project

AIORNOR

AI-or-human text detection

Built with
Python · NLP
Source
Public on GitHub
Author
Samir Sengupta

Detects whether a piece of text is AI-generated or human-written using a perplexity-based metric, served through a lightweight web app.

The problem

Most AI-text detectors are opaque classifiers that give you a confidence score and no reason to believe it. If a tool is going to make an accusation about someone’s writing, the mechanism should be inspectable.

How it works

  • Uses perplexity — how surprised a language model is by the text — as the discriminating signal, rather than a black-box classifier head.
  • Machine-generated text tends to sit in low-perplexity regions because it was sampled from the very distribution being measured; human writing is lumpier.
  • Served through a lightweight web app so the metric is something you can try on a paragraph rather than read about.

Why it exists

A small, honest tool: the method is simple enough to explain in a sentence, which is exactly what a claim like this should be.

More work

Hiring for AI or ML?

I am open to AI/ML Engineering, Data Science, and Python roles, plus research collaborations and consulting. New York based, shipping worldwide.