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.
More from Models
- Testing unreleased Gemini 3.8 Flash: no citations shown for top-of-funnel queries — gaganghotra_ · 2026-09-03
- Hidden-bug eval across 105 issues: Fable 5.1 finds 43, none fixes all — cost per model compared — PawelHuryn · 2026-09-03
- X open-sources new For You algorithm code: long dwell drives retrieval, bots can trigger account review — Kyrannio · 2026-09-03
- Gemini 3.8 Flash reverse-engineers Kerbal save files to build and land a Mun rocket — dosco · 2026-09-03
- User calls out model for double-standard answers on gendered scenario questions — Ribbitz_bow_tie27 · 2026-09-03
- Redditor Predicts Astra Model Release Tomorrow at 1pm PT Based on X Teasers — dolo937 · 2026-09-03