1 paper
Ido Amit, Ido Galil, Ran El-Yaniv
As LLMs generate increasingly long outputs, effective uncertainty estimation must identify errors at fine-grained levels rather than discard entire responses. While such methods ex…