By combining fine-tuning and in-context learning, you get LLMs that can learn tasks that would be too difficult or expensive for either method ...