GPTZero set to counter AI plagiarism caused by services like ChatGPT

With the advent of AI, numerous applications and services have surfaced which are having a profound impact on our lives. Now that the advancements in AI technology are expanding exponentially, everybody is on the lookout for shortcuts and answers to even fairly simple questions on the internet to "save time." Those associated with academia, particularly, have been using online services such as AI chatbots to find quick solutions, remove plagiarism, and more.

Now, with the arrival of platforms like ChatGPT by OpenAI, which has been trained to generate long-form answers, many are using it to spin well-referenced and uncannily natural essays. Although OpenAI said that it would watermark ChatGPT output, people like Edward Tian are determined to solve this issue of AI plagiarism for good.

I spent New Years building GPTZero — an app that can quickly and efficiently detect whether an essay is ChatGPT or human written

— Edward Tian (@edward_the6) January 3, 2023

Over the new year holidays, the computer science student at Princeton developed an app called GPTZero. As the name suggests, this app can "quickly and efficiently" determine whether your essay is a ChatGPT creation. Tian showed off his creation in a series of tweets where he showcased examples of GPTZero in action. One of the examples showed that the New Yorker essay, "Frame of Reference" by John McPhee was composed by a person while a LinkedIn post was created by a bot.

here"s a quick demo with john mcphee"s "frame of reference" pic.twitter.com/WphxfxxFdr

— Edward Tian (@edward_the6) January 3, 2023

GPTZero makes use of "perplexity" and "burstiness" to determine a bot-written text. Perplexity, in machine learning, refers to how surprised a language model is when it sees new data while the phenomenon that the use of a term once in a document is likely to happen again is called burstiness. Human writing has more burstiness leading to more sentence variation, and language models can predict text better when the perplexity is lower.

As pointed out by Tian in a Substack newsletter, over 10,000 people have tested out the publicly available version of GPTZero on Streamlit to date. He also mentioned that he had updated the GPTZero model to reduce false positives. Furthermore, he is working on further updates to GPTZero including new capabilities and scalability.

While Microsoft has plans for using ChatGPT"s capabilities to its advantage, the New York City (NYC) education department has blocked access to ChatGPT on school devices and networks due to concerns regarding plagiarism.

Source: Polygon

Report a problem with article
Next Article

Microsoft backtracks one of the worst Start menu changes in Windows 11

Previous Article

Seagate 8TB, 12TB, 14TB CMR HDDs for NAS, Plex are once again selling at lowest prices