Users of ChatGPT and other competing large language models can be forgiven if they feel the responses are like magic. MIT's Technology Review even recently published an article titled "Nobody Knows How AI Works."