Saved in:
Bibliographic Details
Main Authors: Ji, Jiazhou, Guo, Jie, Qiu, Weidong, Huang, Zheng, Xu, Yang, Lu, Xinru, Jiang, Xiaoyu, Li, Ruizhe, Li, Shujun
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2502.12743
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913908829192192
author Ji, Jiazhou
Guo, Jie
Qiu, Weidong
Huang, Zheng
Xu, Yang
Lu, Xinru
Jiang, Xiaoyu
Li, Ruizhe
Li, Shujun
author_facet Ji, Jiazhou
Guo, Jie
Qiu, Weidong
Huang, Zheng
Xu, Yang
Lu, Xinru
Jiang, Xiaoyu
Li, Ruizhe
Li, Shujun
contents Distinguishing between human- and LLM-generated texts is crucial given the risks associated with misuse of LLMs. This paper investigates detection and explanation capabilities of current LLMs across two settings: binary (human vs. LLM-generated) and ternary classification (including an ``undecided'' class). We evaluate 6 close- and open-source LLMs of varying sizes and find that self-detection (LLMs identifying their own outputs) consistently outperforms cross-detection (identifying outputs from other LLMs), though both remain suboptimal. Introducing a ternary classification framework improves both detection accuracy and explanation quality across all models. Through comprehensive quantitative and qualitative analyses using our human-annotated dataset, we identify key explanation failures, primarily reliance on inaccurate features, hallucinations, and flawed reasoning. Our findings underscore the limitations of current LLMs in self-detection and self-explanation, highlighting the need for further research to address overfitting and enhance generalizability.
format Preprint
id arxiv_https___arxiv_org_abs_2502_12743
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle "I know myself better, but not really greatly": How Well Can LLMs Detect and Explain LLM-Generated Texts?
Ji, Jiazhou
Guo, Jie
Qiu, Weidong
Huang, Zheng
Xu, Yang
Lu, Xinru
Jiang, Xiaoyu
Li, Ruizhe
Li, Shujun
Computation and Language
Artificial Intelligence
Distinguishing between human- and LLM-generated texts is crucial given the risks associated with misuse of LLMs. This paper investigates detection and explanation capabilities of current LLMs across two settings: binary (human vs. LLM-generated) and ternary classification (including an ``undecided'' class). We evaluate 6 close- and open-source LLMs of varying sizes and find that self-detection (LLMs identifying their own outputs) consistently outperforms cross-detection (identifying outputs from other LLMs), though both remain suboptimal. Introducing a ternary classification framework improves both detection accuracy and explanation quality across all models. Through comprehensive quantitative and qualitative analyses using our human-annotated dataset, we identify key explanation failures, primarily reliance on inaccurate features, hallucinations, and flawed reasoning. Our findings underscore the limitations of current LLMs in self-detection and self-explanation, highlighting the need for further research to address overfitting and enhance generalizability.
title "I know myself better, but not really greatly": How Well Can LLMs Detect and Explain LLM-Generated Texts?
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2502.12743