Wireless signals are a unique form of time-series data that encode rich physical characteristics of the propagation environment. As large volumes of signals are continuously generated in wireless sensing systems, efficient compression is essential for scalable transmission under limited system resources. However, wireless signals exhibit strong variability across environments and time, making it difficult to design universal compression schemes that generalize across domains while preserving the physical fidelity required for accurate reconstruction and sensing. Meanwhile, recent advances in large language models (LLMs) have demonstrated remarkable reasoning and adaptation capabilities across diverse modalities and domains. This raises an intriguing question: can LLMs compress wireless signals in a training-free manner while preserving physical fidelity? To the best of our knowledge, this problem remains unexplored. To answer this question, we propose a physics-informed prompting framework for training-free wireless signal compression. The framework incorporates environmental context and scenario-specific knowledge as physical priors in the prompt. We further introduce a dictionary module with self-learned patterns to stabilize inference and a key prompt module to align the semantics between the LLMs used for compression and reconstruction. Extensive experiments demonstrate that the proposed framework achieves high-fidelity reconstruction under cross-domain settings, highlighting the promise of physics-informed prompting as a new paradigm for processing and compressing low-level physical signals.

Mental health risk is a critical global public health challenge, necessitating innovative and reliable assessment methods. With the development of large language models (LLMs), they stand out to be a promising tool for explainable mental health care applications. Nevertheless, existing approaches predominantly rely on subjective textual mental records, which can be distorted by inherent mental uncertainties, leading to inconsistent and unreliable predictions. To address these limitations, this paper introduces ProMind-LLM. We investigate an innovative approach integrating objective behavior data as complementary information alongside subjective mental records for robust mental health risk assessment. Specifically, ProMind-LLM incorporates a comprehensive pipeline that includes domain-specific pretraining to tailor the LLM for mental health contexts, a self-refine mechanism to optimize the processing of numerical behavioral data, and causal chain-of-thought reasoning to enhance the reliability and interpretability of its predictions. Evaluations of two real-world datasets, PMData and Globem, demonstrate the effectiveness of our proposed methods, achieving substantial improvements over general LLMs. We anticipate that ProMind-LLM will pave the way for more dependable, interpretable, and scalable mental health case solutions.