With AI chatbots like ChatGPT and Google Bard, it is so easy to just relax and let AI do the work for you. This is a problem most teachers are having – students not actually putting in the work. You know, sitting down to write their own papers by themselves. They instead depend on ChatGPT to do the work.
So to ensure that students are doing their work themselves, a solution to ChatGPT has arrived in the form of GPTZero. I tested it out and I was amazed by how accurate it was. It actually could tell the difference between an article written by a human and that written by an AI chatbot.
Before I explain why it is possible for GPTZero to distinguish between AI and human writing: let me show you my test.
My Test
I had two blocks of text: one I wrote by myself, and the other I used chatGPT to write.
Text A:
Sugar is an orphaned rich kid who loves her escorts. She doesn’t care about being in a relationship. All the wants to get hers and go. Also, she is very rude to the men she patronises. Her aunt, Kate is tired of her non-committal behaviour and hopes that one day, Sugar will find a good man to settle down with.
On the other side of town is Spencer, he is an orphaned writer who is experiencing writer’s block. There is a competition worth 20 million naira and he has no idea of what to write. His friend and housemate, Bob suggests that he goes undercover to write about the male escort industry from the POV of a customer.
Text B:
Sugar, an affluent orphan, finds solace in the company of escorts, uninterested in committed relationships. She prioritizes her own desires and remains indifferent to the men she engages. Frustrated by her non-committal attitude, her aunt Kate hopes Sugar will eventually settle down with a good man.
On the flip side, Spencer, an orphaned writer grappling with writer's block, faces a 20 million naira competition without a clear idea. His friend and housemate, Bob, proposes an unconventional solution: delving undercover to explore the male escort industry and document it from a customer's perspective.
When I ran Text A on GPTZero, my result was a 2% probability of AI writing the text:
When I ran Text B on GPTZero, my result was a 95% probability of AI writing the text:
Answer
If you want the answer, well, GPTZero was right, Text B was an AI-generated version of Text A.
But how was GPTZero able to figure this out? Well, I went researching.
How GPTZero distinguishes AI content from Human content
Using my trusty Beloga research assistant, I asked it this very same question and it stated that GPTZero uses two parameters to distinguish between a sentence written by a human from one done by an AI.
The parameters are Perplexity and Burstiness. PERPLEXITY has to do with the complexity of the sentence. If the sentence is complex, it is human. You know when Grammarly tells you that a sentence is hard to read or something like that? Yeah, that is complexity and your humanness shining through.
As for BURSTINESS – this is strictly about the length of the sentence. When writing, as a human being, the length of your sentences varies. Some are longer than others and some are shorter than others. For AI, however, everything is pretty uniform.
To conclude:
I don’t think I have much to say other than GPTZero works and if you are a teacher who wants to ensure that your student is actually putting in the work, this tool is your friend.
I hope you try it out for yourselves and see.
good!!!