Probabilities of Chat LLMs Are Miscalibrated but Still Predict Correctness on Multiple-Choice Q&A

Fuente: arXiv
Enregistré dans:
Détails bibliographiques
Auteurs principaux: Plaut, Benjamin, Khanh, Nguyen X., Trinh, Tu
Format: Preprint
Publié: 2024
Sujets:
Accès en ligne:
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866908480452952064
author Plaut, Benjamin
Khanh, Nguyen X.
Trinh, Tu
author_facet Plaut, Benjamin
Khanh, Nguyen X.
Trinh, Tu
contents We study 15 large language models (LLMs) fine-tuned for chat and find that their maximum softmax probabilities (MSPs) are consistently miscalibrated on multiple-choice Q&A. However, those MSPs might still encode useful uncertainty information. Specifically, we hypothesized that wrong answers would be associated with smaller MSPs compared to correct answers. Via rigorous statistical testing, we show that this hypothesis holds for models which perform well on the underlying Q&A task. We also find a strong direction correlation between Q&A accuracy and MSP correctness prediction, while finding no correlation between Q&A accuracy and calibration error. This suggests that within the current fine-tuning paradigm, we can expect correctness prediction but not calibration to improve as LLM capabilities progress. To demonstrate the utility of correctness prediction, we show that when models have the option to abstain, performance can be improved by selectively abstaining based on the MSP of the initial model response, using only a small amount of labeled data to choose the MSP threshold.
format Preprint
id arxiv_https___arxiv_org_abs_2402_13213
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Probabilities of Chat LLMs Are Miscalibrated but Still Predict Correctness on Multiple-Choice Q&A
Plaut, Benjamin
Khanh, Nguyen X.
Trinh, Tu
Computation and Language
Artificial Intelligence
Machine Learning
We study 15 large language models (LLMs) fine-tuned for chat and find that their maximum softmax probabilities (MSPs) are consistently miscalibrated on multiple-choice Q&A. However, those MSPs might still encode useful uncertainty information. Specifically, we hypothesized that wrong answers would be associated with smaller MSPs compared to correct answers. Via rigorous statistical testing, we show that this hypothesis holds for models which perform well on the underlying Q&A task. We also find a strong direction correlation between Q&A accuracy and MSP correctness prediction, while finding no correlation between Q&A accuracy and calibration error. This suggests that within the current fine-tuning paradigm, we can expect correctness prediction but not calibration to improve as LLM capabilities progress. To demonstrate the utility of correctness prediction, we show that when models have the option to abstain, performance can be improved by selectively abstaining based on the MSP of the initial model response, using only a small amount of labeled data to choose the MSP threshold.
title Probabilities of Chat LLMs Are Miscalibrated but Still Predict Correctness on Multiple-Choice Q&A
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2402.13213