Confidence Under the Hood: An Investigation into the Confidence-Probability Alignment in Large Language Models

TL;DR

This paper investigates how well large language models' internal token probabilities align with their expressed confidence, revealing GPT-4's relatively strong alignment and contributing to risk assessment and trustworthiness evaluation.

Contribution

It introduces the concept of Confidence-Probability Alignment and evaluates this alignment across different models and prompting techniques, highlighting GPT-4's superior performance.

Findings

GPT-4 exhibits the strongest confidence-probability alignment among models tested.

Various prompting techniques can influence the perceived confidence of LLMs.

Alignment scores vary significantly across different models and tasks.

Abstract

As the use of Large Language Models (LLMs) becomes more widespread, understanding their self-evaluation of confidence in generated responses becomes increasingly important as it is integral to the reliability of the output of these models. We introduce the concept of Confidence-Probability Alignment, that connects an LLM's internal confidence, quantified by token probabilities, to the confidence conveyed in the model's response when explicitly asked about its certainty. Using various datasets and prompting techniques that encourage model introspection, we probe the alignment between models' internal and expressed confidence. These techniques encompass using structured evaluation scales to rate confidence, including answer options when prompting, and eliciting the model's confidence level for outputs it does not recognize as its own. Notably, among the models analyzed, OpenAI's GPT-4 showed the strongest confidence-probability alignment, with an average Spearman's $\hat{\rho}$ of 0.42, across a wide range of tasks. Our work contributes to the ongoing efforts to facilitate risk assessment in the application of LLMs and to further our understanding of model trustworthiness.