LLMs can draft DAGs but can't guarantee d-separation
LLMs can specify Bayesian network structures from examples, but struggle to respect conditional independence and d-separation constraints. A related notebook's DAG search guarantees factorization, not the correct causal graph.
2026-09-21 ~ 2026-09-21 · 2 related posts
- LLMs can draw DAGs but can't guarantee d-separation, notebook discussion — fdellaert · 2026-09-21
- Search over DAGs guarantees d-separation factorization, not the correct causal graph — fdellaert · 2026-09-21