Tag
MultiGhostBench is a multilingual benchmark for long-form LLM-generated text attribution under distribution shifts, featuring 928 books in six languages and highlighting performance degrades and no single method consistently best across settings.
The article describes a tool that distinguishes AI-generated comments from human-written ones in code using linguistic features like character frequency, and discusses its methodology and implications.
A preprint estimates that by the end of 2025, 89% of open-access biomedical papers show signs of LLM-assisted writing, using a new word-frequency-based method.
This paper presents EVIL-Detect, a multi-signal ensemble framework with conflict-aware fusion for detecting LLM-generated, refined, and human-written Chinese text in the NLPCC 2026 Shared Task 6, achieving first place with a macro-F1 of 0.8888.
MD-ProTector is a research paper proposing a prototype-based method for LLM-generated text detection, using multiple trainable reference vectors per class to capture within-class variation and improve robustness across domains, generators, languages, and adversarial edits.
This paper proposes EchoPrompt, a training-free detector for LLM-generated text that restores a latent prompt dependency by prepending a generic prefix and measuring likelihood gain differences between instruction-tuned and base models, achieving state-of-the-art zero-shot detection performance.
A reviewer recounts flagging two papers with fabricated authors that were accepted as orals, and reports that 68% of 22 reviewed submissions contained fabricated citations, LLM-generated content, or fake author lists. Multiple studies confirm tens of thousands of such papers in 2025 alone, with peer reviews increasingly AI-generated.
Pangram Labs announces Pangram 4, its most powerful AI detector yet, with significantly reduced false positive and false negative rates, robust detection across frontier models, and new image scan features.
Introduces DWT-Fusion, a training-free framework using discrete wavelet analysis of token log-probabilities for detecting LLM-generated text, achieving strong AUROC results on multiple datasets.
This paper investigates how well LLMs can detect their own generated content in educational contexts, finding that detection accuracy varies by task type and is unreliable for short-answer questions.
A developer explores using classical machine learning to detect LLM-generated web fiction, creating an open-source demo and model with ~85% single-sentence accuracy.
The paper introduces Telescope Perplexity, a metric that measures token repetition probability to detect LLM-generated text in a zero-shot manner, achieving state-of-the-art or competitive performance across diverse datasets.
This blog post explores using reinforcement learning to train a small model to detect writing styles of different LLMs, providing code and traces.
Pangram, a tool for detecting LLM use, reveals that many academics do not understand the difference between false positives and false negatives.
This paper evaluates the extent of undisclosed LLM-generated content in parliamentary texts from the UK and Sweden by training an interpretable text classifier, finding a steady increase in undisclosed LLM use from 2022 onwards.
Rodrigo Arias Mallo proposes using asciinema recordings as proof of human authorship for FOSS contributions to Dillo, arguing that LLMs struggle to convincingly generate such recordings.
DetectRL-X is a comprehensive multilingual benchmark for evaluating LLM-generated text detectors across 8 languages and 6 domains, including stress testing with AI-assisted writing operations and perturbations. It reveals strengths and limitations of current detectors in multilingual scenarios.
This paper addresses the degradation of likelihood-based machine-generated text detectors by identifying a Simpson's paradox in token-score aggregation. It proposes a learned local calibration step that significantly improves detection performance across various models and datasets.
Proposes LiSCP, a lightweight stylistic consistency profiling method for robust detection of LLM-generated textual content, focusing on feature stability under adversarial manipulation. Achieves superior performance on in-domain and cross-domain detection with notable robustness.
Andrew Kelley, creator of Zig, argues that LLM-assisted contributions are detectable through distinct mistakes and a 'digital smell,' comparing it to smoking in a non-smoking house.