Detecting AI-Generated Text

How is AI-generated text detected?

Various experimental tools have been developed to detect AI-generated text and assess its quality. These tools can be useful in identifying patterns that mimic natural writing. One such tool, Giant Language Model Test Room (GLTR), released by Harvard and the MIT-IBM Watson AI Lab, scans text and highlights words based on their level of randomness.

Another tool, called GPTZero, developed by Edward Tian at Princeton, measures the likelihood that ChatGPT generated a piece of content by analyzing its perplexity (randomness) and burstiness (variance). OpenAI has also introduced a tool called OpenAI AI Text Classifier that can judge the nature of text longer than 1,000 characters.