Synthetic Personalities: How Well Can LLMs Mimic Individual Respondents Using Socio-Economic Microdata?
Researchers have developed a method to create detailed individual-level digital twins of respondents using socio-economic microdata. These twins, built from existing heterogeneous panel data, can accurately mimic individual responses across various LLMs and information depths. The study found that twin quality improves with more data, but with diminishing returns past a certain point, and suggests that constructing these twins is now more dependent on item volume and model selection than data design. AI
IMPACT This research demonstrates LLMs' capability to create detailed individual digital twins from existing data, potentially transforming market research by enabling more scalable and accurate respondent mimicry.