Stepwise Alignment for Constrained Language Model Policy Optimization

arXiv:2404.11049v1 Announce Type: cross
Abstract: Safety and trustworthiness are indispensable requirements for applying AI systems based on large language models (LLMs) in real-world applications. This paper formulates a human value alignment as a language model policy optimization problem to maximize reward under a safety constraint and then proposes an algorithm called Stepwise Alignment for Constrained Policy Optimization (SACPO). A key idea behind SACPO, supported by theory, is that the optimal policy incorporating both reward and safety can be directly obtained …

abstract ai systems algorithm alignment applications arxiv cs.ai cs.cl cs.lg human language language model language models large language large language models llms optimization paper policy requirements safety systems type value world