DeepMind's April 2026 paper diagnoses transformers as depth-bounded TC^0 circuits, pointing toward a continuous-depth ODE-based recurrent architecture with O(1) training memory.
Google Deep Research Max costs $4.80/report and uses MCP to connect to private data stores. Independent 7-task testing shows the cheaper tier wins 5 of 7.
Google DeepMind's Decoupled DiLoCo trains a 12B Gemma model across four US regions with mixed TPU generations and self-healing failure recovery.
Google DeepMind's Vision Banana shows image generation models can perform segmentation, depth, and normal prediction at SOTA without architectural changes.
Curated AI insights — sent when there's something worth your inbox.