What are common failure modes in prompting and how do you debug them?
LLM & AILevel: IntermediateProduction Prompt Design
What are common failure modes in prompting and how do you debug them?
Hints
- Identify the stage: pretraining, adaptation, generation or evaluation.
- Name the inputs, constraints and a case where the approach fails.
Answer
| Failure | Diagnosis | Fix |
|---|---|---|
| Wrong format | Model ignored format instructions | Add few-shot format example; use JSON mode |
| Too verbose | No length constraint | "Answer in 2–3 sentences maximum" |
| Refuses valid requests | Over-cautious safety training | Reframe request; clarify legitimate use case in system prompt |
| Inconsistent across inputs | Prompt ambiguous or too short | Add examples that cover edge cases |
| Hallucinating | Relying on parametric knowledge | Add "only use provided context; say 'I don't know' if unsure" |
| Language wrong | Default language not specified | Specify target language explicitly |
Source and licence: JustInCache & contributors — Awesome AI Interviews · Apache-2.0