5141 articles across all categories
Latent Space · Apr 3
Google DeepMind released Gemma 4, a 31B dense multimodal model that ties performance with much larger competitors (Kimi K2.5, GLM-5) while using far fewer parameters. Native video, image, and audio su
Axios · Apr 3
California is escalating AI regulation on multiple fronts—Gov. Newsom signed an executive order this week while state legislators advance multiple AI bills—positioning itself as the de facto national
Axios · Apr 3
Chevron and Microsoft are building a dedicated natural gas plant for a Texas data center—a sign that on-site power is rapidly replacing grid connections for AI infrastructure. Roughly 30% of planned d
CNBC Tech · Apr 3
Microsoft pledges $10 billion for Japan AI infrastructure buildout through 2029, partnering with SoftBank—a significant regional investment that sparked a 20% rally in local datacenter operator Sakura
CNBC · Apr 3
Chinese chip firms posted record revenue as AI demand surges domestically and U.S. export restrictions cut them off from Western markets, forcing a strategic pivot toward self-sufficiency in AI silico
NY Times · Apr 3
Economists are reversing their long-held skepticism about AI's ability to displace workers, signaling a fundamental shift in how the profession understands generative AI's labor market impact.
MarkTechPost · Apr 3
Step-by-step guide to building a complete model optimization pipeline with NVIDIA Model Optimizer, covering ResNet training on CIFAR-10, FastNAS pruning for complexity reduction under FLOPs constraint
DeepMind · Apr 2
Google DeepMind releases Gemma 4, claiming their most capable open-source models yet, optimized for advanced reasoning and agent workflows. Positions Gemma as a competitive alternative to closed-sourc
Engadget · Apr 2
Google released Gemma 4, a family of four open-weight models (2B to 31B parameters) claiming unprecedented intelligence-per-parameter efficiency. The 31B and 26B variants rank 3rd and 6th on Arena AI'
Meta AI · Apr 2
Meta's Ranking Engineer Agent now includes KernelEvolve, an autonomous system that generates and optimizes GPU/CPU kernels for different hardware without manual engineering. The agent automatically tu
Tom's Guide · Apr 3
Seven prompt engineering strategies—including few-shot prompting and persona setting—that shift Claude from generic chatbot to expert-level tool, cutting draft revision time by 10 hours weekly.
Latent Space · Apr 2
Moonlake AI is building interactive, multiplayer world models bootstrapped from game engines—enabling indefinite simulation lifetimes and physics-accurate interactions, contrasting sharply with single
Mashable · Apr 2
Google released Gemma 4, an open-source LLM licensed under Apache 2.0 that can run on billions of Android devices and consumer GPUs, giving developers complete control over their data and deployment e
Ars Technica · Apr 2
Researchers demonstrate two novel Rowhammer attacks (GDDRHammer and GeForceHammer) that exploit Nvidia GPU memory to gain full root control of host machines, turning a decade-old DRAM vulnerability in
Bloomberg Tech · Apr 3
Microsoft pledged $10 billion over four years to expand AI infrastructure and services in Japan, signaling intensified competition for regional cloud dominance as tech giants race to capture demand fo
Google Security Blog · Apr 2
Google's GenAI Security Team details its continuous defense strategy against indirect prompt injection (IPI) attacks, where malicious instructions injected into data or tools compromise LLM behavior w
Fortune · Apr 2
Mercor, a $10B startup supplying training data to OpenAI, Anthropic, and Meta, confirmed a data breach via supply chain attack on LiteLLM, an open-source AI library. Extortion gang Lapsus$ claims 4TB
MarkTechPost · Apr 2
Arcee AI released Trinity Large Thinking, a 400B-parameter sparse MoE reasoning model under Apache 2.0 license optimized for autonomous agents and multi-turn tool use. The model activates only 13B par
LangChain · Apr 2
LangChain's evals show open-weight models like GLM-5 and MiniMax M2.7 now match closed frontier models on core agentic tasks (file ops, tool use, instruction following) while costing 8-10x less and ru
arXiv ML · Apr 3
Researchers trained LLMs to forecast supply chain disruptions with calibrated probabilistic outputs, outperforming GPT-5 on accuracy and calibration. The team open-sourced their evaluation dataset to