News archive.
Earlier AI news and analysis, kept for reference. Current coverage lives in News & Analysis.
Gartner Names DevSecOps Platform Leaders for 2024
Gartner's latest Magic Quadrant ranks DevSecOps vendors on execution and vision. See which platforms lead and where the gaps are for your team.
Ad value shifts from views to actions as AI picks what you buy
McKinsey says advertising power moves to platforms controlling discovery and purchase decisions as AI agents reshape how consumers find products. What this means for your media spend.
NVIDIA Fused MoE Kernels Cut Training Time 8-93% for DeepSeek, GPT
NVIDIA's custom CuTe kernels fuse activation, quantization, and GEMM operations to eliminate GPU memory bottlenecks in mixture-of-experts training. Available now in cuDNN Frontend, Transformer Engine, and Megatron-Core.
Fine-tune protein models on one GPU, 1% of parameters, full accuracy
NVIDIA BioNeMo Recipes cut protein model adaptation from impractical to workstation-scale using LoRA. ESM2-3B hits 84.8% accuracy on secondary structure; Evo2-1B reaches 96.6% on splice sites with under 2% trainable parameters.
SoftBank deploys OpenAI cybersecurity tool in Japan amid US model restrictions
SoftBank is rolling out an OpenAI-powered cybersecurity offering in Japan as the US tightens access to competing AI models. What SoftBank's move signals about regional AI policy divergence.
Anthropic lost White House backing over safety stance, sources say
The Washington Post reports Anthropic faced pressure from U.S. government officials over AI safety positions. What happened and why it matters for enterprise AI adoption.
Ornn Launches Token Benchmarks for Claude and GPT Models
Index startup Ornn published benchmarks comparing Anthropic and OpenAI token costs and performance. Teams evaluating LLM providers now have a third-party reference point beyond vendor claims.
SoftBank launches cybersecurity product powered by OpenAI models
SoftBank announced a new cybersecurity offering built on OpenAI's frontier models. The product aims to help enterprises detect and respond to threats faster.
SoftBank launches AI patching service for Japan's critical infrastructure
SoftBank Group is deploying an automated patching service powered by OpenAI to secure critical infrastructure across Japan. The offering combines vulnerability detection with rapid patch deployment.
ALS patient uses brain implant to work, parent, email after 3,800 hours of independent use
Casey Harrell has spent nearly 3,800 hours controlling a speech BCI implant without researchers present, expanding his vocabulary to 125,000 words with 97.5% accuracy. The system now runs at 99% accuracy and lets him work, email, and control his computer.
South Korea ranks third in AI models while workers fear job loss
Only 16% of South Koreans worry more than excite about AI—lowest globally. But 64% fear displacement as government pushes AI-first policy over safety guardrails.
Data Centers Can Run Faster If They Flex Power Use, Study Finds
A Virginia facility will test whether data centers can dial down electricity during grid peaks. Duke research suggests the US grid has 76 gigawatts of spare capacity for flexible facilities.