Reproduce “GLOSS: Group of LLMs for Open-Ended Sensemaking of Passive Sensing Data for Health and Wellbeing” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Akshat Choube, Ha Le, Jiachen Li, Kaixin Ji, Vedant Das Swain, Varun Mishra
GLOSS, a multi-LLM system from Northeastern University, enables open-ended sensemaking of raw, multimodal passive sensing data for health and well-being using natural language queries. It achieved 87.93% accuracy on objective queries, significantly outperforming a Retrieval-Augmented Generation (RAG) baseline, and supports personalized health insights without requiring technical expertise.
Key insights
- An LLM-based agentic framework that generates and executes code for data processing can substantially improve accuracy for objective data analysis queries compared to direct LLM interpretation of textual data.
- Open-ended sensemaking of complex, raw multimodal passive sensing data can be democratized through a natural language interface, making it accessible to diverse stakeholders without specialized technical skills.
- While code execution significantly enhances accuracy, ensuring consistency across multiple runs for ambiguous objective queries remains a challenge, as different but logically valid data paths can be taken by the system.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.