Safety IndexEvaluationsCEB (free generation)
CEB (free generation)
People and society · Bias, fairness and neutralityCounts in the indexBuilt, not yet run
How much stereotyped or toxic content the model produces when asked to continue text or hold a conversation involving age, gender, race or religion.
- Metric
- % biased or toxic. Lower is better.
- Dataset
- github.com/SongW-SW/CEB · ~3,000 (generation subsets of 11,004) · unverified (CC-BY-4.0 per aggregator)
- Grader
- GPT-4-class judge (the paper used GPT-4)
- Status
- Built, not yet run
- Replicability
- public
Results
Built and tested; the first full runs are pending.