The field of caregiving and human activity recognition is moving towards leveraging large language models (LLMs) to deliver empathetic and tailored support for family caregivers and to generate diverse datasets for training robust Human Activity Recognition (HAR) systems. LLMs are being explored for their potential to enhance user requests with visual cues from surroundings for improved intent understanding in robotic assistance scenarios. Studies have shown that LLM-powered conversational agents can effectively deliver Problem-Solving Therapy (PST) and provide emotional support to caregivers, while also enabling the generation of rich, virtual sensor datasets that represent a wide range of users and home settings. The use of LLMs in caregiving is also being examined to design dynamic, stage-sensitive interventions that holistically support caregivers' mental wellbeing. Noteworthy papers include:

• A study on a large language model-powered conversational agent that delivers Problem-Solving Therapy for family caregivers, which highlights the potential of LLMs in delivering empathetic and tailored support.
• A paper introducing AgentSense, a virtual data generation pipeline that leverages LLMs to create daily routines and action sequences for simulated home environments, which substantially improves performance in HAR systems.
• Research on ASMR, a framework that uses large generative models to simulate conversations and environmental contexts for robotic action reflection, which achieves state-of-the-art performance in multimodal classification tasks.