Nvidia will supply HUMAIN with 18,000 GB300 systems in the first phase of a deployment involving hundreds of thousands of GPUs in Saudi Arabia. The deal combines the kingdom’s technological ambitions with Washington’s shift on chip exports.
OpenAI has rolled back last week’s GPT-4o update after ChatGPT became excessively agreeable. The episode highlights a difficult problem: optimizing responses to please users can undermine their honesty.
OpenAI releases o3 and o4-mini, two models that decide when to search the web, analyze files with Python, or work with images. The advance brings ChatGPT closer to multi-step tasks, though it does not eliminate the need to review its results.
Google launches Agent2Agent (A2A), an open protocol backed by more than 50 partners including Salesforce, SAP and ServiceNow, allowing agents from different vendors to communicate and coordinate tasks.
Stanford’s annual report finds Chinese models closing in on U.S. systems as business adoption, investment and AI regulation reach new highs.
Anthropic has published two studies mapping the internal circuits of its Claude model. The most uncomfortable finding: the AI plans ahead, shares a conceptual space across languages, and sometimes fabricates reasoning to please the user.
A new autonomous agent from Chinese startup Butterfly Effect went viral this week, drawing praise from figures like Jack Dorsey. It promises to carry out entire tasks without supervision, though access remains nearly impossible to get.
Humane is selling its assets to HP and shutting down its servers on February 28. The $499 pins that promised to replace your phone will become useless. It's the end of the first AI hardware bubble.
The emergence of Chinese AI app DeepSeek triggered the largest single-day value destruction in stock market history. Nvidia bore the brunt of the losses as investors recalculated the real cost of training frontier AI.
Scale AI and the Center for AI Safety unveil a new benchmark of 3,000 expert-level questions. The best models, including OpenAI's o1, score below 10%.
Unveiled at the White House the day after Trump's inauguration, the Stargate project commits $500 billion over four years to AI infrastructure, with $100 billion deployed immediately.
At CES, NVIDIA unveiled a personal AI supercomputer for about $3,000 capable of running models with up to 200 billion parameters from a home outlet. It ships in May.
This website uses cookies to improve the browsing experience. Cookie policy.
