AI chatbots sometimes guess when they don’t know an answer. I use a simple “cupcake prompt” to spot when an AI might be hallucinating — and it works surprisingly well.