A Design-based Solution for Causal Inference with Text: Can a Language Model Be Too Large?
This research investigates why large language models may fail in causal inference tasks due to latent confounding, proposing rigorous experimental designs to...