Strength in Numbers: Estimating Confidence of Large Language Models by Prompt Agreement

Gwenyth Portillo Wightman, Alexandra DeLucia, Mark Dredze

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Large language models have achieved impressive few-shot performance on a wide variety of tasks. However, in many settings, users require confidence estimates for model predictions. While traditional classifiers produce scores for each label, language models instead produce scores for the generation which may not be well calibrated. We compare generations across diverse prompts and show that these can be used to create confidence scores. By utilizing more prompts we can get more precise confidence estimates and use response diversity as a proxy for confidence. We evaluate this approach across ten multiple-choice questionanswering datasets using three models: T0, FLAN-T5, and GPT-3. In addition to analyzing multiple human written prompts, we automatically generate more prompts using a language model in order to produce finer-grained confidence estimates. Our method produces more calibrated confidence estimates compared to the log probability of the answer to a single prompt. These improvements could benefit users who rely on prediction confidence for integration into a larger system or in decisionmaking processes.

Original languageEnglish (US)
Title of host publication3rd Workshop on Trustworthy Natural Language Processing, TrustNLP 2023 - Proceedings of the Workshop
EditorsAnaelia Ovalle, Kai-Wei Chang, Kai-Wei Chang, Ninareh Mehrabi, Yada Pruksachatkun, Aram Galystan, Aram Galystan, Jwala Dhamala, Apurv Verma, Trista Cao, Anoop Kumar, Rahul Gupta
PublisherAssociation for Computational Linguistics (ACL)
Pages326-362
Number of pages37
ISBN (Electronic)9781959429869
StatePublished - 2023
Externally publishedYes
Event3rd Workshop on Trustworthy Natural Language Processing, TrustNLP 2023, co-located with ACL 2023 - Toronto, Canada
Duration: Jul 14 2023 → …

Publication series

NameProceedings of the Annual Meeting of the Association for Computational Linguistics
ISSN (Print)0736-587X

Conference

Conference3rd Workshop on Trustworthy Natural Language Processing, TrustNLP 2023, co-located with ACL 2023
Country/TerritoryCanada
CityToronto
Period7/14/23 → …

ASJC Scopus subject areas

  • Computer Science Applications
  • Linguistics and Language
  • Language and Linguistics

Fingerprint

Dive into the research topics of 'Strength in Numbers: Estimating Confidence of Large Language Models by Prompt Agreement'. Together they form a unique fingerprint.

Cite this