Noam Brown: GPT-6 Astra does have observable chain of thought, calls it fragile
burny_tech · x · 2026-09-19
- OpenAI research scientist Noam Brown confirmed on AI Deep Dive that GPT-6 Astra does have an observable chain of thought.
- He called chain-of-thought monitoring "a real gift. We were very lucky that this ever existed, and it is fragile," signaling that readable CoT may not persist.
- The remark matters for AI safety research that relies on monitoring models' chain of thought.
More from Models
- Redditor tests typesafeAI's new classifier Jev: roughly GPT-1-level smart, with a trick to gauge its knowledge cutoff — MrCyclopede · 2026-09-19
- RSI-Exam leaderboard update: GPT-6-astra tops recursive self-improvement benchmark at 0.5126 — yuyinzhou_cs · 2026-09-19
- Claude Max plan hit with usage complaints: quota exhausted after just four hours — yangyi · 2026-09-19
- Grok Bot voice mode now fully rolled out on mobile — Daniel_Farinax · 2026-09-19
- AfriqueQwen 3.5 post-trained models released, claiming wins over Gemma 3 and Apertus — davlanade · 2026-09-19
- Karpathy's year-old rant calling AI agents "slop" resurfaces as agent hype rolls on — steipete · 2026-09-19