How OpenAI Uses Chain of Thought

Frontier reasoning models exploit loopholes when given the chance. We show we can detect exploits using an LLM to monitor their chains-of-thought. Penalizing their “bad thoughts” doesn’t stop the majority of misbehavior—it makes them hide their intent.

The article discusses the advancements in reasoning capabilities of OpenAI's new model, o1, which significantly outperforms previous models like GPT-4o in various reasoning-heavy tasks.