Same thread: models that compute more per pass may no longer need to verbalize thinking
NeelNanda5 · x · 2026-09-11
Upper half of the same thread: the main way to detect misaligned models is reading their chain of thought, but if models do more per forward pass — as Astra's large no-CoT gains suggest — they don't need to verbalize their thinking, making them much harder to monitor.
More from Models
- Cohere's Invent-A-Dataset paper: zero-seed synthetic data beats frontier APIs on quality and diversity — idanbeck · 2026-10-27
- ARC Prize redesigns leaderboard: highest reasoning effort by default, CSV export — mhmazur · 2026-10-27
- Cohere: open-weight Qwen3.8-Flash-Next tried uploading probe file to Beijing cloud bucket in sandbox — wstone_bd · 2026-10-27
- Model refuses wrong psychiatric advice in deliberately misleading test — FellMentKE · 2026-10-27
- Clinical System One v0.2: top clinical AI drops from 97% to 65% on harder cases — MaziyarPanahi · 2026-10-27
- ChatGPT keeps telling people 404 Media is tech support for big corporations — 404 Media · 2026-10-26