在人工智能领域，如何引导大语言模型产出贴合人类偏好的内容，已成为备受瞩目的研究焦点。强化学习从人类反馈中学习（RLHF）作为该领域的重要方法之一，虽成效显著，但也暴露出多阶段优化流程复杂、计算负担沉重等弊端。而直接偏好优化（DPO）及其衍生 ...

Supply path optimisation (SPO) and its cousin, demand path optimisation (DPO) had a bit of a moment in the calm before the cookiepocalypse storm, but have subsequently not been getting the time or ...