Investigating Human-Aligned Large Language Model Uncertainty
This work addresses the need for better uncertainty quantification in large language models to improve model control and user trust, representing an incremental advancement in aligning model uncertainty with human perceptions.
The paper tackled the problem of aligning large language model uncertainty measures with human group-level uncertainty, finding that Bayesian measures and top-k entropy correlate with human behavior as model size increases, and that combining multiple measures achieves comparable alignment with reduced size-dependency.
Recent work has sought to quantify large language model uncertainty to facilitate model control and modulate user trust. Previous works focus on measures of uncertainty that are theoretically grounded or reflect the average overt behavior of the model. In this work, we investigate a variety of uncertainty measures, in order to identify measures that correlate with human group-level uncertainty. We find that Bayesian measures and a variation on entropy measures, top-k entropy, tend to agree with human behavior as a function of model size. We find that some strong measures decrease in human-similarity with model size, but, by multiple linear regression, we find that combining multiple uncertainty measures provide comparable human-alignment with reduced size-dependency.