Which evaluators check for harmful content?
Why D is the answer
Risk and safety evaluators
Risk and safety evaluators detect harmful content.
In detail
- Risk and safety evaluators check for harmful content.
- Run them on adversarial prompts as well.
Use case
An illustrative scenario, not a real incident.
A public assistant must avoid harmful output. Risk and safety evaluators test for it before release.
See it in different ways
flowchart LR A["Test dataset"] --> B["Map columns to evaluator inputs"] B --> C["Quality and safety evaluators"] C --> D["Compare with the baseline"] D --> E["Release decision"]
sequenceDiagram participant C as Change participant P as CI/CD pipeline participant R as Production participant T as Traces C->>P: Runs evaluators on a test dataset P->>R: Releases if it passes R->>T: Records real traffic T->>P: New failing cases
mindmap
root(("Evaluators"))
n1["Quality"]
n2["Relevance"]
n3["Coherence"]
n4["Fluency"]
n5["Risk and safety"]
n6["Harmful content"]Key point: Risk and safety evaluators check harmful content.
Why the other options are wrong
- A. Coherence and fluency
- Those measure writing quality.
- B. Bicep what-if checks
- That checks Bicep changes.
- C. MLflow autologging
- That records training runs.