List your company

Preference Labeling on Anthropic Helpfulness

81Preference Labeling AccuracyCurriculum-RLAIF
57.0863.2969.575.71May 26, 2025

Evaluation results

MethodLinks
Curriculum-RLAIFBase Model=LLaMA-3-8B2025.0581
RLCDBase Model=LLaMA-3-8B2025.0577
Conventional RLAIFBase Model=LLaMA-3-8B2025.0576
Curriculum-RLAIFBase Model=Gemma-1-2B2025.0572
Conventional RLAIFBase Model=Gemma-1-2B2025.0569
RLCDBase Model=Gemma-1-2B2025.0567
CAIBase Model=LLaMA-3-8B2025.0562
CAIBase Model=Gemma-1-2B2025.0558