AI RESEARCH
Don't Throw Away Your Beams: Improving Consistency-based Uncertainties in LLMs via Beam Search
arXiv CS.LG
•
ArXi:2512.09538v2 Announce Type: replace-cross Consistency-based methods have emerged as an effective approach to uncertainty quantification (UQ) in large language models. These methods typically rely on several generations obtained via multinomial sampling, measuring their agreement level. However, in short-form QA, multinomial sampling is prone to producing duplicates due to peaked distributions, and its stochasticity