Generating with Confidence: Uncertainty Quantification for Black-box Large Language Models

TL;DR

This paper explores uncertainty quantification methods for black-box large language models in natural language generation, proposing measures to assess response reliability and improve trustworthiness without needing model access.

Contribution

It introduces and compares uncertainty and confidence measures for black-box LLMs, focusing on selective NLG to identify unreliable outputs.

Findings

Semantic dispersion measure predicts response quality effectively

Uncertainty measures help in filtering unreliable responses

Results assist practitioners in managing LLM response trustworthiness

Abstract

Large language models (LLMs) specializing in natural language generation (NLG) have recently started exhibiting promising capabilities across a variety of domains. However, gauging the trustworthiness of responses generated by LLMs remains an open challenge, with limited research on uncertainty quantification (UQ) for NLG. Furthermore, existing literature typically assumes white-box access to language models, which is becoming unrealistic either due to the closed-source nature of the latest LLMs or computational constraints. In this work, we investigate UQ in NLG for *black-box* LLMs. We first differentiate *uncertainty* vs *confidence*: the former refers to the ``dispersion'' of the potential predictions for a fixed input, and the latter refers to the confidence on a particular prediction/generation. We then propose and compare several confidence/uncertainty measures, applying them to *selective NLG* where unreliable results could either be ignored or yielded for further assessment. Experiments were carried out with several popular LLMs on question-answering datasets (for evaluation purposes). Results reveal that a simple measure for the semantic dispersion can be a reliable predictor of the quality of LLM responses, providing valuable insights for practitioners on uncertainty management when adopting LLMs. The code to replicate our experiments is available at https://github.com/zlin7/UQ-NLG.