Support replies must match a new domain's terms, with no retraining. What do you add?
Why B is the answer
Domain examples and context in the prompt
Prompt-level adaptation changes behavior without touching weights.
In detail
- Examples and context adapt behavior with no change to weights.
- Weights stay the same.
Use case
An illustrative scenario, not a real incident.
A support bot moves to a new domain, and terms in the prompt adapt it with no retraining.
See it in different ways
flowchart TD
A{"What goes wrong?"} -->|"Wrong format"| B["Few-shot and structured output"]
A -->|"Logic errors"| C["Chain of thought"]
A -->|"Made-up facts"| D["Retrieval and grounding"]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["Add domain terms"] --> S1["Add context"] --> S2["Test"]
Key point: Prompts can adapt behavior.
Why the other options are wrong
- A. More attention heads in the model
- Heads do not teach terms.
- C. A larger batch size for the serving layer
- Batch size does not teach terms.
- D. A new quantization level for the model
- Quantization does not teach terms.