#llm-alignment

Code-Preference Optimization: A New Recipe for Aligning LLMs with Developer Intent

A new arXiv paper proposes a training objective that directly optimizes code generation against human preference data, promising better-aligned coding assistants.