In-depth analysis of the trends, breakthroughs, and decisions shaping artificial intelligence. Published every Monday.
As context windows scale past 1M tokens, we analyze how this changes RAG architectures, agent workflows, and the economics of inference. Plus: benchmarks that actually matter for long-context performance.
A survey of 50+ companies deploying AI agents in production environments. We break down the patterns that succeed, the failure modes to watch for, and why tool-use reliability is the real bottleneck.
Cutting through the legal jargon — a practical guide to how the EU AI Act's first enforcement phase affects model providers, deployers, and open-source contributors. Includes a compliance checklist.
Open-weight models are closing the gap with proprietary ones faster than anyone predicted. We map the ecosystem — who's funding it, where the gaps remain, and what it means for the industry's power dynamics.
Recent papers suggest the compute-performance curve may be bending. We examine the evidence, the counter-arguments, and what alternative scaling strategies (MoE, test-time compute, data quality) offer.
Video understanding, audio generation, and 3D reasoning are the new frontier. We survey the latest multimodal architectures and explore which modality combinations unlock genuinely new capabilities.