Conformal Prediction with Paraphrase-Aware Scoring for LLM Uncertainty Quantification
arXiv:2610.04239v1 Announce Type: new Abstract: Uncertainty quantification (UQ) for large language models (LLMs) aims to provide reliable measures of predictive confidence, yet current methods are often unstable under meaning-preserving perturbations. Semantically equivalent paraphrases can induce…
Read the full story at arXiv cs.CL ↗
Timeline · 1 report
- 2026-10-06 04:00 · arXiv cs.CL
Conformal Prediction with Paraphrase-Aware Scoring for LLM Uncertainty Quantification