What LLMs explain is not what they believe: Evaluating explanation sufficiency under models' own input beliefs

Open in new window