Reverse Probing: Supervised Token-level Uncertainty Quantification for Large Language Models in Clinical Text | ArxivCSExplorer