Android

GPTZero Origin: browser plugin exposes AI content

Spam lords and lazy learners, take note. Edward Tian, ​​the Canadian developer of GPTZero Origin, is doing well. His eponymous startup already has more than 3,000 paying subscribers and has rolled out the first plugin to expose AI-generated text.

Robot apocalypse put in education – for a while

Edward Tian. Source: Twitter handle

Before ChatGPT, students’ language skills drove their teachers to despair, but suddenly the quality of paper submissions skyrocketed. Had aliens struck with an IQ-raising ray? I wish it was true. After a few days it became clear why: smart students had discovered the joys of ChatGPT en masse.

Would ChatGPT mean the end of the essay and paper as a way to test students’ insights? The teaching part of humanity was close to despair, but AI experts struck back against the robot apocalypse. With tools such as Edward Tian’s GPTZero, which we previously wrote about on Apparata in January 2023.

GPTZero Origin, a browser plugin that exposes AI

A few months later it appears that Tian has not been idle. In the meantime, he has set up a startup around his immensely popular tool and his team has developed a browser plugin, for the time being only for the Google Chrome browser (and the Unix variant Chromium). You can find these here.

The plugin works as follows. After installing, log in. To scan specific text on a web page, select it and ask GPTZero Origin to scan it. This option can be found in the menu that you see when you press the right mouse button. By default, you get ten times the opportunity to scan a text. Of course we put it to the test. The Apparata articles I scanned all appear to be 100% human-written. Which is also true, unless some advanced AI hacked into my brain.

GPTZero Origin
According to the plugin, this text is “entirely human”, but you can see from the very low scores for Perplexity and Burstiness that this is most likely coming from an AI. Which is also true in this case.

Fraudsters escape (for the time being) with GPTZero Origin…

Unfortunately, the tool does not seem to really work. I put it to the test with ChatGPT 3.5 and ChatGPT 4.0. In both cases, the tool indicated that this text was written by humans. So it is necessary to look at the details. That is not possible with GPTZero Origin, but only with GPTZero itself.

GPTZero Origin
Applied to my own text, the score for Perplexity and Burstiness appears to be two orders of magnitude higher. Phew, I’m not a robot after all.

But not if you pay attention to Perplexity and Burstiness yourself

The text written by an AI scores relatively low on the parameters Perplexity Score (a measure of the chance content in a text: unusual word combinations) and Burstiness Score (variations in Perplexity).

Our provisional final verdict: Use the GPTZero site yourself and pay attention to the scores for Perplexity and Burstiness. Hopefully Tian and his team will manage to fix this. Because otherwise the plugin is quite user-friendly.

Leave a Reply

Your email address will not be published. Required fields are marked *