Abstract Recent advances in Large Language Models have led to Large Reasoning Models, which produce step‐by‐step reasoning traces. Such traces may offer insight into how models think, improving explainability and clarifying the underlying process. These traces, however, are often verbose and complex, making them cognitively demanding to comprehend. With this in mind, we propose R e T race , an interactive system that structures and visualizes textual reasoning traces to support understanding. We use a validated reasoning taxonomy to produce structured reasoning data and investigate two types of interactive visualizations thereof. In a controlled human‐subject study, the visualizations provided more accurate comprehension of the model's reasoning and less perceived effort than a raw text baseline. The results of this study could have design implications for making long and complex machine‐generated reasoning processes more usable and transparent, an important step in AI explainability.

In this paper, we test whether Multimodal Large Language Models (MLLMs) can match human-subject performance in tasks involving the perception of properties in network layouts. Specifically, we replicate a human-subject experiment about perceiving quality (namely stress) in network layouts using GPT-4o, Gemini-2.5 and Qwen2.5. Our experiments show that giving MLLMs the same study information as trained human participants yields performance comparable to that of human experts and exceeds that of untrained non-experts. Additionally, we show that prompt engineering that deviates from the human-subject experiment can lead to better-than-human performance in some settings. Interestingly, like human subjects, the MLLMs seem to rely on visual proxies rather than computing the actual value of stress, indicating some sense or facsimile of perception. Explanations from the models are similar to those used by the human participants (e.g., an even distribution of nodes and uniform edge lengths).

Human-human handovers are natural and seamless. To be able to do this, humans optimize towards many factors. One of them is the timing when receiving an object. However, the preferred robot reaction time in Human-Robot handovers is currently unclear. To understand the preferred robot reaction time, we trained an Space-Time-Separable Graph Convolutional Network (STS-GCN) model using motion capture data of human-human handovers. We deployed this system on a robotic arm with live depth camera data. We conducted a user study (N=20) with five robot reaction times. We found that users perceived an early prediction as preferred. Furthermore, we found that designers can adapt this timing to their needs based on six sub-components of user perception. We contribute a ready-to-deploy hand over classification model, a preferred handover time for our system, and an approach to determine the preferred robot reaction time for robotic systems.