This dataset contains adversarial prompts and evaluation data reflecting African cultural contexts, designed to assess cultural bias and safety risks in large language models (LLMs).
Potential Use Cases
- Cultural Bias Evaluation: Evaluation data reflecting African regional and cultural contexts can be used to analyze whether LLMs generate biased responses to culturally specific questions or scenarios.
- LLM Safety Vulnerability Assessment: Adversarial prompts reflecting African cultural contexts can be used to identify potential safety vulnerabilities and risks in LLM-generated responses.