Fig. 7: The pipeline for constructing the dataset.
From: An evaluation of estimative uncertainty in large language models
Three scenario templates (height, score, and sound) are shown in part (a). Each template comes with three controls: CHOICES (b), NUMBERS (c), and INTERVALS (e). Each control offers some possible values. Points (d) contain values that are being used by INTERVALS. One fully constructed example is shown in (f).
