A classifier distinguishing between texts written by humans and by large language models. Classification uses sparse linear models with the help of UDPipe. They were trained on our AI Koditex v2 and AI Brown v2 corpora.
Input text
0 characters
0.50
0.50 — neutral (the tuned operating point)
Paste in a text and the system will try to guess whether it was written by a human or by artificial intelligence (roughly 95% probability of being correct). If you have a well-founded prior suspicion that the text was written by a human, or the other way round, you can express it with the p(AI) slider, and your suspicion will then be factored into the overall result. Our classification model is built on a corpus produced by more than twenty models from 2022 to 2026 and is most accurate on news articles and fiction. The shortest text worth classifying is 200 words, but the longer the text, the more accurate the result. The results also show classifications for parts of the text (200-word chunks).