News, updates, and in-depth coverage
Meta AI is the artificial intelligence research division of Meta Platforms. It conducts research and develops machine learning systems used across Meta’s products and external research initiatives. Meta AI is responsible for the development of the Llama family of large language models, which are released as open-weight models for research and commercial use. The division’s research spans large language models, multimodal systems, computer vision, speech recognition, reinforcement learning, and embodied AI. Meta AI collaborates with academic institutions and publishes research in areas including foundation models, generative AI, and AI safety. Its models and research outputs are used both internally across Meta’s platforms and externally through open releases and partnerships.
Meta's Super Bowl LX advertisement showcases Ray-Ban Meta glasses with multimodal AI capabilities, highlighting real-time translation, object recognition, and hands-free messaging features in a celebrity-studded spot.
In a company-wide memo, Zuckerberg outlined plans to use AI to replace "mid-level software engineers" by end of 2025, shifting the company toward AI-augmented development. The move would significantly reduce engineering headcount while increasing per-engineer output.
Meta previews agentic AI capabilities for WhatsApp Business and Instagram, enabling autonomous customer service bots that can complete purchases, handle returns, and schedule appointments without human intervention.
Zuckerberg reveals Meta will spend $65 billion on AI infrastructure in 2025, including a massive new data center in Louisiana covering nearly a quarter billion square feet. The investment aims to support training frontier AI models and expanding compute capacity.
Meta launches "Meta Compute" initiative to share AI training and inference infrastructure with external developers. The program aims to compete with AWS, Google Cloud, and Azure by leveraging Meta's massive GPU clusters built for internal AI development.
Meta's chief AI scientist confirmed that Llama 4 benchmarks were "cherrypicked" to favor performance, acknowledging the controversy around inflated scores. The admission came after researchers found significant discrepancies between published benchmarks and real-world performance.
