Paired human and LLM free text survey data

Here we present two paired datasets modelled after public consultations. In the human dataset, approximately 1,000 UK residents were surveyed on five public consultation-like questions. In addition to providing a free-text response, respondents were also asked to thematically tag their own response from a list of pre-selected themes. Additionally, each response was annotated by three crowdworkers. In the LLM dataset, the human responses were leveraged to create a synthetic version of the dataset on the same five questions. Here, LLM responses were seeded with pre-selected themes, a persona, and three examples of human responses to the question.