A study conducted by four researchers from across the globe and recently published on the Cornell Tech-owned arXiv, has declared Copyleaks AI Content Detector the most accurate for detecting Large Language Models (LLM) generated text.
In the study, the researchers collected 124 submissions from computer science students written before ChatGPT was created. They then generated 40 ChatGPT submissions and used the data to evaluate eight publicly-available LLM-generated text detectors.
“The purpose of this work is to inform the community of what LLM-generated text detectors work and which do not, but also to provide insights for educators to better maintain academic integrity in their courses.”
The researchers compiled historical assignment data dating back to 2016 from two published funded research-oriented institutions, one based in North America and another in South America, from upper-year undergraduate computer science and engineering students.
From the data, they analyzed a total of 164 submissions (124 submitted by humans, 30 were generated using ChatGPT, and 10 were generated by ChatGPT and altered using the Quillbot paraphrasing tool) and then compared each submission against eight LLM-generated text detectors, resulting in a total of 1,312 prediction results.
The eight LLM-generated detectors used for the study were found between January and February 2023 via social media (i.e., Facebook, Twitter, blog posts, etc.), online news articles, and previously available literature on LLM-generated text detection.
The following chart details the eight LLM-generated text detectors selected for the study:
All eight detectors were tested in April of 2023. Here are the key results.
The overall accuracy of the LLM-generated text detectors, measured using thresholds, sorted from best to worst.
The accuracy of LLM-generated text detectors, measured using weighted averages, sorted from best to worst.
The false positive readings on LLM-generated text, sorted from best to worst.
Maintaining content integrity post-ChatGPT and having the most accurate solutions makes a difference, especially regarding false positives.
It is crucial that AI model detectors continue to evolve and improve along with AI technology. At Copyleaks, we train our technology in real-time to improve the overall accuracy, foreign language detection, and, most importantly, to keep moving our .2% false positive rate as close to 0 as possible.
“Our results find that Copyleaks is the
most accurate LLM-generated text detector.”
That’s why with Copyleaks, you can have complete confidence in a solution that has all of your AI detection bases covered, evolves with the ever-changing technology, and ensures originality while inspiring authenticity. To read the full study, click here.