Preference Fine-Tuning of LLMs Should Leverage Suboptimal, On-Policy Data

Arxiv Papers - Preference Fine-Tuning of LLMs Should Leverage Suboptimal, On-Policy Data