1、 Large Language Models for Official Statistics HLG-MOS White Paper December 2023 2 Foreword Late 2022,ChatGPT 3.5 took the world by storm.The news of its ability to create good-looking text,based on a natural language request from a user,quickly spread.All over the world people from quite different

2、backgrounds started experimenting and shared their experiences with friends and colleagues.When the first wave of euphoria had calmed down a bit,people realised some aspects of ChatGPT might require critical examination.Moreover,other Large Language Models(LLMs)had appeared that required attention.A

3、ll this led to more systematic and organised activities in specific communities.The worldwide official statistics community was no exception.From early 2023,in many of the meetings of the community the topic of ChatGPT was raised and participants wanted to share views and experiences.Two groups unde

4、r the umbrella of the UNECE High-Level Group for the Modernisation of Official Statistics(HLG-MOS)then joined forces,the Blue Skies Thinking Network and the Modernisation Group on Advancing Data Science and Modern Methods.They organised a series of(mostly online)meetings to discuss aspects of ChatGP

5、T and other LLM.Mid 2023 it was decided to bring a group together to draft a white paper on this topic.A draft of the white paper was presented at the annual HLG-MOS workshop in Geneva(November 2023).Moreover,at this workshop it was decided to launch a formal HLG-MOS project on Generative AI includi

6、ng LLMs in 2024,which is open to active participants from the official statistics community,and supporters from the research world and other domains,who have an interest in and enthusiasm for this topic.The white paper is formally released in December 2023.It provides a broad overview of LLM aspects