Jooyoung Lee

Jooyoung Lee, Thai Le, Jinghui Chen et al.

Past literature has illustrated that language models (LMs) often memorize parts of training instances and reproduce them in natural language generation (NLG) processes. However, it is unclear to what extent LMs "reuse" a training corpus. For instance, models can generate paraphrased sentences that are contextually similar to training samples. In this work, therefore, we study three types of plagiarism (i.e., verbatim, paraphrase, and idea) among GPT-2 generated texts, in comparison to its training data, and further analyze the plagiarism patterns of fine-tuned LMs with domain-specific corpora which are extensively used in practice. Our results suggest that (1) three types of plagiarism widely exist in LMs beyond memorization, (2) both size and decoding methods of LMs are strongly associated with the degrees of plagiarism they exhibit, and (3) fine-tuned LMs' plagiarism patterns vary based on their corpus similarity and homogeneity. Given that a majority of LMs' training data is scraped from the Web without informing content owners, their reiteration of words, phrases, and even core ideas from training sets into generated texts has ethical implications. Their patterns are likely to exacerbate as both the size of LMs and their training data increase, raising concerns about indiscriminately pursuing larger models with larger training corpora. Plagiarized content can also contain individuals' personal and sensitive information. These findings overall cast doubt on the practicality of current LMs in mission-critical writing tasks and urge more discussions around the observed phenomena. Data and source code are available at https://github.com/Brit7777/LM-plagiarism.

Adaku Uchendu, Jooyoung Lee, Hua Shen et al.

Advances in Large Language Models (e.g., GPT-4, LLaMA) have improved the generation of coherent sentences resembling human writing on a large scale, resulting in the creation of so-called deepfake texts. However, this progress poses security and privacy concerns, necessitating effective solutions for distinguishing deepfake texts from human-written ones. Although prior works studied humans' ability to detect deepfake texts, none has examined whether "collaboration" among humans improves the detection of deepfake texts. In this study, to address this gap of understanding on deepfake texts, we conducted experiments with two groups: (1) nonexpert individuals from the AMT platform and (2) writing experts from the Upwork platform. The results demonstrate that collaboration among humans can potentially improve the detection of deepfake texts for both groups, increasing detection accuracies by 6.36% for non-experts and 12.76% for experts, respectively, compared to individuals' detection accuracies. We further analyze the explanations that humans used for detecting a piece of text as deepfake text, and find that the strongest indicator of deepfake texts is their lack of coherence and consistency. Our study provides useful insights for future tools and framework designs to facilitate the collaborative human detection of deepfake texts. The experiment datasets and AMT implementations are available at: https://github.com/huashen218/llm-deepfake-human-study.git

Jason Lucas, Adaku Uchendu, Michiharu Yamashita et al.

Recent ubiquity and disruptive impacts of large language models (LLMs) have raised concerns about their potential to be misused (.i.e, generating large-scale harmful and misleading content). To combat this emerging risk of LLMs, we propose a novel "Fighting Fire with Fire" (F3) strategy that harnesses modern LLMs' generative and emergent reasoning capabilities to counter human-written and LLM-generated disinformation. First, we leverage GPT-3.5-turbo to synthesize authentic and deceptive LLM-generated content through paraphrase-based and perturbation-based prefix-style prompts, respectively. Second, we apply zero-shot in-context semantic reasoning techniques with cloze-style prompts to discern genuine from deceptive posts and news articles. In our extensive experiments, we observe GPT-3.5-turbo's zero-shot superiority for both in-distribution and out-of-distribution datasets, where GPT-3.5-turbo consistently achieved accuracy at 68-72%, unlike the decline observed in previous customized and fine-tuned disinformation detectors. Our codebase and dataset are available at https://github.com/mickeymst/F3.

Eun Jung Yeo, Hyungshin Ryu, Jooyoung Lee et al. · cmu

This paper presents a large-scale analysis of L2 Korean pronunciation error patterns from five different language backgrounds, Chinese, Vietnamese, Japanese, Thai, and English, by using automatic phonetic transcription. For the analysis, confusion matrices are generated for each L1, by aligning canonical phone sequences and automatically transcribed phone sequences obtained from fine-tuned Wav2Vec2 XLS-R phone recognizer. Each value in the confusion matrices is compared to capture frequent common error patterns and to specify patterns unique to a certain language background. Using the Foreign Speakers' Voice Data of Korean for Artificial Intelligence Learning dataset, common error pattern types are found to be (1) substitutions of aspirated or tense consonants with plain consonants, (2) deletions of syllable-final consonants, and (3) substitutions of diphthongs with monophthongs. On the other hand, thirty-nine patterns including (1) syllable-final /l/ substitutions with /n/ for Vietnamese and (2) /\textturnm/ insertions for Japanese are discovered as language-dependent.

Thai Le, Jooyoung Lee, Kevin Yen et al.

We proposes a novel algorithm, ANTHRO, that inductively extracts over 600K human-written text perturbations in the wild and leverages them for realistic adversarial attack. Unlike existing character-based attacks which often deductively hypothesize a set of manipulation strategies, our work is grounded on actual observations from real-world texts. We find that adversarial texts generated by ANTHRO achieve the best trade-off between (1) attack success rate, (2) semantic preservation of the original text, and (3) stealthiness--i.e. indistinguishable from human writings hence harder to be flagged as suspicious. Specifically, our attacks accomplished around 83% and 91% attack success rates on BERT and RoBERTa, respectively. Moreover, it outperformed the TextBugger baseline with an increase of 50% and 40% in terms of semantic preservation and stealthiness when evaluated by both layperson and professional human workers. ANTHRO can further enhance a BERT classifier's performance in understanding different variations of human-written toxic texts via adversarial training when compared to the Perspective API.

Jooyoung Lee, Toshini Agrawal, Adaku Uchendu et al.

Recent studies have raised concerns about the potential threats large language models (LLMs) pose to academic integrity and copyright protection. Yet, their investigation is predominantly focused on literal copies of original texts. Also, how LLMs can facilitate the detection of LLM-generated plagiarism remains largely unexplored. To address these gaps, we introduce \textbf{\sf PlagBench}, a dataset of 46.5K synthetic text pairs that represent three major types of plagiarism: verbatim copying, paraphrasing, and summarization. These samples are generated by three advanced LLMs. We rigorously validate the quality of PlagBench through a combination of fine-grained automatic evaluation and human annotation. We then utilize this dataset for two purposes: (1) to examine LLMs' ability to transform original content into accurate paraphrases and summaries, and (2) to evaluate the plagiarism detection performance of five modern LLMs alongside three specialized plagiarism checkers. Our results show that GPT-3.5 Turbo can produce high-quality paraphrases and summaries without significantly increasing text complexity compared to GPT-4 Turbo. However, in terms of detection, GPT-4 outperforms other LLMs and commercial detection tools by 20%, highlights the evolving capabilities of LLMs not only in content generation but also in plagiarism detection. Data and source code are available at https://github.com/Brit7777/plagbench.

Jooyoung Lee, Fan Yang, Thanh Tran et al.

We introduce a novel framework, LM-Guided CoT, that leverages a lightweight (i.e., <1B) language model (LM) for guiding a black-box large (i.e., >10B) LM in reasoning tasks. Specifically, the lightweight LM first generates a rationale for each input instance. The Frozen large LM is then prompted to predict a task output based on the rationale generated by the lightweight LM. Our approach is resource-efficient in the sense that it only requires training the lightweight LM. We optimize the model through 1) knowledge distillation and 2) reinforcement learning from rationale-oriented and task-oriented reward signals. We assess our method with multi-hop extractive question answering (QA) benchmarks, HotpotQA, and 2WikiMultiHopQA. Experimental results show that our approach outperforms all baselines regarding answer prediction accuracy. We also find that reinforcement learning helps the model to produce higher-quality rationales with improved QA performance.

JooYoung Lee, Emily Booth, Hany Farid et al.

What if misinformation is not an information problem at all? To understand the role of news publishers in potentially unintentionally propagating misinformation, we examine how far-right and fringe online groups share and leverage established legacy news media articles to advance their narratives. Our findings suggest that online fringe ideologies spread through the use of content that is consensus-based and "factually correct". We found that Australian news publishers with both moderate and far-right political leanings contain comparable levels of information completeness and quality; and furthermore, that far-right Twitter users often share from moderate sources. However, a stark difference emerges when we consider two additional factors: 1) the narrow topic selection of articles by far-right users, suggesting that they cherry pick only news articles that engage with their preexisting worldviews and specific topics of concern, and 2) the difference between moderate and far-right publishers when we examine the writing style of their articles. Furthermore, we can identify users prone to sharing misinformation based on their communication style. These findings have important implications for countering online misinformation, as they highlight the powerful role that personal biases towards specific topics and publishers' writing styles have in amplifying fringe ideologies online.

Konstantin Sozykin, Stanislav Protasov, Adil Khan et al.

Automatic analysis of the video is one of most complex problems in the fields of computer vision and machine learning. A significant part of this research deals with (human) activity recognition (HAR) since humans, and the activities that they perform, generate most of the video semantics. Video-based HAR has applications in various domains, but one of the most important and challenging is HAR in sports videos. Some of the major issues include high inter- and intra-class variations, large class imbalance, the presence of both group actions and single player actions, and recognizing simultaneous actions, i.e., the multi-label learning problem. Keeping in mind these challenges and the recent success of CNNs in solving various computer vision problems, in this work, we implement a 3D CNN based multi-label deep HAR system for multi-label class-imbalanced action recognition in hockey videos. We test our system for two different scenarios: an ensemble of $k$ binary networks vs. a single $k$-output network, on a publicly available dataset. We also compare our results with the system that was originally designed for the chosen dataset. Experimental results show that the proposed approach performs better than the existing solution.

Leonard Johard, Victor Rivera, Manuel Mazzara et al.

In this paper we propose an algorithm, Simple Hebbian PCA, and prove that it is able to calculate the principal component analysis (PCA) in a distributed fashion across nodes. It simplifies existing network structures by removing intralayer weights, essentially cutting the number of weights that need to be trained in half.

Rasul Tumyrkin, Manuel Mazzara, Mohammad Kassab et al.

It is well known that the software process in place impacts the quality of the resulting product. However, the specific way in which this effect occurs is still mostly unknown and reported through anecdotes. To gather a better understanding of such relationship, a very large survey has been conducted during the last year and has been completed by more than 100 software developers and engineers from 21 countries. We have used the percentage of satisfied customers estimated by the software developers and engineers as the main dependent variable. The results evidence some interesting patterns, like that quality attribute of which customers are more satisfied appears functionality, architectural styles may not have a significant influence on quality, agile methodologies might result in happier customers, larger companies and shorter projects seems to produce better products.