Decoupled DiLoCo: A new frontier for resilient, distributed AI training

OpenAI’s o1 correctly diagnosed 67% of ER patients vs. 50-55% by triage doctors

Accelerating Gemma 4: faster inference with multi-token prediction drafters

A couple million lines of Haskell: Production engineering at Mercury

Using “underdrawings” for accurate text and numbers

ProgramBench: Can language models rebuild programs from scratch?