Frontiers: Reducing Disparity Between LLMs and Humans: Optimal LLM Sample Calibration

Large language models (LLMs) are increasingly used to measure human preferences and opinions, but their outputs can diverge from human responses. We introduce human mimicry calibration (HuMCal), a post hoc, model-agnostic method that uses aggregate human benchmark data to learn how to combine LLM outputs from different prompts, personas, and model checkpoints. In applications to consumer transaction data and public opinion surveys, HuMCal improves agreement with human benchmarks over standard aggregation rules, and the calibrated weights often remain useful across time, geography, industry, and question clusters.