Breakthrough in AI Plagiarism: New Model with 99.7% Accuracy Rate of Detecting ChatGPT Writing
Feb 22, 2023
Sydney, New South Wales Feb 22, 2023 (Issuewire.com) - Edukado AI, has announced a breakthrough in AI text detection with the release of a latest version of their AICheatCheck model that achieved an accuracy rate of 99.7% and a false positive rate of only 0.046%. This represents a significant improvement over existing detection models.
The technical paper, called Keeping AI Honest in Education: Identifying GPT-Generated Text which can be found here, aims to provide educators and the public confidence that there can be a reliable method of detection of AI written text.
"What inspired the release of our white paper was that we found that most universities are not willing to trust a black-box system. For highly reputable institutions to entrust something as sensitive as Academic Integrity, we need to be transparent with educators about how our model works and enable them to verify the reliability of our systems." Arend Groot Bleumink, Chief Technology Officer.
To achieve a model with99.7%accuracy and the other validation metrics, Edukado AI conducted hundreds of experiments testing various models, data sources, parameters, and other variables.This new AICheatCheck model consists of over 100 million parameters which were fine-tuned for the purpose of classifying GPT-generated text specifically in the education context.
"Up until now, universities have been reluctant to officially implement any of the existing AI text detection tools due to the high false positive rates." - said Aaron Shikhule, Chief Operating Officer of Edukado AI.
Another yet-to-be-released tool has claimed to be able to detect GPT generated text 97% of the time with less than a 1/100 false positive rate. However, the team at Edukado AI is convinced that these metrics are not feasible for educational institutions.
A false positive rate of 1/100 would mean that in a typical university with 30,000 students, 300 students would be falsely accused of AI plagiarism and put through the academic misconduct process - which would be a catastrophe. On the other hand, the breakthrough false positive rate of 0.046% would translate to a manageable 13 students.
The company's Chief Strategy Officer, Kash Imam, currently in law school, says that "If a law school or medical school student is accused of plagiarism, the consequences would be career-ending as they would be considered unfit to practice.".
The new AICheatCheck model has been rigorously tested and validated, with results demonstrating its superior performance compared to existing text detection models. The model is also highly scalable with an API, making it suitable for use in large-scale academic environments. The reliability and scalability mean that educational institutions can finally deploy AI text detection within their Learning Management Systems (LMS) and implement it for a large number of students.
"We aim to build a future where education and AI go hand in hand without having a negative impact on the quality of education, but rather to improve the quality of education," said David Cyrus, the company's CEO.
Edukado AI is undertaking further research to enhance the reliability of the model and possibly additional methods of confirming the authenticity of a submitted text.
Source :Edukado AI
This article was originally published by IssueWire. Read the original article here.
More News
YOUR NEWS, OUR NETWORK.
Do you have Great News you want to tell the world?
Be it updates about your business or your community, you can make sure that it’s heard by submitting your story to our network reaching hundreds of news sites across 6 verticals.