Students using ChatGPT is making teachers more anxious. Does an antidote exist for this?
Anti ChatGPT, GPTZero, is an AI-powered writing detection tool. The software employs complex algorithms to determine if an article was written by a human or created by an AI chatbot such as ChatGPT. It considers numerous aspects of the writing, such as word choice, sentence structure, and coherence, as well as any changes made to the essay. ChatGPT is an artificial intelligence-based chatbot that has lately been prohibited from various school devices and Wi-Fi networks due to its potential to facilitate cheating. GPTZero, on the other hand, is designed to assist teachers in rapidly and precisely determining if a student’s essay was written by them or created via ChatGPT.
GPTZero, an antidote for ChatGPT:
While ChatGPT is still far from being a trustworthy source of information, students have already found a practical application for it: plagiarism in essays. The AI chatbot can write essays in a matter of seconds that would normally take a human writer hour to complete. A Princeton University student studying computer science, Edward Tian, obviously did not enjoy this and spent his New Year’s holidays developing an app that can assist in determining if a paragraph was written by a person or using ChatGPT.
The GPTZero software, according to Tian, can “quickly and efficiently detect whether an essay is ChatGPT or human-produced.” A recent restriction on ChatGPT from school computers and Wi-Fi networks in New York City’s public schools stems from concerns about “AI plagiarism.” Although it may be a step in the right direction, kids can still get around it by writing their essays at home, making this measure ineffective. GPTZero, on the other hand, allows teachers to distinguish between essays produced by AI and those produced by humans, which is a more intelligent approach.
Tian posted a few videos comparing GPTZero’s analysis of a New Yorker piece with a LinkedIn post produced by ChatGPT in his tweets. The tool awarded the article 114 points and accurately identified it as having been written by a human, whereas the LinkedIn post received 40 points and was once more correctly identified as having been “AI generated” by the tool. According to GPTZero’s developer, they will be spending the coming weeks strengthening the model and analysis. GPTZero is now in its beta phase. From its official website, gptzero.me, you can try it out.
How does GPTZero work?
GPTZero primarily uses two factors to identify whether the text is AI-generated or not:
Burstiness, which assesses the variety of sentences, is the periodic ups and downs in the activity or frequency of an event. A significant characteristic of a bot is that it tends to produce text with consistently long sentences, whereas when we write, we tend to ignore such details and write short sentences next to large sentences. AI-generated text is distinguished from non-AI-generated text using this concept.
Perplexity is the capacity of a probability distribution to correctly forecast a sample. The text’s level of complexity is returned. It operates under the premise that the more complex the text, the more probable it is to have been produced by a human. Additionally, because a bot is educated on such data, each text it creates follows a specific pattern. Since GPTZero was trained on text data, it can immediately identify artificial intelligence (AI)-generated text when it encounters subject matter that is comparable to its training data.
Conclusion: GPTZero can readily detect plagiarism because ChatGPT and GPTZero were both trained on almost identical types of text data, which is the simplest way to explain how it works. Some users found flaws in the program after trying the bot, and Tian recognized that the bot is still under development and needs more effort to improve its functionality.
