One of AI’s most annoying writing habits has a 2,000-year-old history
“We do not sell products. We build experiences.” If that sounds like a chatbot, there’s a name for it. A new preprint examines why AI models use epanorthosis, a self-correcting figure from classical rhetoric.
Artificial intelligence's tendency to express itself in a particular way has deep historical roots. A new preprint study explores this phenomenon, known as epanorthosis. This rhetorical device involves a speaker or writer revisiting a statement they've just made to correct, qualify, or replace it. Epanorthosis comes from Greek roots meaning "a further straightening" and was cataloged in Latin by Roman orators and rhetoricians like Cicero and Quintilian.
There are three main functional forms: an emphatic upgrade, a softening downgrade, and a substantive revision. While humans often use epanorthosis with phrases like "I mean" or "or rather," chatbots frequently favor the dramatic upward version. A preprint paper examines how large language models utilize this ancient rhetorical figure and finds evidence of its overuse in some genres.
The study also tests methods to reduce this habit, which in small pilot tests brought it closer to typical human rates.
Written by urgent.news from PsyPost's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
