The researchers argue that CoT monitoring can help researchers detect when models begin to exploit flaws in their training, ...

At a Hill hearing, policy experts offered a grim assessment about the potential for criminal uses of artificial intelligence.