Suggests specialized behavior-prediction models can outperform general LLMs on human response forecasting.
A newly released AI model built to predict human behavior rather than generate text reportedly beats nine leading models at predicting how people answer surveys, with 74% less error.