When does zero-shot prompting work well?
Why C is the answer
When the task is simple and clearly described
Simple tasks need no examples. Unusual formats benefit from them.
In detail
- Simple, clearly described tasks need no examples.
- Unusual formats benefit from examples.
Use case
An illustrative scenario, not a real incident.
"Translate this sentence to French" needs no examples, because it is simple and clear.
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 TD
Q{"How unusual is the task?"}
Q -->|"Simple and clear"| A0["Zero-shot"]
Q -->|"Unusual format"| A1["Few-shot"]Key point: Zero-shot suits simple, clear tasks.
Why the other options are wrong
- A. When the model has never seen any similar language
- Models need some exposure.
- B. When the prompt must be longer than the context window
- That cannot work at all.
- D. When the output must follow an unusual, strict format
- Unusual formats benefit from examples.