List your company

Preference Labeling on Anthropic Harmlessness

77Preference Labeling AccuracyCurriculum-RLAIF
54.1260.066671.94May 26, 2025

Evaluation results

MethodLinks
Curriculum-RLAIFBase Model=LLaMA-3-8B2025.0577
Conventional RLAIFBase Model=LLaMA-3-8B2025.0571
Curriculum-RLAIFBase Model=Gemma-1-2B2025.0568
RLCDBase Model=LLaMA-3-8B2025.0565
RLCDBase Model=Gemma-1-2B2025.0561
Conventional RLAIFBase Model=Gemma-1-2B2025.0559
CAIBase Model=LLaMA-3-8B2025.0557
CAIBase Model=Gemma-1-2B2025.0555