Just trained a model on 5 use cases and zero shot an entirely unrelated one. It's still amazing to me how fast LLMs can generalize.