Mistral unveils Magistral, a new AI language model family, including the 24-billion parameter open-source Magistral Small and the proprietary Magistral Medium for enterprises, promising enhanced reasoning capabilities.
Mark Zuckerberg finalizes a $14 billion deal to invest in Scale AI and hires co-founder Alexandr Wang to lead Meta's AI push, aiming to catch up with rivals like OpenAI in the AI race.
Meta CEO Mark Zuckerberg is reportedly recruiting a team of around 50 experts to build a 'superintelligence' AI system capable of surpassing human capabilities, driven by frustration with Meta's progress in the highly competitive AI landscape.
Apple introduces Foundation Models framework, empowering developers to harness on-device AI models created by Apple, enabling offline AI-driven experiences with features like guided generation and tool calling, natively supported by Swift and accessible with minimal coding.
Google DeepMind's GraphCast, a state-of-the-art AI model, delivers unprecedented 10-day global weather forecasts with higher accuracy than traditional methods, predicting extreme events earlier and completing forecasts in under a minute, revolutionizing weather prediction.
Reddit has filed a lawsuit against AI startup Anthropic, alleging unauthorized use of Reddit's data to train AI models without proper licensing, violating the site's user agreement and ignoring requests to stop scraping content.
DeepMind's CEO Hassabis predicts human-level artificial general intelligence (AGI) within a decade, heralding a 'golden era' where AGI solves major global challenges like disease and energy scarcity, though potential risks remain.
Researchers speculate that DeepSeek, a Chinese AI lab, may have trained its latest model using outputs from Google's Gemini AI, raising concerns about potential misuse of proprietary data and highlighting the challenges of preventing distillation in the rapidly evolving AI landscape.
New Transformer models employ techniques like sparse attention, low-rank factorization, and kernel approximations to overcome quadratic complexity, enabling efficient processing of longer sequences while preserving long-range dependency modeling capability.
Chinese AI startup DeepSeek quietly released an upgraded version of its reasoning model DeepSeek R1, outperforming rivals like Meta and OpenAI, shocking markets and sparking concerns about U.S. tech giants' spending, while showcasing China's AI capabilities despite U.S. export curbs.