Listen to this story
|
After weeks of anticipation, Edward Tian, the developer of GPTZero, finally announced the launch of GPTZero–GPTZeroX. This application can successfully differentiate a text generated by human from large language models, particularly ChatGPT and alike.
The new GPTZero highlights the portions of text that are most likely to be AI-generated. As per Tian, this feature was heavily requested by educators.
Additionally, the model has a pipeline to handle file batch uploads in PDF, Word or in .txt format, allowing users to run multiple files through GPTZero.
Tian asserts that the new model is faster since it is processed by GPUs that are supported by AWS Lambda servers. The developer further asserts that they have created a Python API for businesses that have undergone stress testing to manage huge request volumes.
When checked, the GPTZero could identify a text written by ChatGPT; however, multiple users have complained about the model not working properly.
Problems with GPTZero
A Twitter user, while replying to the thread, claimed that the model detected his self-written paper as written by AI. According to him, the schools are implementing the model without testing its accuracy.
Another Twitter user claims that “The model is still flawed”. According to him, the model still shows some parts of tech as written by AI, which he wrote entirely.
To check the claims, we checked Nature.com’s article from 2015, when generative AI was not mainstream and found that GPTZero still detected some parts of that article as AI-generated.