A decision affects someone's loan. Which practice fits?
Why D is the answer
Human in the loop
Human in the loop keeps a person responsible for high-impact decisions, and it helps with hallucinations and edge cases.
In detail
- A high-impact decision needs a human in the loop.
- Reserve this for decisions with real impact.
Use case
An illustrative scenario, not a real incident.
A model recommends a loan decision, and a person reviews it before it takes effect.
See it in different ways
flowchart TD A["Knowledge cutoff or hallucination"] --> B["Grounding and RAG"] C["Vague output"] --> D["Prompt engineering"] E["Specialized style"] --> F["Fine-tuning"] G["High-stakes decisions"] --> H["Human in the loop"]
stateDiagram-v2 [*] --> SourceOfFacts SourceOfFacts --> GoogleSearch: current public facts SourceOfFacts --> AgentSearch: company documents GoogleSearch --> GroundedAnswer AgentSearch --> GroundedAnswer GroundedAnswer --> HumanReview: high impact GroundedAnswer --> [*]
flowchart TD
Q{"What is the impact?"}
Q -->|"Serious, such as a loan"| A0["Human in the loop"]
Q -->|"Low"| A1["Lighter review"]Key point: People review high-impact decisions.
Why the other options are wrong
- A. Higher temperature
- Temperature adds randomness.
- B. No review to save time
- Saving time is not worth the harm.
- C. Zero-shot prompting
- That is a prompt style.

