This research addresses the critical gap in using Large Language Models (LLMs) for detecting fine-grained human activities from motion sensor data.

• Standard LLMs fail at detailed motion detection tasks (near-random accuracy)
• Researchers developed specialized fine-tuning methods for flat-surface writing recognition
• Demonstrates significant potential for advanced biometric authentication systems
• Enables more precise monitoring capabilities beyond basic movement detection

Security Implications: This advancement creates new possibilities for behavioral biometrics, allowing systems to recognize unique movement signatures that could strengthen authentication methods and enhance surveillance capabilities.

Exploring the Capabilities of LLMs for IMU-based Fine-grained Human Activity Understanding