Caglar Yildirim

Caglar Yildirim

Large language models (LLMs) are increasingly deployed as tool-using agents, shifting safety concerns from harmful text generation to harmful task completion. Deployed systems often condition on user profiles or persistent memory, yet agent safety evaluations typically ignore personalization signals. To address this gap, we investigated how mental health disclosure, a sensitive and realistic user-context cue, affects harmful behavior in agentic settings. Building on the AgentHarm benchmark, we evaluated frontier and open-source LLMs on multi-step malicious tasks (and their benign counterparts) under controlled prompt conditions that vary user-context personalization (no bio, bio-only, bio+mental health disclosure) and include a lightweight jailbreak injection. Our results reveal that harmful task completion is non-trivial across models: frontier lab models (e.g., GPT 5.2, Claude Sonnet 4.5, Gemini 3-Pro) still complete a measurable fraction of harmful tasks, while an open model (DeepSeek 3.2) exhibits substantially higher harmful completion. Adding a bio-only context generally reduces harm scores and increases refusals. Adding an explicit mental health disclosure often shifts outcomes further in the same direction, though effects are modest and not uniformly reliable after multiple-testing correction. Importantly, the refusal increase also appears on benign tasks, indicating a safety--utility trade-off via over-refusal. Finally, jailbreak prompting sharply elevates harm relative to benign conditions and can weaken or override the protective shift induced by personalization. Taken together, our results indicate that personalization can act as a weak protective factor in agentic misuse settings, but it is fragile under minimal adversarial pressure, highlighting the need for personalization-aware evaluations and safeguards that remain robust across user-context conditions.

Mehmet Akhoroz, Caglar Yildirim

Conversational AI interfaces powered by large language models (LLMs) are increasingly used as coding assistants. However, questions remain about how programmers interact with LLM-based conversational agents, the challenges they encounter, and the factors influencing adoption. This study investigates programmers' usage patterns, perceptions, and interaction strategies when engaging with LLM-driven coding assistants. Through a survey, participants reported both the benefits, such as efficiency and clarity of explanations, and the limitations, including inaccuracies, lack of contextual awareness, and concerns about over-reliance. Notably, some programmers actively avoid LLMs due to a preference for independent learning, distrust in AI-generated code, and ethical considerations. Based on our findings, we propose design guidelines for improving conversational coding assistants, emphasizing context retention, transparency, multimodal support, and adaptability to user preferences. These insights contribute to the broader understanding of how LLM-based conversational agents can be effectively integrated into software development workflows while addressing adoption barriers and enhancing usability.

Caglar Yildirim

Cybersickness is an unpleasant side effect of exposure to a virtual reality (VR) experience and refers to such physiological repercussions as nausea and dizziness triggered in response to VR exposure. Given the debilitating effect of cybersickness on the user experience in VR, academic interest in the automatic detection of cybersickness from physiological measurements has crested in recent years. Electroencephalography (EEG) has been extensively used to capture changes in electrical activity in the brain and to automatically classify cybersickness from brainwaves using a variety of machine learning algorithms. Recent advances in deep learning (DL) algorithms and increasing availability of computational resources for DL have paved the way for a new area of research into the application of DL frameworks to EEG-based detection of cybersickness. Accordingly, this review involved a systematic review of the peer-reviewed papers concerned with the application of DL frameworks to the classification of cybersickness from EEG signals. The relevant literature was identified through exhaustive database searches, and the papers were scrutinized with respect to experimental protocols for data collection, data preprocessing, and DL architectures. The review revealed a limited number of studies in this nascent area of research and showed that the DL frameworks reported in these studies (i.e., DNN, CNN, and RNN) could classify cybersickness with an average accuracy rate of 93%. This review provides a summary of the trends and issues in the application of DL frameworks to the EEG-based detection of cybersickness, with some guidelines for future research.

David Saffo, Sara Di Bartolomeo, Caglar Yildirim et al.

As immersive analytics continues to grow as a discipline, so too should its underlying methodological support. Taxonomies play an important role for information visualization and human computer interaction. They provide an organization of the techniques used in a particular domain that better enable researchers to describe their work, discover existing methods, and identify gaps in the literature. Existing taxonomies in related fields do not capture or describe the unique paradigms employed in immersive analytics. We conceptualize a taxonomy that organizes immersive analytics according to two dimensions: spatial and visual presentation. Each intersection of this taxonomy represents a unique design paradigm which, when thoroughly explored, can aid in the design and research of new immersive analytic applications.