An Evaluation on Large Language Model Outputs: Discourse and Memorization
This is an incremental study that addresses the issue of output evaluation for users of large language models, highlighting trade-offs between memorization and perceived quality.
The paper tackles the problem of evaluating large language model outputs by analyzing memorization and discourse quality, finding that 80% of outputs contained memorized data, and those with more memorized content were often rated higher in quality.
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.