Language models can replicate human opinions based on demographic data, revealing societal biases and replicability challenges.