A method constructs interpretable causal graphs of LLM concept reasoning via MCMC-inspired counterfactual chains and applies it to diagnosis, sentiment, and judging tasks across three models.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
LLM Explainability with Counterfactual Chains and Causal Graphs
A method constructs interpretable causal graphs of LLM concept reasoning via MCMC-inspired counterfactual chains and applies it to diagnosis, sentiment, and judging tasks across three models.