Answers are correct, but too long for your interface. What is the cheapest fix?
Why A is the answer
State the length limit in the prompt
Output control through the prompt comes before retraining.
In detail
- A length limit in the prompt is the cheapest fix.
- Change the prompt before the model.
Use case
An illustrative scenario, not a real incident.
Replies are correct but too long, so a length limit in the prompt fixes it at no cost.
See it in different ways
mindmap
root((Prompting))
Format drift
Few-shot examples
Structured output
Reasoning errors
Chain of thought
Simple tasks
Zero-shot
Domain words
Context in the promptflowchart LR S0["Answers too long"] --> S1["State a limit"] --> S2["Check results"]
Key point: Prompt changes come first.
Why the other options are wrong
- B. Fine-tune a new model on shorter example answers
- Fine-tuning is far more costly.
- C. Add more GPUs to the serving cluster
- GPUs do not change length.
- D. Move the model to a different data center
- Location does not change length.