Large language models already read, write, and answer questions with striking skill. They do this by training on vast libraries of text. Once that training ends, though, the model’s knowledge largely ...