FlashAttention-4 retools attention for Blackwell GPUs by maximizing on‑chip reuse and alleviating memory bottlenecks, delivering petaflop‑class throughput and meaningful speedups for long‑context LLM inference.
The all‑electric Mercedes‑Benz CLA earned a five‑star Euro NCAP rating and was named 2025’s overall Best Performer, underpinned by NVIDIA’s DRIVE AV software on DRIVE AGX integrated with MB.OS. Here’s what the scores mean, how the dual‑stack safety architecture works, and why it matters for OEM strategies.
NVIDIA DRIVE AV Mercedes‑Benz CLA Tops Euro NCAP 2025 With Best‑in‑Class Safety Read Post »
NVFP4 on Blackwell-class data center GPUs slashes memory footprints and boosts throughput for FLUX.2, while preserving accuracy with mixed precision and scaling cleanly to rack-level NVL72 systems.
Scaling NVFP4 Inference on NVIDIA Blackwell GPUs for FLUX.2 Read Post »
Anthropic’s terminal-based coding agent has moved from experiment to core infrastructure, with reports of rapid adoption, major revenue momentum, and a shift toward agentic development workflows.
How Claude Code Developer Productivity Is Reshaping Software—and Anthropic Read Post »
Google DeepMind struck a licensing deal with Hume AI and hired CEO Alan Cowen and key engineers. The move brings emotion-aware voice capabilities to Gemini and signals a broader push toward empathetic AI interfaces.
Google licensing Hume AI technology: DeepMind taps emotion-aware voice talent Read Post »
Reports say Meta attempted to recruit researchers from Mira Murati’s Thinking Machines Lab with unprecedented pay packages—including an alleged $1 billion offer—but none have accepted. Here’s what’s known, what’s contested, and why it matters.
Inside the Thinking Machines Lab talent poaching fight Read Post »
