'Probably' doesn't mean the same thing to your AI as it does to you

As large language models are increasingly used in high-stakes fields like health care, government policy and scientific reporting, the way they communicate risk becomes a matter of public trust.