Just Ask for Calibration: Strategies for Eliciting Calibrated Confidence Scores from Language Models Fine-Tuned with Human Feedback

202372 citationsJournal Articlegold Open Access

Authors

Katherine Tian · Harvard University

Eric Mitchell · Harvard University

Allan Zhou · Harvard University

Archit Sharma · Harvard University

Rafael Rafailov · Harvard University

Huaxiu Yao · Stanford University

Chelsea Finn · Stanford University

Abstract

Katherine Tian, Eric Mitchell, Allan Zhou, Archit Sharma, Rafael Rafailov, Huaxiu Yao, Chelsea Finn, Christopher Manning. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. 2023.

Topics & Keywords

Topic Modeling Natural Language Processing Techniques

Publication Details

DOI: 10.18653/v1/2023.emnlp-main.330

Field-Weighted Citation Impact: 12.61

Command Palette

Just Ask for Calibration: Strategies for Eliciting Calibrated Confidence Scores from Language Models Fine-Tuned with Human Feedback

Authors

Abstract

Topics & Keywords

Publication Details