about as effective as don't make mistakes... For anthropic's models at least