Guide to Evaluation Perspectives on AI Safety
Japan AI Safety Institute guide listing evaluation perspectives for LLM safety
Organises AI safety evaluation into ten perspectives (toxicity, bias, controllability, robustness, data quality, privacy, security, explainability, verifiability, high-risk use) with methods and dataset examples for each.
Contribute
Spot an error, or know a record we should add? Members can propose additions and submit corrections.