Multimodal Large Language Models (MLLMs) are advancing the ability to reason about complex sports scenarios by integrating textual and visual information. To comprehensively evaluate their capabilities, we introduce SPORTU, a benchmark designed to assess MLLMs across multi-level sports reasoning tasks. SPORTU comprises two key components: SPORTU-text, featuring 900 multiple-choice questions with human-annotated explanations for rule comprehension and strategy understanding. This component focuses on testing models' ability to reason about sports solely through question-answering (QA), without requiring visual inputs; SPORTU-video, consisting of 1,701 slow-motion video clips across 7 different sports and 12,048 QA pairs, designed to assess multi-level reasoning, from simple sports recognition to complex tasks like foul detection and rule application. We evaluated four prevalent LLMs mainly utilizing few-shot learning paradigms supplemented by chain-of-thought (CoT) prompting on the SPORTU-text part. GPT-4o achieves the highest accuracy of 71\%, but still falls short of human-level performance, highlighting room for improvement in rule comprehension and reasoning. The evaluation for the SPORTU-video part includes 6 proprietary and 8 open-source MLLMs. Experiments show that models fall short on hard tasks that require deep reasoning and rule-based understanding. GPT-4o performs the best with only 57.8\% accuracy on the hard task, showing large room for improvement. We hope that SPORTU will serve as a critical step toward evaluating models' capabilities in sports understanding and reasoning. The dataset is available at [https://github.com/chili-lab/SPORTU](https://github.com/chili-lab/SPORTU).

Reconstructing keyboard input through side-channel attacks has posed significant threats to user security. While conventional keystroke eavesdropping attacks have demonstrated effectiveness using side channels such as acoustic signals, they are usually shorter in range and can be significantly affected by environmental noises. In this paper, we proposeLaserKey, a novel keystroke eavesdropping technique that leverages the long-range and noise-resistant nature of lasers to achieve a more stealthy side-channel attack. We utilize laser sensors to accurately capture the subtle vibrations induced on laptop screens by keystrokes, and innovatively design a laser-driven deep learning-based keystroke recognition model with the inputs being the Mel-frequency Cepstral Coefficien (MFCC), Time Difference of Arrival (TDoA), and amplitude features extracted from such vibration signals. Through systematic experiments, we demonstrate thatLaserKeyachieves a 92.2% single-key recognition accuracy. By combining multiple single-key recognition capabilities based on this, we then realize the end-to-end word-level recognition. Moreover, to mitigate the recognition errors caused by the changes in keystroke positions, we introduce a meta-learning based domain generalization approach for achieving robust laser position calibration. Results show thatLaserKeyachieves as low as 3% character error rate (CER) for word-level recognition, proving its effectiveness for long-range and high-accuracy keystroke eavesdropping, and highlighting the necessity for countermeasures in the future.