A model confidently states a fact that is false. What is this?
Why D is the answer
A hallucination
A hallucination is a confident answer that is false, and it is a known disadvantage of generative AI.
In detail
- A confident false statement is a hallucination.
- RAG and human review reduce the risk, but do not remove it.
Use case
An illustrative scenario, not a real incident.
An assistant cites a return policy that does not exist, and it sounds confident. That is a hallucination.
See it in different ways
mindmap
root((Generative AI))
Token
Unit of text
Drives cost
Embedding
Meaning as numbers
Chunking
Focused passages
Hallucination
Confident and false
Temperature
Varietyflowchart TD
Q{"Can the claim be verified?"}
Q -->|"Yes"| A0["Use it"]
Q -->|"No"| A1["Treat it as suspect"]Key point: Hallucinations are confident false statements.
Why the other options are wrong
- A. A guardrail
- A guardrail is a control.
- B. An embedding
- An embedding is a vector.
- C. A prompt template
- A template is a reusable prompt.