Doha – March 26, 2025: Researchers at Weill Cornell Medicine-Qatar (WCM-Q) have published a study that utilized large language models and data from wearable smart devices to generate personalized recommendations to enhance the wellbeing and academic performance for a group of school students.

The research, led by Dr. Arfan Ahmed, assistant professor of research in population health sciences, collected data from 12 students from academic reports (grades and teachers' observations), sleep quality surveys, and wearable Fitbit devices which recorded activity levels (step counts), sleep minutes and stress scores.

A large language model (LLM) called Llama 3, by Meta, was then used to analyze the data and create personalized recommendations for each student. The data from the students, who were monitored for a six-week period, was interpreted using Llama 3, which was programmed with detailed prompts constructed by the research team to guide the analysis of the data and the automatic generation of personalized recommendations for each student.

The advice generated by the LLM was then scored and validated by three independent human raters (including healthcare workers and academics) who evaluated all recommendations based on the criteria of clarity, actionability, and alignment with student data. Analysis of the scores showed that the LLM had the ability to analyze and integrate diverse data sources and to then generate personalized recommendations that were not only contextually relevant but also sensitive to the emotional state of the student. However, some recommendations received low or average scores for clarity, actionability and relevance. All student data and recommendations were collected anonymously to protect the participants' privacy. The researchers noted that care must be taken to avoid overwhelming students with excessive feedback, and highlighted the risk of 'model hallucinations,' which is when the LLM generates plausible but incorrect information. They also stated that future iterations of this type of research will need to incorporate safeguards against harmful recommendations and ensure that outputs are critically evaluated for accuracy and appropriateness.

Dr. Ahmed said: 'The research gave us clear indications that large language models have the capacity to interpret different types of data from multiple sources and generate personalized recommendations that could help improve students' wellbeing and academic performance. While there is a great deal of room for improvement, this proof-of-concept study shows there is potential for developing very useful tools that could not only help students reach their full potential but also relieve pressure on teachers, school administrators and student counselors.'

Other researchers who worked on the paper, which has been published in Scientific Reports, a leading journal, were Dr. Javaid Sheikh, dean of WCM-Q, professor of psychiatry and professor of population health sciences; Dr. Alaa Abd-Alrazaq, assistant professor of research in population health sciences; Dr. Rawan Al Saad, postdoctoral associate, and Ms. Sarah Aziz, research specialist.

Dr. Sheikh said: 'This research demonstrates the great potential of AI and large language models in particular to boost academic performance and wellbeing for students. The study also highlights the need for AI approaches to be deployed extremely carefully to protect the safety and wellbeing of users.'