GPTZero aims to detect ChatGPT-generated text and catch plagiarism

A report on TechSpot highlights a tool designed to detect the use of ChatGPT in plagiarism. Called ‘GPTZero’, the tool was created by 22-year-old Princeton University student Edward Tian and uses two key signals to identify text generated by ChatGPT.

These signals are called ‘perplexity’ and ‘burstiness’. Perplexity measures the complexity of a text – in this case, how unpredictable GPTZero finds it. High perplexity suggests the text was written by a human. Low perplexity points to AI-generated content – GPTZero looks for features such as text patterns to make that assessment.

Burstiness measures variation between sentences. Human writing typically mixes longer and shorter sentences, making this variation a useful signal. Edward Tian acknowledges that GPTZero is far from perfect (the tool is still in beta), but says it should provide a quick check when a text is suspected of having been generated by ChatGPT.

ChatGPT

Use of ChatGPT, OpenAI's AI chat platform, has been growing recently. The service frequently reports that it is ‘at capacity’, reflecting strong demand worldwide. Microsoft reportedly intends to invest $10 billion in the company, with potential applications including integrating the technology into its office software portfolio.