Game developers need upscaling, AI characters and debugging tools to work on Nvidia's new Arm-based RTX Spark laptops and small desktops.

Nvidia brought DLSS 5, a small on-device model for in-game AI characters, neural rendering kits and Arm-ready Nsight debugging tools to the platform.

Speculative decoding, where a small model guesses words for a big one to check, is only as fast as the draft model is good.

AMD trained a draft model from scratch for a large mixture-of-experts model, using the big model's own outputs, entirely on AMD GPUs, and packaged it as reusable recipes.

General models answer local questions in the right language but get local procedures wrong, such as how to apply for a passport if you cannot travel.

Synthetic resident profiles based on census data, paired with official procedure text, generate training chats to fine-tune a Nemotron model.

Answers in big public Discord servers are hard to find from the web.

Server web pages from November 16, with some public messages appearing from mid-December, plus admin and user settings to opt out now.