DE-GPT: A MODEL DETECTOR TO DISTINGUISH BETWEEN LLMS AND HUMAN TEXT

SETIAWAN, ALEXANDRO (2023) DE-GPT: A MODEL DETECTOR TO DISTINGUISH BETWEEN LLMS AND HUMAN TEXT. Other thesis, UNIVERSITAS KHATOLIK SOEGIJAPRANATA.

	Text 19.K1.0037-ALEXANDRO SETIAWAN-COVER_a.pdf Download (641kB)
	Text 19.K1.0037-ALEXANDRO SETIAWAN-BAB I_a.pdf Restricted to Registered users only Download (91kB)
	Text 19.K1.0037-ALEXANDRO SETIAWAN-BAB II_a.pdf Restricted to Registered users only Download (164kB)
	Text 19.K1.0037-ALEXANDRO SETIAWAN-BAB III_a.pdf Restricted to Registered users only Download (583kB)
	Text 19.K1.0037-ALEXANDRO SETIAWAN-BAB IV_a.pdf Restricted to Registered users only Download (805kB)
	Text 19.K1.0037-ALEXANDRO SETIAWAN-BAB V_a.pdf Restricted to Registered users only Download (87kB)
	Text 19.K1.0037-ALEXANDRO SETIAWAN-DAPUS_a.pdf Download (151kB)
	Text 19.K1.0037-ALEXANDRO SETIAWAN-LAMP_a.pdf Restricted to Registered users only Download (188kB)

Abstract

The advent of large language models (LLMs), including ChatGPT Google Bard, and Bing AI, has had a transformative impact on natural language processing. However, the rise of these models has also introduced a new challenge: distinguishing between text produced by humans and that generated by LLMs. Accurate detection of LLM-generated text is vital due to the risk of disinformation, fake news, and automated spam, and has significant implications across various fields such as journalism and social media. This study aims to tackle this challenge by fine-tuning the GPT-Neo model with variant of 1.3B parameters to detect and identify text generated by LLMs or text produced by human. The objectives include developing a fine-tuned GPT-Neo 1.3b model for generated text detection, evaluating its performance, and conducting a comparative analysis with other existing models. This research will utilize a customized dataset comprising both LLM-generated and human-authored texts. The results of this research could provide valuable insights into the practical applications and potential implications of accurate text generation detection in real-world settings.

Item Type:	Thesis (Other)
Subjects:	000 Computer Science, Information and General Works > 004 Data processing & computer science
Divisions:	Faculty of Computer Science > Department of Informatics Engineering
Depositing User:	Mr Yosua Norman Rumondor
Date Deposited:	05 Oct 2023 06:51
Last Modified:	05 Oct 2023 06:51
URI:	http://repository.unika.ac.id/id/eprint/32973

Actions (login required)

View Item