Published at 11/12/2026, 5:22:00 AM · 104 views
📰 OpenAI Alignment Research: Smarter Models Are Getting Harder to Monitor Through Chain of Thought, as CoT Controllability Rises During RL on Astra-Class Models…
OpenAI published "What affects chain-of-thought monitorability across the model lifecycle?" on its Alignment Research Blog, tracking CoT monitorability from GPT-5 to Astra-class models through two proxies: no-CoT capability and CoT controllability.
💬 What's your take? Join the discussion in the comments.
Read the original Chinese version
📰 OpenAI 对齐研究：模型越强越难靠思维链监控，GPT-6 Astra 级模型的 CoT 可控性在强化学习中显著上升，还能经 SFT 蒸馏传给下一代
OpenAI 在对齐研究博客发布《What affects chain-of-thought monitorability across the model lifecycle?》，用「无 CoT 推理能力」和「CoT 可控性」两个代理指标，追踪从 GPT-5 到 Astra 级模型的思维链可监控性。
💬 你怎么看？欢迎在评论区讨论。
#OpenAI #AI Safety #Alignment
0 Comments