Prompt Engineering & Evaluation Flashcards
HardSpaced repetition with the SM-2 algorithm — grade each card and PlayPrepHQ schedules it to resurface right before you'd forget it. Progress saves in this browser.
Loading your deck…
Terms in this set
- Prompt Engineering The practice of designing prompts to get reliable, high-quality model output.
- Zero-shot Prompting a model to perform a task with no examples provided.
- Few-shot Prompting a model with a few examples to demonstrate the desired output.
- Chain-of-Thought Prompting a model to show step-by-step reasoning before its final answer.
- System Prompt A high-level instruction that sets a model's role, tone, and rules for a conversation.
- Prompt Injection An attack that smuggles instructions into input to override the model's intended behavior.
- Jailbreaking Crafting prompts that bypass a model's safety guardrails to elicit disallowed output.
- ROUGE A recall-oriented metric that compares generated summaries to reference text by overlap.
- BLEU A precision-oriented metric for evaluating machine translation quality.
- BERTScore An evaluation metric that compares texts by embedding similarity rather than exact overlap.
- Human Evaluation Having people rate model output for quality, relevance, or safety.
- Agent An LLM-driven system that plans steps and calls tools or APIs to complete tasks.