Its a certainty that given any prompt there will be likely one gnarly output, especially the longer that output becomes.

What I'd like to see is an estimate of the uncertainty of the output.