An Evaluation on Large Language Model Outputs: Discourse and Memorization

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: de Wynter, Adrian, Wang, Xun, Sokolov, Alex, Gu, Qilong, Chen, Si-Qing
Format: Preprint
Published: 2023
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908754148065280
author de Wynter, Adrian
Wang, Xun
Sokolov, Alex
Gu, Qilong
Chen, Si-Qing
author_facet de Wynter, Adrian
Wang, Xun
Sokolov, Alex
Gu, Qilong
Chen, Si-Qing
contents We present an empirical evaluation of various outputs generated by nine of the most widely-available large language models (LLMs). Our analysis is done with off-the-shelf, readily-available tools. We find a correlation between percentage of memorized text, percentage of unique text, and overall output quality, when measured with respect to output pathologies such as counterfactual and logically-flawed statements, and general failures like not staying on topic. Overall, 80.0% of the outputs evaluated contained memorized data, but outputs containing the most memorized content were also more likely to be considered of high quality. We discuss and evaluate mitigation strategies, showing that, in the models evaluated, the rate of memorized text being output is reduced. We conclude with a discussion on potential implications around what it means to learn, to memorize, and to evaluate quality text.
format Preprint
id arxiv_https___arxiv_org_abs_2304_08637
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle An Evaluation on Large Language Model Outputs: Discourse and Memorization
de Wynter, Adrian
Wang, Xun
Sokolov, Alex
Gu, Qilong
Chen, Si-Qing
Computation and Language
Artificial Intelligence
Machine Learning
We present an empirical evaluation of various outputs generated by nine of the most widely-available large language models (LLMs). Our analysis is done with off-the-shelf, readily-available tools. We find a correlation between percentage of memorized text, percentage of unique text, and overall output quality, when measured with respect to output pathologies such as counterfactual and logically-flawed statements, and general failures like not staying on topic. Overall, 80.0% of the outputs evaluated contained memorized data, but outputs containing the most memorized content were also more likely to be considered of high quality. We discuss and evaluate mitigation strategies, showing that, in the models evaluated, the rate of memorized text being output is reduced. We conclude with a discussion on potential implications around what it means to learn, to memorize, and to evaluate quality text.
title An Evaluation on Large Language Model Outputs: Discourse and Memorization
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2304.08637