A cross-domain benchmark finds LLMs simulating survey respondents lose to simple statistical baselines and overstate how much demographics predict attitudes.
Continue to AI University →