has traditionally open sourced its A.I. models, making the computer code public for other developers to build upon, and any shift toward a closed A.I. model would mark a significant philosophical change for Meta. Meta had completed training its Behemoth model by feeding in data to improve it, but delayed its release due to poor internal performance. After the company announced the formation of the superintelligence lab last month, teams working on the Behemoth model, which is considered a "frontier" model, stopped conducting new tests on it. The discussions within the superintelligence lab remain preliminary, and no decisions have been finalized. Any potential changes would require approval from Meta CEO Mark Zuckerberg.

Read more of this story at Slashdot.

Five EU States To Test Age Verification App To Protect Children

Par :BeauHD
14 juillet 2025 à 23:30
France, Spain, Italy, Denmark, and Greece will pilot an age verification app to better protect children online, as part of the EU's push to enforce its Digital Services Act. Reuters reports: The setup for the age verification app is built on the same technical specifications as the European Digital Identity Wallet which will be rolled out next year. The five countries can customize the model according to their requirements, integrate into a national app or keep it separately. The landmark legislation, which became applicable last year, requires Alphabet's Google, Meta, ByteDance's TikTok and other online companies to do more to tackle illegal and harmful online content. EU regulators said the new guidelines would help online platforms to tackle addictive design, cyberbullying, harmful content and unwanted contact from strangers.

Read more of this story at Slashdot.

China's Moonshot Launches Free AI Model Kimi K2 That Outperforms GPT-4 In Key Benchmarks

Par :BeauHD
14 juillet 2025 à 22:50
Chinese AI startup Moonshot AI has released Kimi K2, a trillion-parameter open-source language model that outperforms GPT-4 in key benchmarks with particularly strong performance on coding and autonomous agent tasks. VentureBeat reports: The new model, called Kimi K2, features 1 trillion total parameters with 32 billion activated parameters in a mixture-of-experts architecture. The company is releasing two versions: a foundation model for researchers and developers, and an instruction-tuned variant optimized for chat and autonomous agent applications. "Kimi K2 does not just answer; it acts," the company stated in its announcement blog. "With Kimi K2, advanced agentic intelligence is more open and accessible than ever. We can't wait to see what you build." The model's standout feature is its optimization for "agentic" capabilities -- the ability to autonomously use tools, write and execute code, and complete complex multi-step tasks without human intervention. In benchmark tests, Kimi K2 achieved 65.8% accuracy on SWE-bench Verified, a challenging software engineering benchmark, outperforming most open-source alternatives and matching some proprietary models. [...] On LiveCodeBench, arguably the most realistic coding benchmark available, Kimi K2 achieved 53.7% accuracy, decisively beating DeepSeek-V3's 46.9% and GPT-4.1's 44.7%. More striking still: it scored 97.4% on MATH-500 compared to GPT-4.1's 92.4%, suggesting Moonshot has cracked something fundamental about mathematical reasoning that has eluded larger, better-funded competitors. But here's what the benchmarks don't capture: Moonshot is achieving these results with a model that costs a fraction of what incumbents spend on training and inference. While OpenAI burns through hundreds of millions on compute for incremental improvements, Moonshot appears to have found a more efficient path to the same destination. It's a classic innovator's dilemma playing out in real time -- the scrappy outsider isn't just matching the incumbent's performance, they're doing it better, faster, and cheaper.

Read more of this story at Slashdot.

Two Guys Hated Using Comcast, So They Built Their Own Fiber ISP

Par :BeauHD
14 juillet 2025 à 22:10
An anonymous reader quotes a report from Ars Technica: Samuel Herman and Alexander Baciu never liked using Comcast's cable broadband. Now, the residents of Saline, Michigan, operate a fiber Internet service provider that competes against Comcast in their neighborhoods and has ambitions to expand. "All throughout my life pretty much, I've had to deal with Xfinity's bullcrap, them not being able to handle the speeds that we need," Herman told Ars. "I lived in a house of 10. I have seven other brothers and sisters, and there's 10 of us in total with my parents." With all those kids using the Internet for school and other needs, "it just doesn't work out," he said. Herman was particularly frustrated with Comcast upload speeds, which are much slower than the cable service's download speeds. "Many times we would have to call Comcast and let them know our bandwidth was slowing down... then they would say, 'OK, we'll refresh the system.' So then it would work again for a week to two weeks, and then again we'd have the same issues," he said. Herman, now 25, got married in 2021 and started building his own house, and he tried to find another ISP to serve the property. He was familiar with local Internet service providers because he worked in construction for his father's company, which contracts with ISPs to build their networks. But no fiber ISP was looking to compete directly against Comcast where he lived, though Metronet and 123NET offer fiber elsewhere in the city, Herman said. He ended up paying Comcast $120 a month for gigabit download service with slower upload speeds. Baciu, who lives about a mile away from Herman, was also stuck with Comcast and was paying about the same amount for gigabit download speeds. Herman said he was the chief operating officer of his father's construction company and that he shifted the business "from doing just directional drilling to be a turnkey contractor for ISPs." Baciu, Herman's brother-in-law (having married Herman's oldest sister), was the chief construction officer. Fueled by their knowledge of the business and their dislike of Comcast, they founded a fiber ISP called Prime-One. Now, Herman is paying $80 a month to his own company for symmetrical gigabit service. Prime-One also offers 500Mbps for $75, 2Gbps for $95, and 5Gbps for $110. The first 30 days are free, and all plans have unlimited data and no contracts. "We are 100 percent fiber optic," Baciu told Ars. "Everything that we're doing is all underground. We're not doing aerial because we really want to protect the infrastructure and make sure we're having a reliable connection." Each customer's Optical Network Terminal (ONT) and other equipment is included in the service plan. Prime-One provides a modem and the ONT, plus a Wi-Fi router if the customer prefers not to use their own router. They don't charge equipment or installation fees, Herman and Baciu said. Prime-One began serving customers in January 2025, and Baciu said the network has been built to about 1,500 homes in Saline with about 75 miles of fiber installed. Prime-One intends to serve nearby towns as well, with the founders saying the plan is to serve 4,000 homes with the initial build and then expand further. [...] A bit more than 100 residents have bought service so far, they said. Herman said the company is looking to sign up about 30 percent of the homes in its network area to make a profit. "I feel fairly confident," Herman said, noting the number of customers who signed up with the initial construction not even halfway finished.

Read more of this story at Slashdot.

Reçu hier — 14 juillet 2025Actualités numériques

Apple Faces Calls To Reboot AI Strategy With Shares Slumping

Par :msmash
14 juillet 2025 à 21:30
Apple is facing pressure to shake up its corporate playbook to invigorate its struggling artificial intelligence efforts. From a report: Alarmed by a share slump that's erased more than $640 billion in market value this year and frustrated with delays in rolling out AI features, investors are calling for Apple to break with long-standing traditions to make a big acquisition and more aggressively pursue talent. "Historically Apple does not do big mergers and acquisitions," said Citigroup Inc. analyst Atif Malik, noting that the last major deal was its takeover of Beats in 2014. But, he argues, "investors would turn more positive if Apple could acquire or invest a meaningful stake in an established AI provider." Apple shares have fallen 16% this year while traders bid up the shares of peers like Meta, which is spending lavishly on AI. While Apple faces other problems, including its exposure to tariffs and regulatory issues, disappointment in bringing compelling AI features to its vast ecosystem of devices has become top of mind for investors.

Read more of this story at Slashdot.

Armagh Observatory Marks 230 Years of Recording Weather

Par :msmash
14 juillet 2025 à 20:50
Armagh Observatory is marking a very special meteorological milestone as the institute celebrates 230 years of continuous weather observation. From a report: The unbroken tradition of handwritten data makes it the longest sequence of continuous weather information gathered anywhere in the UK and Ireland. Events are being held at Armagh Observatory on Monday to mark the significant anniversary. Nowadays, most weather data is gathered only by automated weather stations, but not in Armagh, where the human touch remains. The first handwritten recording was made on the evening of 14 July 1795, when a measurement of the temperature and air pressure was recorded on a graph at the observatory that sits above the city of Armagh. The measurement was repeated the next day and every subsequent day for the next 230 years.

Read more of this story at Slashdot.

Are a Few People Ruining the Internet For the Rest of Us?

Par :msmash
14 juillet 2025 à 20:10
A small fraction of hyperactive social media users generates the vast majority of toxic online content, according to research by New York University psychology professor Jay Van Bavel and colleagues Claire Robertson and Kareena del Rosario. The study found that 10% of users produce roughly 97% of political tweets, while just 0.1% of users share 80% of fake news. Twelve accounts known as the "disinformation dozen" created most vaccine misinformation on Facebook during the pandemic, the research found. In experiments, researchers paid participants to unfollow divisive political accounts on X. After one month, participants reported 23% less animosity toward other political groups. Nearly half declined to refollow hostile accounts after the study ended, and those maintaining healthier newsfeeds reported reduced animosity 11 months later. The research describes social media as a "funhouse mirror" that amplifies extreme voices while muting moderate perspectives.

Read more of this story at Slashdot.

MoonPay Executives May Have Sent $250,000 To Nigerian Scammer, DoJ Filing Suggests

Par :msmash
14 juillet 2025 à 19:30
A Department of Justice filing aiming to recover fraudulently obtained cryptocurrency may have inadvertently revealed the scam's victims as the CEO and CFO of crypto payment firm MoonPay. From a report: The filing, which aims to seize around $40,350 in USDT frozen by Tether, reveals that two victims sent $250,300 in USDT to a person posing as Steve Witkoff, co-chair of the President Trump's inaugural committee. However, records obtained from Binance revealed that the wallet that received the funds was registered to Ehiremen Aigbokhan, a man based in Lagos, Nigeria. The victims are identified in the filing only as "Ivan" and "Mouna." However, as outlet NOTUS noticed, Crypto payment firm Moonpay's CEO is Ivan Soto-Wright and its CFO is Mouna Ammari Siala. Furthermore, a wallet involved in the $250,300 transaction is listed by Etherscan as a MoonPay wallet.

Read more of this story at Slashdot.

Cognition AI Buys Windsurf as AI Frenzy Escalates

Par :msmash
14 juillet 2025 à 18:50
Cognition AI, an artificial intelligence startup that offers a software coding assistant, said on Monday that it had bought rival Windsurf as part of an escalating battle to lead in the technology. From a report: The move follows a $2.4 billion deal by Google to acquire some of Windsurf's top executives and license the start-up's technology, which was revealed on Friday. Google's deal appeared to leave Windsurf in a difficult position as a stand-alone start-up. OpenAI, the maker of the ChatGPT chatbot, had also been in talks to buy Windsurf before the Google deal. "We've long admired the Windsurf team and what they've built," said Scott Wu, a co-founder of Cognition, in an email to employees viewed by The New York Times. "Within our lifetime, engineers will go from bricklayers to architects, focusing on the creativity of designing systems rather than the manual labor of putting them together."

Read more of this story at Slashdot.

Air India Chief Says Preliminary Crash Report Raises Fresh Questions

Par :msmash
14 juillet 2025 à 18:10
Air India's chief executive urged staff to avoid drawing premature conclusions about what caused one of the airline's Boeing triangle jets to crash last month, after a preliminary investigation ruled out mechanical or maintenance issues, turning attention to the pilots' actions. WSJ: Campbell Wilson told staff that the probe into the crash was "far from over," according to an internal memo, reviewed by The Wall Street Journal, in which he set out some of the findings of a report issued by India's Aircraft Accident Investigation Bureau at the end of last week. Wilson's memo didn't mention one of the AAIB's findings: that the airplane's fuel-control switches had been turned off one by one, seconds after takeoff, starving both engines of fuel. The switches, which sit between the two seats in the cockpit, were turned back on about 10 seconds later, but the engines apparently couldn't fully restart and gain thrust fast enough, the report said. The crash of the London-bound Boeing 787 Dreamliner killed all but one of the 242 passengers and crew on board, as well as 19 people on the ground, when the plane slammed into a residential area beyond the airport in the Indian city of Ahmedabad. In the memo, Wilson said "over the past 30 days, we've seen an ongoing cycle of theories, allegations, rumours and sensational headlines, many of which have later been disproven."

Read more of this story at Slashdot.

Quality of Scientific Papers Questioned as Academics 'Overwhelmed' By the Millions Published

Par :msmash
14 juillet 2025 à 17:36
A scientific paper featuring an AI-generated image of a rat with an oversized penis was retracted three days after publication, highlighting broader problems plaguing academic publishing as researchers struggle with an explosion of scientific literature. The paper appeared in Frontiers in Cell and Developmental Biology before widespread mockery forced its withdrawal. Research studies indexed on Clarivate's Web of Science database increased 48% between 2015 and 2024, rising from 1.71 million to 2.53 million papers. Nobel laureate Venki Ramakrishnan called the publishing system "broken and unsustainable," while University of Exeter researcher Mark Hanson described scientists as "increasingly overwhelmed" by the volume of articles. The Royal Society plans to release a major review of scientific publishing disruptions at summer's end, with former government chief scientist Mark Walport citing incentives that favor quantity over quality as a fundamental problem.

Read more of this story at Slashdot.

Google Plans To Combine ChromeOS and Android Into Single Platform

Par :msmash
14 juillet 2025 à 16:45
Google will merge ChromeOS and Android into a unified platform, according to Sameer Samat, President of Android Ecosystem at Google. "We're going to be combining ChromeOS and Android into a single platform, and I am very interested in how people are using their laptops these days and what they're getting done," Samat said during a recent interview.

Read more of this story at Slashdot.

Zuckerberg Pledges Hundreds of Billions For AI Data Centers in Superintelligence Push

Par :msmash
14 juillet 2025 à 16:05
Mark Zuckerberg said on Monday that Meta would spend hundreds of billions of dollars to build several massive AI data centers for superintelligence, intensifying his pursuit of a technology that he has chased with a talent war for top AI engineers. From a report: The social media giant is among the large technology companies that have chased high-profile deals and doled out multi-million-dollar pay packages in recent months to fast-track work on machines that can outthink humans on most tasks. Unveiling the spending commitment in a Threads post on Monday, CEO Zuckerberg touted the strength in the company's core advertising business to support the massive spending that has raised concerns among tech investors about potential payoffs. "We have the capital from our business to do this," Zuckerberg said. He also cited a report from a chip industry publication Semianalysis that said Meta is on track to be the first lab to bring online a 1-gigawatt-plus supercluster, which refers to a massive data center built to train advanced AI models.

Read more of this story at Slashdot.

❌