An Evaluation on Large Language Model Outputs: Discourse and Memorization
arXiv:2304.08637 · doi:10.1016/j.nlp.2023.100024
Abstract
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.
Final version at Natural Language Processing Journal
References in corpus (9)
- Training language models to follow instructions with human feedback
- LLaMA: Open and Efficient Foundation Language Models
- The Pile: An 800GB Dataset of Diverse Text for Language Modeling
- Galactica: A Large Language Model for Science
- Check Your Facts and Try Again: Improving Large Language Models with External Knowledge and Automated Feedback
- The Radicalization Risks of GPT-3 and Advanced Neural Language Models
- OPT-IML: Scaling Language Model Instruction Meta Learning through the Lens of Generalization
- Training Data Leakage Analysis in Language Models
- ResMem: Learn what you can and memorize the rest