MUHAYU, ‘GPT Killer’ function update… “Detects even GPT-4o generated sentences”

-

(Photo = Muhayu)

Artificial Intelligence (AI) specialist Muhayu (CEO Shin Dong-ho) announced on the twenty sixth that it has updated ‘GPT Killer’ to have the ability to detect generated sentences of OpenAI’s latest model ‘GPT-4o’.

To enhance detection accuracy, numerous assignments and self-introductions were generated and used for model training using prompts much like those utilized by actual students when creating assignments in each version of ChatGPT. After constructing the task data generated by GPT-4o, additional training was conducted on the present ChatGPT model, improving accuracy for all versions.

Consequently of our own test, the detection accuracy for sentences written by GPT4-o was 0.9695. Sentences generated by ‘GPT-3.5 Turbo’ and ‘GPT-4 Turbo’ also showed high accuracy of 0.9807 and 0.9824, respectively.

We also built an automatic pipeline to reply immediately to each ChatGPT update. We automated the technique of repeatedly monitoring and taking motion on model performance changes to robotically detect and compensate for performance degradation of GPT Killer because of ChatGPT model updates.

Currently, GPT Killer is being applied to Muhayu’s AI plagiarism check service ‘Copy Killer’ and AI document evaluation service ‘Prism’. Determine whether GPT-generated sentences are included in students’ theses and assignments, and job seekers’ self-introductions. It’s getting used by universities reminiscent of Chung-Ang University, University of Seoul, and Sungkyunkwan University, in addition to research institutes reminiscent of the National Research Foundation of Korea and the Korea Institute of Civil Engineering and Constructing Technology.

CEO Shin Dong-ho said, “Overseas, our DetectGPT solution not only helps teachers detect AI-generated student work, but has recently expanded its scope to incorporate government procurement agencies, grant writing agencies, hiring managers, and AI education data labelers. “GPT Killer may also function a guide to stop the indiscriminate use of generative AI and to put it to use properly by expanding the use cases to a wide range of ways,” he said.

Reporter Park Soo-bin sbin08@aitimes.com

ASK DUKE

What are your thoughts on this topic?
Let us know in the comments below.

0 0 votes
Article Rating
guest
0 Comments
Inline Feedbacks
View all comments

Share this article

Recent posts

0
Would love your thoughts, please comment.x
()
x