Benchmarking Machine-Generated Text Detection with Expert and LLM Edited Outputs: The Beemo Dataset
Expert editing can significantly reduce the detectability of machine-generated text, while LLM-based editing is less effective at evading detection.
Detecting Machine-Generated Text at the Word Level in Partially Machine-Generated Documents
This research paper introduces and evaluates new approaches for accurately identifying machine-generated text segments within partially machine-generated documents, demonstrating significant improvements over existing methods and highlighting potential applications for detecting AI-generated content.
Benchmarking Neural News Detection in English, Turkish, Hungarian, and Persian
This research introduces a novel multilingual benchmark dataset for detecting machine-generated news, exploring the performance of various classifiers and highlighting the potential of linguistically informed and large language model-based approaches for robust and interpretable detection across languages.
잠재 공간 변수를 사용한 기계 생성 장문 콘텐츠 탐지: 디코딩 설정, 프롬프트 변형 및 적대적 공격에 대한 강력성
본 논문에서는 기계 생성 텍스트와 인간 작성 텍스트를 구별하기 위해 이벤트 전환과 같은 잠재 공간 변수를 활용하는 새로운 탐지 프레임워크를 제안하며, 이는 기존 탐지기가 취약했던 다양한 생성 설정 및 적대적 공격에 대한 강력성을 보여줍니다.
Detecting Machine-Generated Text: A Latent-Space Approach Focusing on Event Transitions
Analyzing event transitions within the latent structure of a text provides a robust method for detecting machine-generated content, even when traditional token-based methods fail.
다국어 기계 생성 텍스트 탐지에서 저자 신원 은닉의 영향 및 대응 방안
고급 언어 모델(LLM)이 생성한 텍스트를 사람이 작성한 텍스트와 구별하는 것은 점점 어려워지고 있으며, 특히 다국어 환경에서는 저자 신원 은닉(AO) 기법을 통해 탐지를 회피하는 것이 가능해짐에 따라, 탐지 모델의 정확도를 높이기 위한 연구가 필요하다.
Detecting Machine-Generated Text Across Diverse Domains and Language Models
Combining linguistic features and language model embeddings can effectively distinguish machine-generated text from human-written text, even across unseen language models and domains.
Detecting Machine-Generated Text: Exploring Contrastive Learning for Robust and Efficient Classification
Contrastive learning can be an effective approach for detecting machine-generated text, even with a single model and without relying on the specific text generation model used.
AIpom: A Two-Stage Pipeline for Detecting Human-Machine Boundary in Mixed Text
AIpom, a novel method for human-machine mixed text detection, leverages a pipeline of decoder and encoder models to accurately identify the boundary between human-written and machine-generated text.
M4: Multi-Generator, Multi-Domain, and Multi-Lingual Black-Box Machine-Generated Text Detection
Large language models have led to an increase in machine-generated content, raising concerns about potential misuse. This study focuses on creating automated systems to detect machine-generated texts and address misuse.
