We empower people with better information to make their best decisions.
From Proof to Program: Characterizing Tool-Induced Reasoning Hallucinations in Large Language Models