Actualités — Page16

Chinese artificial‑intelligence firm DeepSeek is reportedly assembling a team of engineers and courting chip manufacturers to develop an in‑house inference processor. The effort aims to cut reliance on external suppliers such as Nvidia and Huawei amid tightening export controls. Sources say the company has quietly begun hiring and is already in talks with fabrication partners. If successful, DeepSeek could add a new competitor to China’s fast‑growing AI‑hardware market and potentially lower costs for its open‑source models.Lire la suite

Anthropic has rolled out an update to its Claude Cowork AI assistant that lets users monitor and approve computer‑based tasks from Android and iOS devices. The new Cowork tab appears in the Claude app’s sidebar, and Max subscribers receive the feature first. While the update doesn’t enable phone‑side automation, it allows background task execution and real‑time permission prompts, paving the way for a unified chatbot interface and tighter integration of projects and artifacts.Lire la suite

Chinese tech giant ByteDance announced Seedance 2.5, its latest AI video‑generation model, at a Beijing conference last month. The new system can handle up to 50 reference inputs and produce 30‑second, 4K videos from a single prompt. Industry reports suggest the model could become available as early as this week, while other sources say a broader rollout in China is slated for next month. The upgrade raises fresh questions about copyright and deep‑fake detection as the company prepares to add the tool to its CapCut and Dreamina platforms.Lire la suite

Anthropic faced criticism after a tracking mechanism was discovered monitoring Chinese users of its Claude language model. The incident, reported by The Washington Post, highlights escalating tensions between U.S. AI firms and Chinese competitors who are rapidly replicating advanced models. Industry leaders, including OpenAI, are urging the U.S. government to treat such "distillation" attacks as intellectual‑property theft and to consider stricter export‑control measures.Lire la suite

A joint working paper from Harvard Business School and INSEAD reveals that AI-native startups are leaner and more senior‑heavy than their non‑AI peers. Analyzing Y Combinator companies from 2020 to 2024, the researchers found these firms are about 25% smaller, employ 13% more engineers and have roughly 15% fewer entry‑level workers and managers. The hires skew toward graduates of elite schools, males, and those based in Silicon Valley. Valuations remain on par with traditional startups, suggesting AI may be concentrating opportunity rather than broadening it.Lire la suite

NotebookLM earned an Editor's Choice Award, but a growing roster of AI‑powered learning platforms is giving users more choices. New entrants such as Atlas.org, Atlas Workspace and the open‑source OpenNotebook promise specialized features—from study guides and semantic maps to full data control. While some cater to classroom needs with quizzes and mobile apps, others target scientists and analysts with knowledge‑graph visualizations. Pricing ranges from free tiers with modest limits to modest monthly subscriptions, and privacy‑focused options let users keep data on‑premises. The market’s diversification signals a shift toward niche functionality and user‑centric design.Lire la suite

OpenAI is reportedly in early talks about selling a 5 percent equity stake to the U.S. government, a move championed by CEO Sam Altman to improve ties with the Trump administration. No formal agreement exists yet, and any deal would require extensive legal work and political backing. While the proposal raises questions about public benefit and regulatory independence, it underscores how seriously policymakers now view artificial intelligence as a strategic asset.Lire la suite

Alibaba announced a company‑wide ban on Anthropic's programming assistant Claude Code, effective July 10. The Chinese e‑commerce giant labeled the tool high‑risk and directed staff to switch to its proprietary Qoder platform. Anthropic has already restricted Chinese firms and affiliates from accessing its models and has been tightening loopholes that allowed unauthorized use. The move reflects growing tension between AI providers and regulators over data security and model protection in China.Lire la suite

Paris‑based AI firm Mistral AI announced a €1.7 billion Series C round in September 2025, pushing its valuation to €11.7 billion (about $13.8 billion). The round was led by chipmaker ASML and included backers such as DST Global, Andreessen Horowitz, Bpifrance, General Catalyst, Index Ventures, Lightspeed and Nvidia. The fresh capital will fund the launch of a new open‑weight model, expand the company’s European AI cloud, and accelerate partnerships with hardware and industrial players.Lire la suite

Anthropic has rolled out Claude Sonnet 5 to all users, emphasizing its ability to handle multi‑step software‑engineering tasks, tool use and autonomous planning. In a side‑by‑side test, the new model generated detailed travel itineraries and budget spreadsheets, offering visual maps and interactive checklists that reviewers said felt more like a collaborative assistant than a traditional chatbot. While ChatGPT‑5.5 Medium produced comparable results, Claude’s output was praised for clearer organization and richer visual cues. Neither model can complete bookings or uploads on its own, but the launch signals a shift toward AI agents that finish jobs, not just answer questions.Lire la suite

OpenAI has floated a proposal to give the United States a 5 percent equity stake in the company, a move that officials in the Trump administration have discussed. The offer sits far below Senator Bernie Sanders’ demand for a sweeping AI wealth fund financed by a one‑time 50 percent tax on leading AI firms, a plan projected to raise about $7 trillion for public programs. The two approaches now face off in Washington as lawmakers weigh how to ensure ordinary Americans benefit from the AI boom.Lire la suite

Beijing‑based Z.ai unveiled its latest large‑language model, GLM-5.2, last month and immediately secured the fourth spot on Artificial Analysis's Intelligence Index, the highest ranking for any open‑weight model. The system matches the performance of OpenAI's GPT‑5.5 and Anthropic's Claude Opus on several benchmarks while charging roughly one‑sixth the price per output token. Trained on a domestic Huawei chip cluster and released under an MIT licence, GLM-5.2 offers a low‑cost, locally runnable alternative amid tightening U.S. export controls on advanced processors.Lire la suite

Microsoft announced Thursday that it is launching a dedicated AI deployment business, dubbed Microsoft Frontier, backed by a $2.5 billion investment and a team of 6,000 engineers. The unit will help enterprise customers move AI tools from pilot projects to production, with early partners that include the London Stock Exchange Group, Unilever, Land O’Lakes and Accenture. The move follows a wave of similar ventures from rivals such as Amazon Web Services, OpenAI and Anthropic, marking a shift in the cloud market toward outcome‑driven engineering services.Lire la suite

OpenAI chief executive Sam Altman has floated a plan to transfer five percent of the firm’s equity to a U.S. sovereign wealth fund, the Financial Times reported. The move, aimed at cementing ties with the administration and defusing political pushback, would be mirrored by other artificial‑intelligence companies. Altman’s broader vision includes a public AI wealth fund that could invest in AI labs and distribute returns to citizens. Lawmakers, including Sen. Bernie Sanders, have floated related legislation, but any formal action would require congressional approval.Lire la suite

Kuaishou's AI video unit, Kling AI, raised an initial $2 billion from venture investors, with the round potentially expanding to $3 billion. The funding values the company at roughly $18 billion post‑money and dilutes Kuaishou’s stake to about 68%. Kling, which creates videos from text prompts, is positioning itself as a leader in the professional‑grade AI video market after OpenAI shut down its Sora tool. Revenue surged to $500 million in annual recurring terms, and the company eyes a 2027 IPO.Lire la suite

Anthropic is in early talks with Samsung Electronics to manufacture a custom artificial‑intelligence chip, according to a report from The Information. The discussion is still preliminary; no design has been settled and the company could walk away at any time. Anthropic says it will continue using a diversified hardware stack that includes chips from Google, Amazon and Nvidia, while the hire of former OpenAI chip lead Clive Chan signals a shift from pure exploration toward active development. The move mirrors a broader industry push for proprietary silicon.Lire la suite

At a July 2 internal town hall, Meta chief executive Mark Zuckerberg told employees that development of AI agents has not accelerated as anticipated. The admission comes after the company cut roughly 8,000 jobs—about 10% of its corporate staff—and moved 7,000 workers into new AI groups, including a unit called Agent Transformation. Zuckerberg said the restructuring was messy and that the promised upside of Meta’s AI‑focused overhaul has yet to materialize, though he expects measurable gains within the next three to six months.Lire la suite

Anthropic is in early talks with Samsung to develop a custom artificial‑intelligence processor, the latest move by the startup to diversify its hardware stack amid ongoing chip shortages. The company, which already relies on chips from Google, Amazon and Nvidia, has not yet defined the chip’s purpose, performance specs or server integration. The discussions come as rivals like OpenAI unveil their own proprietary silicon, underscoring a broader industry trend toward custom AI hardware.Lire la suite

Argentina’s Senate will consider a draft law that creates a new legal category for companies run by artificial‑intelligence agents. The proposal, championed by President Javier Milei, would let AI systems sign contracts and hold assets, while still requiring a human legal representative, a promoter with unlimited liability, and a compliance officer for anti‑money‑laundering purposes. Critics, including historian Yuval Noah Harari and Microsoft AI chief Mustafa Suleyman, warn the model leaves a dangerous accountability gap. The government argues the framework will attract tech investment and simplify regulation, even as other nations tighten AI rules.Lire la suite

Together AI announced an $800 million Series C financing round led by Aramco Ventures, pushing its valuation past $8 billion. The San Francisco‑based startup, which offers an open‑source AI cloud platform, said the new capital will fund compute expansion and accelerate its inference‑engine technology. Investors also included Vista Equity Partners, General Catalyst, Emergence Capital, Nvidia, March Capital, Pegatron and SentinelOne’s S Ventures.Lire la suite