大语言模型的价值对齐多元性可转移性和可控性

Value Alignment in LLMs: Pluralism, Transferability and Controllability天津大学自然语言处理实验室TheNaturalLanguageProcessingLaboratoryatTianjinUniversityDeyiXiongTianjinUniversitydyxiong@tju.edu.cnhttps://dyxiong.github.iohttps://tjunlp-lab.github.ioYSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024WhatHumanValuesare?1.Values are beliefslinked inextricably to affect.2.Values refer to desirable goals that motivate action.3.Values transcendspecific actions and situations.4.Values serve as standards or criteria.5.Values are ordered by importance relative to one another.6.The relative importanceof multiple values guides action.Shalom H. SchwartzYSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024Value AlignmentYSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024YSSNLP2024Challenges of Value Alignment?oIt is usually difficult to understand and define human values or intentions.oThere are many different fine-grained dimensions of human values. Do we need to align the specified objective