<!-- thumbnail: https://i.sstatic.net/8Q61x.png -->
<!-- version: 0.1 -->
<!-- tag: [moderation] -->
<!-- excerpt: Adds a button to check the probability that a post was written by a bot -->

## Screenshot / Code Snippet

[!["Detect OpenAI" link in post menu; notice with results][1]][1]

## About

This script adds a link to each post on SE, revision on SE revision pages, and post Markdown views on metasmoke, to check it against [Hugging Face's AI detector](https://huggingface.co/openai-detector).

The first click will test the Markdown text using the Hugging Face detector API for the likelihood it was created by GPT and report that as a percentage likelihood that the post is fake.

Lots of users and spammers edit the text provided by ChatGPT a bit, so if you click the button a second time (or use <kbd>Alt</kbd>-<kbd>Click</kbd>) an iframe to the detector demo page will open with the Markdown text pre-filled for you, so you can try some manipulations to see if the content was edited from ChatGPT source text. For example, on metasmoke it looks like:

[![Animation of a post on metasmoke. First the base % FAKE is determined of 24.94%, then an iframe is opened showing that after removal of the spammer's added text the detection is 99.86% FAKE](https://i.sstatic.net/5wyoP.gif)](https://i.sstatic.net/5wyoP.gif)



### Download

The script can be installed via [this link](https://github.com/Glorfindel83/SE-Userscripts/raw/master/openai-detector/openai-detector.user.js); source code can be found [here](https://github.com/Glorfindel83/SE-Userscripts/blob/master/openai-detector/openai-detector.user.js).

### Platform

Tested with Violentmonkey in Firefox on macOS, but it's likely to work on all browsers and platforms. (Please let me know if it doesn't!)

## Contact

Post a comment or answer here, or create an issue/pull request on GitHub.


  [1]: https://i.sstatic.net/mDMTr.png