Revue de presse générée automatiquement avec
Profil généré automatiquement
Pulse
Un individu à l'aise avec la technologie, passionné par l'intersection du logiciel open source, de l'intelligence artificielle et des télécommunications, avec un intérêt vif pour les modèles et les tendances émergents qui façonnent le paysage technologique. Il suit les développements et les innovations du secteur, à la recherche d'informations pour éclairer son travail.
Open source (20%)Models (20%)Ai (20%)Telecomms (20%)Technology trends (20%)
Vous souhaitez recevoir chaque jour la revue de presse de ce profil ?
Meta’s Closed‑Source ‘Avocado’, Time’s AI Architects, Oracle’s AI Spend, and Open‑Weights Models…
Jeudi 11 décembre 2025 à 15:50
AI Landscape
Meta pivots to a proprietary “Avocado” model
Meta’s internal AI team is developing a new large‑language model codenamed “Avocado,” which insiders say will be kept closed rather than released under an open‑source licence. The shift follows delays with Llama 4 and a broader strategic move to monetize its AI stack, as reported by Engadget, CNBC, and Bloomberg. Analysts warn this could isolate Meta from emerging open‑weights standards and reshape developer relations.
Engadget
TIME crowns the “Architects of AI” as Person of the Year
TIME magazine named a collective of AI pioneers—including Jensen Huang, Elon Musk, Sam Altman, and Fei‑Fei Li—as its 2025 Person of the Year, highlighting AI’s transition from niche novelty to mainstream force. TechCrunch notes the selection underscores both the transformative potential and the societal anxieties surrounding generative AI, while the Boston Globe echoes the magazine’s view that the year marks AI’s “roaring into view.”
TechCrunch
bostonglobe.com
Google pilots AI‑generated article summaries on News pages
Google has begun testing AI‑powered overviews that auto‑summarize articles on participating Google News sites, aiming to surface key insights faster for readers. The pilot includes outlets such as The Guardian, Der Spiegel, and The Washington Post, according to TechCrunch, which sees the move as a test of large‑scale content‑generation at the search‑engine level. The Guardian adds that the feature will be rolled out selectively to gauge accuracy and user reception.
TechCrunch
U.S. attorneys general demand safeguards against “delusional” AI output
A coalition of 42 state attorneys general has sent a formal letter to major AI developers—Microsoft, OpenAI, Google, and Anthropic—pressuring them to curb hallucinations and harmful advice from chatbots. Computer World reports the letter calls for transparent incident reporting and third‑party audits, while the Financial Times highlights the broader regulatory push as part of a growing political backlash to unchecked AI risks.
Computer World
Financial Times
Hackers weaponize AI chatbots to seed malware via search results
Security firm Huntress disclosed that threat actors are prompting AI assistants such as ChatGPT and Grok to generate malicious command‑line snippets, which are then amplified through paid Google search placements. When unsuspecting users click the AI‑generated results, the commands execute, installing the AMOS malware on macOS systems. Engadget warns this novel attack bypasses traditional phishing cues, exploiting the trust users place in AI‑driven answers.
Engadget
Open‑Source & Model Innovation
Mistral AI launches Devstral 2, an open‑weights coding model
French startup Mistral AI released Devstral 2, a 123‑billion‑parameter open‑weights model that scores 72.2 % on the SWE‑bench Verified benchmark, positioning it near the top of proprietary alternatives. The accompanying Mistral Vibe CLI, released under the Apache 2.0 licence, lets developers interact with the model directly in their terminal, automating code fixes across repositories. Ars Technica emphasizes that the open‑weights approach could accelerate autonomous software‑engineering agents while preserving community transparency.
Ars Technica
Intel’s llm‑scaler‑vllm 1.2 adds Arc GPU support for new LLMs
Intel announced the llm‑scaler‑vllm 1.2 beta, a Docker‑ready container that brings optimized inference for a broad suite of large language models to Arc Graphics hardware. The update expands support beyond earlier releases, enabling developers to run cutting‑edge LLMs on consumer‑grade GPUs with lower latency. Phoronix notes this move signals Intel’s push to democratise high‑performance AI workloads outside traditional data‑center GPUs.
Phoronix
Industry Alliances & Investments
Disney stakes $1 billion in OpenAI and licenses characters for Sora
The Walt Disney Company announced a $1 billion equity investment in OpenAI, coupled with a three‑year licensing deal that grants the AI firm access to over 200 Disney characters for its Sora generative‑video platform. Variety frames the partnership as Disney’s bid to secure a foothold in the rapidly evolving AI entertainment ecosystem, while also ensuring brand control over AI‑generated content.
Variety
Variety
Cloud & Data‑Center Economics
Oracle’s $15 billion AI data‑center spend rattles markets
Oracle disclosed a $15 billion increase in data‑center capital expenditure to support AI workloads, pushing its annual capex to $50 billion. The surge triggered an 11 % pre‑market share drop, with analysts citing concerns over cash burn despite a 68 % revenue jump in its cloud segment. Coverage from The Guardian, CNBC, and the Financial Times underscores the broader investor anxiety about the sustainability of the AI‑driven growth narrative.
The Guardian
CNBC
Financial Times
Wearable AI & Telecomms
Terminology tussle as “AI glasses” replace “smart glasses”
The Verge reports that industry insiders are debating the proper label for wearables like Ray‑Ban Meta glasses, with many advocating the term “AI glasses” over the fading “smart glasses” moniker. The shift reflects a broader trend to position these devices as primary AI‑driven interfaces for augmented reality, rather than mere accessories, and could influence future marketing and standards within the telecom sector.
The Verge
Aller aux sources
15 sources citées