Multilingual Moral Alignment in Large Language Models

This research examines the moral alignment of large language models (LLMs) across cultures, emphasizing their relevance to global AI ethics. It introduces the MULTITP dataset, evaluates 19 LLMs using MIS scores, and identifies key findings, such as strong alignment in specific models like Llama 3.1 70B and the influence of demographic dimensions. Results underscore the significance of culturally nuanced and multilingual datasets in LLM ethical decision-making. The study highlights the...