Large Causal Models from Large Language Models

TLDR: This paper introduces a pipeline that extracts unstructured causal claims from LLMs and stitches them into a structured, geometric Large Causal Model.

Traditional causal inference usually relies on numerical data from specific experiments. This paper instead proposes mining the latent world model inside an LLM. It prompts the model to generate causal triples (Subject -> Relation -> Object) across diverse domains (archaeology, economics, biology) and uses categorical machine learning methods (Topos theory) to embed these fragments into a coherent manifold, rather than a messy graph "hairball."