Counting Disagreements Between AI Models to Measure Answer Confidence

Original: Probability is Not Enough: Exploring and Counting Divergent Tokens for Reasoning Uncertainty Quantification in LLMs

Source: arxiv.org ↗