Computer Science · Information Systems
In a test of eight open-weight AI language models, factual errors were more common in Uzbek than in Russian or English. Models adapted for Uzbek largely closed the gap, pointing to a reliability concern for students who use AI as a study aid.
A systematic review finds that AI-writing detectors can produce false positives, particularly for non-native English writing and structured academic prose. The authors warn that relying on these tools alone could make academic assessments less fair for some language groups.
Among 380 Chinese undergraduates who had used large language models, stronger prompt-writing ability was strongly associated with perceived academic gains. In a separate test, a five-part prompt structure produced higher-rated responses than basic prompts across four task types and four AI systems.
A systematic review finds that AI-writing detectors can falsely classify human-written student work as AI-generated. The review identifies language and algorithmic bias, limited contextual understanding and insufficiently diverse training data as major contributors.
A systematic review of 72 studies published from 2024 to 2026 examines how generative AI is changing assessment in higher education. The research points to more emphasis on students’ learning processes, authentic tasks and spoken or interactive assessments, while educators are using AI for tasks including assessment design, grading and feedback.
A qualitative study of 24 university students found that neurodivergent participants used generative AI to manage academic demands, energy and cognitive processing. They described it as essential scaffolding but also raised concerns about authenticity, over-reliance and interfaces that may exclude some users.
