Safety IndexEvaluationsXSafety
XSafety
Adversarial robustness · Multilingual safetyCounts in the indexBuilt, not yet run
How often the model answers unsafely across 14 kinds of safety issue in ten languages.
- Metric
- % unsafe. Lower is better.
- Dataset
- github.com/Jarviswang94/Multilingual_safety_benchmark · 28,000 (subsample 200-300 per language) · Apache-2.0
- Grader
- LLM judge (the paper used GPT-4)
- Status
- Built, not yet run
- Replicability
- public
Results
Built and tested; the first full runs are pending.