The rise in malicious usage of large language models, such as fake content creation and academic plagiarism, has motivated the development of approaches that identify AI-generated text, including those based on watermarking or outlier detection. However, the robustness of these detection algorithms to paraphrases of AI-generated text remains unclear. To stress test these detectors, we build a 11B parameter paraphrase generation model (DIPPER) that can paraphrase paragraphs, condition on surrounding context, and control lexical diversity and content reordering. Paraphrasing text generated by three large language models (including GPT3.5-davinci-003) with DIPPER successfully evades several detectors, including watermarking, GPTZero, DetectGPT, and OpenAI's text classifier. For example, DIPPER drops detection accuracy of DetectGPT from 70.3% to 4.6% (at a constant false positive rate of 1%), without appreciably modifying the input semantics.To increase the robustness of AI-generated text detection to paraphrase attacks, we introduce a simple defense that relies on retrieving semantically-similar generations and must be maintained by a language model API provider. Given a candidate text, our algorithm searches a database of sequences previously generated by the API, looking for sequences that match the candidate text within a certain threshold. We empirically verify our defense using a database of 15M generations from a fine-tuned T5-XXL model and find that it can detect 80% to 97% of paraphrased generations across different settings while only classifying 1% of human-written sequences as AI-generated.

Recent advancements in Large Language Models (LLMs) and their increased accessibility have made it easier than ever for students to automatically generate texts, posing new challenges for educational institutions. To enforce norms of academic integrity and ensure students' learning, learning analytics methods to automatically detect LLM-generated text appear increasingly appealing. This paper benchmarks the performance of different state-of-the-art detectors in educational contexts, introducing a novel dataset, called Generative Essay Detection in Education (GEDE), containing over 900 student-written essays and over 12,500 LLM-generated essays from various domains. To capture the diversity of LLM usage practices in generating text, we propose the concept of contribution levels, representing students' contribution to a given assignment. These levels range from purely human-written texts, to slightly LLM-improved versions, to fully LLM-generated texts, and finally to active attacks on the detector by "humanizing" generated texts. We show that most detectors struggle to accurately classify texts of intermediate student contribution levels, like LLM-improved human-written texts. Detectors are particularly likely to produce false positives, which is problematic in educational settings where false suspicions can severely impact students' lives. Our dataset, code, and additional supplementary materials are publicly available at https://github.com/lukasgehring/Assessing-LLM-Text-Detection-in-Educational-Contexts.

A growing fraction of all code is sampled from Large Language Models (LLMs). We investigate the problem of attributing code generated by language models using hypothesis testing to leverage established techniques and guarantees. Given a set of samples $S$ and a suspect model $\mathcal{L}^*$, our goal is to assess the likelihood of $S$ originating from $\mathcal{L}^*$. Due to the curse of dimensionality, this is intractable when only samples from the LLM are given: to circumvent this, we use both samples and density estimates from the LLM, a form of access commonly available. We introduce $\mathsf{Anubis}$, a zero-shot attribution tool that frames attribution as a distribution testing problem. Our experiments on a benchmark of code samples show that $\mathsf{Anubis}$ achieves high AUROC scores ( $\ge0.9$) when distinguishing between LLMs like DeepSeek-Coder, CodeGemma, and Stable-Code using only $\approx 2000$ samples.

The rise in malicious usage of large language models, such as fake content creation and academic plagiarism, has motivated the development of approaches that identify AI-generated text, including those based on watermarking or outlier detection. However, the robustness of these detection algorithms to paraphrases of AI-generated text remains unclear. To stress test these detectors, we build a 11B parameter paraphrase generation model (DIPPER) that can paraphrase paragraphs, condition on surrounding context, and control lexical diversity and content reordering. Paraphrasing text generated by three large language models (including GPT3.5-davinci-003) with DIPPER successfully evades several detectors, including watermarking, GPTZero, DetectGPT, and OpenAI's text classifier. For example, DIPPER drops detection accuracy of DetectGPT from 70.3% to 4.6% (at a constant false positive rate of 1%), without appreciably modifying the input semantics.To increase the robustness of AI-generated text detection to paraphrase attacks, we introduce a simple defense that relies on retrieving semantically-similar generations and must be maintained by a language model API provider.

The widespread adoption of large language models (LLMs) has created an urgent need for robust tools to detect LLM-generated text, especially in light of \textit{paraphrasing} techniques that often evade existing detection methods. To address this challenge, we present a novel semantic-enhanced framework for detecting LLM-generated text (SEFD) that leverages a retrieval-based mechanism to fully utilize text semantics. Our framework improves upon existing detection methods by systematically integrating retrieval-based techniques with traditional detectors, employing a carefully curated retrieval mechanism that strikes a balance between comprehensive coverage and computational efficiency. We showcase the effectiveness of our approach in sequential text scenarios common in real-world applications, such as online forums and Q\&A platforms. Through comprehensive experiments across various LLM-generated texts and detection methods, we demonstrate that our framework substantially enhances detection accuracy in paraphrasing scenarios while maintaining robustness for standard LLM-generated content.

These can range from logistic regression models to convolutional In this paper, we study the problem of detecting neural networks (Weller and Woo, 2019) or LSTM models machine-generated text when the large language model (Kudugunta and Ferrara, 2018). These binary classifiers (LLM) it is possibly derived from is unknown. We do so by can also act as base learners in ensemble methods (Fayaz et apply ensembling methods to the outputs from DetectGPT al., 2020). These features can also be augmented with classifiers (Mitchell et al. 2023), a zero-shot model for additional information such as account data in the context machine-generated text detection which is highly accurate of social media bot detection. However, high classification when the generative (or base) language model is the same accuracy for these methods are reliant on sufficiently-long as the discriminative (or scoring) language model. We find text length and a sufficiently-diverse corpus of training that simple summary statistics of DetectGPT sub-model machine-generated samples in terms of stylometric and outputs yield an AUROC of 0.73 (relative to 0.61) while linguistic characteristics in order to prevent overfitting. As retaining its zero-shot nature, and that supervised learning such, these classifiers need to be continually trained and methods sharply boost the accuracy to an AUROC of 0.94 updated, limiting their usefulness (Pegoraro et al., 2023).

Large Language Models (LLMs) are increasingly used as "content farm" models (CFMs), to generate synthetic text that could pass for real news articles. This is already happening even for languages that do not have high-quality monolingual LLMs. We show that fine-tuning Llama (v1), mostly trained on English, on as little as 40K Italian news articles, is sufficient for producing news-like texts that native speakers of Italian struggle to identify as synthetic. We investigate three LLMs and three methods of detecting synthetic texts (log-likelihood, DetectGPT, and supervised classification), finding that they all perform better than human raters, but they are all impractical in the real world (requiring either access to token likelihood information or a large dataset of CFM texts). We also explore the possibility of creating a proxy CFM: an LLM fine-tuned on a similar dataset to one used by the real "content farm". We find that even a small amount of fine-tuning data suffices for creating a successful detector, but we need to know which base LLM is used, which is a major challenge. Our results suggest that there are currently no practical methods for detecting synthetic news-like texts 'in the wild', while generating them is too easy. We highlight the urgency of more NLP research on this problem.

We present the CheckforAI text classifier, a transformer-based neural network trained to distinguish text written by large language models from text written by humans. CheckforAI outperforms zero-shot methods such as DetectGPT as well as leading commercial AI detection tools with over 9 times lower error rates on a comprehensive benchmark comprised of ten text domains (student writing, creative writing, scientific writing, books, encyclopedias, news, email, scientific papers, short-form Q&A) and 8 open- and closed-source large language models. We propose a training algorithm, hard negative mining with synthetic mirrors, that enables our classifier to achieve orders of magnitude lower false positive rates on high-data domains such as reviews. Finally, we show that CheckforAI is not biased against nonnative English speakers and generalizes to domains and models unseen during training.

The burgeoning capabilities of large language models (LLMs) have raised growing concerns about abuse. DetectGPT, a zero-shot metric-based unsupervised machine-generated text detector, first introduces perturbation and shows great performance improvement. However, DetectGPT's random perturbation strategy might introduce noise, limiting the distinguishability and further performance improvements. Moreover, its logit regression module relies on setting the threshold, which harms the generalizability and applicability of individual or small-batch inputs. Hence, we propose a novel detector, Pecola, which uses selective strategy perturbation to relieve the information loss caused by random masking, and multi-pair contrastive learning to capture the implicit pattern information during perturbation, facilitating few-shot performance. The experiments show that Pecola outperforms the SOTA method by 1.20% in accuracy on average on four public datasets. We further analyze the effectiveness, robustness, and generalization of our perturbation method.

To combat the potential misuse of Natural Language Generation (NLG) technology, a variety of algorithms have been developed for the detection of AI-generated texts. Traditionally, this task is treated as a binary classification problem. Although supervised learning has demonstrated promising results, acquiring labeled data for detection purposes poses real-world challenges and the risk of overfitting. In an effort to address these issues, we delve into the realm of zero-shot machine-generated text detection. Existing zero-shot detectors, typically designed for specific tasks or topics, often assume uniform testing scenarios, limiting their practicality. In our research, we explore various advanced Large Language Models (LLMs) and their specialized variants, contributing to this field in several ways. In empirical studies, we uncover a significant correlation between topics and detection performance. Secondly, we delve into the influence of topic shifts on zero-shot detectors. These investigations shed light on the adaptability and robustness of these detection methods across diverse topics. The code is available at \url{https://github.com/yfzhang114/robustness-detection}.