Glossary
Few-shot prompting
Showing a model a few worked examples of the task in the prompt, so it follows the pattern without retraining. Anthropic calls examples one of the most reliable ways to steer format, tone and structure.
Examples are instructions too: in the ICLAttack research, poisoned examples changed a model's behavior without any fine-tuning, even when each one looked correctly labeled. Take examples from a reviewed set, never straight from user feedback.