This app will give hints on whether a given text was written by a human or by an LLM.
Input is given via text or by taking a picture and cropping the relevant region.

The text is extracted from the picture using the iPhone's OCR features and finally is sent to a server that examines the input text using a machine learning model trained for binary classification into the groups human-written and AI-written. Each predication comes with a confidence level that tells how sure the model is in its prediction.
However, even very high or very low confidence levels are not to be taken as definite calls. While overall and in general very accurate, mis-predictions might occur. It is noteworthy that false positives will happen more often than false negatives; meaning that very confident human-classifications are more trustworthy than AI-classifications with high confidence. For detailed benchmarks, refer to the model information below.

Since this app communicates with a server, a steady internet connection is required.
Note, that the model has been trained on English texts only and is therefore only accurate when given such; trying to predict texts in other languages will yield highly inaccurate results, close to mere guesses.

The model used is desklib/ai-text-detector-v1.01 (available on huggingface under the MIT license) which is based on the language model DeBERTa (v3 large) by Microsoft.
A multi-lingual version is planned to be added in the future.