ChatGPT and other AI apps are surfacing online as more people become reliant on them. Some individuals think that artificial intelligence helps them a lot by simplifying their tasks. However, many also believe that it's unethical to use them.
Educators claim that using chatbots raises a lot of questions. That's why some want their students to refrain from using them.
With this, a group of researchers developed DetectGPT, a special tool that will help teachers detect if the students use an AI-based tool for their homework.
DetectGPT Could be Teacher's Best Friend
According to a report by Business insider, DetectGPT can be a helpful tool in evaluating if a paper uses ChatGPT or other tools with large language models (LLMs).
The Stanford researchers can come up with software that can recognize whether the content relies on AI or not.
According to them, words, phrases, or texts that utilize LLMs can "occupy negative curvature regions of the model's log probability function."
This means that the tool can give "log probabilities" if an article has an element of LLM. The best part about DetectGPT is that it can accurately detect AI content by 95%.
With this tool, institutions and educators can greatly benefit from it. The rampant reliance on AI tools is something that shouldn't be bad at all times, given that humans still surpass artificial intelligence.
What makes AI apps such as ChatGPT look bad is how people abuse them to the point that they entirely depend on their capability.
Some schools in some parts of the world are already notified that AI tools are everywhere. The school heads warned that students who use these tools would receive a sanction.
At the time, we have no idea how DetectGPT works in detail, but soon enough, we might hear a word from the researchers about it.
Related Article: ChatGPT Passes MBA Exam: Achieves B/B Grade
GPTZero vs. ChatGPT
In other news, Polygon reported that a student created GPTZero, an app similar to DetectGPT that can detect if an article is human-written or ChatGPT-generated.
According to Edward Tian, he started developing the AI plagiarism tool during New Year's eve and tested if there was a way to increase its possibility of detecting dubious content.
Tian added that over 10,000 people have already tested his app so far. The public version of GPTZero is currently available on Steamlit.
Somehow, he hopes that the GPTZero model can further improve in the future by improving the "output results" and decreasing the "rate of false positives."
Tian said the criteria for detecting an essay or a passage were through its "burstiness" and "perplexity."
Burstiness tackles the comparison of one sentence to another sentence or phrase. Meanwhile, perplexity is more on the randomness of the text in a paragraph.
