When large language models first came out, most of us were just thinking about what they could do, what problems they could solve, and how far they might go.