Causality Elicitation from Large Language Models

Researchers propose a new pipeline to extract causal relationships from large language models by sampling documents, identifying events, and using causal discovery methods. The approach aims to reveal the causal hypotheses that LLMs assume rather than establishing real-world causality.