How should you finetune a large language model for general-purpose question answering? One intriguing approach is that of supervised finetuning on a small...