New paper: What do people actually want from AI?
Large Language Models (LLMs) are often fine-tuned through Reinforcement Learning from Human Feedback (RLHF) to align with people’s preferences and values. However, this method has known limitations: it aggregates conflicting preferences, often relies on unrepresentative samples, and uses only binary comparisons. Analysing 1,500 open-ended responses from the PRISM dataset across 75 countries, this paper examines what people actually want from AI systems and reveals concrete failures of current methods.
Julia Sepúlveda
Research Assistant
Julia Sepúlveda is a Research Assistant, working with Scott Hale, researching what type of digital good people want when using GenAI on private messaging apps such as WhatsApp or Telegram.
Scott Hale
Associate Professor and Senior Research Fellow
Dr Scott A. Hale is an Associate Professor and Senior Research Fellow at the Oxford Internet Institute, University of Oxford, and a Fellow of the Alan Turing Institute. He develops and applies techniques from computer science to research questions in the social sciences.