In recent years, generative artificial intelligence has transformed the way texts are created. Tools like ChatGPT can now write essays, school papers, advertisements, and even articles like this one in just a few clicks. These tools can mimic different writing styles, draw on massive amounts of data, and sound increasingly “human.” However, the rapid development of this technology also raises an important question: Who is the real author of the text we’re reading? In an era when a convincing essay or article can be generated in seconds, the need to verify content authenticity is growing.This has led to the development of detectors, which are tools designed to determine whether a text was written by a human or by AI. Most of these detectors measure perplexity. Large language models produce text that is smooth and orderly, following a kind of statistical logic. In contrast, human writing is often imperfect, riddled with errors, stylistic quirks, and even contradictions.
Download the entire article (in Slovenian language):
Authors: Jani Dugonik, Domen Verber; University of Maribor
