Artificial intelligence has moved from checking homework to attacking problems that professional mathematicians once treated ...

In a new paper from OpenAI, the company proposes a framework for analyzing AI systems' chain-of-thought reasoning to understand how, when, and why they misbehave.