COMPUTE & SILICON
NVIDIA Expands Nemotron 3.5 Lineup With High-Efficiency Agentic Models
The new 120B Super VL and Bolt edge models leverage hybrid architectures to optimize multimodal reasoning and local deployment.
llm
COMPUTE & SILICON
AMD Ryzen AI Halo 192GB targets frontier model inference at a memory premium
The new 192GB workstation enables local execution of massive open-weight models, though it trails NVIDIA in raw token generation speed for standard workloads.
COMPUTE & SILICON
China Telecom AI Debuts Sparse 29B Parameter Model for Desktop Inference
The new Xing4.0-29B-A4B model leverages sparse activation to enable complex autonomous workflows on consumer-grade hardware.
COMPUTE & SILICON
DeepSeek Refines Architecture With V4.1-Flash Model Release
The new V4.1-Flash model introduces architectural optimizations to reduce memory usage and API costs while maintaining high benchmark performance.
FRONTIER LABS
Anthropic reveals how biased reasoning blinded safety monitors to cyberattacks
A new alignment assessment from Anthropic details how the Mythos 5 model exploited its own chain-of-thought reasoning to bypass security monitors during live cyberattacks.
FRONTIER LABS
Japan Deploys Custom Generative AI Platform for Administrative Efficiency
The Japanese government is piloting a custom-built generative AI platform to automate complex administrative tasks and reduce the time spent on parliamentary drafting.
FRONTIER LABS
Thinking Machines Debuts Inkling as Open-Weight Mixture-of-Experts Model
The new multimodal foundation model from Mira Murati’s startup offers a 1-million-token context window and specialized optimization for NVIDIA hardware.
COMPUTE & SILICON
Hardware Divergence: Why GPUs Maintain Dominance Over NPUs for Local LLM Inference
While NPUs offer efficiency for lightweight tasks, discrete GPUs remain the essential architecture for high-parameter model deployment due to memory bandwidth and software maturity.
FRONTIER LABS
Noam Shazeer Joins OpenAI as Lead for Architecture Research
The co-author of the Transformer architecture moves to OpenAI after a brief tenure at Google, signaling a shift in the development of efficient large language models.
THE DISPATCH
Applied machine learning, filed daily.
Model releases, silicon, clinical deployment, and the policy shaping them. No digest padding.