Yubin Choi

Privacy has emerged as a critical societal concern in large language models (LLMs). Designing an effective and usable privacy experience for end-users requires more than technical mitigation; it necessitates a careful understanding and balancing of the risks and benefits perceived by the public. To identify public perceptions towards LLMs, we surveyed 1,073 individuals from the US, UK, and South Korea in three healthcare scenarios. The findings revealed that privacy concerns related to data leaks and disclosure without consent are major issues for end users. Across the studied countries, the potential for personalization and public benefits outweighed these concerns and positively influenced the willingness to disclose personal data. We found cross-cultural and individual differences in privacy perceptions, highlighting the need for culturally and personally tailored privacy designs. Given the limited privacy awareness among participants, we discuss opportunities for the responsible deployment of privacy-preserving LLMs and the associated legal implications.

Creating games together is both a playful and effective way to develop skills in computational thinking, collaboration, and more. However, game development can be challenging for younger developers who lack formal training. While teenage developers frequently turn to online communities for peer support, their experiences may vary. To better understand the benefits and challenges teens face within online developer communities, we conducted interviews with 18 teenagers who created games or elements in Roblox and received peer support from one or more online Roblox developer communities. Our findings show that developer communities provide teens with valuable resources for technical, social, and career growth. However, teenagers also struggle with inter-user conflicts and a lack of community structure, leading to difficulties in handling complex issues that may arise, such as financial scams. Based on these insights, we propose takeaways for creating positive and safe online spaces for teenage game creators.

International aid aimed at fostering economic growth and improving the welfare of developing countries is essential for achieving sustainable development. This study leverages an extensive dataset comprising over one million aid projects, capturing various attributes such as project descriptions, policy objectives, aid modalities, donor and recipient countries, and annual budgets. By mapping these projects to the United Nations’ Sustainable Development Goals (SDGs) using a multi-label classification approach, we propose an innovative AI model to alleviate the labor-intensive process of manual classification. Our approach demonstrates how integrating prior knowledge from large language models effectively addresses the complexities posed by the heterogeneous nature of development aid projects while respecting the specific contexts of each SDG. Validation confirms strong classification performance, with notable improvements in F1-score and AUROC compared to baseline models, further supported by human expert reviews in the domain of development cooperation. When applied to broader datasets, our model has the potential to uncover unseen trends in the temporal evolution of international development cooperation, stratified by the recipient country’s development status (e.g., low, middle, or upper-middle income). These insights offer valuable guidance to policymakers and non-governmental organizations, enabling more efficient project management and enhanced impact on sustainable development.

Most video-based learning content is designed for desktops without considering mobile environments. We (1) investigate the gap between mobile learners’ challenges and video engineers’ considerations using mixed methods and (2) provide design guidelines for creating mobile-friendly MOOC videos. To uncover learners’ challenges, we conducted a survey (n=134) and interviews (n=21), and evaluated the mobile adequacy of current MOOCs by analyzing 41,722 video frames from 101 video lectures. Interview results revealed low readability and situationally-induced impairments as major challenges. The content analysis showed a low guideline compliance rate for key design factors. We then interviewed 11 video production engineers to investigate design factors they mainly consider. The engineers mainly focus on the size and amount of content while lacking consideration for color, complex images, and situationally-induced impairments. Finally, we present and validate guidelines for designing mobile-friendly MOOCs, such as providing adaptive and customizable visual design and context-aware accessibility support.