The world is being quietly rearranged by people who write very long documents.


The title they went with Extending Puzzle for Mixture-of-Experts Reasoning Models with Application to GPT-OSS Acceleration Noisy translates that to

Researchers shrink reasoning AI models while keeping them smarter


Computer scientists reduced a massive language model's size by 26% while maintaining or improving its ability to solve complex problems by reorganizing how it processes information internally. This means AI reasoning systems could run on cheaper hardware and faster, which matters because these models currently consume enormous electricity and cost money every time someone asks them a question.
If AI reasoning models can deliver the same quality answers with 26% less computation, that directly reduces the infrastructure cost and energy consumption of deploying them at scale — the thing currently preventing widespread adoption of these systems.

If you insist
Read the original →