Estimating LLM Consistency: A User Baseline vs Surrogate Metrics

TL;DR

This paper evaluates how well automated metrics for LLM consistency align with human perceptions, finding current methods often fall short and advocating for more human-involved evaluation approaches.

Contribution

The paper introduces a logit-based ensemble method for estimating LLM consistency and demonstrates its comparable performance to existing metrics in reflecting human judgments.

Findings

Current automated metrics poorly align with human perceptions.

The proposed logit-based ensemble method matches the best existing metrics.

Human evaluation remains crucial for accurate LLM consistency assessment.

Abstract

Large language models (LLMs) are prone to hallucinations and sensitive to prompt perturbations, often resulting in inconsistent or unreliable generated text. Different methods have been proposed to mitigate such hallucinations and fragility, one of which is to measure the consistency of LLM responses -- the model's confidence in the response or likelihood of generating a similar response when resampled. In previous work, measuring LLM response consistency often relied on calculating the probability of a response appearing within a pool of resampled responses, analyzing internal states, or evaluating logits of responses. However, it was not clear how well these approaches approximated users' perceptions of consistency of LLM responses. To find out, we performed a user study ($n=2,976$) demonstrating that current methods for measuring LLM response consistency typically do not align well with humans' perceptions of LLM consistency. We propose a logit-based ensemble method for estimating LLM consistency and show that our method matches the performance of the best-performing existing metric in estimating human ratings of LLM consistency. Our results suggest that methods for estimating LLM consistency without human evaluation are sufficiently imperfect to warrant broader use of evaluation with human input; this would avoid misjudging the adequacy of models because of the imperfections of automated consistency metrics.