Sebastian Raschka debunks The Information's report on OpenAI Astra's looped transformer architecture

rasbt · x · 2026-09-03

Following The Information's report that OpenAI's Astra uses a "recurrent depth or looped transformer" architecture, Sebastian Raschka published a technical breakdown (with YouTube video), building on his article from two months ago on looped-depth architectures — separating what aligns with known research from overinterpretation of the rumor.

Original post →

More from Models

Models channel →