In a novel proposal by Luca Arrotta and his cohort, ContextGPT presents an approach where LLMs infuse contextual common-sense knowledge into Human Activity Recognition (HAR) models. By bypassing complex ontology design, ContextGPT simplifies the process, requiring less human intervention and expertise when compared to traditional NeSy approaches. The experimental results using public datasets affirm that ContextGPT not only matches but sometimes surpasses logic-based models, all while streamlining effort and cost.

• Neuro-symbolic AI to ease label scarcity
• Utilizes LLMs for common-sense knowledge infusion
• Outperforms logic-based models with minimal input
• Potentially simplifies deployment of context-aware HAR systems

ContextGPT is a significant step towards harmonizing human insights with neuro-symbolic models, providing promising directions for pervasive computing and interactive AI systems.