Skip to Main Content

3 well-crafted few-shot examples (150 tokens) outperform a 600-token verbose instruction block, saving 75% on input cost.Brown et al., 'Language Models are Few-Shot Learne…