Microsoft's StudentSim helps AI tutors learn from simulated students

microsoft and the university of illinois developed studentsim, a system that creates realistic replicas of individual students from limited data. these replicas provide rapid feedback, helping researchers improve ai tutors. studentsim trains in two stages, first learning shared patterns from pooled student data, then adapting to an individual student. the system outperforms gpt-5.4 in chess, english, and math, correctly predicting a player's next move about twice as often and following corrective guidance.