On 18th January, we hosted a webinar to practically consider how large language models (LLMs) could be used to surface official statistics. We showed working examples that were the result of experiments that our team carried during an internal hack day.
The prototypes we created addressed three use cases, each tackled by a different member of our team:
Generating geographic area summaries: automatically creating a textual summary for a geographic area, highlighting its distinctive characteristics, by Duncan Harris.
Natural language queries for census data: helping policy makers to obtain Northern Ireland 2021 census data through simple natural language requests, by Peter Hynes.
Data analyst AI assistant: creating an AI assistant who could help a user discover, explore and download Northern Ireland 2021 census data, by Spencer Hedger.
All the prototypes we created worked directly with the Northern Ireland 2021 census data hosted in Cantabular and used LLMs accessed through OpenAI’s APIs.