RLCD: Reinforcement Learning from Contrastive Distillation for Language Model Alignment

arXiv:2307.12950v3 Announce Type: replace
Abstract: We propose Reinforcement Learning from Contrastive Distillation (RLCD), a method for aligning language models to follow principles expressed in natural language (e.g., to be more harmless) without using human feedback. RLCD creates preference pairs from two contrasting model outputs, one using a positive prompt designed to encourage following the given principles, and one using a negative prompt designed to encourage violating them. Using two different prompts causes model outputs to be more differentiated on average, …

abstract alignment arxiv cs.ai cs.cl distillation feedback human human feedback language language model language models natural natural language positive prompt reinforcement reinforcement learning type