GPTZero works by analyzing a piece of text and determining if there is a high or low indication that a bot wrote it. It looks for two signs: “disruption” and “explosion.” “Preplexity” is how likely each word is to be suggested by the bot. A human would be more random. “Burstness” spikes measure how annoying each sentence is. A bot would likely have the same confusing sentence for sentence, but a human is going to write with spikes – maybe a long, complex sentence followed by a shorter one. Such.
To test Tian’s creation, I fed him a short essay written by ChatGPT using a prompt that a high school cheater might try: Describe the central theme of Hamlet. (“The central theme of Shakespeare’s play ‘Hamlet’ is the struggle of the protagonist, Hamlet, to come to terms with the fact that his uncle has usurped the throne by killing his father.“Blah blah etc.)
GPTZero gave the article a perplexity score of 10 and a burst score of 19 (these are fairly low scores, Tian explained, meaning the author was more likely to be a bot). He correctly deduced that it was likely written by an AI.
For comparison, I entered the first half of this article, which I wrote myself, into the tool. confusion: 39; Eruption: 387. “I want people to use ChatGPT,” he said.) Ultimately, GPTZero considered the subject to be human. Right!
However, GPTZero’s exact success rate is unclear. At least one Twitter user said that it failed to catch a few patterns written by their AI. Elsewhere on the platform, the response has been mixed: adults are applauding the effort, and others, mostly teenagers, have called Tian “narc“
Tian said The Daily Beast in an interview That after his tweet about it, his DM Venture Capital was blowing up with interest. For now, though, he plans to keep his creations free and accessible. “I want to support new English teachers everywhere,” he said.
Read full article here