The PRISM Alignment Dataset

Dataset Card for PRISM PRISM is a diverse human feedback dataset for preference and value alignment in Large Language Models (LLMs). It maps the characteristics and stated preferences of humans from a detailed survey onto their real-time interactions with LLMs and contextual preference ratings Dataset Details There are two sequential stages: first, participants complete a Survey where they answer questions about their demographics and stated preferences, then proceed to… See the full description on the dataset page: huggingface.co/datasets/HannahRoseKirk/prism-alignment.