Bend promises GPU-class parallelism without a single thread, lock, or mutex in your code. Two years on, it's still a thought experiment worth taking seriously.
llm-d 1.0 brings production-ready open source distributed inference — cheaper agentic serving, zero-downtime upgrades, 3,000 concurrent agents.
Most AI agents forget. Nous Research's Hermes Agent learns on the job — saving skills from finished sessions, searching its own history, and remembering who you are.
After a decade of feature requests, OpenTofu 1.12 lets a variable guard production resources from deletion. Here's why that matters for Terraform adopters.
Python 3.15 lands with lazy imports, frozendict and sentinel built-ins, UTF-8 by default, and a profiler you can attach to a live process.
Every LLM weight squeezed into -1, 0, or +1: the 1.58-bit model story, from the 2024 paper to bitnet.cpp's 100B-on-one-CPU trick.