Reproducible Experiment Asking ChatGPT to Create JSON; Includes data set and Python code to reproduce

This is an experiment in the accuracy of ChatGPT 3.5 and 4.0 to construct a JSON concordance from a list of words.

The results show that ChatGPT is unreliable at this task for any list above a trivial size, and that the nature of the list of words provided, temperature settings, and version of ChatGPT all have a strong effect on the output. Therefore, the only strong generalization that can be made is that ChatGPT is unreliable.