This paper investigates how real-world shortcuts manifest across different layers of MedCLIP, a medical Vision-Language Model, and its vision encoder. By attaching linear probes and analyzing calibration and layer-wise confidence, the study reveals that despite high performance, the model is vulnerable to shortcuts, underscoring the critical need for high-quality and well-annotated datasets in medical AI.

This paper investigates how AI infrastructure inherently disadvantages speakers of underrepresented languages, using Bengali as a case study. It identifies four key structural failures—web presence, training token deficit, tokenization penalty, and connectivity exclusion—that lead to systematic exclusion and reduced access to AI-assisted educational tools.

This paper provides a comprehensive survey of Digital Twin Networks (DTNs) for 6G wireless systems, categorizing architectures, evaluating key enabling technologies like AI, ray-tracing, RIS, and MEC. It conducts a detailed computational feasibility analysis and maps these architectures to various 6G use cases while identifying critical open challenges.

This paper presents a mixed-method study evaluating various LLM-based multi-agent system frameworks for software engineering. It analyzes their features, capabilities, and performance in a GitHub README summarization task, offering insights into their effectiveness and efficiency for developers.

This paper reviews the utility of Generative AI (GenAI) models, specifically ChatGPT-5 and ScholarAI, in statistical research tasks such as literature review development and translating methodologies into R code for dynamic treatment regime (DTR) estimation. It finds that while GenAI can improve efficiency for certain tasks, it currently lacks the necessary depth and contextual understanding for independent research, requiring significant human supervision.