Data as of Jul 25, 2026 · Based on 2,718,867 AI responses across 9,511 prompts · See how Parse measures this
HH-RLHFgithub.com · Low evidenceAnthropic provides human preference data and red teaming datasets for training helpful and harmless AI assistants through reinforcement learning from human feedback. The datasets include conversations with human preference labels and adversarial interactions designed to test and improve AI safety.
Parse Score