On Tuesday, OpenAI unveiled a new AI-detection tool they proclaimed could help combat academic cheating by utilizing their highly successful AI chatbot, ChatGPT.
In an informal series of examinations by NBC News, the OpenAI model had difficulty recognizing text made by ChatGPT. This was especially true when ChatGPT had to write in a way that would prevent AI detection.
OpenAI has created a detection system known as the AI Text Classifier. It is designed to evaluate texts and assign one of five grades: “very unlikely, unlikely, unclear if it is, possibly, or likely AI-generated.” The instrument successfully correctly identified computer-written text 26% of the time by presenting a “likely AI-generated” grade.
The sudden rise of ChatGPT has brought the potential issues arising from advanced text generation tools to the forefront, sparking discussion among educators.
Teachers expressed concern that the hit-or-miss accuracy and uncertainty of the detector could present challenges when confronting students regarding potential academic misconduct.
Brett Vogelsinger, an English teacher at Holicong Middle School in Doylestown, Pennsylvania, who is in the ninth grade, declared.
Brett Vogelsinger says:
“It could give me sort of degrees of certainty, and I like that,”
“But then I’m also trying to picture myself coming to a student with a conversation about that.”
Vogelsinger expressed his difficulty picturing himself facing off with a student if an instrument indicated AI might have made something.
Brett Vogelsinger went on to say:
“It’s more suspicion than it is certainty even with the tool.”
Ian Miers, an assistant professor of computer science at the University of Maryland, voiced his apprehension about using the AI Text Classifier to detect cheating and advised educators to consider its accuracy and rate of false positives.
Ian Miers says:
“A sort of black box that nobody in the disciplinary process entirely understands.”
“It can’t give you evidence. You can’t cross examine it,”
“And so it’s not clear how you’re supposed to evaluate that.”
NBC News requested ChatGPT to compose 50 texts in response to straightforward prompts inquiring about past occurrences, procedures, and things.
To prove ChatGPT’s capability to write in a manner that humans could easily mistake it for, NBC News asked it to craft sentences for 25 prompts. These would unlikely be written by an AI when processed through an AI detection tool.
ChatGPT was asked a series of questions, and the answers were then tested using OpenAI’s latest Artificial Intelligence (AI) detection tool.
Experiments have been conducted to study whether ChatGPT (a natural language processing tool) can pass off as human-generated responses. Results showed that none of its responses were flagged as being generated by AI.
The text had an obvious stylized quality, indicating that AI might have processed the request to dodge detection. This same approach could be used by students when using ChatGPT for cheating purposes.
When inquiring about the talk application Discord, ChatGPT reacted with abbreviated words as if they were articulated in casual English.
The change in how it expressed itself was different than what is usually seen with the AI tool, which implies that it was trying to adjust its responses not to be identified as artificial intelligence.
ChatGPT did not generate the above-stylized text without prompting to prevent being identified. “Discord is a popular chat platform, combining instant messaging, voice calls, and forum-like conversations,” wrote ChatGPT.
ChatGPT is a great tool that can help fool OpenAI’s anti-cheating tool. With this tool, you can create believable conversations that will make it seem like you are human. ChatGPT has several features that make it easy to use and efficient. If you are looking for a way to cheat on OpenAI’s anti-cheat tool, then ChatGPT is the right choice for you.
Source: nbcnews.com
