Perplexity, Burstiness, and Responsible Rewriting
People search for an "AI humanizer" because a draft sounds uniform: similar sentence length, predictable word choice, and few surprises. Two statistical ideas show up in that conversation. Perplexity describes how unsurprising the next word is. Burstiness describes how much the rhythm changes from sentence to sentence. A flat draft is easy to skim and, to a classifier, easy to cluster with other machine-written text. What those measures are, and what they are not Perplexity is a model-relative score. A sentence that is ordinary to one language model can look unusual to another. Burstiness is about variation,...