Can Large Language Models Build Causal Inference