Meta's TLX Attention Kernel Outperforms FlashAttention-4 on Blackwell GPUs
Meta disclosed a Triton-based attention kernel called Jagged Flash Attention (JFA), built with Triton Low-level Extensions (TLX), that outperforms FlashAttention-4 (May 2026 version) on NVIDIA B200 GPUs for the variable-length sequences use…
Lambda Outlines Path From World Models to Active Physical AI Agents
Lambda published a technical blog post describing a progression from generative AI toward physical AI, stating that emerging world models are beginning to capture geometry, motion, and interaction beyond the photorealistic scene synthesis a…
CoreWeave Adds Interactive Notebooks to Forge Platform
CoreWeave Notebooks are now available in CoreWeave Forge, providing reactive Python notebooks for interactive AI development . The notebooks connect to users' existing data and persist within projects, enabling team collaboration .…
Cloudflare Adds EuroLLM, Swiss Apertus Models to Workers AI
Cloudflare announced it is bringing two publicly developed multilingual models to its Workers AI platform: EuroLLM, which supports all 24 official EU languages, and Apertus, Switzerland's fully open model trained on more than 15 trillion to…
ALL BRIEFS
RANKED · 5 THIS DAYProduction failures become training data as fine-tuning shifts to live traces
LangChain's LangSmith Fine-Tuning entered public beta with training running on Baseten Loops, joining Shopify's daily retraining flywheel and a growing set of tools that treat live deployment artifacts as the primary signal for model improv…