Uncertainty in Language Models: Assessment through Rank-Calibration

arXiv:2404.03163v1 Announce Type: cross
Abstract: Language Models (LMs) have shown promising performance in natural language generation. However, as LMs often generate incorrect or hallucinated responses, it is crucial to correctly quantify their uncertainty in responding to given inputs. In addition to verbalized confidence elicited via prompting, many uncertainty measures ($e.g.$, semantic entropy and affinity-graph-based measures) have been proposed. However, these measures can differ greatly, and it is unclear how to compare them, partly because they take values over different ranges …

abstract arxiv assessment confidence cs.ai cs.cl cs.lg entropy generate however inputs language language generation language models lms natural natural language natural language generation performance prompting responses semantic stat.ml through type uncertainty via